Penicillin tolerance genes of Streptococcus pneumoniae: the ABC-type manganese permease complex Psa

DNA sequence and organization of the putative ABC transporter

The entire nucleotide sequence of the 3.141 kb putative Psa ABC transporter was determined. Analysis of the sequence revealed four ORFs (Fig. 1A) encoding an ATP-binding protein (PsaB), a hydrophobic membrane protein (PsaC), a lipoprotein (PsaA) and a protein with similarity to a periplasmic thiol peroxidase of E. coli (PsaD). The amino acid sequence of the ATP-binding protein (PsaB) included two consensus binding sites for ATP. The first nucleotide binding site (GPAGAGKST) starting at amino acid 34 of PsaB conformed to the consensus sequence (GxxGxGKS/T) in the glycine-rich loop of ATP-binding enzymes (Walker et al., 1982). In addition, a glutamine/glycine-rich motif (LSGGQFQR) was identified between amino acids 97 and 104. This region may function as a peptide linker joining different domains of the protein. A second nucleotide binding site was located between amino acids 109 and 123 (RCLVQEADYILLDEP; Ames et al., 1990; Hyde et al., 1990). The second aspartic acid residue of this motif is 100% conserved and is thought to bind MgATP (Higgins et al., 1988; Ames et al., 1990). The protein encoded by psaC is a putative transmembrane protein. This assumption was supported by sequence comparison with other described transmembrane proteins. PsaC demonstrated 81% identity with a putative transmembrane protein in S. gordonii (Kolenbrander et al., 1994), 76% identity with a transmembrane protein in S. parasanguis (Fenno et al., 1989) and 40% identity with a transmembrane protein that is part of a Mn transporter in a cyanobacterium (Bartesevich and Pakrasi, 1995). Hydropathy analysis using the method of Kyte and Doolittle (1982) revealed that the protein encoded by psaC contains six potential membrane-spanning domains (data not shown). Furthermore, at amino acid 190, a 17-amino-acid sequence (AMQSVGTILIVAMLITP) was identified. Several hydrophobic membrane proteins known to be involved in peptide transport as well as other small molecule transport functions share this motif (Dassa and Hofnung, 1985; Gilson et al., 1988; Alloing et al., 1990). The two subsequent ORFs, psaA and psaD (formerly orf3 ), have been described previously (Berry and Paton, 1996). PsaD showed 25–50% similarity with gene products downstream of psaA homologues, ssaB, fimA and scaA (Ganeshkumar et al., 1991; Andersen et al., 1993; Oligino and Fives, 1993; Berry and Paton, 1996). psaD is predicted to encode a protein of 173 amino acid residues with an amino acid sequence that is 46% identical to a periplasmic thiol peroxidase of Escherichia coli (Cha et al., 1996), which scavenges superoxide and peroxide ions.

Northern blot analysis using a 379 bp probe specific for psaA revealed a single transcript, slightly larger than 3 kb (Fig. 1B), indicating that the ORFs psaB, psaC, psaA and psaD are organized as an operon. Using a 311 bp probe specific for psaD, two transcripts were obtained. The first transcript was 3 kb, matching the size of the transcript obtained with the probe specific for psaA. The second transcript of 0.6 kb was more weakly expressed. These results suggest that transcription of the Psa ABC transporter locus was initiated from a promoter located upstream of psaB. psaD was transcribed either from its own promoter or by readthrough of the psa operon. Similar organization of ABC loci is found in S. parasanguis (Fenno et al., 1995) and suggested for S. gordonii (Kolenbrander et al., 1998). Insertional duplication mutagenesis was used to knock out all four genes of the psa locus.

Chain formation

Pneumococci characteristically grow in diplococcal units (60–70% of a population), with a minority of cells forming short chains of three to six cells (Tomasz, 1968). Compared with the parent strain R6, mutants with defects in psaC and psaD showed altered morphological features. The mutant psaC ⁻ grew in chains up to a length of 40–50 cells, whereas the mutant psaD ⁻ demonstrated chain formation of 200–300 pneumococci in stationary phase (Fig. 2). Forming small conglomerates, the mutants psaB ⁻ and psaA⁻ showed only a slightly aberrant morphology.

