AppA, a Redox Regulator of Photosystem Formation in Rhodobacter sphaeroides 2.4.1, Is a Flavoprotein IDENTIFICATION OF A NOVEL FAD BINDING DOMAIN*

The AppA protein is required for increased photosystem gene expression upon transition of the facultatively photoheterotrophic bacterium Rhodobacter sphaeroides 2.4.1 from aerobic to anaerobic photosynthetic conditions. AppA shows no obvious similarity to proteins with established function. Genetic evidence suggests that its effect is exerted through modulation of the activity of the repressor PpsR, which controls expression of multiple photosystem genes. To gain insight into the nature of AppA involvement in redox-dependent photosystem gene expression, the appA gene was overexpressed in Escherichia coli . AppA was produced as insoluble inclusion bodies. The purified inclusion bodies were found to contain FAD. By overexpressing various deletion derivatives, we were able to localize the region of AppA sufficient for FAD binding to approximately 120 amino-terminal residues. To assess the role of FAD binding in AppA function, we constructed an AppA derivative lacking the entire FAD binding domain. Surprisingly, this derivative complemented the AppA null mutant under-going transition from aerobic to anaerobic photosynthetic growth conditions almost to the same extent as the full-length AppA protein. When the sequence of the amino-terminal portion of AppA was examined, it was shown not to contain any known flavin binding motifs. However, two open reading frames of unknown function, showing significant similarity to the amino terminus of AppA, were identified, i.e. Synechocystis sp. Srl1694 and E. coli F403. The latter gene was amplified and overexpressed in E.

the photosystem whose function is to capture photons and to convert light energy into electron flow (1,2). The transition from aerobic to an anaerobic light environment requires, among other things, a significant increase in expression of the photosystem genes, i.e. genes encoding structural and assembly proteins of the light harvesting and reaction center complexes as well as genes encoding enzymes for bacteriochlorophyll and carotenoid biosynthesis. Under highly aerobic conditions, these genes are expressed at low basal levels. Several transcription factors coordinate the regulation of photosystem gene expression (reviewed in Refs. 3 and 4). Some of these factors are specific to photosystem gene expression, e.g. a redox-sensitive transcriptional repressor PpsR (5-7), whereas others have a broader range of targets comprised of photosystem as well as nonphotosystem genes, e.g. the PrrBA twocomponent activation system (8,9) and the anaerobic activator, FnrL (10).
The AppA protein has been identified as yet another critical component required for activation of photosystem gene expression (11,12). An R. sphaeroides 2.4.1 AppA null mutant is impaired in its transition from aerobic to anaerobic photosynthetic growth because of defects in the production of both photopigments and structural protein components of the photosystem. In contrast to the transcriptional factors mentioned above, the primary sequence of AppA gives no clue as to its possible role in photosystem gene expression. AppA contains no DNA binding motif(s) and shows no obvious similarity to proteins with known function. Using a variety of molecular genetic approaches, we have established that AppA exerts its regulatory effect independently of the FnrL and PrrBA regulatory pathways, but it does appear to affect PpsR repressor activity (7). At present, neither the nature of the redox sensitivity of PpsR (13,14) nor the mechanism of the AppA-PpsR interactions is known. AppA could function as a hypothetical redoxsensing partner directly interacting with PpsR. Alternatively, it could be involved in modulation of a critical redox carrier, which in turn interacts with or is a cofactor of the PpsR repressor. Until now, we have remained unaware of what features, if any, allow AppA to participate in redox-dependent processes.
In this work, we present the first biochemical characterization of AppA, which has revealed that this protein is likely to contain several redox-responsive cofactors. We have identified one of these compounds as FAD. We have located the FAD binding domain within AppA and tested the significance of this domain for protein function. We have also uncovered, as a result of this study, a small group of bacterial flavoproteins showing similarity to the FAD binding domain of AppA.
