Genomewide Transcriptional Responses of Iron-Starved Chlamydia trachomatis Reveal Prioritization of Metabolic Precursor Synthesis over Protein Translation

By utilizing an experimental approach that monitors the immediate global response of Chlamydia trachomatis to iron starvation, clues to long-standing issues in Chlamydia biology are revealed, including how Chlamydia adapts to this stress. We determined that this pathogen initiates a transcriptional program that prioritizes replenishment of nutrient stores over replication, possibly in preparation for rapid growth once optimal iron levels are restored. Transcription of genes for biosynthesis of metabolic precursors was generally upregulated, while those involved in multiple steps of translation were downregulated. We also observed an increase in transcription of genes involved in DNA repair and neutralizing oxidative stress, indicating that Chlamydia employs an “all-or-nothing” strategy. Its small genome limits its ability to tailor a specific response to a particular stress. Therefore, the “all-or-nothing” strategy may be the most efficient way of surviving within the host, where the pathogen likely encounters multiple simultaneous immunological and nutritional insults.

T he sexually transmitted bacterium Chlamydia trachomatis infects the mucosal epithelium of the endocervix, urethra, and anogenital tract. These infections usually resolve spontaneously, and most are asymptomatic and thus underreported. Over 1.5 million cases of C. trachomatis genital infections were reported in the United States in 2015 alone (1). As many as 17% of females infected with C. trachomatis develop long-term infections in the genital tract, which can result in serious complications such as pelvic inflammatory disease (PID), fallopian-tube scarring, and ectopic pregnancy, all of which are major risk factors for tubal factor infertility (TFI) (2). Rectal infections with lymphogranuloma venereum (LGV) serovars of C. trachomatis can be invasive and, if untreated, can lead to complications such as proctocolitis, inguinal adenopathy, reactive arthropathy, and colorectal ulcers (3). In some patients, infection persists even after antibiotic treatment (4,5). The ability of C. trachomatis to survive over the long term in some individuals despite host immunity and antibiotic treatment is not well understood and may be associated with Chlamydia's ability to become persistent (6). While aberrant chlamydial forms have been identified in cervical samples, the clinical relevance of this phenomenon is not well understood (6,7).
Chlamydiae are obligate intracellular Gram-negative bacteria that undergo a biphasic developmental cycle that includes both nonreplicative and replicative forms (8). Infection begins when the small, metabolically quiescent chlamydial elementary body (EB) binds to mucosal epithelial cells and translocates virulence factors that induce its endocytic uptake. Within 2 h of entry, the EB differentiates into its replicative form, the reticulate body (RB). Continued secretion of effectors leads to modification of the endocytic vesicle such that it avoids fusion with the lysosome and enables capture of nutrient-rich vesicles. This unique intracellular niche, called the inclusion, continues to expand as RBs replicate. In response to unknown signals present at around 24 h postinfection (p.i.). RBs then differentiate into infectious EBs, followed by EB release 36 to 72 h postinfection (8). Under conditions of exposure to certain forms of stress in cell culture (e.g., penicillin treatment, interferon gamma [IFN-␥] treatment, iron depletion, or tryptophan [Trp] depletion), RBs do not differentiate into EBs but instead enter into a state of persistence characterized by aberrant, enlarged morphology (9)(10)(11)(12)(13). Persistent Chlamydia bacteria are resistant to both antibiotics and host immunity mechanisms and can recover from this state upon removal of stress or addition of missing nutrients (14)(15)(16)(17).
Chlamydiae have undergone reductive evolution as they have adapted to intracellular growth in mammalian cells, discarding metabolic genes responsible for synthesizing factors that could be acquired from the host (18). The core genome of C. trachomatis serovar L2 includes only 889 open reading frames, making Chlamydia dependent on its host for lipids, nucleotides, amino acids, and metal cofactors (18). Exposure of Chlamydia-infected cells to immune mediators, such as IFN-␥, reduces the availability of these factors and results in reduced RB division and differentiation (12,14). IFN-␥ induces intracellular depletion of tryptophan by increasing levels of indolamine 2,3-dioxygenase (IDO), which is responsible for catabolizing tryptophan into kynurenines, which cannot be utilized in tryptophan metabolism (19).