Chain formation is known to occur in pneumococcal cultures grown in medium in which choline is replaced by ethanolamine (Tomasz, 1968). These cells also fail to autolyse and do not undergo natural transformation. In contrast to the parent strain, the mutants psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻ did not undergo autolysis in the stationary phase of growth; the loss of culture turbidity was less than 10% over 24 h. Pneumococci undergo rapid lysis upon exposure to low concentrations of deoxycholate (DOC) (0.005% ml⁻¹), which triggers autolysin activity. Pneumococcal cultures of the mutants psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻ could not be lysed, even by 10 × higher concentrations of DOC (data not shown).

Rate of penicillin-induced killing in mutants psaB⁻, psaC⁻, psaA⁻ and  psaD⁻

Cultures of the mutants psaB ⁻, psaC ⁻, psaA⁻, psaD ⁻, Lyt-4-4 and R6 were exposed at an OD₆₂₀ of 0.2–0.3 to 10 times the MIC of penicillin. The drug effect on turbidity and viability was followed. After 1 h, R6 underwent penicillin-induced lysis, while over a period of 6 h, the psaB ⁻, psaC ⁻, psaA⁻, psaD ⁻ and Lyt-4-4 mutants demonstrated no significant indication of lysis (Fig. 3A). A substantial loss of viability was demonstrated for the parent strain R6, which was reduced in the autolysin-defective strain Lyt-4-4. These results correspond with the findings of Moreillon et al. (1990), who observed that mutation of the major autolytic amidase resulted in decreased penicillin-induced killing of pneumococci (3–4 log units per 6 h compared with 4–5 log units of killing per 6 h in R6). Although psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻ mutants also underwent loss of viability, the number of surviving cfu exceeded that of Lyt-4-4 (Fig. 3B). Thus, psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻ mutants had a greatly reduced sensitivity to the antibacterial effect of penicillin, including lytic and killing effects.

Changes in the autolytic amidase in psaB⁻, psaC⁻, psaA⁻ and psaD⁻ mutants

The reduced lytic and killing effects of penicillin in the permease mutants could be caused by changes in the expression or activity of the autolysin. The addition of exogenous recombinant autolysin to cultured Lyt-4-4 results in the reconstitution of lysis. This property formed the basis of an in vitro assay to analyse the functionality of the autolysin of the penicillin-tolerant mutants. As a positive control, a crude autolysin preparation of R6 restored autolysis of Lyt-4-4. In contrast, a crude autolysin preparation of psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻ mutants failed to render Lyt-4-4 sensitive to penicillin (Fig. 4). Furthermore, Western blot analysis of psaC ⁻, psaA⁻ and psaD ⁻ mutants showed no translational product of autolysin (Fig. 5). These results suggest that penicillin tolerance and reduction in killing in this set of mutants resulted from alteration of the expression of autolysin. Northern blot analysis of the autolysin locus would have addressed the level of dysregulation, but all attempts to lyse the bacteria to obtain mRNA failed, including treatments with DOC, exogenous autolysin, lysozyme and hot phenol extraction. The fact that the psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻ mutants failed to lyse indicated that major changes in the cell wall composition had taken place.

A link between autolysis and transformation?