* This work was supported by National Institute of Health Grant GM 15590. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
R. sphaeroides strains were grown at 30°C in Sistrom's medium A (1), supplemented, when required, with tetracycline, 1 g/ml. An aerobic-to-anaerobic transition was performed as follows. Liquid cultures (5 ml) of the AppA null strain, APP11 (12), carrying the appropriate plasmid(s) in trans, were grown aerobically in culture tubes until late exponential phase and placed at 4°C overnight. These cultures were then diluted 1:50 with fresh medium to approximately the same optical density at 660 nm, placed into fully filled screw-cap tubes, and grown anaerobically under incandescent illumination at a light intensity of approximately 10 W/m 2 .
Plasmids were introduced into E. coli strains by either electroporation (Gene Pulser, Bio-Rad) or CaCl 2 transformation (15) and into R. sphaeroides by conjugation as described (16).
DNA Manipulations-Standard recombinant DNA techniques (15) and molecular biological enzymes and reagents were used according to specifications provided by the manufacturers. DNA sequencing with either the universal or sequence-specific primers was performed on an ABI377 automatic sequencer (Applied Biosystems) at the DNA Core Facility of the Department of Microbiology and Molecular Genetics. All recombinant DNA constructs described here were verified by DNA sequencing.
Sequence Analysis-Sequence analyses were performed using BLASTϩBEAUTY, WU-BLASTϩBEAUTY, ClustalW 1.7 (multiple alignment) programs, available at the Baylor College of Medicine Search Launcher. 1 Analysis of protein secondary structures was performed using the consensus secondary structure prediction developed by James Cuff and available as a part of the Jpred package. 2 Construction of appA-overexpression Plasmids-The 1.6-kilobase pair XhoI-AgeI fragment from plasmid p484Nco1 (12), carrying the appA gene, was first cloned into the multiple cloning site of vector LITMUS28 (NEB Biolabs), digested with XhoI and AgeI, to produce plasmid pLapp. The XhoI [513] site is positioned 4 codons downstream of the translational start of AppA; the AgeI site is positioned 270 base pairs downstream of the appA stop codon. The number in brackets corresponds to the sequence coordinates deposited in GenBank L42555. To construct a translational fusion of AppA to the amino terminus of LacZ␣ (LacZЈ::AppA) expressed from LITMUS28, pLapp was digested with SpeI and NsiI, treated with T4 polymerase and religated. The resulting plasmid, pLappA, expresses 19 residues of LacZ␣ fused inframe with the AppA protein under control of P lac (see Fig. 1). The deletion derivatives of pLappA, expressing the amino-terminal portions of AppA, were constructed by removing the 3Ј-terminal portions of appA using the Ecl136II site of the vector positioned downstream of appA and the following internal sites: plasmid pLapNae-NaeI [868]; pLapNar-NarI [748] (see Fig. 1).
To construct the glutathione-S-transferase (GST) 3 fusion with AppA (GST-AppA), the 1.6-kilobase pair BglII-Ecl136II fragment from pLapp containing the appA gene was first cloned into vector pGEX-2TK (Amersham Pharmacia Biotech) digested with BamHI and SmaI to produce plasmid pGapp5. The translational frame was reconstructed by digesting pGapp5 with NsiI, making ends blunt with T4 polymerase and religation. The resulting plasmid, pGappN, expressed GST fused to the fifth codon of AppA under control of P tac (see Fig. 1 Fig. 1).
Plasmids Expressing AppA Derivatives in R. sphaeroides-All plasmids used for expression in R. sphaeroides were based on vector pRK415 (17). Plasmid p484-Nco5 contains the full-length appA gene with its own promoter (12). To construct the appA gene derivative lacking the flavin binding domain, plasmid p484Nco5 (12) was digested with XhoI (partial digest) and StyI; the derivative containing the desired deletion of the XhoI [513]-StyI [1066] fragment located within the appA gene was gel-purified, blunt-ended with T4 DNA polymerase, and religated, resulting in plasmid p484Nco5⌬. The 2.1-kilobase pair Hin-dIII-SacI fragment from p484Nco5⌬ was cloned into vector pRK415 digested with HindIII and SacI, resulting in plasmid p484-Nco5⌬. The latter plasmid differs from p484-Nco5 only by deletion of the XhoI [513]-StyI [1066] fragment, resulting in an in-frame deletion of amino acid residues 5-190 of the AppA protein.
