Genome-Wide Dynamic Transcriptional Profiling of the Light-to-Dark Transition in Synechocystis sp. Strain PCC 6803

Abstract

We report the results of whole-genome transcriptional profiling of the light-to-dark transition with the model photosynthetic prokaryote Synechocystis sp. strain PCC 6803 (Synechocystis). Experiments were conducted by growing Synechocystis cultures to mid-exponential phase and then exposing them to two cycles of light/dark conditions, during which RNA samples were obtained. These samples were probed with a full-genome DNA microarray (3,169 genes, 20 samples) as well as a partial-genome microarray (88 genes, 29 samples). We concluded that (i) 30-min sampling intervals accurately captured transcriptional dynamics throughout the light/dark transition, (ii) 25% of the Synechocystis genes (783 genes) responded positively to the presence of light, and (iii) the response dynamics varied greatly for individual genes, with a delay of up to 120 to 150 min for some genes. Four classes of genes were identified on the basis of their dynamic gene expression profiles: class I (108 genes, 30-min response time), class II (279 genes, 60 to 90 min), class III (258 genes, 120 to 150 min), and class IV (138 genes, 180 min). The dynamics of several transcripts from genes involved in photosynthesis and primary energy generation are discussed. Finally, we applied Fisher discriminant analysis to better visualize the progression of the overall transcriptional program throughout the light/dark transition and to determine those genes most indicative of the lighting conditions during growth.

The extent and dynamics of cellular changes in response to changing light conditions are areas of fundamental importance to biology. Light plays a crucial role in all photosynthetic organisms, ranging from bacteria to plants, by regulating growth and photomorphogenesis, initiating phototropism or phototaxis, altering gene expression, and resetting circadian rhythms, among other things (3, 19, 21, 28, 37, 39, 47, 56). In this study, we developed full-genome microarrays and used them to analyze the transcriptional profiles of the entire genome of the photosynthetic cyanobacterium Synechocystis sp. strain PCC 6803 (Synechocystis) throughout repeated light-to-dark transitions. Synechocystis is often used as a model organism for the study of photosynthesis and other important physiological processes of photosynthetic organisms and was one of the first prokaryotes to be fully sequenced (32). It was our intention to (i) characterize the dynamics of light-regulated gene expression for this organism, (ii) determine the portion of the genome that is light responsive, (iii) provide a complete list of light-responsive genes for future studies of light-regulated gene expression and physiology, and (iv) reveal those genes whose expression characteristics most strongly reflect lighting conditions.

Cyanobacteria are prokaryotic photosynthetic microalgae that exhibit a well-studied response to light intensity and quality (3, 25, 26, 30, 44, 56, 65). This response is reflected at multiple regulatory scales throughout the cell, including rapid alterations in transcript levels (25, 29, 34, 46, 65). Most recently, Hihara et al. reported more than 160 genes that responded to changes in the intensity of light in Synechocystis (30). Other reports detail alterations in fatty acid desaturase gene accumulation (34), increased transcript levels of pho regulon genes (4), altered levels of light harvesting phycobilisomes by chromatic adaptation (20, 24), induced expression of stress genes (8, 23), and the differential expression of photosynthesis-related genes (4, 12, 17, 37, 38, 46, 59). Other physiological mechanisms, such as central carbon metabolism (Calvin cycle) and the generation of cellular building blocks, have also been studied in detail (reviewed in reference 5).

The advent of DNA microarrays has enabled researchers to survey the expression levels of thousands of individual genes simultaneously (22, 55). Many studies have since been published with as many as 40,000 genes monitored on a single microarray (1, 11, 14, 15, 40, 41, 60, 66). The great majority of this work has been applied to eukaryotic systems, while DNA microarray applications for prokaryotic systems have progressed much less rapidly. Previous bacterial microarray studies have primarily taken static samples, with only a few providing data of a scale and complexity comparable to that of their eukaryotic counterparts (15, 30, 33, 53, 64). As a result, determining an appropriate time frame (total time and sampling frequency) to study whole-genome prokaryotic transcriptional dynamics with DNA microarrays remains a question yet to be fully addressed. Of particular interest in profiling the extent and dynamics of the light/dark transition in this photosynthetic prokaryote was whether DNA microarrays would generate reproducible and accurate measurements of dynamic gene expression. The positive answer to this question enabled the experimental design and subsequent analysis of our light/dark transcriptional profiling studies aimed at discovering the particular genes of the entire Synechocystis genome (3,169 genes) that respond during the light/dark transition. We report in this study these results along with the response dynamics for every gene in the Synechocystis genome. In this paper we report some new findings about photosynthesis and respiration genes, but we also use these data to validate our transcriptional results by placing them in context with what is presently known about light/dark regulation of these genes. Although the focus of this paper is not to elucidate fundamental mechanisms involved in light/dark-regulated cyanobacterial physiology, the results reported herein establish a framework for discovering links between gene expression and overall cell physiology for Synechocystis. This framework comprises accurate transcriptional profiles and a projection method whereby quantified transcript levels are linearly combined to create projection variables allowing for better visualization of the progression in the overall transcriptional program throughout light/dark transitions. This projection method also facilitates the determination of genes that discriminate between light and dark growth conditions.