Induction of inflammatory cytokines such as IFN-␥ and interleukin-6 (IL-6) in response to chlamydial infection likely causes sequestration of free iron by the activity of the mononuclear phagocytic system, which includes both cellular and systemic regulatory pathways (20)(21)(22)(23)(24)(25)(26)(27). Readers are referred to two comprehensive reviews of the coordinated regulation of iron homeostasis by systemic and cellular mechanisms (26,28). In the context of Chlamydia infection of the genital epithelium, iron availability in and it induces the development of aberrant forms more consistently and homogenously than the previously used ferric iron chelator, deferoxamine (43). Here, we determined the optimal duration of BPDL treatment to induce iron-responsive transcription without inducing morphological abnormalities in Chlamydia. We chose to begin starvation during midcycle development (12 h p.i.) instead of at the beginning of infection for two reasons: (i) to test the response of actively replicating Chlamydia bacteria that are able to maximally respond to stress and (ii) to ensure that the treated and mock-treated Chlamydia bacteria remained in the same stage of development (RB). We monitored chlamydial morphology, growth, and transcriptional responses after 3, 6, and 12 h of BPDL treatment (Fig. 1A). Indirect immunofluorescent confocal microscopy revealed similar morphologies for the mock-treated and BPDL-treated forms for up to 12 h of BPDL treatment (Fig. 1B). Interestingly, observation of BPDL-treated cultures by light microscopy revealed an obvious decrease in Brownian movement within inclusions after 6 or more hours of treatment (data not shown). This observation is consistent with our findings showing that chlamydial growth is reduced compared to that seen with mock treatment after only 6 h of BPDL treatment, as determined by quantitative PCR (qPCR) analysis of chlamydial genomes (Fig. 1C).
We monitored the transcriptional response of the known iron-responsive genes ytgA and ahpC by reverse transcriptase quantitative PCR (RT-qPCR) to validate the iron starvation protocol (40,41,43,49). Elevated (1.5-fold and 1.7-fold) transcription of both iron starvation markers was detected after only 6 h of BPDL treatment compared to the mock treatment results. Maximum differences in transcription of both markers (2.8-fold and 3.7-fold) were detected after 12 h of BPDL treatment (Fig. 1D). In the same experiment, we monitored the transcriptional profile of the early-stage gene euo, whose transcription decreases during late stages of normal development. Multiple persistence models have demonstrated dysregulated euo transcription, with high levels of euo mRNA detected late in development under persistence-inducing conditions (15,43,45). After 12 h of BPDL treatment, we observed that euo transcript levels remained elevated relative to the mock-treated control results, indicating dysregulated transcription or a possible delay in development (Fig. 1E, top). The idea that a delay in development occurs during longer BPDL treatment is supported by the lack of recoverable inclusion-forming units (IFUs) detected after 12 or 24 h of BPDL treatment compared to mock-treated controls (Fig. 1F), indicating a possible lack of RB-to-EB differentiation. Because 6 h of BPDL treatment is sufficient to induce iron-responsive transcription without inducing the patterns of morphology and transcription associated with persistence, we chose that as the optimal duration of iron starvation for our genomewide transcriptional studies. We also included 3 h of BPDL treatment to detect the earliest possible response to iron starvation prior to BPDL-induced changes in growth.
Global transcriptional response of C. trachomatis to iron starvation during midcycle development. The primary global response of C. trachomatis to midcycle iron starvation was determined by RNA sequencing (RNA-seq). We utilized an Ion Proton chip for sequencing, which allowed easy and rapid scaling of time points based on observed yields of mapped reads. This approach is relevant to the study of Chlamydia transcription because chlamydial mRNA represents a small proportion of the total RNA at the time points analyzed, even after significant enrichment steps. For midcycle iron starvation studies, we aimed for greater than 10ϫ coverage of 100% of the C. trachomatis genome, with a minimum of 3 biological replicates per sample. The sequencing reads were trimmed to exclude adaptor sequences and polyclonal reads, followed by exclusion of reads less than 30 nucleotides (nt) in length. The remaining reads were aligned to the C. trachomatis genome and plasmid, with 2% to 23% of trimmed reads mapping. Average read lengths ranged from 92 to 134 nucleotides, requiring an average of 108,837 mapped reads to reach our coverage goal. A summary of the read and mapping statistics for all of our samples can be found in Table S1 in the supplemental material. Alignments, comparisons, and normalization of aligned reads were done with CLC Genomics Workbench version 9.0 according to default settings. All midcycle conditions (12 h untreated, 12 ϩ 3 h BPDL, 12 ϩ 3 h mock, 12 h ϩ 6 h BPDL, 12 ϩ 6 h mock) were compared using the CLC Genomics experiment tool, normalized by quantile scaling, and analyzed for differential gene expression levels using EdgeR statistical analysis with false-discovery-rate (FDR) calculation. Because we included ribosomal rRNA, eukaryotic mRNA, and small (Ͻ100-nt) RNA depletion steps when preparing chlamydial mRNA for RNA-seq, we also excluded tRNAs, rRNAs, and genes with Ͻ10 mean reads in a sample group prior to normalization and analysis.
The genomewide profile of mock-treated and BPDL-treated gene expression during midcycle development (12 h to 18 h postinfection) is displayed as a heat map of normalized expression values ( Fig. 2A). The raw and normalized data for these individual replicates can be found in Table S4. The mock treatment (left) and BPDL treatment FIG 2 Global and differential gene expression of the midcycle response to iron starvation. The global response of C. trachomatis to iron starvation was detected by RNA sequencing, and reads were aligned to the genome and plasmid. (A) (Left) The untreated expression profile is displayed for all genes that changed significantly (P value, Ͻ0.01) during midcycle development as a heat map of log 10 -transformed normalized expression means. (Right) Levels of expression across the same genes are displayed for BPDL-treated samples. The highest and lowest expression values are displayed in green and red, respectively. (B) Genes whose expression was significantly changed in response to iron starvation, with a P value Ͻ 0.01, are displayed as a heat map of fold changes for BPDL-treated samples compared to mock-treated equivalent samples. The most highly upregulated and downregulated transcripts are displayed in green and red, respectively.