Construction of the f403-overexpression Plasmid-The gene encoding the E. coli F403 protein was PCR-amplified using synthetic primers: F403-S, 5Ј-ttgttaggcCTTACCACCCTTATTTATCGTAGCC-3Ј, and F403-R, 5Ј-ACTGAAtTCATGTGAAAGAATGTGCTG-3Ј, whose design was based upon the sequence deposited in GenBank under the accession number AE000215. The nucleotides shown in lower case represent deviations from the GenBank file sequence. These were introduced to create the StuI restriction site (F403-S) positioned at the translational start codon and EcoRI downstream of f403 (F403-R). Chromosomal DNA isolated (15) from E. coli DH5␣ was used as a PCR template. PCR was performed with Pfu DNA polymerase (Stratagene) as follows: 97°C, 5 min; 30 ϫ (95°C, 45 s; 55°C, 45 s; 68°C, 3 min); 72°C, 6 min. The 1.3-kilobase pair PCR fragment corresponding to f403 was gelpurified, digested with StuI, and cloned into the SmaI site of vector pGEX-2TK to produce plasmid pGORF403. The latter plasmid expresses the GST-F403 translational fusion under control of P tac .
Overexpression of AppA Protein Derivatives and Inclusion Body Purification-The plasmids overexpressing various portions of AppA were routinely maintained in E. coli DH5␣. Standard protocols for induction of expression were used according to recommendations of the manufacturer (Amersham Pharmacia Biotech). Proteins were separated by SDS-PAGE (18). Because full-length AppA, expressed from either pGappN (as a GST-AppA fusion) or pLappA, was always found in the insoluble fraction, various modifications of the standard protocol were attempted to achieve production of AppA in soluble form. These modifications included lower growth temperatures (12 to 37°C), lower IPTG concentration (0.05 to 1.0 mM), variable times of induction (1 to 20 h (for low temperature-growing cultures)), alternative E. coli hosts in place of DH5␣, e.g. JM109, HB101, solubilization of cell extracts with 0.1% Triton X-100 or sodium lauryl sarcosyl, addition of 1-5 mM dithiothreitol, etc. These modifications did not result in an increased production of soluble AppA. The AppA containing inclusion bodies from 1 liter of DH5␣(pLappA) culture were partially purified by multiple rounds of washing (PBS buffer containing 0.05% Triton X-100) followed by low speed centrifugation (45 min, 3,500 ϫ g) of the insoluble portion of crude cell extract.
Overexpression and Purification of the Amino Terminus of AppA and the E. coli F403 Protein-Strains DH5␣(pGapNae) and DH5␣(pG-ORF403), overexpressing the GST fusions to the 120-residue aminoterminal fragment of AppA and F403, respectively, produced significant amounts of soluble proteins when grown at room temperature and induced by 0.4 mM IPTG (final concentration) for 8 h. These proteins were purified by affinity chromatography according to recommendations of the manufacturer (Amersham Pharmacia Biotech).
Identification of the Flavin Cofactors Associated with AppA and F403-AppA-enriched inclusion bodies were originally extracted with acetone/concentrated HCl (1000: 1.3 v/v) and pelleted by centrifugation, and the supernatant was subjected to spectral analysis. Flavins were extracted from the yellow-pigmented protein preparations by heating at 95°C for 15 min in PBS buffer. The released flavins were separated from the insoluble fraction by centrifugation followed by ultrafiltration through Microcon-3 (cut-off size, approximately 3 kDa) columns (Amicon). The flavins present in the filtrate were either further purified by extraction into phenol and returned to the aqueous phase by adding ether as described (19) or analyzed without phenol extraction. The flavin solution was dried with a Speed-Vac concentrator, dissolved in 50% methanol, and analyzed by TLC on a silica gel 60 plate (Selecto Scientific) using 5% Na 2 HPO 4 as a developing solvent (20). 50% methanol solutions of FAD, FMN, and riboflavin (Sigma) were used as reference compounds. Flavin spots were detected by UV illumination.