MATERIALS AND METHODS

Strains, maintenance, and growth conditions.

Batch cultures of Synechocystis sp. strain PCC 6803 (wild type) (Pasteur Culture Collection) were maintained at 30°C in BG11 medium (Sigma, St. Louis, Mo.). Throughout each experiment, continuous irradiance of ca. 250 μmol of photons m⁻² s⁻¹ was provided by cool white fluorescent bulbs (light conditions). All light-to-dark experiments were performed in a 10-liter glass vessel (6-liter working volume) in BG11 containing 10 mM acetate and 10 mM HEPES (pH 8.0) and continuously sparged with 3% CO₂ (in air) in a light-tight incubator (Percival, Perry, Iowa).

Partial DNA microarray design and production.

The partial-genome microarray contained 88 genes comprising 21 functional categories (as defined by the Kyoto Encyclopedia of Genes and Genomes), including light harvesting and photosynthesis, membrane fluidity, energy generation, the Calvin cycle, amino acid biosynthesis, and CO₂ fixation, among others. DNA microarrays were produced as previously described (cmgm.stanford.edu/pbrown, www.nhgri.nih.gov/DIR/LCG/15K/HTML, and sequence.aecom.yu.edu/bioinf/funcgenomic) with the following modifications. Primers were designed using the Primer 3.0 software (www-genome.wi.mit.edu; Whitehead Institute, Cambridge, Mass.) set to produce 500-bp DNA fragments with a T_m of 60°C and 18 nucleotides in length. All primers were purchased from Sigma-Genosys (The Woodlands, Tex.) (all primer sequences can be obtained from the authors upon request). All genes were PCR amplified with a 96-well plate thermacycler (MJ Research, Incline Village, Nev.) and an AmpliTaq Gold PCR buffer solution containing 1.5 U of AmpliTaq Gold (PE Biosystems, Foster City, Calif.), 0.25 mM deoxynucleoside triphosphates (Roche Molecular Biochemicals, Indianapolis, Ind.), 1 μM primers, and 0.5 μg of Synechocystis chromosomal DNA. Synechocystis chromosomal DNA was purified with DNAzol (Life Technologies, Gaithersburg, Md.). Amplification was performed for 30 cycles at an annealing temperature of 53°C for 45 s, elongation at 72°C for 90 s, and denaturation at 95°C for 30 s. All PCR products were verified by using 1% agarose gel electrophoresis prior to purification by the Qiagen 96-well PCR cleanup kit (Valencia, Calif.). Genes were reamplified under altered PCR conditions until a single band was visible on the agarose gel (three rounds produced 99% success). Purified PCR products were adjusted to 3× SSC (1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate) and 0.1% sarcosyl prior to spotting. The final concentrations for PCR products ranged from 0.2 to 0.6 μg/μl.

PCR products were spotted in triplicate with 3 taps per spot by using the GMS 417 microarrayer (Affymetrix, Santa Clara, Calif.) in array mode. Glass slides were purchased from Corning (CMT-GAP; Acton, Mass.). After printing, all arrays were snap dried, UV cross-linked for 10 min with a long-wavelength UV lamp, blocked in a solution containing 0.2 M boric acid, 1-methyl-2-pyrollidinone, and succinic anhydride (7:63:1 [vol/vol]) for 20 min, rinsed in H₂O, boiled for 2 min, flash denatured in ice-cold 100% ethyl alcohol, and stored at room temperature in the dark. Positive and negative control DNA included human genomic DNA and PCR-amplified β-actin and human GAPDH from the pTRI series of in vitro transcription vectors from Ambion, Inc. (Austin, Tex.).

Whole-genome DNA microarray design and production.

Full-length PCR-amplified gene products for nearly every gene in the Synechocystis genome were provided by Dupont Co. PCR products were resuspended in 50% dimethyl sulfoxide, spotted with a Cartesian Technologies quill pin microarrayer (Irvine, Calif.), cross-linked with a UV stratalinker, and stored in the dark until use.

RNA purification.

RNA was purified by using Qiagen Mini, Midi, and Maxi kits. Immediately after transfer from the growth culture into 50-ml polypropylene centrifuge tubes, cells were placed into liquid N₂ and chilled to <5°C within 20 s. Chilled cells were immediately centrifuged at 4,000 × g for 3 min in a precooled centrifuge (4°C), supernatant was discarded, and cell pellets were immediately frozen in liquid N₂ prior to permanent storage at −20°C. Cell pellets were resuspended in RLT buffer (Qiagen) and an equal volume of 0.1-mm-diameter glass beads (B. Braun Biotech, Inc., Allentown, Pa.) and were ground in a bead mill (B. Braun) for four cycles of 1 min of grinding and 1 min on ice. All grinding was performed in a walk-in 4°C cold room. Lysed cells were then purified by following the exact protocols of the Qiagen RNA purification kit. To remove carbohydrates and chromosomal DNA, a final precipitation using 4 M LiCl was performed. This modified RNA purification procedure produced results comparable to those previously described for light-dark studies of Synechocystis (46).

cDNA creation and labeling.