Recently, a recA competence-specific transcript has been shown to include the gene for autolysin lytA (Fig. 6). The cin (rec) operon includes cinA, recA, dinF and lytA (Martin et al., 1995). Transcription of lytA, which is located downstream of dinF, can be induced by three different promoters (Mortier-Barriere et al., 1998). During competence, the transcription of the autolysin gene is strongly induced by the competence-inducible promoter Pc located upstream of cinA. In non-competent cells, lytA is transcribed from the promoter Pb, located upstream of recA, and from the promoter PI, located directly upstream of the autolysin gene lytA (Fig. 6). The location of the three promoters described allows three different transcriptional products. Non-induction of two of these promoters, Pb and PI, leads to an impact on two physiological events, transformation and autolysis (Mortier-Barriere et al., 1998). To address the link between autolysis and transformation in the permease mutants, the transformation efficiency of the mutants psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻ was determined. In this assay, strains were transformed with a selectable marker, Str^R DNA, through a complete natural competence cycle. The psaB ⁻, psaC ⁻ and psaD ⁻ mutants grown to an OD₆₂₀ of 0.1 showed a reduction in transformation of > 90% compared with the parent strain R6; however, the mutant psaA⁻ was not transformation deficient. To determine if a shift in timing of competence occurred in the mutants (Pearce et al., 1994), we tested the transformation efficiency of the set of mutants at OD₆₂₀ of 0.3, 0.5 and 0.8 (Table 1). None of the mutants showed increased transformation efficiency at these higher ODs. Although the frequency of transformation was greatly reduced in the psaB ⁻, psaC ⁻ and psaD ⁻ mutants, a few transformants were always detected, a finding distinct from the complete absence of transformation of a comA-defective mutant (Hui and Morrison, 1991). Complementation of the cultures of psaB ⁻, psaC ⁻ and psaD ⁻ mutants with the competence-stimulating peptide (CSP) did not restore transformation at any time during growth. As the RecA protein is required for transformation and the autolysin gene is co-transcribed with recA, Western blot analysis was performed for the RecA protein. The mutants psaC ⁻, psaA⁻ and psaD ⁻ showed wild-type levels of translational products of recA (Fig. 7), demonstrating that the lack of expression of autolysin was not caused by non-initiation of the promoter upstream of recA. The size of RecA was slightly different in the mutants psaC ⁻, psaA⁻ and psaD ⁻. These differences did not correlate with transformation deficiency, as parent strain R6 and mutant psaA⁻ demonstrated different sizes of RecA but transformed normally.

Table 1. . Transformation efficiencies of parent strain R6, mutants psaB ⁻, psaC ⁻, psaA⁻ and psaD ⁻. Transformation was scored by the acquisition of a streptomycin resistance marker in the presence and in the absence of synthetic CSP (5 ng μl⁻¹). Numbers indicate the percentage of transformed bacteria.

Adhesion deficiency and detection of different CBP patterns in psaC⁻, psaA⁻ and  psaD⁻ mutants

Adherence properties of the mutants were assessed using the A-549 lung cell line. Adhesion of the parent strain R6 and the mutants was expressed in total numbers (cfu/cell) as well as in percentage compared with the parent strain (100%). At a dose of 10⁷ cfu ml⁻¹, adherence of R6 was 1.62 × 10⁵ cfu per monolayer (SD 100 ± 15.5), compared with 1.64 × 10⁴ cfu (25.8% adherence; SD 100 ± 9.2) for psaC, 4.3 × 10⁴ cfu (40.2% adherence; SD 100 ± 10.5) for psaA and 6.5 × 10⁴ cfu (50.7% adherence; SD 100 ± 14.06) for psaD. Knowing that the mutagenesis of the genes psaC, psaA and psaD led to loss of expression of the CBP autolysin, the expression of other CBPs on the surface of each mutant was determined (Fig. 8A). The identity of the different CBPs in strain R6 was determined by referring to a CBP preparation consisting of eight recombinant purified CBPs (K. Gosink, manuscript in preparation). This CBP ladder served as a marker for Western blot analysis. The psaC ⁻ and psaA⁻ mutants demonstrated an almost complete loss of CBPs. The structural adhesin CbpA was absent in all psa mutants. This was confirmed by Western blot analysis using a polyclonal CbpA-specific antibody (Fig. 8B). In order to exclude the possibility that the mutations in psaC, psaA and psaD affected the production of cell wall choline to which CBPs bind, the teichoic acid was analysed by immunoblotting; mAb TEPC-15, which recognizes the phosphorylcholine, failed to detect differences in the wall teichoic acid (data not shown).

