The Complex Transcriptional Response of Acaryochloris marina to Different Oxygen Levels

Ancient oxygenic photosynthetic prokaryotes produced oxygen as a waste product, but existed for a long time under an oxygen-free (anoxic) atmosphere, before an oxic atmosphere emerged. The change in oxygen levels in the atmosphere influenced the chemistry and structure of many enzymes that contained prosthetic groups that were inactivated by oxygen. In the genome of Acaryochloris marina, multiple gene copies exist for proteins that are normally encoded by a single gene copy in other cyanobacteria. Using high throughput RNA sequencing to profile transcriptome responses from cells grown under microoxic and hyperoxic conditions, we detected 8446 transcripts out of the 8462 annotated genes in the Cyanobase database. Two-thirds of the 50 most abundant transcripts are key proteins in photosynthesis. Microoxic conditions negatively affected the levels of expression of genes encoding photosynthetic complexes, with the exception of some subunits. In addition to the known regulation of the multiple copies of psbA, we detected a similar transcriptional pattern for psbJ and psbU, which might play a key role in the altered components of photosystem II. Furthermore, regulation of genes encoding proteins important for reactive oxygen species-scavenging is discussed at genome level, including, for the first time, specific small RNAs having possible regulatory roles under varying oxygen levels.

KEYWORDS cyanobacteria oxygen levels transcriptome response chlorophyll biosynthesis reactive oxygen species Photosynthesis uses solar energy, and transforms it into chemical energy, which is stored within the organic molecules of the organism. In essence, it provides the energy for all life on our planet. During oxygenic photosynthesis, sunlight is funneled toward a special pair of chlorophyll molecules, which produce a charge separation that results in the extraction of electrons from water. This initiates a chain of redox reactions that power the fixation of inorganic carbon into 3-phosphoglycerate, with oxygen (O 2 ) generated as a side product of this reaction. In fact, before photosynthesis occurred in ancient cyanobacteria around 3.5-2.4 billion yr ago, the atmosphere was largely anaerobic (Blankenship and Hartman 1998). Over billions of years, oxygenic photosynthetic organisms changed the Earth's atmosphere, steadily increasing its O 2 levels to over 21% (v/v). This permitted the rise of multicellular organisms, dependent upon aerobic respiration (Dismukes et al. 2001;Blankenship 2008;Hedges et al. 2004). Aerobic respiration is highly efficient in recovering the energy contained within the chemical bonds of organic molecules through oxidative phosphorylation. However, organisms living in aerobic environments also run the risk of being damaged by oxidants and reactive oxygen species (ROS). ROS include a number of reactive molecules derived from O 2 . Clearly, O 2 in its ground state is harmless, as it has two unpaired electrons with parallel spin, making it paramagnetic. In this form, it is unlikely to participate in reactions with organic molecules, unless it is enzymatically or chemically activated by other reactions (Apel and Hirt 2004;Sharma et al. 2012). However, oxygen-derived ROS comprise superoxide, hydrogen peroxide, and hydroxyl radicals, which are a threat to the cell. Organisms mitigate ROS deleterious effects in various ways: by scavenging pathways ( Slesak et al. 2016;Pospí sil 2012), by changing the regulation of affected genes (Tamagnini et al. 2007), by separating the location of their product to oxygen-free compartments like heterocysts (Murry et al. 1984), or by evolving an alternative pathway resistant to oxidation (Busch and Montgomery 2015). Many metabolic pathways functioning today still contain enzymes sensitive to O 2 levels, as illustrated by the coexisting oxygen-dependent, or oxygen-independent, reactions within the tetrapyrrole biosynthetic pathway (Busch and Montgomery 2015;Raymond and Blankenship 2004), and the activation of a counterpart D1 subunit of photosystem II (PSII) in response to changed O 2 levels (Summerfield et al. 2008).
Acaryochloris marina (hereafter Acaryochloris) is a unicellular cyanobacterium, using chlorophyll d (Chl d), instead of chlorophyll a (Chl a), as the major pigment in its photosystems (Chen et al. 2002a(Chen et al. , 2005b. Similar to all cyanobacteria, the thylakoids of Acaryochloris need to mitigate not only the oxidative stress generated by oxygenic photosynthetic activities, but also the oxidative stress produced because of aerobic respiration. Therefore, it is not surprising that under illumination, especially high-intensity light, singlet oxygen is mainly produced because of the interaction of unquenched Chl triplets with O 2 generated within PSII, and the water-splitting complex. In photosystem I (PSI), the univalent reduction of O 2 generates mainly superoxide anion radicals (reviewed in Latifi et al. 2009;Rutherford et al. 2012). To reduce the effects of ROS, a constant diffusion of O 2 through the cell and photosynthetic membranes under illumination is crucial. This diffusion of O 2 , along with antioxidant enzymes that prevent accumulation of ROS, and the existence of remediation metabolites, such as ascorbic acid, glutathione, tocopherols, carotenoids, and flavonoids, prevent the interaction of O 2 with electrons, other than those in the normal electron transfer pathways to O 2 , avoiding any possible oxidative stress (Latifi et al. 2009).