RNA was reverse transcribed to fluorescently labeled cDNA by using 15 U of Superscript II/μg of RNA; 1× superscript buffer; 1× dithiothreitol; 0.5 mM each dCTP, dATP, and dGTP; 0.2 mM dTTP; and 0.1 mM Cy-dUTP (Amersham-Pharmacia Biotech, Upsala, Sweden). Reverse transcription was performed for 2 h at 42°C. A volume of 1.5 μl of 1 N NaOH was added, and RNA template was degraded at 65°C for 10 min followed by neutralization with 1 N HCl. Cy3- and Cy5-labeled sample and control cDNA were mixed and ethyl alcohol precipitated. Precipitated cDNA was resuspended in 16 μl (32 μl for whole-genome arrays) of prewarmed (65°C) hybridization buffer (Roche Molecular Biochemicals for partial-genome arrays, Clontech for whole-genome arrays) and denatured for 10 min at 95°C prior to being applied to the microarrays.

Hybridization and scanning.

Microarrays were denatured for 2 min in 95°C H₂O and flash-cooled in −20°C ethyl alcohol. After heat denaturation, labeled cDNA was flash-cooled in an ice slurry and was briefly spun at 7,000 × g to collect evaporated liquid. A small aliquot (1 μl) of cDNA was removed for spectrophotometric diagnostics, and the remainder was carefully pipetted over the microarray. A glass slide was placed over the hybridization solution (Roche Molecular Biochemicals [partial-genome array] or Clontech [full-genome array]), and hybridization was performed in a water bath overnight at 62°C (partial genome) or 50°C (whole genome) (as recommended by the manufacturer) in water-tight humidified hybridization chambers (Corning). Arrays were washed in an excess of 1× SSC + 0.1% sodium dodecyl sulfate for 5 min, 0.2× SSC for 3 min, and 0.1× SSC for 5 min. Cleaned arrays were briefly washed with 1 M ammonium acetate and were immediately spun at 500 × g for 4 min to remove all salt deposits prior to scanning. Clean slides were scanned with the ArrayWorx scanner from Applied Precision, Inc. (Issaquah, Wash.), for all partial-genome arrays or the Axon Instruments GenePix 4000B (Foster City, Calif.) for all full-genome arrays.

Partial-genome data acquisition, filtering, and adjustments.

Scanned slide images were generated automatically with the software provided with the ArrayWorx Scanner (Applied Precision, Inc.). Images were quantified with the ArrayWorx Software, and these raw data were used for all subsequent calculations. Raw data were filtered to remove any spots with local background exceeding the average background (over all spots on the slide) by one standard deviation (SD) (σ_background for all spots across the entire slide): if spot_background was greater than average_background plus σ_background, then the spot was removed. Additionally, the filter removed those spots with signal values exceeding the background by less than one SD: if spot_signal was less than spot_{background (local)} plus σ_{background (local)}, then the spot was removed. To account for differences in Cy3 and Cy5 incorporation, brightness, and photobleaching characteristics, all data were adjusted by the average Cy3 background/Cy5 background ratio and by the specific brightness of each fluorophore in the original cDNA hybridization solution. The specific brightness was calculated by measuring the absorbance of the cDNA solution at 260 nm, 550 nm (Cy3_{extinction coefficient} = 150,000 M⁻¹ cm⁻¹; Amersham Pharmacia BioTech), and 649 nm (Cy5_{extinction coefficient} = 250,000 M⁻¹ cm⁻¹; Amersham Pharmacia BioTech) and by using Beer's law to calculate the specific fluorophore incorporation (Cy-dye/NT). Brightness is the product of the extinction coefficient and the quantum efficiency (QE) of each fluorophore (QE_Cy3 = 0.38; QE_Cy5 = 0.28).

Full-genome data acquisition, filtering, and analysis.

Microarrays were quantified by using the GenePix Pro software from Axon Instruments. Erroneous spots were manually flagged and removed from the final data set. All microarray data were filtered to remove any spots in which less than 60% of the signal pixels exceeded the local background value for both lasers (532 and 632 nm) by more than one SD. The median pixel ratio of the filtered data for each spot was used for all subsequent analysis. This ratio was normalized by the median Cy3/Cy5 ratio automatically generated by the GenePix Pro software [normalized ratio = (raw Cy3/Cy5)/(median Cy3/Cy5)].

Bioinformatic analysis.

Normalized expression data were analyzed by a variety of techniques. Hierarchical clustering analysis was performed by using the Cluster and TreeView programs developed by M. Eisen as well as by using the hierarchical clustering algorithm within MATLAB. Pearson (genes) and Spearman (samples) correlation coefficients, as well as Euclidean distance, were used as measures of similarity. The samples were grouped by using the average linkage technique. Self-organizing map algorithms were implemented in MATLAB and developed essentially as described by Tamayo et al. (62), with the number of nodes ranging from 2 by 2 to 4 by 8. Fisher discriminant analysis (FDA) was also performed in MATLAB.