(right) profiles were remarkably similar across all genes whose expression significantly changed during normal midcycle development of Chlamydia (based on comparisons of  data from 12 h versus 15 h, 15 h versus 18 h, or 12 h versus 18 h; P value of Յ0.01). The annotated expression heat map and EdgeR comparisons for normal growth can be found in Fig. S1 and Table S2 in the supplemental material, respectively. The entire RNA-seq data set of normal development can be found in Table S3. The similarity between global expression profiles indicates that the normal development of Chlamydia is not dysregulated after only 3 h or 6 h of BPDL treatment. However, EdgeR analysis of BPDL-treated cultures compared to mock-treated samples (at equivalent time points postinfection) revealed that 8% (76/889) of the genome was differentially expressed after 3 h BPDL treatment and 1% (12/889) was differentially expressed after 6 h of BPDL treatment. Genes that were differentially expressed with a maximum P value of 0.01 are displayed in a heat map of fold change differences between BPDL-treated and mock-treated samples (Fig. 2B). Examples of decreased transcription after 3 h and 6 h of BPDL treatment include the ribosomal subunit genes rpsO and rpsT and the type III secretion genes copB and scc2, respectively. Transcription of the tryptophan salvage pathway operon trpBA and the ribonucleotide reductase operon nrdAB was significantly increased after both 3 and 6 h of treatment. Iron-responsive genes that were differentially expressed with a P value of Ͻ0.01 after 3 or 6 h of BPDL treatment are listed in Tables 1 and 2, respectively. The fully annotated heat map can be found in Fig. S2, and the full set of RNA-seq results for midcycle iron starvation can be found in Table S4.
To independently confirm the midcycle response detected by RNA sequencing, we utilized RT-qPCR. Increased transcription in response to iron starvation was confirmed for all of the transcripts tested by RT-qPCR, with the exception of recA (Fig. 4A). None of the tested downregulated genes were significantly reduced in expression compared to controls as determined by RT-qPCR, likely due to the fact that the genes tested were very low in abundance at the time points tested (Fig. 4B).
Functional categorization of genes differentially expressed during early-cycle response to iron starvation. Chlamydia infections of the genital tract are asynchronous. Thus, Chlamydia could be exposed to host-induced stress at any point in the developmental cycle. For this reason, we extended our analysis of the immediate response to iron starvation to an earlier point in the developmental cycle. Chlamydiainfected cells were treated with BPDL starting at 6 h postinfection, which is a time point after the initial EB-to-RB differentiation and at the beginning of the logarithmic-growth phase. RNA and genomic DNA (gDNA) were collected at 9 h postinfection for both treated and mock-treated samples. RNA-seq analyses and alignments were performed as described above. A summary of mapped reads and coverage can be found in Table S1. Genes differentially expressed during the early-cycle response (6 ϩ 3 h BPDL treatment versus 6 ϩ 3 h mock treatment), with a maximum P value of 0.01, are grouped by functional categories of induced and reduced transcripts (Fig. 5). Data corresponding to the full set of differentially expressed genes and their annotations can be found in Table 3. Similarly to the results of analysis of the midcycle response, transcription of 4% of the genome, including genes involved in DNA replication and repair (nrdA, nrdB, mutS, dnaQ, and recA), amino acid biosynthesis (trpB, trpA, aspC_1, and glyA), and translation (CTL0111, trpS, thrS, and aspS), was induced during the early-cycle response to iron starvation. Uniquely, genes involved in redox homeostasis  Table S4. Data from upregulated and downregulated genes are shown in the top and bottom halves of the table, respectively. These data were exported from CLC Genomics Workbench 9.5.3. rRNAs, tRNAs, and features (genes) with fewer than 10 reads in all samples were eliminated from the data set prior to normalization and EDGE analysis, and the data include only those genes that were differentially expressed with a significance P value of Յ0.01. ID, identifier.

FIG 3
Functional categorization of midcycle response to iron starvation. Transcripts that were significantly upregulated (left) or downregulated (right) after 3 h of BPDL treatment, starting at 12 h p.i., are organized in pie charts by their functional categories. Shown adjacently to each pie slice is the number of genes in that category, and the percentages of differentially expressed genes in the category are indicated in parentheses. N ϭ number of differentially expressed genes, and the percentage of the total genome that is represented is indicated.