Because of the iconic character of Acaryochloris, in which the function of Chl a has been largely replaced by red-shifted Chl d, most of the research on this organism has been directed toward understanding Chl d biosynthesis (Schliep et al. 2010;Loughlin et al. 2014;Yoneda et al. 2016). In particular, this has focused on its role in photosynthesis (Chen et al. 2002b(Chen et al. , 2005bHu et al. 1998;Tomo et al. 2007), and in farred light acclimation (Duxbury et al. 2009). These studies have revealed direct oxidation of Chl a to Chl d, with participation of O 2 (Schliep et al. 2010;Loughlin et al. 2013). Although the structural difference between Chl a and Chl d has a significant effect on its spectral characteristics, it does not affect the binding of Chl d with typical Chl a binding-peptides, as shown in in vitro reconstitution experiments (Chen et al. 2005a;Hoober et al. 2007;Chen and Cai 2007). In fact, so far, none of the studies carried out on the photosystems of Acaryochloris have revealed any significant difference to Chl a-containing photosystems, besides their distinct spectral characteristics, related to their pigment substitutions (Chen et al. 2005b).
In this study, we grew Acaryochloris under different O 2 levels to test the effects of O 2 on photo-pigment biosynthesis, photosynthetic reactions, and on their relationship with other essential metabolic reactions (including DNA and protein metabolism). Using high-throughput RNA sequencing (RNAseq), we obtained genome-level information on all expressed transcripts, under microoxic, normal air (control), and hyperoxic conditions. We detected genes coding for key proteins in photosynthesis and synthesis of chlorophyll, which were preferentially expressed under microoxic conditions. As expected, proteins involved in oxidative stress remediation also were induced under hyperoxic conditions. We also generated the first inventory of previously unknown small RNAs (sRNA), including untranslated regions (UTRs) and intergenic noncoding RNAs (ncRNAs), many of them differentially expressed upon O 2 perturbation. The sRNAs that showed a strong induction were further investigated to predict their potential targets and their involvement in adaptation to these stress conditions. In this first systems-level study to include sRNAs performed on Acaryochloris, we uncover new insights on the particu-larities of "oxygenic" photosynthesis and its coevolutionary "anaerobic" metabolism.

Culture conditions
A. marina MBIC11017 was routinely kept in a culture room at 27°u nder 15-30 mmol photons m 22 s 21 of cool white light. Sterilized K +ES (artificial seawater), buffered with 25 mM TES at pH 8.0 was used as the culture medium for all three treatment groups. To make sure photosynthesis was not limited by CO 2 , NaHCO 3 was dissolved in a small volume of autoclaved medium, and injected into the enclosed culture flasks every 2 d (yielding an initial concentration of 0.375 mM). The initial cell density of all culture groups was adjusted to an optical density at 750 nm of 0.2. The cultures were shaken on an orbital flat-bed shaker at 90 rpm. Cultures under normal O 2 levels were inoculated in 1-liter Erlenmeyer glass flasks containing 500 ml of medium, capped with a cotton stopper that permitted gas exchange. Thus, the O 2 concentration of the gas phase inside the bottle was similar to atmospheric levels 21% (v/v).
Microoxic conditions were achieved by using a 500-ml two-necked round-bottom flask sealed tightly by a rubber stopper. Cultures were vacuumed, and refilled with pure nitrogen gas (99.95% purity) to ensure normal atmospheric pressure. We repeated this process several times, yielding a final O 2 concentration of ,0.2% inside the sealed culture flask. To maintain a microoxic condition, a positive pressure was created by bubbling nitrogen through the culture.
A similar set-up was used for hyperoxic conditions, with the exception that pure O 2 gas was used to refill the flask after vacuuming. Because the cells generate O 2 under illumination, ongoing input of O 2 gas to maintain the high-oxygen concentration was not required. However, to avoid pressure build-up, the flask was revacuumed and refilled with O 2 gas every 48 hr, as described above. The O 2 concentration inside the flask remained within the range of 65-75% (v/v) during the experiment.
Total RNA extraction Acaryochloris cultures were harvested after 7 d from all three different treatments. The harvested cell pellets were mixed with TRIzol (TRIzol Reagent, Life Technologies, Australia), and frozen immediately using liquid nitrogen. They were stored at 280°for at least 60 min. The frozen samples were thawed in a water bath at 37°, and spun down at 16,000 · g for 5 min to eliminate cellular debris. This supernatant was mixed at a volume ratio of 4:1, with chloroform, and spun down at 16,000 · g for 10 min at 4°. The upper layer, containing RNA and DNA, was carefully transferred to a new tube, without disturbing the white middle layer of solid components. The RNA was precipitated by addition of an equal volume of isopropanol, and incubated at 220°for at least 45 min. RNA pellets were washed with 70% (v/v) ethanol, and the DNA removed using the Baseline-ZERO DNase kit (Epicentre, WI), following the manufacturer's instructions. Prior to RNA quality assessment, the absence of DNA was confirmed by polymerase chain reaction (data not shown).
The quality of RNA was assessed on a 2100 Bioanalyzer (Agilent Technologies, CA), using a RNA 6000 Nano RNA Kit (Agilent Technologies), to obtain an RNA integrity number of .8.0. Transcripts corresponding to rRNAs (5S, 16S, and 23S) were reduced from the samples with a Ribo-Zero Kit (Epicentre), following the manufacturer's instructions. Further processing, including quality assessment, was undertaken prior to RNAseq analysis by Beijing Genomics Institution, China.