Spectroscopy-Spectral analyses were performed using a Shimadzu UV-1601PC diode array spectrophotometer and 1-cm quartz cuvettes. A Klett-Summerson photoelectric colorimeter (Klett Manufacturing Co.) with red filter was used to measure the absorbance of anaerobically grown R. sphaeroides cultures.
Stoichiometry of FAD Binding-The amounts of protein-bound FAD were determined based upon the assumption that absorptivity of 11.3 mM Ϫ1 cm Ϫ1 at 450 nm, which is characteristic of free FAD (19), is equal to that of the protein-bound FAD. The protein concentration was measured with the BCA protein assay kit (Pierce). The protein purity was estimated based on SDS-PAGE.
The stoichiometry of binding of FAD to purified AppA was estimated as follows. Exogenous FAD was incubated with the equimolar amount (38.4 M) of the purified amino-terminal fragment of AppA (150-l reaction volume) for 30 min at room temperature. The mixture was diluted with 250 l of PBS, loaded into a Microcon-3 concentrator and centrifuged at 14,000 ϫ g for 25 min. The concentrations of FAD were determined in the filtrate and retentate, based upon absorption at 450 nm. The amount of bound FAD (first estimate) was calculated based upon the amounts of FAD in the original protein sample, filtrate, and retentate. The retentate (200 l) was further diluted with 4 volumes of PBS and concentrated as above-described, and the amount of bound FAD was calculated (second estimate). The average of the first and the second estimates from each experiment was taken.

RESULTS
Overexpression of the Full-Length AppA Protein in E. coli-To gain insight into the biochemical properties of AppA, we intended to purify this protein by using an E. coli overexpression system. For this purpose, the appA gene was cloned into the overexpression vector pGEX-2TK to produce plasmid pGappN (Fig. 1). The latter construct encodes the GST fused to the fifth amino acid residue of the AppA protein, expression of the fusion being controlled by P tac . The GST::AppA fusion protein was overproduced in E. coli strain DH5␣ following induction by IPTG (Fig. 2, lane 1). Employing the standard protocol resulted in the production of insoluble protein (not shown). Numerous attempts to obtain the protein in a soluble form were undertaken (see "Materials and Methods"); however the GST::AppA fusion was always produced in insoluble form, i.e. as inclusion bodies. To test whether the GST tag adversely affected solubility, the AppA protein was translationally fused to the amino-terminal residues of LacZ␣ and overexpressed under the control of P lac (Fig. 1, plasmid pLappA). The LacZЈ::AppA protein was overproduced in strain DH5␣ (pLappA) and found only as inclusion bodies (Fig. 3A). Several attempts to renature AppA from the denatured inclusion bodies were undertaken (21). However, none of the various protocols produced satisfactory results (data not shown), possibly because proper refolding of denatured AppA required the simultaneous presence of associated cofactors, should they exist. We therefore set about to identify these possible cofactors.
Extraction of the Cofactors Associated with AppA-We observed that E. coli cells overproducing AppA turned yellow upon induction. To investigate the basis of this phenomenon, we partially purified the AppA-containing inclusion bodies from strain DH5␣(pLappA) to 80 -85% purity as judged by SDS-PAGE (Fig. 3A). These inclusion bodies retained their yellow coloration, suggesting the presence of a cofactor(s) associated with AppA. The inclusion bodies were extracted with acidic acetone solution, and the spectrum of the extract consisted of multiple, apparently overlapping peaks, indicative of the presence of several potential cofactors associated with AppA (Fig. 3B). The major absorption peaks observed were at 365, 412, 440, and 466 nm. We anticipated that some of these peaks could belong to a flavin(s), e.g. the absorption peaks of FMN dissolved in acidic acetone are at approximately 362, 442, and a shoulder at 466 nm (data not shown). To test for the presence of noncovalently bound flavin, the AppA-enriched inclusion bodies were resuspended in PBS buffer and boiled. The yellow pigment was released and subjected to spectroscopic analysis. The absorption peaks characteristic of a flavin were observed, thus confirming the presence of a noncovalently bound flavin (see below). Additional compounds were present in the extract of the AppA-enriched inclusion bodies, and their analysis will be presented elsewhere. Below we focus on the characterization of the flavin associated with AppA.