Global Response of C. trachomatis to Iron Starvation (pdi, ahpC, and sodM) were also upregulated in response to iron starvation starting at 6 h postinfection but not during the midcycle response. Of the 23 genes with reduced transcription during the early-cycle response to iron starvation (3% of the genome), 17% are associated with translation (rplW, prfA, rplC, and ftsY) and 13% with DNA replication and repair (pGP8D, amn, and dnaX_1). Upregulation of trpA transcription during early-cycle iron starvation was confirmed by RT-qPCR, while only modest increases were observed for the other upregulated genes tested (Fig. 6A). Downregulation of CTL0430, CTL0063, and incD during iron starvation could not be confirmed by RT-qPCR (Fig. 6B). We reasoned that early-cycle responses were not detected by RT-qPCR for most of our tested genes due to the limit of detection of the technique. The raw values detected for most of our early-cycle transcripts fell at or below the lowest concentrations in our standard curves. Between 6 and 9 h postinfection, chlamydial mRNA represents a very small proportion of the total RNA. This limitation was overcome for RNA-seq experiments by depleting rRNAs and eukaryotic RNA prior to synthesizing cDNA. However, cDNA used in RT-qPCR was prepared from total RNA because mRNA enrichment would have made it impossible for us to normalize our RT-qPCR data to chlamydial genomes. The overwhelming proportion of eukaryotic RNA present in the undiluted cDNA used as the template may have impeded accurate detection of transcripts.
Network and biological pathway analysis. To further analyze the relevance of these gene expression changes to chlamydial survival, we utilized the bioinformatics tool STRING-db v.10.5 to generate networks of functionally associated genes (51). The  Table S4. These data were exported from CLC Genomics Workbench 9.5.3. rRNAs, tRNAs, and features (genes) with fewer than 10 reads in all samples were eliminated from the data set prior to normalization and EDGE analysis. These data include only genes that were differentially expressed with a significance P value of Յ0.01. representation of differentially expressed gene sets (P value, Յ0.05) corresponding to short-term iron starvation (3 h) reveals gene networks with intersecting pathway clusters (see the manually added gray circles). We chose to use the less stringent P value to allow entire pathways to emerge (the pathways would not have been quite as apparent with a more stringent cutoff value). Consistent with our predicted functional categories, network analysis of both early (Fig. 7A) and midcycle (Fig. 7B) responses to iron starvation revealed clusters that include amino acid biosynthesis, DNA replication and repair, and translation. Functional clustering of the midcycle response also revealed the entire cluster of genes necessary to convert pyruvate to acetyl coenzyme A (acetyl-CoA), as well as gene clusters involved in tRNA modification and charging. The locus identifiers of genes in each identified cluster were submitted to KEGG-Mapper v2.8 to determine possible roles in specific biological pathways (52). For example, genes from the early-cycle DNA replication and repair cluster (Fig. 7A) were mapped to multiple pathways, including the purine metabolism (5), pyrimidine metabolism (5), mismatch repair (5), replication (4), homologous recombination (3), double-stranded break repair (2), and base excision repair (1) pathways.
Nucleotide metabolism. We modified the KEGGMapper output for pyrimidine metabolism to indicate the direction of change in midcycle gene expression during iron starvation (Fig. 7C). Under all iron starvation conditions, ribonucleotide reductase gene nrdA was upregulated. Ribonucleotide diphosphates (NDPs) bound to NrdA are converted by NrdB to deoxynucleotide diphosphates (dNDPs). These dNDPs are not likely further converted to deoxynucleotide triphosphates (dNDPs), as indicated by the downregulation of the ndk nucleotide diphosphate kinase gene. Available dUMP would likely be derived from the UDP pool, instead of from the dUTP pool, since transcription of the dUTP pyrophosphatase gene, dut, is downregulated during iron starvation. Taking the data together, these transcriptional changes would result in a net increase FIG 5 Functional categorization of the early-cycle response to iron starvation. Transcripts that were significantly upregulated (left) or downregulated (right) after 3 h of BPDL treatment, starting at 6 h p.i., are organized in pie charts by their functional categories. Adjacent to each pie slice is the number of genes in that category, and the percentages of differentially expressed genes in the category that make up the pie are indicated in parentheses. N ϭ number of differentially expressed genes, and the percentage of the total genome that is represented is indicated.
Global Response of C. trachomatis to Iron Starvation in levels of dNDPs, enabling rapid DNA replication when iron levels and ndk expression return to normal (Fig. 7C).