RNAseq data analysis FASTQ files of the resulting sequences were processed using open source software. FASTQ files were aligned against the A. marina MBIC11017 reference genome on NCBI (http://www.ncbi.nlm.nih.gov/), using the "Tophat for Illumina" tool available in the Galaxy suite (Afgan et al. 2016). The BAM files obtained were superposed on the genome. and visualized in Artemis (Rutherford et al. 2000) to facilitate annotation of the predicted transcriptional units (TUs). The Java-based Rockhopper system (McClure et al. 2013) was used to process mapped sequence reads for differential analysis. The Rockhopper report, containing a summary of the number of reads aligned. is available in Supplemental Material, File S1. sRNAs target prediction All sRNA sequences (including ncRNAs and UTRs) were obtained from the transcriptional coordinates generated by the Rockhopper software, after mapping the obtained reads to unannotated regions of the Acaryochloris genome. The target protein-coding genes were predicted using the IntaRNA algorithm (Busch et al. 2008), with a window of 275 nucleotides around the respective start codons (200 upstream and 75 downstream).

Functional enrichment analysis
A standard functional enrichment analysis of differentially expressed genes (EADEG) was applied using hypergeometric tests, after Hernández-Prieto et al. (2012). Derived p-values were adjusted for multiple testing, while false discovery rates (FDR) were calculated using the Benjamini-Hochberg method. We used the gene annotation given in Cyanobase (Fujisawa et al. 2014) (http://genome.microbedb.jp/cyanobase/AM1), while gene associations with cellular functions are from the KEGG database (Kanehisa et al. 2016), and Gene Ontology (GO) terms in Uniprot (UniProt Consortium 2015) (Table S1). Our lists included genes associated with 116 KEGG pathways, and with 1196 GO terms. Only lists having a minimum of five genes annotated from the Acaryochloris genome (89 of 116 KEGG pathways, and 320 of 1196 GO terms) were investigated. To determine the functional composition of differentially expressed genes, enrichment was separately assessed for upregulated and downregulated genes.

Data availability
The Acaryochloris strain used in this study is available as an axenic culture through the NBRC culture collection (NBRC 102967). Raw gene expression data, and processed information, is available at GEO with the accession number GSE89387. File S1 contains detailed information of all supplemental files.

Culture growth
Monitoring of the O 2 concentrations of the gas phase inside the culture bottles was performed daily with a Clark-type electrode. The concentration of O 2 in the hyperoxic culture was maintained within the range of 65-75% (v/v), while in the microoxic culture, it was ,0.2% (v/v). Thus, O 2 concentration in the medium was equal to 350 and 1% that of air saturation at 25°for hyperoxic and microoxic cultures, respectively. These deviations from atmospheric conditions negatively affected Acaryochloris cells, as reflected in their low apparent growth rates (Figure 1). The control culture doubled its OD 750nm every 57 hr during the exponential growth phase, while treated cultures had doubling times of . 72 hr ( Figure 1). Thus, both treatment conditions had detrimental effects on apparent cell growth, given their 80% decrease in growth rate ( Figure 1).
Full transcriptome profiling of Acaryochloris RNA was extracted from cells collected after 7 d under their respective treatments, at an OD 750nm between 0.4 and 0.6 for all cultures. The gene expression profiling at genomic level was achieved by paired-end highthroughput sequencing of RNA, isolated from Acaryochloris exposed to different O 2 concentrations. The experiment was duplicated for both test conditions, and triplicated for control conditions. To assess differential gene expression, we used the algorithms available in Rockhopper, because this software has been optimized for the analysis of RNAseq data obtained from prokaryotes (McClure et al. 2013). Transcript units (TUs) for 8446 of the 8462 TUs annotated for Acaryochloris (NCBI BioProject: PRJNA12997) were detected in both control and treated samples. Of the 16 undetected transcripts (Table S2), only one (AM1_A0163) has an annotated function in Cyanobase (as at June 2016), five were localized in the main chromosome, and the rest in the plasmids (two in pREB1, two in pREB2, three in pREB3, two in pREB4, and two in pREB5). RNA extracted from Acaryochloris cells grown under control culture conditions was used as a reference to evaluate transcript changes related to altered O 2 levels. Under control conditions, the 50 most abundant transcripts correspond to 21 sRNAs (16 newly described in this study), and 29 to protein-coding genes, of which 12 encode subunits of PSI or PSII. Eight of these genes encoding photosystem subunits were among the 50 top transcripts in all three samples: five of them (psaA, psaB, psaC, psaJ, and psaM) encoding PSI subunits, two (psbA and psbK) encoding PSII subunits, and, intriguingly, a high-light induced protein (HLIP) involved in the incorporation of chlorophyll in newly assembled photosystems (Hernandez-Prieto et al. 2011) ( Table 1).

Identification and classification of differentially expressed transcripts
A total of 8446 transcripts (99.8% of the genes annotated in the Cyanobase database) were detected as expressed in at least one of the test conditions. Imposing a stringent threshold for minimum expression in 50 reads, we identified 6635 protein-coding and 523 noncoding TUs as expressed, under at least one of the test conditions. We considered a transcript to be differentially expressed when it had an absolute log 2 FC (fold change) value $1.0 (i.e., a minimum twofold up or downregulation change). It is important to remark here that, since RNA sequencing Figure 1 Optical density curves of A. marina cultures under the three different oxygen concentrations (control, microoxic, and hyperoxic). Apparent growth was monitored daily by measuring the optical density of the cultures at 750 nm (OD 750nm ). Data were averaged from quadruplicate cultures; variability in these results is represented by error bars. Apparent growth was negatively affected under both treatment conditions. data do not provide information on whether the differences in expression reflect induction or repression of transcription or changes in RNA stability under the new conditions, we use the term expression to indicate the number of detected transcripts. Using this FC criterion, 2896 TUs were identified as differentially expressed for at least one of the test conditions. These TUs consisted of 2536 mRNAs, 41 tRNAs, six rRNAs, three RNAs involved in RNA processing, and 310 unannotated sRNA, of which 248 were encoded in the chromosome (Table S3). Since the samples were treated to remove rRNAs (Materials and Methods), we eliminated data corresponding to 5S, 16S, and 23S rRNAs from our analysis, on the basis that differences in rRNA may reflect processing.