Addition of Mn to psaA⁻ and  psaD⁻mutant cells did not restore the expression of autolysin or CbpA but was required for growth

The addition of MnSO₄ has been shown to be required for the growth of the mutant psaA⁻ (Dintilhac et al., 1997). Consistent with this observation, the mutants psaA⁻ and psaD ⁻ failed to grow in chemically defined medium without MnSO₄ or MnCl₂. The addition of MnSO₄ at concentrations > 0.5 μM led to a reconstitution of growth with an optimum at 2 μM MnSO₄. As the addition of MnSO₄ at concentrations> 0.5 μM restored the growth of psaA⁻ and psaD ⁻ mutants, we investigated whether the addition of different concentrations of MnSO₄ (0.5–100 μM) was also able to restore the expression of autolysin and CbpA. Western blot analysis failed to demonstrate expression of autolysin or CbpA at any concentration of added MnSO₄. Furthermore, growth of the parent strain R6 in C + Y without Mn²⁺, Zn²⁺ or Mn²⁺/Zn²⁺ neither changed the pattern of CBP distribution nor affected expression of autolysin or CbpA. However, aggregation and growth in chains up to a length of 40 was observed in R6 grown in medium deprived of these cations. The addition of Mn²⁺, Zn²⁺ or Mn²⁺/Zn²⁺ to the mutants psaC ⁻ and psaD ⁻ reduced chain formation but failed to reconstitute transformation (data not shown).

Strains of pneumococci and growth conditions

S. pneumoniae strain R6 (Tiraby and Fox, 1973) was obtained from the Rockefeller University collection. The autolysin-deficient strain Lyt-4-4 was provided from the collection of Dr A. Tomasz, Rockefeller University. This strain is a stable point mutant created by chemical mutagenesis. S. pneumoniae was cultured on tryptic soy agar (TSA; Difco) supplemented with 3% (v/v) sheep blood. For growth in liquid culture, the bacteria were grown at 37°C without aeration in 5% CO₂ using a semi-synthetic casein hydrolysate medium supplemented with yeast extract (C + Y medium; concentration of Mn²⁺ = 0.5 μM) (Lacks and Hotchkiss, 1960). For Mn²⁺/Zn²⁺ experiments, a defined medium (C + Y without Mn²⁺/Zn²⁺) with adjustment to the concentrations of Mn²⁺ and Zn²⁺ was used. For the selection and maintenance of pneumococci containing chromosomally integrated plasmids, bacteria were grown in the presence of 1 μg ml⁻¹ erythromycin (Sigma).

Recombinant DNA methods

DNA ligations, restriction endonuclease digestions, agarose gel electophoresis and DNA amplification by polymerase chain reaction (PCR) were performed according to standard techniques (Sambrook et al., 1990). DNA purification and plasmid preparations were performed using kits from Qiagen and Promega/Wizard according to the manufacturer's instructions. Transformation of E. coli with plasmid DNA was carried out with CaCl₂-treated cells as described previously (Brown et al., 1979). Transformation of S. pneumoniae was performed according to standard protocols (Pearce et al., 1993).

Inverse PCR and DNA walking

Chromosomal DNA was prepared according to an established protocol (Pearce et al., 1993). Parts of the gene psaC were obtained by sequencing an amplification product, which was recovered using the primer psaCEcoB (5′-TATTGGGAATTCTCGACCACCAAGGCC-3′), chosen to be similar to the described putative ABC transporter in S. gordonii PK488 (Kolenbrander et al., 1994), and primer psaA2 (5′-CGTGTGGATCCTCTTTTCCTTTTTCA-3′). Amplifying sequences further upstream by this technique failed, therefore we used inverse PCR to obtain the missing ORFs. Genomic DNA of R6 was digested with Sau3A, generating fragments of about 0.8–1.0 kb in size containing the target sequence. Recircularization was performed by ligation of the DNA fragments. Amplification of DNA was carried out using primers INVpsaC1 (5′-TTCACCACGACCTCAGCAAGATTCCC-3′) and INVpsaC2 (5′-AATCAAGACCCGCTGGAATTGACC-3′) complementary to sequences located near the 5′ termini of the target DNA (Maniatis et al., 1982). A 1.6 kb fragment including psaB and the 5′ portion of psaC was obtained, gel purified and sequenced at the Center for Biotechnology at St Jude's Hospital. Both strands were sequenced independently from each other, and the redundancy was greater than 99.99%.