Amino acid biosynthesis. Functional clustering also indicates that Chlamydia prioritizes maintenance of amino acid pools during iron starvation. Multiple amino acid  Table S5. These data were exported from CLC Genomics Workbench 9.5.3. rRNAs, tRNA, and features (genes) with fewer than 10 reads in all samples were eliminated from the data set prior to normalization and EDGE analysis. These data include only genes that were differentially expressed with a significance P value of Յ0.01. synthesis, interconversion, and uptake mechanisms were upregulated in response to short-term iron starvation. Transcriptional upregulation of the branched-chain amino acid transporter gene brnQ, the aspartate aminotransferase gene aspC, and the serine hydroxymethyltransferase gene glyA may increase the diversity of the amino acid pool such that Chlamydia can quickly adapt to fluctuations in amino acids. Surprisingly, the tryptophan salvage pathway genes, trpB and trpA, were consistently upregulated during short-term iron starvation. Tryptophan synthase subunit TrpB catalyzes the beta-replacement of indole with serine to form tryptophan (Trp), while TrpA facilitates the interaction of TrpB with indole (53). Their role in recovery from IFN-␥ and Trp starvation stresses is well documented, but differential regulation in response to iron starvation is novel (54,55). While the biological relevance of trpBA induction during iron starvation is unclear, we reason that Chlamydia could in fact prepare for further immune insult (e.g., IFN-␥ induction of indoleamine 2,3-dioxygenase expression) by increasing intracellular Trp levels. Taking the data together, iron starvation may increase levels of serine, aspartate, glutamate, branched-chain amino acids, and tryptophan, many of which are essential for normal development (56)(57)(58)(59)(60). Amino acid biosynthetic genes were significantly overrepresented (4.38-fold; P value ϭ 0.0464) in the set of differentially expressed midcycle genes as determined by the PANTHER overexpression test (61). Translation. The largest cluster generated from STRING-db included translation factors of the midcycle response (Fig. 7B). Based on protein annotations in Uniprot and Biocyc databases, it is evident that C. trachomatis responds to iron starvation by shutting down factors involved in every step of translation: ribosome assembly, initiation, elongation, termination, ribosome recycling, and peptide targeting (50, 90) ( Table 4). While preventing the assembly and function of translational machinery, Chlamydia also responds to iron starvation by increasing the levels of factors important for synthesis and modification of tRNAs, in addition to increasing transcription of rnC, the product of which cleaves rRNA transcripts into ribosomal subunit precursors (Table 4). Translation genes were significantly overrepresented (3.26-fold, P value ϭ 0.0243) in the set of midcycle differentially expressed genes as determined by the PANTHER overexpression test (61).
Acetyl-CoA synthesis. Transcription of the entire set of genes necessary for conversion of pyruvate to acetyl-CoA was induced during the midcycle response to BPDL treatment (Fig. 7D). This set includes the lipoylation enzyme genes lipA and lpdA and the genes corresponding to the entire pyruvate dehydrogenase complex, pdhABC. In addition, transcription of the tricarboxylic acid (TCA) cycle gene mdhC and the glycolysis gene eno was induced, likely driving formation of pyruvate from different carbon sources. Acetyl-CoA can be converted to malonyl-CoA for fatty acid biosynthesis or Global Response of C. trachomatis to Iron Starvation utilized in the formation of N-acetylglucosamine-1-phosphate for peptidoglycan synthesis, both of which are required for rapid growth of Chlamydia (62,63). Since the levels of transcription of the peptidoglycan-modifying enzymes encoded by glmS and murB were also increased during iron starvation, acetyl-CoA is likely used to form new peptidoglycan. Expression of fatty acid synthesis genes was unchanged during iron starvation.
Pathway analysis of both early and midcycle responses to iron starvation ( Fig. 7A  and B) revealed that downregulation of translation and upregulation of amino acid  , and pGP8-D) showed differential expression after 3 h of BPDL treatment, during both the early and midcycle responses, while trpB, trpA, and nrdA were upregulated in all BPDL treatments (Fig. S3). This overlap in differential gene expression data is displayed as a Venn diagram in Fig. S3.

DISCUSSION
We monitored the immediate global transcriptional response of Chlamydia trachomatis serovar L2 to short-term iron starvation during early and midcycle (RB-phase) development. In contrast to previous studies of iron starvation in Chlamydia, our short-term treatment performed with BPDL did not cause the hallmark changes in morphology and euo transcription associated with persistence. This approach enabled us to detect a response specific to iron starvation as Chlamydia tries to adapt to stress, rather than detect the transcriptome of the aberrant bacterium. By deep RNA sequencing, we were able to identify novel primary transcriptional responses, representing 7% to 8% of the genome, after only 3 h of iron starvation with BPDL. It is possible that a portion of the detected BPDL-responsive regulon was actually due to chelation of metals other than iron. Cu 2ϩ is chelated at affinities similar to those seen with Fe 2ϩ and Fe 3ϩ , while Zn 2ϩ is chelated at a level of affinity 2 to 3 logs lower than that seen with iron ions (64). We suspect that Zn 2ϩ was not efficiently depleted during the short-term BPDL treatments used in this study but cannot exclude the possibility that we had detected transcriptional changes that represent responses to altered availability of other metals. It is also possible that a more immediate response could be detected with even shorter-term BPDL treatments, though we expect a longer duration is required to chelate both free iron and iron bound to protein complexes in intracellular Chlamydia. Since only 12 genes were differentially expressed after 6 h of BPDL treatment, a longer duration of treatment may be necessary to detect the full secondary response, which may not be obvious until the effects of the primary transcriptional response are realized at the protein level. This conjecture is supported by the fact that 6 h of BPDL treatment maintains induction of the primary response operons, trpBA and nrdAB, while reducing or delaying expression of some late-cycle genes (scc2 CTL0840 copB copD, tsp). De- creased or delayed late gene expression has also been observed during long-term iron starvation (43,49,65,66).