Localization of the Flavin Binding Domain of AppA-No significant similarity between AppA and known flavoproteins present in the data bases was detected by primary sequence comparisons. We therefore set about to identify the flavin binding domain(s) of AppA by constructing a series of carboxylterminal deletion derivatives differing in length by approximately 25-65 amino acids (Fig. 1). These derivatives were overexpressed as GST fusion proteins, GST::AppAЈ (Fig. 2), and their coloration was noted. The proteins, progressively shortened from the AppA carboxyl terminus, were all yellow up to residue 124 (Fig. 1, plasmid pGapNae). Deletion of the 46 amino-terminal residues of AppA from the GST::AppA fusion (Fig. 1, plasmid pGap⌬Sm) resulted in a colorless protein. We therefore concluded that the flavin binding domain of AppA is confined to residues 5-124.
To rule out the possible involvement of the GST moiety in flavin binding or in the stabilization of an AppA-flavin complex, we constructed and overproduced the 120-residue AppA derivative lacking the GST-tag (Fig. 1, plasmid pLapNae). This derivative retained the yellow coloration, suggesting that GST was not involved in flavin association and that the AppA domain between residues 5 and 124 was sufficient for flavin binding. To further define the boundary of this domain, we constructed an AppA derivative truncated at residue 84 ( Fig. 1, plasmid pLapNar). This protein was found to be colorless, which suggested that the boundary of the flavin binding domain lies between residues 84 and 124.
Purification of the Flavin Binding Domain and Identification of the Flavin Moiety-The GST::AppA fusion protein containing the minimal flavin binding domain was overproduced in strain DH5␣(pGapNae) grown at room temperature under moderate induction (Fig. 4A, lanes 1-2). Under these conditions, the fusion protein was produced predominantly in the soluble form (Fig. 4A, lane 3). It was purified by one-step affinity chromatography with glutathione-Sepharose. The amino-terminal derivative of AppA was released from the resin by thrombin, whose cleavage site is positioned at the junction of GST and AppA. The resulting amino-terminal derivative appeared as two closely positioned bands after separation by SDS-PAGE (Fig. 4A, lane 4). The band sizes are in the range of 15-17 kDa, i.e. slightly larger than the 14 kDa, which was anticipated, relying on the amino acid sequence. The reason for the double band was not investigated. We estimated that the amino-terminal domain of AppA was purified to at least 90% purity. The spectrum of this protein preparation appeared to be typical of a flavoprotein, with absorption maxima at 266, 280, 290, 356, 380, 412, and 452 nm. The minor absorption peak at 412 nm suggested that some portion of the flavin is present in the partially reduced, semiquinone form, despite the presence of an aerobic environment and the absence of exogenously added reductants (Fig. 4B).
The nature of the flavin cofactor was determined by TLC of the flavin extracted from the heat-denatured protein. The extracted flavin had a mobility identical to that of FAD (data not shown). FAD content in the purified amino-terminal fragment of AppA was 0.58 Ϯ 0.09 mol/mol of protein. Because repetitive resin washes resulted in visible loss of yellow pigment, we determined the stoichiometry of FAD binding using exogenously added FAD. At saturating concentration of FAD, 1 mol of the amino-terminal domain of AppA was found to bind 0.89 Ϯ 0.10 mol of FAD, suggesting that 1 mol of the aminoterminal fragment of AppA is capable of binding 1 mol of FAD.