Insertional inactivation of psaB, psaC, psaA and psaD

To create the knockout mutants, the method of insertional duplication mutagenesis, which is a homology-directed insertion of foreign DNA, was used (Haldenwang et al., 1980; Mejean et al., 1981) (Fig. 1). For insertional duplication mutagenesis of psaA, an internal 379 bp fragment (bp 1704–2084) was amplified using total DNA of R6 and psaA1 (5′-TGCAAGAATTCTTGTAGCATGTGCTA-3′) and psaA2 as primers. The PCR product was digested with BamHI and EcoRI. An identical strategy was used to knock out the other genes. The insertional inactivation of psaB was performed using the primers psaB1 (5′-AGTTTAGGATCCGTATCGAAAACC-3′) and psaB2 (5′-AGCAAAGAATTCATCCAAGAGGATATAGTCG-3′). The resulting fragment was 490 bp long (bp 1–491). For the insertional inactivation of psaC, an integral 420 bp fragment (bp 890–1310) was amplified using the primers psaC1 (5′-TTTCAGAATTCTACGCGGGATGTCAC-3′) and psaC2 (5′-GACAGGGATCCCCATGGCTTTAGCCA-3′). The insertional inactivation of psaD was performed using the primers psaD1 (5′-TTGGAGAATTCCGACTGTTTCAGCTA-3′) and psaD2 (5′-AAGGTGGATCCATTGAAACAGTCAAT-3′). The resulting fragment was 311 bp long (bp 2710–3021). The amplified fragments were cloned in pJDC9 (Chen and Morrison, 1987). The resulting recombinant plasmids were then transformed into R6. Mutations were confirmed by Southern blot analysis (data not shown).

Penicillin susceptibility and autolysis rates

Autolysis rates of the strains were determined using 10 ml cultures of S. pneumoniae exposed to 10 times the MIC of benzylpenicillin (0.1 μg ml⁻¹) when the OD₆₂₀ reached 0.25–0.3. Autolysis rates were calculated as the first-order rate constant K = ln (A₀/A₁₂₀) × min⁻¹, where A₀ represents the peak of absorbance reading at 620 nm and A₁₂₀ the reading after a further 120 min of incubation (Liu and Tomasz, 1985). The effect of penicillin treatment on the viability was determined by exposing 10 ml cultures in the early exponential phase of growth (OD₆₂₀ = 0.3, corresponding 5 × 10⁷ cfu ml⁻¹) to 10 times the MIC of benzylpenicillin. After various times of exposure, 100 μl portions were removed, serially diluted in C + Y supplemented with 100 units of penicillinase (Sigma) and plated on TSA supplemented with 3% sheep blood (v/v).

Crude preparation of autolysin and assay for autolytic activity

Purification of pneumococcus autolysin was performed on the basis of an established protocol (Mosser and Tomasz, 1970). Late exponential phase bacteria (500 ml) were harvested by centrifugation at 5000 × g for 10 min. The bacterial pellet was resuspended in 10 ml of ice-cold 20 mM KH₂PO₄ and shock frozen in an ethanol–ice mixture. After the bacterial pellet was thawed slowly, the samples were sonicated to break the cell wall. In order to fractionate cell wall and autolysin, the samples were centrifuged at 30 000 × g for 30 min. The supernatant comprised a crude preparation of autolysin; the pellet contained most of the solid cell wall components. The amount of protein in the supernatant was determined by the Bradford test. The activity of autolysin obtained by the method described above was determined in a biological assay. The lysis-defective strain Lyt-4-4 was grown to an OD₆₂₀ of 0.25 and exposed to 10 times the MIC of penicillin and crude autolysin preparations. The turbidity readings of the cultures were followed for 8 h, incubating the strains at 37°C and 5% C0₂.