In agreement with proteomic observations of deferoxamine-treated C. trachomatis after 24 h and C. pneumoniae after 48 h postinfection, we observed upregulation of CTL0874 (CADD gene), ahpC, eno, and htrA during short-term BPDL treatment (45,67). In contrast to previous iron starvation studies, we did not detect a significant increase in ytgA expression in our RNA-seq analyses. We expected the ytgABCD iron acquisition operon to be induced immediately in response to iron starvation, since its repression by YtgR is dependent on the presence of available iron (42). Expression of the ytgABCD operon peaks during midcycle development, indicating that the iron-dependent repressor YtgR may be inactive or present at low levels during the early cycle and midcycle (15,42). It is possible that we did not observe significant differences in the expression of the operon during iron starvation because it was already maximally expressed in the mock-treated controls (see Table S3 in the supplemental material). Global detection of YtgR repression by chromatin immunoprecipitation (ChIP) sequencing or targeted analysis of specific promoters will be necessary to delineate the contribution of YtgR activity to that of the detected iron-responsive regulon. Recent work to define targets of known transcription factors in Waddlia chondophila discovered binding sites of YtgC by ChIP sequencing (68). Interestingly, the level of the most frequent target, hrcA, was also increased during iron starvation in our study, indicating that it may also be a target of YtgC in C. trachomatis.
Transcriptional responses to iron starvation in most bacteria typically include upregulation of iron acquisition systems and virulence factors (69)(70)(71). While expression of the ytgABCD operon was not upregulated during short-term iron starvation, other unidentified iron uptake and iron-dependent repression mechanisms may exist and thus could be represented in our set of iron starvation-induced genes. The virulence factor CADD (CTL0874) gene and the MACPF (CTL0408) gene were induced in both the early and midcycle responses to iron starvation. CADD overexpression induces apoptosis under conditions of expression in cultured human epithelial cells but has also been demonstrated to play a role in folate biosynthesis (72,73). MACPF contains a domain that may enable perforin activity but so far has only been shown to undergo cleavage upon infection and become inserted into bacterial membranes (74). Several type III secretion structural components (mcsC, sctJ, sctR, fliF, cdsN [CTL0043], and cdsD [CTL0033]) and effectors (CTL0884, CTL0476, CTL0184, and CTL0081) were also transcriptionally upregulated during iron starvation, which could potentially alter interactions between the host and chlamydial inclusion.
Similarly to the upregulation of the ribonucleotide reductase operon nrdHIEF seen during iron starvation in Escherichia coli and Yersinia pestis, the ribonucleotide nrdA and nrdB reductase genes are consistently upregulated during under short-term iron starvation (70,75). This upregulation indicates that deoxynucleotides may be important for Chlamydia to survive this stress. However, since NrdB requires iron for its function, deoxynucleotide levels may not increase until iron becomes available. Instead, high levels of inactive NrdA-B complexes may actually impede replication and development by inducing stalling at replication forks, providing a possible explanation for the decreased replication observed during iron starvation (76,77).
The immediate transcriptional response of Chlamydia to iron starvation is remarkably similar to the stringent responses seen in other bacteria, which enable rapid adaptation to various stresses by diverting resources from macromolecular biosynthesis, e.g., translation, and from growth to immediate survival, often resulting in a quiescent state (78,79). This rapid transcriptional response is achieved through synthesis of the chemical alarmone (p)ppGpp, which interacts with RNA polymerase and DksA to globally modify transcriptional activity (80,81). During amino acid starvation in bacteria, uncharged tRNAs in the A-site of ribosomes are sensed by RelA, which responds by synthesizing (p)ppGpp from ATP and GDP or GTP (82,83). (p)ppGpp can also be synthesized and hydrolyzed by SpoT under other stress conditions. However, since Chlamydia lacks the RelA and SpoT homologues necessary for (p)ppGpp synthesis, it likely evolved alternative mechanisms to reduce growth and increase survival responses during stress (17,84,85). Iron starvation has been shown to induce a stringent response in Bacillus subtilis that upregulates transcription of amino acid biosynthesis genes (86).
Multiple amino acid synthesis, interconversion, and uptake mechanisms were upregulated in response to short-term iron starvation. Surprisingly, the primary response included an increase in expression of transcripts involved in tryptophan salvage, trpB and trpA, but not in expression of the tryptophan-dependent repressor trpR gene. TrpR-dependent regulation of the polycistronic transcript trpRBA has been extensively studied during tryptophan starvation and IFN-␥ treatment but rarely, if ever, in the context of iron starvation (54,87,88). Notably, trpB levels, but not trpR levels, were also increased under conditions of estradiol-induced persistence, suggesting that a trpRindependent mechanism for inducing tryptophan salvage transcription may exist (89).