Of these 2896 TUs (1234 in microoxic, and 1662 in hyperoxic), 2119 had a significant expression change in only one of the test conditions, 643 showed a similar response under both, while 134 had n opposing expression under microoxic vs. hyperoxic environments (Figure 2). Interestingly, among the genes with opposed expression profiles, five genes (two of them colocalized in the same genomic region) encode proteins involved in the metabolism of tetrapyrrole molecules (Table  2). Two out of five accumulated mainly under microoxic conditions (Log 2 FC .5), including AM1_0465 encoding the oxygen-dependent Mg-protoporphyrin IX monomethyl ester cyclase (AcsF), and AM1_0466 encoding a heme oxygenase (HO). In contrast, the transcripts of the genes encoding the three subunits of the light-independent protochlorophyllide reductase (ChlN, AM1_1444, ChlL, AM1_1445, and ChlB, AM1_1539) were significantly reduced under microoxic conditions. Similarly, the gene AM1_4366, encoding the uroporphyrin-III C-methyltransferase (cysG) at the branching point for the B12 (cobalamin) synthetic pathway, was significantly reduced under microoxic conditions. Another group of genes (AM1_1222, AM1_1223, and AM1_1224) in which expression was reduced under microoxic conditions, was the operon (SufBCD) coding for the proteins involved in the assembly of iron-sulfur clusters (Shen et al. 2007) ( Table 2).
Functional composition of the set of differentially expressed genes An EADEG was applied to identify functional categories significantly affected by our treatments. After a preliminary examination of the categorical classifications available in the Cyanobase, KEGG, and GO databases, we elected to use the KEGG and GO databases for categorization, because they cover a larger number of genes compared to Cyanobase (Table S1). The results for both databases indicated that, under microoxic conditions, a significant number of genes encoding subunits of both photosystems, as well as phycobilisomes (PBS) are downregulated, having a FDR ,0.05 (Table S4). The results using the KEGG database categories did not show any significant upregulation for categories, while using the GO database classification, DNA processing reactions showed a significant upregulation under microoxic conditions (Figure 3). Although increased concentrations of O 2 in the medium caused a significant downregulation of genes encoding subunits of the ribosome, and proteins involved in RNA translation, none of these categories were significantly upregulated ( Figure 3). Nevertheless, these results should be viewed with caution, since .57% of the proteins-coding genes in Acaryochloris lack any annotated function, i.e., they are not categorized under any biological function.
Protein-coding genes differentially expressed within significantly affected categories Based on results of our EADEG, genes encoding proteins involved in light harvesting, and subunits of both photosystems, were among the most affected by altered O 2 levels in the cultures. None of the genes encoding PSI subunits showed a positive regulation in either microoxic or hyperoxic environments. In fact, only three genes (ycf3, ycf4, and ycf37) involved in PSI assembly/stability (Dühring et al. 2006a;Ozawa et al. 2009;Boudreau et al. 1997) showed stable expression under microoxic conditions (Table 2). A very similar downregulation was observed for genes localized in the plasmid pREB3, encoding the PBS subunits that form the phycocyanin units, and the linker proteins. Only one (AM1_1558, apcA) of the genes encoding the allophycocyanin subunits showed increased expression levels, while expression of others (AM1_4469, apcA, AM1_5810, apcA, and AM1_2376, apcB) did not change. It is important to note that expression of apcA under all test conditions is 100-1000 times lower than that of apcB, and .10,000 times lower than some of the phycocyanin-binding apo-proteins ( Table 2).
The expression patterns of genes coding PSII subunits were similar to those of PSI and phycobiliprotein complexes under microoxic conditions ( Figure 4). Exceptions were noted for the induction of the normally cryptic psbA1 (AM1_0448) gene encoding the D1 protein (Summerfield et al. 2008), and the expression of one of the genes (AM1_G0114) encoding the PsbU subunit of the oxygen-evolving complex ( Figure 4). Under hyperoxic conditions, the expression of psbA2 and psbA3 (encoding the D1 protein) increased, as did that of the gene encoding the PsbJ subunit, and AM1_D0138, encoding another homolog of the PsbU subunit ( Table 2).
Expression of ROS-scavenging genes ROS are by-products of both respiration and photosynthesis. Thus, efficient scavenging is crucial to prevent photo-oxidative damage, especially in cyanobacteria, in which both processes occur simultaneously. Enzymes, like superoxide dismutase (SOD), efficiently scavenge the superoxide (O 2 2 ). In Acaryochloris, four open reading frames, AM1_5239 (Cu 2+ /Zn 2+ -SOD), AM1_2962 (Mn 2+ /Fe 2+ -SOD), AM1_3669 (Mn 2+ /Fe 2+ -SOD), and AM1_0511 (Ni-SOD), encode proteins resembling SOD. Of these, only AM1_3669 and AM1_0511 accumulated under hyperoxic conditions, but decreased under microoxic conditions ( Figure 5). The highest induction of genes encoding scavenging proteins under hyperoxic conditions was for AM1_3715 (catalase, katG) (log 2 FC . 3), which participates in the elimination of H 2 O 2 . Other genes, related to H 2 O 2 scavenging, such as AM1_A0300 (thiol-specific peroxidase), increased their expression under hyperoxic conditions, but had decreased expression under microoxic conditions. Interestingly, the genes encoding enzymes involved in glutathione metabolism either did not change markedly, or else their transcripts increased more under microoxic conditions than under hyperoxic conditions. The only gene for which expression under hyperoxic conditions was significantly higher than under control conditions was AM1_3681, encoding glutathione-disulfide reductase; its expression decreased with respect to the control under microoxic conditions ( Figure 5).