Role of the FAD Binding Domain of AppA Following an Aerobic-to-Anaerobic Transition-To gain insight into the role of the FAD binding on AppA function, we constructed an inframe deletion (residues 5 to 190) derivative of AppA, which lacked the entire FAD binding domain (Fig. 5A, plasmid p484-Nco5⌬). The truncated protein was expressed from the intact upstream regulatory region of appA to enable a direct comparison with the plasmid containing the full-length appA.
Plasmid p484-Nco5⌬ was introduced into the AppA null mutant, APP11. The resulting strain as well as strains carrying in trans the full-length appA gene in plasmid p484-Nco5 or the vector pRK415 alone (Fig. 5A) were tested for their ability to transit from aerobic to anaerobic photosynthetic conditions. Strain APP11(pRK415) has an extensive lag phase before resuming growth under anaerobic photosynthetic conditions, and its subsequent growth is very slow (Fig. 5B, trace 3), in line with our earlier observations (7). In contrast, strain APP11(p484-Nco5) performed the aerobic-to-anaerobic transition at least as efficiently as the wild type 2.4.1 (Fig. 5B, trace  1). Surprisingly, the transition of strain APP11(p484-Nco5⌬) was only slightly impaired as compared with that of APP11(p484-Nco5) (Fig. 5B, trace 2). This suggested that the function of AppA in the transition of R. sphaeroides from aerobic to anaerobic photosynthetic conditions is generally preserved despite the absence of the entire FAD binding domain. Therefore FAD binding does not appear to be essential for AppA function, at least under these experimental conditions. It should be mentioned that the appA gene derivatives were present in 4 to 6 copies/cell. Therefore, an increased abundance of these derivatives over the physiological concentration of AppA could serve to have masked the extent of the impairment resulting from the absence of the FAD binding domain in the deletion derivative (see also "Discussion").
A Novel Flavin Binding Domain-Searching the data bases for proteins similar to the FAD binding domain of AppA, i.e. residues 5-124, revealed two entries present in the sequences of the genomes of E. coli and Synechocystis sp., i.e. F403 and Srl1694, respectively (Fig. 6). The sequence identity over the 94-residue fragment between AppA and Srl1694 is 39%, between AppA and F403 is 33%, and between F403 and Srl1694 is 30%. By using similarity search programs based upon the multiple alignment present in Fig. 6, we could not identify any other members of this protein group.
The three proteins, AppA, F403, and Srl1694, differ in size, i.e. 450, 403, and 150 residues, respectively. Sequence similarities among these proteins are clearly limited to their aminoterminal regions, i.e. residues 16 -108 of AppA, residues 1-94 of F403, and residues 1-95 of Srl1694 (Fig. 6). Both the differences in size and confinement of the similarity to only the amino-terminal portion of AppA implied that AppA, F403, and Srl1694 perform different functions. We suggest that the region of similarity among these three proteins represents a novel flavin binding domain. This domain does not contain an obvious pattern of similarity to known flavoproteins; however, cer- tain features common to flavin binding domains can be deduced based upon multiple alignment (see "Discussion").
The secondary structure of the flavin binding domain was determined by several prediction algorithms based upon the multiple alignment shown in Fig. 6. According to the derived consensus, the FAD binding domain contains predominantly ␣-helices and only one evident ␤-sheet.
Amplification, Overexpression, and Flavin Binding of E. coli F403-We set about to test the hypothesis that F403 and Srl1694, which show similarity to the FAD binding domain of AppA, could be involved in flavin binding. For this, we chose the protein with the least similarity to AppA, i.e. E. coli F403. The f403 gene was PCR-amplified and cloned into pGEX-2TK to generate the GST::f403 translational gene fusion (plasmid pGORF403). The GST::F403 fusion protein was overproduced in strain DH5␣(pGORF403) (Fig. 7A, lanes 1 and 2). Similar to cells overproducing AppA, cells overproducing F403 appeared yellow following the course of induction. The GST::F403 fusion migrated as a 71-kDa protein, i.e. in good agreement with the calculated value. The GST::F403 protein was produced under these conditions in soluble form (Fig. 7A, lane 3). It was absorbed to glutathione-Sepharose and washed, and the F403 portion was released from the GST tag by thrombin digestion. The resulting 48-kDa F403 protein appeared to be at least 50% pure on SDS-PAGE (Fig. 7A, lane 4). The nature of the two major contaminating proteins was not investigated further because they did not interfere with the analysis of F403.