Pathway analysis clearly indicates that transcripts involved in all steps of translation from initiation to ribosome recycling are downregulated during iron starvation. This reduction in translation factors might lead to an eventual shutdown or modification of translation activity that could increase survival during stress. By shutting down energyexpensive protein synthesis, ATP and GTP pools can be rerouted to immediate survival responses (tRNA charging, transcription). Similarly, iron starvation reduces the transcription of several ABC transporter genes which require ATP for their function. Uncoupled RNA and protein levels in Chlamydia have also been observed during IFN-␥ stress (17). The apparent decrease in translation during IFN-␥ exposure could be exacerbated by decreases in the levels of components of the translation machinery in response to simultaneous iron starvation. However, decreased expression of translation factors during the primary response to iron starvation may not be apparent until preexisting ribosome-protein complexes are degraded or destabilized. This may explain why Ն24 h of iron starvation is required to induce the development of aberrant RBs (43). Downregulation of translation factors during iron starvation will have to be examined at the protein level to determine its contribution to adaptation to iron starvation and development of persistence.
In contrast to downregulation of translation, iron starvation increases transcription of amino-acyl synthesis genes (cysS, pheT, glyQ, aspS, thrS), which are responsible for charging tRNAs with amino acids. The apparent disconnect between increased levels of aminoacyl-tRNA pools and decreased translation indicates possible survival mechanisms. Charged tRNAs might be utilized in an immediate survival response to iron starvation, prior to the turnover of ribosomal subunits. Alternatively, Chlamydia might accumulate charged tRNAs for recovery and resumption of development when normal levels of iron and translation factors are restored.
A major theme that emerged from our gene expression analysis is that Chlamydia likely perceives iron starvation as a signal to prepare for further nutrient deprivation and immune insult. Transcriptional upregulations of tryptophan salvage pathway (trpB, trpA), oxidative stress (ahpC, pdi, and sodM), and DNA repair (mutS, mutL, ssb, ung, recA) genes indicate a protective response to antimicrobial insults of the inflammatory immune response (e.g., IDO activation, reactive oxygen species). As an obligate intracellular pathogen, Chlamydia has undergone reductive evolution with constant selective pressure from the host immune system and its multiple antichlamydial effectors. Due to its small (~1-Mbp) genome, Chlamydia may not have the capability to induce a specific transcriptional response to each particular stressor, and the simultaneous deployment of stress responses may have been the most parsimonious route of adaptation to immune insult. In this case, we would expect that iron-starved Chlamydia would be better protected from damage by antimicrobial insults than mock-treated Chlamydia. Immediate transcriptional responses to other stress conditions will need to be monitored to determine if this coordination of antimicrobial responses is unique to iron starvation.
This report provides the first evidence of a global iron-dependent regulon for C. trachomatis. By using a system approach to delineate Chlamydia's transcriptional response to iron starvation, we have been able to detect biological pathways and place them in the context of chlamydial development. These findings are novel and add to previous studies of iron-dependent transcriptional and proteomic profiling in aberrant RBs, revealing transcriptional adaptive strategies prior to the development of a persistent state. Additionally, our results include a high-resolution profile of midcycle development of C. trachomatis serovar L2, including relevant time points for monitoring shifts in gene expression in the early, middle, and late cycles. We expect that this data set will prove useful for future studies that seek to determine the immediate transcriptional response of Chlamydia to other chemical and/or nutrient stresses. Our findings include previously unrecognized shifts in energy utilization and downregulation of translation that resemble a stringency-like survival response. Chlamydia may utilize a two-stage approach of increasing transcription of survival genes in the short term to delay development and survive during iron starvation, followed by an eventual shutdown of translation at later times of sustained stress. The latter might account for the observed irreversibility of the persistent state during long-term starvation for iron or tryptophan.

MATERIALS AND METHODS
Cell culture and infection. HeLa monolayers were infected with C. trachomatis strain L2 434/Bu in 6-well plates at a multiplicity of infection (MOI) of 2 for RNA and genomic DNA (gDNA) collection experiments and on coverslips in 24-well plates for morphology studies. Cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 2 mM glutamine, and 10 g/ml gentamycin in 5% CO 2 at 37°C. HeLa cells used in this study were started from P1 stocks from ATCC and were regularly checked for contamination by DAPI (4=,6-diamidino-2-phenylindole) staining and the use of a Universal Mycoplasma Detection kit (ATCC).