Terminal oxidases also play an important role in ROS prevention. The presence of terminal oxidases in the thylakoid membrane is key to balancing metabolic flow between the respiratory and photosynthetic electron transport chains, as well as reducing the amount of O 2 in the vicinity of the thylakoid membrane thus, preventing ROS   (Schmetterer 2016). In Acaryochloris, genes encoding four terminal oxidases have been previously annotated: (i) AM1_4621 (coxB), AM1_4620 (coxA), and AM1_4619 (coxC), encoding subunits of a mitochondrial-type cytochrome c oxidase (cox) complex; (ii) AM1_A0138; (iii) AM1_0843; and (iv) AM1_1551, encoding plastidic-type terminal oxidases (ptox) (Schmetterer 2016). In Anabaena variabilis, coxB, the first gene within the cox locus (coxBAC), was apparently transcribed more often than the other two genes in the operon (Schmetterer et al. 2001). A similar expression pattern is apparent in our results, with coxB (AM1_4621) transcript levels at least three times higher than the other two subunits ( Figure S1A). In fact, coxB expression level increased up to eight times under hyperoxic conditions. The expression of coxA and coxC showed no sig-nificant changes under altered O 2 conditions. Intriguingly, the expression of two (AM1_A0138 and AM1_0483) of the three ptox genes was very low under all conditions, while the expression level of AM1_1551 was .45 times higher under microoxic than under control conditions ( Figure S1B).

Potential transacting sRNAs involved in the adaptation to aerobic variations
Alignment of the reads resulted in a large number of sRNAs corresponding to either UTRs or intergenic ncRNAs. We differentiated sRNAs within these two main groups, based on whether the distance between the detected sRNA and the closest annotated mRNA was #20 nucleotides (in the case of UTRs), or .20 nucleotides (in the case of n ncRNAs). The transcription start site and orientation of the transcript were determined from predictions obtained using algorithms embedded in PePPER (de Jong et al. 2012). Using this criterion for the 248 differentially expressed chromosome-detected transcripts, we identified 190 sRNAs as UTRs and 58 as ncRNAs. The UTR expression level was mostly correlated (Spearman correlation coefficient, r S . 0.6) with the closest gene, except for four of the sRNAs (AM1_NC24, AM1_NC96, AM1_NC169, and AM1_NC181) ( Figure S2). Some of the intergenic ncRNAs were among the most highly expressed transcripts under all three conditions profiled (Table 1). Of the 58 differentially expressed ncRNAs localized in the chromosome, only six showed significant opposing expression profiles for microoxic and hyperoxic conditions. Three were less abundant under microoxic conditions (AM1_NC12, AM1_NC161, and AM1_NC254), while transcripts for the other three (AM1_NC270, AM1_NC256, and AM1_NC315) were enhanced in the hyperoxic culture ( Figure 6).
Given that most bacterial ncRNAs act through sequence-specific binding to regions close to the ribosome-binding site of mRNAs, we sought to predict potential RNA targets using a window of 275 nucleotides around their respective start codons. Setting a p-value threshold of 0.01 for results obtained from the IntaRNA server, the number of predicted targets was 95 for AM1_NC6, 96 for AM1_NC246, 50 for AM1_NC137, 89 for AM1_NC276, 45 for AM1_NC249, and 84 for AM1_NC323. The number of targets was much larger than expected based on available literature. To reduce these targets, as well as the number of false positives, we calculated the correlation between the numbers of reads obtained for each of the three conditions. We reasoned that any potential target should show an inverse correlation with the particular ncRNA, given that most bacterial ncRNAs act as negative regulators of gene expression, even when other mechanisms cannot be disregarded (Storz et al. 2011). Defining a threshold of r S # 20.5 for inverse correlation, the list of candidates was reduced to 38 potential  Table S4). The size of the circles is proportional to the number of genes in that category, reflecting differential expression, while color indicates their confidence level or FDR value. The graph was generated using the R package ggplot2.

DISCUSSION
In the laboratory, Acaryochloris can grow as a free-living form, under conditions very different from those in which it was initially isolated. In nature, it forms part of an algal mat, associated with colonial ascidians (Miyashita et al. 1996). It has been speculated that the multiplicity of homologous genes, and the relatively large genome size, of Acaryochloris reflect its evolutionary adaptation to its niche (Swingley et al. 2008), in contrast to the reduced genome size of the picoplanktonic Prochlorococcus genus (Delaye and Moya 2010;Dufresne et al. 2005). Here, we have shown that, under different O 2 conditions, expression levels among homologous genes varies. Given such adaptive capability, it is tempting to speculate on the flexibility of Acaryochloris to adapt to changing environmental conditions. This would be one of the benefits obtained from having coexisting multiple copies of genes, in spite of the cost of such a large genome size.