The spectrum of the F403 protein preparation was typical of FIG. 5. A, schematic representation of the appA gene derivatives expressed in strain APP11, the AppA null mutant. 1, plasmid p484-Nco5 carrying the fulllength appA gene; 2, plasmid p484-Nco5⌬ carrying the appA gene that lacks the amino terminus; 3, pRK415, vector used as control. B, aerobic-to-anaerobic light transition of strain APP11 containing various plasmids. Time zero corresponds to the transfer of the inoculi to anaerobic light conditions. The experiment was repeated three times essentially with the same outcome. Representative curves from one experiment are shown. a flavoprotein (Fig. 7B). The flavin moiety was extracted from the heat-denatured protein and identified by TLC as FAD (data not shown). Therefore, similar to AppA, F403 is a FAD binding protein. Although the ability of Synechocystis sp. Slr1694 to bind flavin has not been tested experimentally, we believe that it would be likely to do so, because the amino terminus of Slr1694 shares more similarity to the FAD binding domain of AppA than does the amino terminus of F403. DISCUSSION The AppA protein plays a critical, albeit as yet biochemically undefined, role in the ability of the facultative photoheterotroph R. sphaeroides 2.4.1 to transit from aerobic to anaerobic photosynthetic conditions (12) (Fig. 5B). It is also required for anaerobic photosynthetic growth per se; however, it is dispensable for anaerobic growth in the dark with Me 2 SO as a terminal electron acceptor, suggesting a specific involvement of AppA in photosynthesis-related processes. The impairment imposed by the absence of AppA lies, at least in part, in inefficient expression of photosystem genes (12). We have presented genetic evidence that AppA can affect photosystem gene expression through modification of the repressor PpsR, which coordinates expression of multiple photosystem genes (7). However, it is unclear at this time whether this is the only, or for that matter the main, function of the AppA protein. Because the primary structure of AppA has no obvious similarity to proteins with known function, we were unaware of what features of AppA allow it to affect PpsR activity.
Numerous attempts to overproduce and purify AppA in soluble form from E. coli have been undertaken; however, the full-length AppA protein was always present as yellow-colored inclusion bodies. The yellow pigment associated with AppA was identified as FAD. Because the primary sequence of AppA bears no similarity to known flavoproteins, we were interested in localizing the FAD binding region(s) within AppA. We constructed a number of AppA deletion derivatives, and each was overproduced in E. coli. By monitoring the coloration of these deletion derivatives, we were able to localize the minimal domain sufficient for FAD binding to approximately 120 aminoterminal residues of AppA, i.e. between residues 5 and 124. This domain is capable of noncovalent FAD binding at an apparent 1:1 molar ratio.
Flavoproteins with similar flavin binding domains are usually grouped into families. When employing the FAD binding domain of AppA, we found two homologues in the public data bases. An approximately 94-residue-long region of E. coli F403 and Synechocystis sp. Srl1694 were significantly similar to the FAD binding domain of AppA. To prove that flavin binding is not unique to AppA but common to this group of proteins, we amplified, overexpressed, and partially purified F403. As anticipated, this protein was found to noncovalently bind FAD.