FDA is essentially a projection from the original space of gene expression data to a reduced space obtained so as to maximize the ratio of the variance between groups to the variance within groups. This is mathematically equivalent to maximizing the mean separation between the various groups or classes of samples in the reduced dimensional space. If there are c classes of samples in the data, the within-group variance, W, and the between-group variance, B, are defined by the following two equations, respectively:

(1)

(2)

where T is the total variance and W is the sum of the variances of each of the c sample classes considered separately. X_k and X are gene expression data matrices for samples in class k and the entire expression set, respectively. These matrices are organized such that X(i,j) is the expression of gene j in sample i. The vector _k (1 × g) is the group mean of gene expression for each of the g genes for class k, while is the mean of gene expression for all the data organized in a single group. It can be proved that the separation between predefined groups in a reduced dimensional space is maximized when the space is defined by the eigenvectors of the matrix W⁻¹B (16). Mathematically, the eigenvalue decomposition of the matrix is given by the equation W⁻¹BL = LΛ. The eigenvector matrix (L) defines the dimensions of the reduced space. Each column of L defines an axis or discriminant function (DF) of the FDA space with the diagonal entries of the eigenvalue matrix (Λ) providing a measure of the discriminant power of the corresponding DF.

Very simply, FDA is just a linear combination of the original gene expression data yielding a projection from the gene expression space onto each DF that is called the discriminant score and is calculated by the equation

(3)

where y_j is the discriminant score of the actual sample x on the jth DF. The coefficients in this linear projection, as indicated in the above equation, are the entries of eigenvector L and are called the discriminant weight for each gene. The discriminant weights determine the contribution of each gene in defining the DF, i.e., the projection. Dillon and Goldstein (16) can be consulted for more details of the procedure used in our analysis. To facilitate the application of FDA to other sets of gene expression data, the software GeneProjection is freely available on our website. The program first rank orders all genes in terms of class discriminating ability by using Wilk's lamda and univariate F distribution. This allows selection of a smaller number of discriminatory genes that are subsequently used in the FDA projection.

Microarray quality control.

The partial-genome DNA microarrays had, on average, a signal-to-noise (S/N) ratio [S/N = average (signal_i − background_i)/σ_background] of greater than 3, with several values greater than 10. Full-genome microarray S/N ratios were routinely greater than 10, with several arrays greater than 30. For each spot on the array the S/N ratio was calculated by using the spot signal, the local background value, and the background variation across the entire slide. The S/N ratio, therefore, represents the number of SD of the background that the spot signal exceeds the local background value.

RESULTS AND DISCUSSION

Development of experimental protocols using full- and partial-genome microarrays.

The first objective of this study was to determine the potential of DNA microarrays to reliably measure the transcriptional state of the prokaryote Synechocystis sp. strain PCC 6803. To do so, RNA samples were obtained at frequent and evenly spaced intervals from cultures exposed to steady and transient light and dark conditions (see Fig. 1) and probed by using a partial-genome DNA microarray. Specifically, Synechocystis was grown to mid-exponential phase (A₇₃₀ = 0.6 to 0.8), at which point a large sample of RNA was isolated (full light conditions). The lights were extinguished in the light-tight growth incubator, and RNA samples were taken after 23 and 24 h in the dark (full dark). The lights were then turned back on for 100 min (transient light), followed immediately by an additional 100 min in the dark (transient dark). This experimental design followed closely the design of Kis et al. (34), in which the desABCD desaturase genes were found to be transcriptionally regulated by alternating light and dark conditions. Partial-genome arrays were used to develop experimental protocols, design full-genome experiments, and determine the reproducibility of the microarray data. We present below validation of the expression data on the basis of an assessment of the variability of background noise (σ_background) in 29 partial- and 20 full-genome arrays and the reproducibility of fluorescent signal within microarrays, as well as among experimental samples from parallel cultures. Furthermore, we examine the extent of correlation among bicistronic gene expression profiles and compare our results to those for previously described light-regulated genes.

FIG. 1.

Description of the growth experiment. Synechocystis was grown to mid-exponential phase (A₇₃₀ = 0.6 to 0.8), with a doubling time of approximately 12 h, and a large sample of RNA was isolated (full light). The lights were extinguished in the light-tight growth incubator, and RNA samples were taken after 23 and 24 h in the dark (full dark). The lights were then turned back on for 100 min (transient light) followed immediately by an additional 100 min in the dark (transient dark). After the lights were extinguished, the cells did not grow (as measured by A₇₃₀ and cell counts) but did remain in a viable state (as indicated by growth recovery in the light). For partial-genome arrays, RNA samples were taken every 10 min for the 200 min of the transient-light-to-transient-dark alternating conditions. For full-genome arrays, samples were taken every 30 min and the transient phases were shortened to 90 min. To assess the repeatability and accuracy of the DNA microarray techniques, additional RNA samples from parallel experiments were evaluated for a total of 29 and 20 total microarrays for the partial and full genome, respectively.