RNA sequencing. RNA was collected and pooled from 2 or 4 T75 flasks of C. trachomatis-infected HeLa monolayers that had been treated with 100 M 2,2-bipyridyl (BPDL) starting at 6 or 12 h postinfection (6 h ϩ 3 h BPDL, 12 h ϩ 3 h BPDL, 12 h ϩ 6 h BPDL) and from mock-treated samples at equivalent time points postinfection (9 h, 12 h, 15 h, 18 h). RNA was purified using a RiboPure Bacteria (Ambion) kit per the instructions of the manufacturer. Total RNA was further enriched for transcripts over 100 nucleotides in length by the use of a MegaClear kit (Ambion). Mammalian transcripts and rRNAs were removed using a MicrobEnrich kit (Ambion), and bacterial rRNAs were removed using a MicrobExpress kit (Ambion), repeating 2 to 3 times. The integrity and quantity of total and depleted RNA were monitored with an AATI fragment analyzer. cDNA libraries were prepared with Ion Total RNA-seq kit V2, sequencing beads were prepared using an Ion Chef system, and sequencing was performed on an Ion Proton chip with HiQ chemistry. Primary sequence analysis and trimming and binning of reads were performed using Torrent Suite Software version 5.0.5. Remaining reads were mapped to the combined core genome of C. trachomatis strain L2 434/Bu (GenBank accession no. AM884176) and the plasmid of C. trachomatis L2b CS784/08 (NZ_CP009926) using CLC Genomics Workbench 9, requiring reads be at least 30 nucleotides in length, with default alignment parameters.
The EdgeR algorithm in CLC Genomics was used to determine differential gene expression levels during development and iron starvation, assuming a false-discovery rate of 10% and P values of Յ0.05. tRNAs and ribosomal RNAs were filtered from the reads to account for differences in depletion efficiency, and only genes with at least 5 mapped reads were included in the analysis. Differentially expressed genes were confirmed for selected transcripts by RT-qPCR. qPCR and RT-qPCR. C. trachomatis-infected HeLa monolayers were treated with 100 M BPDL starting at 6 or 12 h hours postinfection (6 h ϩ 3 h BPDL, 12 h ϩ 3 h BPDL, 12 h ϩ 6 h BPDL) and mock-treated samples at equivalent time points postinfection (6 h, 9 h, 12 h, 15 h, 18 h). RNA and gDNA were collected with RiboPure Bacteria and the DNeasy Blood and Tissue (Qiagen) kits, respectively. cDNA was generated with Superscript IV reverse transcriptase (Life Technologies, Inc.) using 200 to 500 ng RNA per the instructions of the manufacturer, except with the use of random nonamers instead of hexamers. Transcripts were amplified with a PowerUp SYBR green system from undiluted cDNA for early-cycle samples (6 to 9 h) or diluted 1:10 in 10 mM Tris for midcycle samples (12 to 18 h) and detected with an Applied Biosystems 7300 RT-qPCR system.
Chlamydial morphology. Chlamydiae were monitored for 3, 6, or 12 h for changes in morphology in response to mock treatment or treatment with 100 M BPDL starting at 12 h postinfection. Infected cultures were fixed on coverslips and stained with pooled human serum (Sigma; H4522) at 1:750 followed by goat anti-human antibody conjugated to Alexa Fluor 488 (Thermo Fisher) at 1:1,000. DNA was stained with DAPI at 5 g/ml. Images were taken on a Leica SP8 confocal microscope with a 63ϫ oil objective and 4ϫ zoom.
IFU assay. Chlamydiae were monitored starting at 12 h postinfection for 12 or 24 h for changes in infectivity in response to mock treatment or treatment with 100 M BPDL at an MOI of 1. Infected cultures were scraped into 300 l SPG (succinic acid, sodium dihydrogen phosphate, glycine) and stored at Ϫ80 C for later testing. Thawed lysates were serially diluted into complete DMEM, centrifuged onto HeLa monolayers in 24-well plates, washed with Hanks balanced salt solution (HBSS), and allowed to infect for 24 h. Infected cultures were fixed and stained with pooled human serum at 1:750 followed by goat anti-human antibody conjugated to Alexa Fluor 488 (Thermo Fisher) at 1:1,000. Inclusions were counted by fluorescence microscopy, and levels of inclusion-forming units (IFU) were calculated as previously described.
Visual analysis of differentially expressed genes. Functional categories were assigned for all genes differentially regulated with a P value of Յ0.01 by referring to the GO terms listed on UniProt. Pie charts were generated using the "pie" function in Rstudio. Heat maps were generated in Rstudio using the package "pheatmaps," with parameters set to average clustering and Euclidean distance. The PANTHER overexpression test was done in PANTHER v12.0 on differentially regulated gene sets (total) with P values of Յ0.01, using the default parameters and Bonferroni correction. Pathway analysis was performed on differentially regulated genesets with P values of Յ0.05 and a minimum of 10 mapped reads, with STRING-db v.10.5 set to a confidence value Ն0.7. StringDB maps were slightly modified to make space to increase font size, to indicate the direction of change by color coding, and to add pathway labels without altering network relationships. Clustered genes detected with StringDB were further analyzed using KeggMapper v.2.8, and pathway maps were generated based on KeggMapper output using Affinity Designer v1.4.1.
Data availability. Raw and processed sequencing files were submitted to the NCBI Gene Expression Omnibus (GEO) as a Superseries, and the midcycle and early-cycle projects can be found using accession number GSE106763.