In general, structural components required for binding flavins are poorly defined. However, in many flavoprotein families, a characteristic signature for dinucleotide binding is readily apparent, i.e. GXGX 2 G, where X is any amino acid (22). The sequence surrounding the conserved glycine 52 of AppA is reminiscent of this signature, i.e. ARAQLTG, with two alanine residues in place of the glycines and an extended distance between the residues (Fig. 6). The corresponding sequence of Srl1694 differs from the dinucleotide binding motif by only one residue, i.e. a conserved glycine-to-alanine substitution, AN-GITG. However, in the sequence of F403, only one glycine residue is present. The secondary structure prediction places the putative dinucleotide binding site between the ␣-helix and ␤-sheet, which would be consistent with its proposed role. Tyrosine residues are often found to interact with the isoalloxazine ring of FAD. Perhaps the best candidate for such a role would be the conserved tyrosine 21 of AppA (Fig. 6). The local sequence surrounding tyrosine 21, YRS, is conserved in AppA, Srl1694, and F403. Similar residues, albeit in a different order, RXYS, surround the conserved tyrosine in the isoalloxazine ring binding sites in a number of flavoproteins (23).
It is possible that the residues that lie outside of the minimal FAD binding domain of AppA can participate in interactions with FAD. However, as few as 94 to 120 residues of AppA may be sufficient for stable flavin binding. In light of the unorthodox structure and compactness of this FAD binding domain, it would seem fair to suggest that its structural analysis could extend our understanding of the means by which proteins bind flavins.
It is worth mentioning that the newly identified FAD binding domain must be of a relatively ancient origin. Because it is present in members of both the Proteobacteria and Cyanobacteria, it may have arisen before the divergence of these two bacterial branches, albeit horizontal gene transfer can not be ruled out. It is somewhat surprising that this domain is found only once in the genomes of both E. coli and Synechocystis sp., whereas many other known flavin binding domains occur more frequently. Furthermore, we were unable to find homologues of this domain in other bacterial, archaeal, or eukaryotic species whose complete genome sequences are presently available. Whether the rare occurrence of this domain is associated with any peculiarity of function or origin has yet to be determined.
We believe that FAD binding is the only feature common to the three proteins. In each protein, the FAD binding domain is positioned at the amino terminus. The carboxyl-terminal portion of F403, i.e. the portion downstream of residue 95, shares the most noticeable similarity with the proteins belonging to the YJCC/YEGA/YHDA/YHJK family, e.g. E. coli YJCC (SwissProtein P32701). Representatives of this protein family are abundant among various bacterial species. However, to our knowledge, the function of this family of proteins has not yet been elucidated. It will be interesting to determine whether any of the other members of this family bind flavin(s) or if this is a feature unique to E. coli F403.
We could not find similarity of the carboxyl terminus of Synechocystis sp. Srl1694 to any of the proteins in the data bases. Srl1694 is annotated, apparently after AppA, as an activator of photopigment and puc expression. Given the limits in similarity to, specifically, the FAD binding domain of AppA and the absence of a puc operon in Synechocystis sp., such annotation would seem no longer appropriate.
As far as the function of AppA is concerned, our finding that AppA contains FAD provides the first clue as to how AppA could be involved in the redox-dependent regulation of photosystem gene expression. Previously, we showed that AppA exerts its effect through the repressor PpsR, which controls expression of multiple photosystem genes (7). One can envision several possibilities as to the mode of AppA-PpsR interaction. AppA could directly interact with PpsR, thus making the DNA binding ability of PpsR dependent on the changes in redox state of AppA. This scenario would be similar to the recently described NifL-NifA system of Azotobacter vinelandii, where NifL is an FAD-binding protein that directly binds the DNA binding response regulator of nitrogen fixation genes, NifA (24). Alternatively, AppA could affect PpsR indirectly, e.g. by modulating the redox state of a possible cofactor (if any) associated with PpsR or an additional redox carrier interacting with PpsR. These possibilities will be explored in the future.
We constructed a derivative of AppA lacking the entire FAD binding domain, and this derivative was able to complement the lesion present in the AppA null mutant upon a gradual transition from aerobic to anaerobic photosynthetic growth. At this point, it is unclear why deletion of the FAD binding domain resulted in only a minor effect upon its function under these conditions. One obvious possibility is the gene dosage effect, i.e. presence of the truncated appA gene in several copies could partially mask the absence of the intact AppA. Other explanations are possible and are currently under investigation.