Dynamic transcriptional profiling depends on the ability to distinguish noise (random or systematic experimental variation) from dynamic features (oscillations, lag times). For this purpose, a robust estimation of experimental error across microarrays rather than within sample variability is required. Repeat experiments (five RNA samples from parallel cultures) were performed and analyzed by microarrays for which expression ratio means and SD were calculated for each of the genes located on the array. For each gene across the considered samples, the measured ratio between the Cy3- and Cy5-labeled samples varied by an average of 44.9% (average across all genes of σ_genei/[average ratio of gene i]). This value was used to calculate the minimum ratio required for statistically significant expression differences. Using the 95% confidence interval, a transcript ratio difference of 88% (1.96 × 44.9%) was determined as the threshold above which the transcript level was deemed to be significantly different in the Cy3-labeled sample relative to that in the Cy5-labeled sample.

We determined the sampling frequency such that the variation in gene transcript levels between two samples exceeded the experimental error determined above. We found that 10 min after a change in light conditions, the average change (SD) in transcript levels was 47% of the mean. This is expressed as follows: average across all genes of [(transcript level_{time 2} − transcript level_{time 1})/average transcript level]_{gene i} = 0.47. This average change in transcript levels increased to 56% after 20 min and to 73% after 30 min. We concluded that expression ratio changes could be reliably detected at least 20 min after a perturbation in light conditions. We chose a 30-min sampling interval as the best balance between accuracy and experimental resolution when studying whole-genome dynamic transcriptional changes throughout the light/dark transition. The major features of the dynamic response were satisfactorily captured by the 30-min interval (Fig. 2B ). Specifically, changes in the transcript levels began between 20 to 40 min, peaked at 50 to 70 min, and settled at 80 to 100 min. Figures 2A and C display typical partial- and full-genome microarrays from this study.

FIG. 2.

Representative partial-genome (A) and full-genome (C) microarrays and full-genome experimental design (B). Partial genome arrays contained 88 genes spotted in triplicate and were used to develop microarraying methods and to design full-genome experiments. Full-genome arrays contained PCR products for nearly all of the coding regions of the Synechocystis genome (3,169 regions). Transcriptional dynamics were compared to the variation of an individual sample taken from parallel cultures (44.9%). Thirty minutes was required for transcript levels to change to a level substantially different from experimental variation.

The expression profiles obtained from partial- and full-genome microarrays were consistent. In Fig. 2B, the profiles for apcAB and cpcBA from the partial-genome microarrays are displayed. These matched the profiles obtained from the full-genome microarrays very closely (see Fig. 4, class II genes) where transcripts accumulated when exposed to the light followed by a decrease in the dark. Similar correspondence between partial- and full-genome microarray results were observed for rbcS, rbcL, psaA, psaB, petF, psbA123, and atpB. Moreover, these results all matched previously described light- and dark-responsive data (38, 46, 59).

FIG. 4.

Dynamic transcriptional profiling of photosynthesis and respiration genes throughout the light/dark transition. Genes are separated by dynamic (classes I to IV) and were color coded by gene category. Z values (gene_i − average_i)/(SD_i) were used to adjust for differences in magnitude for each gene.

Extent and dynamics of the light/dark transition.

Whole-genome microarrays were used for the rest of this study. After the 24-h dark incubation phase (Fig. 1), cells were exposed to a 90-min transient light phase followed by a 90-min transient dark phase, during which RNA samples were taken at 30-min intervals (a total of 20 arrays were performed, including replicates). The microarray data were quantified, filtered, and normalized (see Materials and Methods), and genes that did not change appreciably after the 24 h of the dark incubation phase were removed (ca. 2,100 genes). For the full-genome arrays, the average variation of identically prepared samples was 24% of the mean, meaning that a threshold of 48% (1.96 × 24%) in the gene expression ratio was required for statistically significant expression differences.

The filtered gene expression data were clustered into groups on the basis of similarity among expression ratio profiles by using a self-organizing map (SOM) algorithm (Fig. 3A and B). To determine the total number of distinct expression patterns, we varied the number of possible expression patterns (nodes) in the SOM between 12 and 32 (2 by 6 to 4 by 8 nodes). Thus, by increasing the number of possible expression patterns, only those genes with a high degree of similarity were grouped together. Equivalently, the SOM forced nondistinct expression patterns into neighboring nodes as a means of accommodating all 32 nodes. Upon examination of the expression patterns corresponding to a 32-node SOM, we found that only four to five were substantially different. This analysis also generated upper and lower bounds on the number of genes that were classified into each of these dynamic patterns. The number of genes reported in each class was determined as the average of the number of genes found in each class when 12 or 32 initial expression patterns were used in the SOM. To further increase the stringency and to eliminate possible jackknife expression patterns (a jackknife pattern has a single peak or trough and is indicative of possible experimental anomalies), each gene expression vector was required to have more than one value significantly greater than its initial 24-h dark level.

FIG. 3.

The extent and dynamics of gene regulation during the light/dark transition. SOM methods were used to visualize the primary dynamic patterns followed throughout the light/dark transition (A and B). Class I peaked within the first 30 min of transient light, class II peaked after 60 to 120 min, class III peaked after 120 to 150 min, and class IV peaked at 180 min. (C) One-dimensional clustering results for all named genes within the Synechocystis genome (ca. 1,050 genes). Individual gene values were normalized by each gene's average 24-h dark transcript accumulation level and log transformed prior to running the clustering algorithm. Red indicates genes that accumulated in the light, and green indicates genes which decreased relative to the 24-h dark values. About 25% of the genome was found to increase in the light.