The combination of porphyrin-containing molecules, O 2 , and light often results in photo-oxidative damage to cellular structures. Hence, it has been speculated that divergence of the biosynthesis of bacteriochlorophyll and chlorophyll occurred to reduce photo-oxidative damage under an increasingly oxic atmosphere (Reinbothe et al. 1996). The syntheses of various tetrapyrrole molecules (heme, bilins, cobalamin, and chlorophyll) share a common pathway from ALA to uroporphyrinogen III. At this point, the cobalamin biosynthesis pathway branches from the uroporphyrinogen III pathway via a methylation reaction, catalyzed by the multifunctional chelatase CysG. The downregulation of cysG expression under microoxic conditions indicates that O 2 levels are an important regulatory element for the synthesis of cobalamin. In fact, CysG directs the uroporphyrinogen III pathway toward the synthesis of cobalamin either via an oxygen-independent or dependent pathway (Figure 7). Alternatively, it can be redirected toward the heme or chlorophyll biosynthetic pathways. In these pathways, O 2 levels influence the expression of genes encoding enzymes involved in oxidation. The part common to the heme and chlorophyll biosynthetic pathways, from uroporphyrinogen III to protoporphyrin IX, contains several oxidation reactions. The first oxidation step converts coproporphyrinogen III to protoporphyrinogen IX. This oxidation is catalyzed by either HemF or HemN, using O 2 or a 59deoxyadenosil radical generated from S-adenosylmethionine, respectively. HemN contains a [4Fe-4S] cluster, as a prosthetic group, sensitive to the presence of O 2 . In Synechocystis, a mutant lacking hemF was able to grow under microoxic conditions, but did not grow under aerobic conditions. In contrast, an hemN knockout mutant showed impaired growth only under microoxic conditions (Goto et al. 2010). Thus, in the presence of O 2 , it could be expected that hemN expression would be minimal compared with hemF. In Acaryochloris, the product of two genes (AM1_0467 and AM1_1283) resembles HemN, but only AM1_0467 had a strong induction (Log 2 FC 6.8) under microoxic conditions (Figure 7). Expression of HemF (AM1_0615) did not show any significant change. This suggests that AM1_0467 is the homolog to hemN in Acaryochloris, based on its expression profile. The product of the HemN/HemF reaction (protoporphyrinogen IX) is converted to protoporphyrin IX, becoming the final precursor common to Chl and heme, as well as heme-derived bilins. The protoporphyrinogen IX oxidation to protoporphyrin IX can be catalyzed by three enzymes, namely HemG, HemY, and HemJ (Kato et al. 2010). Only HemG, which is absent in most cyanobacteria, seems to be oxygen-independent (Boynton et al. 2009). Most cyanobacteria use the oxygen-dependent HemJ, although a few use HemY (Kobayashi et al. 2014). Interestingly, Acaryochloris contains both a HemJ-and a HemY-like enzyme, and, like most cyanobacteria, it lacks a gene homologous to HemG. AM1_5767, having enhanced expression in the microoxic environment, encodes a protein containing a HemY-like domain. A BLAST search, using the sequence for the Synechocystis HemJ (encoded by slr1790) (Kato et al. 2010), returned AM1_2801 as being highly homologous to HemJ. Unlike AM1_5767, the expression of AM1_2801 was not influenced by altered O 2 levels, suggesting that the main pathway for the oxidation of protoporphyrinogen IX in Acaryochloris under microoxic conditions is through HemY (AM1_5767). After this step, iron or magnesium is incorporated into protoporphyrin IX by a ferro-or magnesium chelatase, leading to the synthesis of heme and bilins (in the case of iron insertion), or Chl (in the case of magnesium insertion) (Chen 2014). In the Chl branch, the next substrate that is oxidized is Mg-protoporphyrin IX monomethyl ester (MPE). This reaction is catalyzed by MPE cyclase, which converts MPE into protochlorophyllide (Beale 1999). MPE can follow two pathways: one through an oxygen-dependent MPE-cyclase (AcsF, encoded by AM1_0465, or AM1_2295); and the other through an oxygenindependent MPE-cyclase (BchE) (Raymond and Blankenship 2004). BLAST results did not return any gene homologous to BchE in Acaryochloris. Nevertheless, our results show that the expression of both ascF (AM1_0465, AM1_2295) homologs differ. AM1_0465 was strongly induced under microoxic conditions (log 2 FC . 6), while the expression of AM1_2295 was reduced by almost half (Figure 7). In Synechocystis, a similar expression profile was described for two homologous genes encoding AcsF; deletion of these genes impaired growth under aerobic conditions (Minamizaki et al. 2008). Based on the conclusions drawn for Synechocystis, it is likely that AM1_0465 is the main MPE-cyclase in Acaryochloris under microoxic conditions, while, under aerobic conditions, our results show its expression is null, compared to that of AM1_2295 (Figure 7).
Based on published work, the FeS cluster within the ChlL subunit of the light-independent protochlorophyllide reductase (D-POR) shows a high vulnerability to O 2 Yamazaki et al. 2006). Such susceptibility might explain why this multimeric enzyme functions in the dark, when O 2 levels are low. Intriguingly, downregulation of the genes encoding the subunits of D-POR under microoxic conditions was observed, confirming that expression levels of this gene are controlled by the reduction state of the photosynthetic electron trans-port chain, and not by O 2 levels (Horiuchi et al. 2010). In contrast, the expression level of the light-dependent protochlorophyllide reductase (L-POR) did not change under microoxic conditions, but decreased under hyperoxic conditions ( Table 2).