Class I contained a total of 108 ± 20 genes, or about 3% of the genome, that responded immediately following exposure to light. Classes II and III contain genes that were induced between 60 to 90 or 120 to 150 min, respectively, after initiation of the light phase, which was the majority of light-responsive genes (537 ± 100). The remaining 138 genes exhibited a very delayed response (class IV). A total of 783 ± 148 genes (25% of the Synechocystis genome) were estimated to be light responsive. In a similar study with Arabidopsis, Harmer et al. (28) found that about 500 genes responded throughout repeated light/dark cycles. Also, Laub et al. (36) found 553 genes (19% of the genome) that altered throughout the Caulobacter crescentus cell cycle. Although genes in classes I, II, and III clearly begin to increase prior to the transient dark phase (at 90 min), the cause of the pattern exhibited by class IV genes remains unclear. It is possible that class IV includes both delayed light-responsive genes as well as genes responding to the absence of certain molecules available during growth in the light. Additional patterns in the 4-by-8 SOM were observed in which changes in expression values did not fit clearly into any of the overall classes. These results could be explained by an experimental anomaly at secondary peaks or may represent oscillatory behavior caused by repression of member genes.

Figure 3C displays the clustering results for the whole-genome data from this study (only genes with assigned names [ca. 1,050 genes] are displayed; full genome expression data are available from the authors' web page). We report these results to provide a comprehensive view of the structure of the transcriptional data obtained in this study. Genes were grouped according to the similarity of their expression profiles across all experiments by using correlation. The groupings that were obtained were consistent with the SOM results and with previously published results for light-responsive genes.

Transcriptional profiles of the photosynthetic and respiration family of genes.

All transcriptional data are available from http://bioinformatics.mit.edu. Here we narrow our discussion on the dynamics of genes related to photosynthesis and respiration, both important processes for the generation of energy for growth. Transcriptional profiles for each of the significantly induced genes from this large gene family are displayed in Fig. 4. Each gene is classified by its dynamics (classes I to IV) and is color-coded by its subgrouping within the photosynthesis and respiration gene family. Although the light regulation of some of these genes has been previously described, we report our results for these same genes to both corroborate what is known and to further validate our microarraying protocols.

The primary light harvesting apparatus of Synechocystis is concentrated within the phycobilisomes as allophycocyanin and phycocyanin (24, 27). The genes responsible for the protein structures of these apparatuses responded rapidly, and transcripts differentially accumulated to some of the highest levels of all genes compared to data for the 24-h dark levels. Transcripts from both of the core allophycocyanin genes, apcA and apcB (Fig. 4C), significantly accumulated within 30 to 90 min, as did transcripts for the core β-subunit gene apcF (Fig. 4C). The L_CM core membrane linker gene apcE (Fig. 4C) and the L_C core linker apcC gene (Fig. 4B) also responded rapidly to the presence of light. The apcABC operon is transcribed as a polycistron in Synechococcus sp. strain PCC 6301. Therefore, the coordinated transcription of these genes in Synechocystis was not unexpected (6, 58).

Cells acclimated to dark conditions typically reside in state 2, where energy is directly transferred from the allophycocyanin core to photosystem I through the apcD-encoded AP-B subunit. As cells transition to state 1, energy is transferred via apcF and apcE directly to photosystem II (57). Our results are consistent with this model, where apcD transcripts did not increase over the dark-acclimated levels while apcE and apcF transcripts accumulated. Similar rapid accumulation of transcripts was observed for the rod element phycocyanin genes. The cpcBAC1C2D polycistronic operon (51) as well as both cpcG genes were significantly induced within the first hour of turning on the lights. Interestingly, both phycobilisome degradation genes (nblA; Fig. 4D) significantly decreased after the lights were turned on (2, 13).

The light energy absorbed by the phycobilisomes is transferred directly to photosystem II and either directly or indirectly to photosystem I. Photosystem I genes responded rapidly to the introduction of light. In the psaAB operon (7, 59), corresponding to photosystem I P700 apoprotein subunits Ia and Ib, both messages increased dramatically within the first hour, followed by a dynamically similar decrease after the lights were extinguished (Fig. 4B). These results are consistent with previous studies that indicated psaAB transcripts were highly degraded in cells grown in the dark without glucose (59). The transcript levels of additional photosystem I subunit genes psaC and psaE as well as psaJ, psaF, psaL, and psaI followed dynamics similar to those of psaAB (Fig. 4A to C) (9, 10, 48, 61).