Because of the potential deleterious effects of a misregulation of the tetrapyrrole biosynthetic pathway, it is expected that several regulatory factors (including sRNAs) control its activity. However, how a cell senses O 2 levels, and controls the expression of multiple genes is not fully understood. In Synechocystis, O 2 levels are sensed by the transcriptional factor ChlR (sll1512), which positively regulates acsF, ho2, and hemN expression (Aoki et al. 2012). The homolog to ChlR in Acaryochloris is encoded by AM1_1552. Expression of AM1_1552, as expected for a  positive regulator of genes sensitive to O 2 , increased under microoxic conditions (log 2 FC 1.9). A search using the FIMO tool within the MEME suite using the ChlR recognition motif (TTMCC-N 4/3 -GGWAA) provided by Aoki et al. (2012) returned a putative site (p-value ,0.0005), located 22 bp upstream of AM1_0466 (ho2). Furthermore, the expression control performed by regulatory factors, the synthesis of the final products, and their assembly into the apoprotein moiety, have to be tightly regulated to avoid their accumulation as free pigments in the membrane. For example, members of the HLIP family appear to mediate between both pathways in the assembly of photosynthetic complexes (Hernandez-Prieto et al. 2011;Yao et al. 2012;Sobotka et al. 2008;Adamska et al. 2001). In Synechocystis, HLIPs accumulate under multiple-stress conditions, while, under laboratory growth conditions, they are expressed at a low level (He et al. 2001). In Acaryochloris, there are 13 hlip genes; our data showed that AM1_3193 is among the most expressed transcripts in the control sample, contrary to what was observed in Synechocystis (He et al. 2001). In general, hlip expression levels decreased both under microoxic and hyperoxic conditions, following the trend observed for both photosystems. The regulatory role of HLIPs is most evident when examining the C-terminal extension of the ferrochelatase gene in cyanobacteria and chloroplasts, which shares a high homology with HLIPs (Funk and Vermaas 1999). This HLIP-like extension appears to induce or repress ferrochelatase activity, depending on the amount of free chlorophyll in the thylakoid membrane (Sobotka et al. 2008). In most cyanobacteria, only one gene encodes for ferrochelatase, while in Acaryochloris there are three ferrochelatase gene copies (AM1_1959, AM1_C0107, and AM1_C0204), and all of them have HLIP-like C-terminal extensions. AM1_C0107 and AM1_C0204, localized in the plasmid pRBE3, encode identical ferrochelatases; and their transcripts are .100 times more abundant than those generated from the copy (AM1_1959) localized in the chromosome, under all test conditions. The identification of ferrochelatase, as a pivotal enzyme at the intersection between Chl and heme syntheses (Figure 7), makes the regulation of its function by the HLIP-like extension essential to the flux of metabolites toward one or other of its final products.
In cyanobacteria, a large portion of the heme generated by ferrochelatase is funneled toward heme oxygenase. This is the first step in the synthesis of bilins, the pigments bound in the phycobiliproteins of the PBS. The PBS antenna found in Acaryochloris consists of a single rod structure Hu et al. 1999), in which phycocyanin (pc)containing subunits are the main component, and allophycocyanin (Apc)-containing subunits are only a minor component of the bottom disc of the rod-structured phycobiliprotein complex (Marquardt et al. 1997). This structure explains the larger number of reads for the genes encoding the pc-binding apoproteins (Table 2), compared with the Apc-containing subunits (ApcA and ApcB). Furthermore, their gene loci are physically separated from each other, with the genes encoding the Apc proteins, ApcA (AM1_1558, AM1_4469, and AM1_5810) and ApcB (AM1_2376), localized in the main chromosome, and the ones encoding the pc-binding apoproteins localized in the pREB3 plasmid. Our results and previously published work  show that, under microoxic conditions, the expression of genes encoding PBS subunits and their assembly in the antenna complex, are significantly decreased. Similarly, experimental conditions that limit access of cultures to essential nutrients also induce a downregulation of PBS encoding genes (Foster et al. 2007;Wang et al. 2004;Zhang et al. 2008). Downregulation of PBS encoding genes under microoxic conditions also has been observed in Synechocystis sp. PCC6803 (Summerfield et al. 2011). This downregulation may reflect decreased formation of bilins by the oxygen-dependent heme oxygenase (HO). In most cyanobacteria, two genes encode for heme oxygenases (ho1, and ho2), with HO2 having a higher affinity for O 2 , and being most active under microoxic conditions (Aoki et al. 2011). In Acaryochloris, two genes encode proteins with high similarity to HO1 (AM1_C0205 and AM1_C0108), while two others encode HO2 (AM1_0850 and AM1_0466). The expression of both genes encoding for HO1 decreased under microoxic conditions, while the expression of AM1_0466 increased under microoxic conditions (log 2 FC . 5), revealing this gene as most probably HO2.
In Synechocystis, it was noted that the expression of genes encoding PBS subunits is highly correlated with genes encoding subunits of the ATP synthase (ATPase) complex (Hernández-Prieto et al. 2016;Summerfield et al. 2011). Hence, a similar downregulation of ATPase subunits would also be expected for Acaryochloris under microoxic conditions. In Acaryochloris, two sets of genes encoding for ATPase subunits exist (one set is localized in the chromosome, and another in the pREB4 plasmid) (Swingley et al. 2008). The expression of the genes encoding ATPase located in the plasmid increased slightly under Figure 7 Diagram representing the tetrapyrrole biosynthetic pathway. Only genes encoding proteins discussed in the text are shown. Genes differentially expressed are highlighted in green and red, indicating downregulation and upregulation under microoxic conditions, respectively. Specific values are given in Table 2. microoxic conditions, while the ATPase from the main chromosome showed decreased expression under microoxic conditions, consistent with the results for Synechocystis (Hernández-Prieto et al. 2016). In fact, the ATPase encoding genes localized in the chromosome are phylogenetically closer to ATPase from other cyanobacteria (Swingley et al. 2008) than those localized in the plasmid. In addition to their downregulation under microoxic conditions, the ATPase genes localized in the chromosome are transcribed more frequently under all test conditions than the copies localized in the plasmid. These results indicate that the ATPase gene copies localized in the plasmid might be cryptic, or function in conditions different from the ones tested here.