Photosystem II genes showed various dynamics in their response to light. The important structural subunit genes psbA2 and psbD2 both increased significantly within the first 30 min and maintained high transcript levels until 60 min after the lights were turned off (Fig. 4A and B). These two genes encode the essential subunit polypeptides D1 and D2 of photosystem II. Other essential subunit polypeptides with increased transcript accumulation levels included CP47 and CP43 from psbB and psbC (7), respectively, as well as the cytochrome b₅₅₉ α and β subunits encoded by psbE and psbF (7). Synechocystis contains multiple multigene families, including two encoding photosystem II polypeptides (43, 45, 46, 52). The transcript level of the psbD gene, from the psbDC operon, was markedly increased after the transition to light conditions, following dynamics similar to those for the psbD2 gene. Of note, transcript levels of both psbD genes are known to accumulate in Synechocystis sp. strain PCC 6803 under standard growth conditions (25). The psbA multigene family contains three different loci, psbA1, psbA2, and psbA3 (43, 45, 46). The psbA2 and psbA3 genes code for functional D1 protein; however, psbA1 is thought to be a cryptic gene that does not normally code for a functional product (54). Under illuminated conditions, psbA2 and psbA3 transcripts are known to accumulate while psbA1 transcripts do not accumulate (46). These results are consistent with our experiments where psbA2 transcripts accumulated dramatically, psbA3 transcripts accumulated at 120 min to levels roughly 1.6-fold greater than the 24-h dark value, and psbA1 transcript levels did not change significantly from the 24-h dark level (data not shown). It is of note that cross-hybridization among these transcripts is a possible confounder of these results; however, this is unlikely in the case of psbA1 due to the lack of any substantial change in accumulation of its transcripts.

Synechocystis contains multiple copies of different ndh genes of the NADPH dehydrogenase complex. Overall, the ndh genes did not respond dramatically over the course of this experiment. In fact, of the 21 different ndh genes only gene expression from multigene families was significantly altered. Specifically, ndbB (slr1743; 3 total ndb genes [31]), ndhD (slr2007; 5 ndhD genes [50]), ndhD2 (slr1291; 5 genes [50]), ndhF1 (sll1732; 4 genes), and ndhF2 (slr0844; 4 genes) transcripts all increased, while solitary genes, such as ndhA, ndhB, ndhC, and ndhE, did not show transcript accumulation in the presence of light. Also, expression from the operon ndhAIGE (18) and gene clusters ndhCKJ and ndhFD were internally consistent and did not change from dark levels (data not shown).

Finally, substantial up-regulation of genes involved in ATP synthesis and the cytochrome b₆/f complex was observed (35, 39, 42). This result is consistent with the general theme of the increased expression of other photosynthetic genes. Also, CO₂ fixation genes were observed to increase after 30 to 90 min of light growth. Specifically, multiple carbon dioxide concentrating proteins (i.e., ccmA [49]; Fig. 4A) were rapidly increased after the lights were turned on. The rubisco gene operon rbcSXL followed internally consistent patterns of expression, even though the actual increase in transcript levels was moderate for both rbcS and rbcL (44 and 35% greater than those for dark phase, respectively). Cell densities during the 24-h dark acclimation phase did not change, but growth did recover once the lights were left continuously on (Fig. 1). Increased expression of these genes is consistent with the need for growing autotrophs to fix carbon dioxide.

Additional genes up-regulated after the shift to light.

An additional concern of this study was stress induced by the rapid transition from full-dark to light growth conditions (30). To this end we examined changes in several stress-induced gene transcript accumulation profiles after this transition. Transcripts for the groESL operon (slr2075-2076) rapidly accumulated after 30 min in the light; however, groEL-2 transcripts did not change noticeably. Transcripts from the clpB (slr1641) gene also accumulated rapidly and had some of the highest fold increases of all of the genes evaluated. ftsH transcript accumulation was observed for both the slr1370 and slr0228 loci, about threefold after 1 h in the light. In contrast, dnaK and dnaJ transcripts did not accumulate upon exposure to the light. On the basis of these data, we concluded that the rapid introduction to light did result in cell stress as reflected in the increased transcript accumulation of stress genes (8, 23). Therefore, some of the genes indicated as light induced were likely determined to be so as an indirect result of increased cell stress during the light/dark transition rather than being directly regulated by a light-specific signaling mechanism.

Additional reports of genes regulated by light include the pho operon (4), the des operon (34), and three different operons implicated in Synechocystis chemotaxis (3). In all of these cases except that of phoH (slr2047), we did not observe any noticeable trends in transcript accumulation. This result appears to be in contrast to results of previous studies, except for that for phoH, and deserves further investigation

After 30 min in the light, ccmA, clpB, cpcB, psaF, groESL, and rpoD1 were all among the most significantly induced genes. These genes are involved in photosynthesis, carbon fixation, or cell stress, all pathways related to the light/dark transition. The rpoD1 gene encodes a principle sigma factor of Synechocystis (63). Interestingly, rpoD (sll0306) transcripts substantially accumulated after 24 h in the dark, while rpoD1 (slr0653) transcript levels substantially accumulated after 30 min in the light. Moreover, transcript accumulation levels for rpoD (sll0306) decreased about twofold after cells were exposed to light and did not increase again during the transient dark phase. These two sigma factors deserve further study for their apparently complementary role in managing the cell's response to rapidly changing lighting conditions. As the transient light phase progressed, a noticeable trend towards genes involved in photosynthesis and respiration was apparent among the most strongly induced genes. By 90 min, petH, cpcA, cpcB, cpcC, apcB, ccmL, rbcX, psbA2, and psbX were all among the most strongly induced genes. This would suggest a gradual buildup of these genes following the initial transition that included increased stress genes and possibly a redistribution of sigma factors to restructure the expression landscape in the light.

Linking transcriptional profiles with physiological states.

A common concern in transcriptional profiling studies is identifying those features that best describe the ultrahigh dimensional data sets and doing so in a way that can accommodate the large volume of generated data. Most often the approach has been to focus on those genes or gene classes thought to be of relevance to the environmental conditions under investigation (the approach reported in Fig. 4). What is desired, however, is a data-driven approach to the analysis of large gene expression databases (16). The SOM and clustering results of Fig. 3 facilitate the visualization of transcriptional patterns. FDA (16, 61a) was also used in order to facilitate data visualization in a reduced dimensional space. FDA additionally allowed sample classification and identification of discriminating genes. The method is outlined in Materials and Methods and can be implemented by using the GeneProjection program.

Figure 5A shows the discriminant weights of the FDA linear combinations for each of the top 30 discriminatory genes (out of a total of 3,169 genes). Genes with negative canonical variable 1 (CV1) loadings were significantly increased during the initial 90-min light phase, and those with positive loadings were significantly increased during the final 90 min. Importantly, the strongly light-induced psbA2, rbcL, cpcB, and apcE genes were among this group of highly discriminatory genes. These genes are all part of the photosynthetic and respiration family of genes and can clearly be seen to respond early and dramatically to changing light conditions (see Fig. 4). It is thus seen that FDA rapidly produces an initial list of genes whose expression patterns were consistently altered across the conditions studied without any a priori assumptions or knowledge about the relevance of these genes to the process under study. The majority of discriminatory genes had positive discriminant weights and exhibited class III dynamics (peaking at 120 to 150 min). For example, the NADPH dehydrogenase gene, ndhG, was discriminatory due to higher transcript levels in transient-dark phase compared to those in transient-light phase (class III). This is a result of early-light-induced genes remaining elevated into the transient dark phase.

FIG. 5.

Class separation and data reduction by FDA. Three distinct states were assumed (full dark [24 h], transient light, and transient dark). (A) Gene loadings for the top 30 discriminatory genes. (B) Projections for samples using the top 30 discriminatory genes displayed in panel A. Negative CV1 values indicate early-light-responsive genes (A) or light samples (B), and positive CV1 values indicate delayed-light-responsive genes (A) or dark samples (B). Note the following equation: CV_j = L_1jgene₁ + L_2jgene₂ + L_3jgene₃…… L_njgene_n, where L_1j equals weight of gene i in CV_j.

Figure 5B displays the discriminant scores (i.e., sample projections) of the transcriptional profiles measured for all samples obtained during the light-to-dark transitions. It can be seen that (i) all samples from the same physiological state (i.e., full dark, transient light, and transient dark) cluster in the same area of the projection space, thus defining the corresponding physiological state, and (ii) a progression from one cluster to another reflects the physiological changes occurring during the light-to-dark transitions. Similar to the loadings shown in Fig. 5A, light and dark samples were easily separated along the CV1 axis. Diagrams of discriminant scores like that of Fig. 5B can be used for sample classification purposes once the projection (i.e., the discriminant weights) has been defined. Additionally, such projection suggests ways by which physiological states can be rationally linked to gene expression data.

Conclusions.

The primary objective of this study was to establish the use of microarrays as a means for measuring full-genome transcriptional dynamics during the light/dark transition in Synechocystis. This objective was accomplished by first applying partial-genome microarrays to RNA samples taken at frequent intervals from cells exposed to alternating light and dark growth conditions. From this analysis we determined that a 30-min sampling frequency was sufficient to capture the dynamics of this transition beyond experimental error. This was followed with a detailed transcriptional analysis using whole-genome DNA microarrays. Major findings include the following: (i) 25% of the Synechocystis genome is light responsive, (ii) the dynamics of the light/dark transition followed four waves of transcript accumulation with the majority of genes responding in 60 to 150 min, (iii) many photosynthetic and respiration gene transcripts increased rapidly in the light, and (iv) dimensional reduction approaches confirmed the dynamics of the transcriptional program and allowed the visualization of the progression of physiological states throughout the light/dark transition.

The quality of the data obtained in this study justifies the use of microarrays for the systematic study of transcriptional changes during states of ecological or biotechnological interest, such as effects of carbon dioxide and light or conditions maximizing biopolymer accumulation, among others. Such data will reveal important aspects of the expression regulatory mechanisms and help direct efforts of genetic modification for cell property improvement or maximization of product accumulation. It should be noted that, important as expression data might be, they represent only a portion of the information needed to fully define the cellular phenotype. Therefore, conclusions from gene expression studies need to be validated experimentally, as overinterpretation of such results is possible. In this context, protein profiles and fluxes of intracellular pathways are important, as they provide complementary information about the actual cellular state.