A well-documented effect observed under microoxic conditions in cyanobacteria is the induction of the psbA1 gene encoding a homolog of the D1 protein of PSII (Summerfield et al. 2008), with repression of psbA2 and psbA3. D1 differential expression also was confirmed in this study, consistent with previously published results (Kiss et al. 2012). Noticeably, a similar expression profile was observed here for one (AM1_G0114) of the four genes (AM1_G0114, AM1_D0138, AM1_3966, and AM1_5046) encoding the PsbU subunit of PSII ( Figure  4). The PsbU subunit is associated with the oxygen-evolving complex, and functions to stabilize the PSII complex under high-intensity light conditions, protecting it from ROS . The differential induction of this gene is interesting, since it has been assumed that, because of differences in the sequence of the psbA1 encoded D1 protein, the PSII complexes assembled with this protein lack the capacity to evolve O 2 (Kiss et al. 2012;Murray 2012). The expression of the rieske-containing subunit (PetC) (Wenk et al. 2005) of the cytochrome b 6 f subunits is affected in a similar manner to psbA1. Similar to D1, PetC can be transcribed from three genes (AM1_4450, AM1_0450, and AM1_1961), with AM1_0450 being the only one induced under microoxic conditions (Table 2). It is important to note that AM1_0450 is upstream of psbA1. Thus, it is likely that they are cotranscribed in Acaryochloris, together with another two genes encoding proteins of unknown function: one (AM1_0449) containing a bacteria-conserved domain (DUF2892), and the other (AM1_0451) containing a rhodaneselike domain linked with assimilation of thiosulfate under anaerobic conditions in other bacteria (Schedel and Trüper 1980). A similar gene cluster is observed in Synechocystis genome, where psbA1 (slr1181) is the first gene of a set of 10 genes orientated in the same direction, including slr1184 encoding a rhodanese-like protein, and slr1185 encoding PetC; expression levels in these genes were also higher under microoxic conditions (Summerfield et al. 2008). Based on the microarray meta-analysis presented in CyanoEXpress, only the expression of slr1182, slr1183, and slr1184 seem to follow the same trend under different environmental conditions, as expected for genes in an operon (Hernández-Prieto et al. 2016). Nevertheless, results obtained from genes for which expression levels are low under most conditions (as is the case for psbA1) should be viewed with caution, especially when interpreting correlations. Further investigation to confirm whether expression of these proteins results in a restructuring of the photosynthetic complexes, and rerouting of the electron transport chain under microoxic conditions, is needed, but is beyond the scope this paper.
Transcriptome data (Mitschke et al. 2011;Kopf et al. 2014;Voigt et al. 2014;Hernández-Prieto et al. 2012), as well as computational predictions (Voß et al. 2009;Voigt et al. 2014), of sRNAs in cyanobacteria have revealed a large number of previously unidentified noncoding protein transcripts, exceeding all previous predictions. Although the roles of some of these sRNAs have been partially characterized in cyanobacteria (Nakamura et al. 2007;Voß et al. 2007;Dühring et al. 2006b), the function of most of them is still unknown. A well understood process in Escherichia coli is the degradation of complementarily paired ncRNAs and mRNAs, mediated by the protein Hfq (Massé et al. 2003). Such processes affect protein synthesis at the transcriptional level, saving valuable resources that can be used to synthesize a different protein complement that is more suitable to the new environmental conditions. In cyanobacteria, a gene encoding a homolog to Hfq has been identified, but its role in ncRNA/mRNA degradation has not yet been demonstrated (Bøggild et al. 2009;Dienst et al. 2008). Of the 58 ncRNAs localized in the chromosome (Table 1), only six (AM1_NC12, AM1_NC161, AM1_NC254, AM1_NC270, AM1_NC256, and AM1_NC315) showed significant opposing expression profiles for microoxic and hyperoxic conditions, as would be expected for ncRNAs involved in adaptation to different O 2 levels ( Figure 6). Of these, functional enrichment analyses of their potential targets returned significant results (FDR , 0.05) for only two ncRNAs: AM1_NC6 and AM1_NC276. The targets predicted for AM1_NC276 comprised genes within the category "aerobic respiration" ( Table 3), indicating that expression of this ncRNA might be relevant for adaptation to altered O 2 levels. In E. coli, several ncRNAs have been shown to play a role in adaptation to oxidative stress (Berghoff and Klug 2012), but further experimental work is necessary to determine whether they have the same functions in Acaryochloris.
In conclusion, the large number of protein-coding genes and sRNAs detected as differentially expressed under our test conditions revealed that there is a high level of regulation related to O 2 in cyanobacteria. The multiplicity of genes encoding homologous proteins in Acaryochloris exceeds that of many cyanobacteria, indicating a complex regulatory network in this organism. Multiple ROS-scavenging pathways, and their different transcriptomic responses, may represent the history of alternative ROS-scavenging mechanisms, which has evolved and developed in parallel to new metabolic pathways that produce ROS.
Ultimately, the lack of efficient methods to generate mutants in Acaryochloris makes environmental studies, such as this one, key to understanding its regulation and annotating its yet unknown gene functions.