Transformation assay

The transformation efficiency of the permease mutants was assessed according to an established protocol (Pearce et al., 1994). Cultures of mutants were grown to the peak of competence at an OD₆₂₀ of 0.1 and transformed with the selectable marker Str^R DNA. Additional transformation assays were performed at OD₆₂₀ of 0.3, 0.5 and 0.8. Transformation efficiency was calculated as the percentage of Str^R transformants/total number of bacteria and compared with the parent strain, R6. Transformations were performed with and without adding exogenous CSP (5 ng μl⁻¹ final concentration).

Choline-binding protein purification

Choline-binding protein purification was performed according to the protocol of Rosenow et al. (1997). Vinylsulphone-activated agarose beads (Sigma) served as the immobilized choline affinity matrix. For this purpose, they were washed twice with 10 ml of distilled water and then rotated overnight at room temperature in 10 mM choline in a sodium carbonate buffer, pH 11.5. The beads were washed with 10 ml of phosphate-buffered saline (PBS) to remove unbound choline. The choline agarose beads were added to a 400 ml culture of pneumococci (10⁸ cfu ml⁻¹) and incubated for 30 min. The bacteria bead mixture was isolated by centrifugation at 8000 × g for 15 min, then lysed in 50 ml of C + Y medium containing Triton X-100 (0.5% w/v) in the presence of leupeptin (20 μg ml⁻¹) and phenylmethylsulphonyl fluoride (PMSF; 100 μg ml⁻¹). The lysate was centrifuged at 1000 × g for 5 min to pellet the beads. The beads were washed once with three volumes of PBS, 0.5 M NaCl to remove non-specifically bound material. Choline-bound material was eluted with three volumes of PBS adjusted to a final concentration of 10% choline (w/v). This eluate was dialysed against PBS and then passed through a Centricon 10 ultrafilter (Amicon).

Analytical and immunological methods

The CBPs and RecA were analysed by running 8% and 10% SDS–PAGE and by Western blotting using Immobilon-P membranes (Millipore Corporation). The membranes were incubated with polyclonal rabbit anti-CBP antibody (1:400) (Rosenow et al., 1997), polyclonal rabbit anti-autolysin antibody (1:1000), polyclonal rabbit anti-CbpA antibody (1:1000) and polyclonal rabbit anti-RecA antibody (1:1000). The membranes were developed using goat anti-rabbit horseradish peroxidase (ECL chemiluminescence kit; Amersham).

Northern blot

Total RNA was prepared according to the manufacturer's instructions (Qiagen). Approximately 10–20 μg of total RNA was separated in a 1.2% formaldehyde gel. The gel was rinsed in 20 × SSC buffer, and RNA was transferred to nylon membranes (Hybond N−; Amersham) by capillary blotting (Sambrook et al., 1990). A 379 bp PCR fragment generated by primers psaA1 and psaA2 was used as the psaA-specific probe. A 311 bp PCR fragment generated by primers psaD1 and psaD2 was used as the psaD-specific probe. The probes were labelled with [α-³²P]-dCTP (Amersham). Hybridization was performed under stringent conditions according to standard protocols.

Adhesion assay

The human lung carcinoma cell line A-549 has an epithelial-like morphology and resembles most type II pneumocytes. A-549 cells were grown to confluence in 24-well plates (Corning Costar) coated with collagen type 1 (Sigma). Pneumococcal cultures grown in C + Y medium to an OD₆₂₀ of 0.4 were centrifuged and resuspended in DMEM plus 1% FBS at a concentration of approximately 10⁷ cfu ml⁻¹. The concentration was confirmed by viable counting on TSA blood plates. Confluent A-549 cells were infected with 1 ml aliquots of bacterial suspension and incubated at 37°C. After an incubation time of 30 min, the cells were washed five times with PBS. To detach the cells, 100 μl of 0.25% trypsin was added to each well, and 400 μl of 0.025% Triton X-100 was used for lysis. To determine the number of adherent bacteria, serial dilutions were performed and, finally, 100 μl of bacteria suspension was plated on TSA blood plates (plus 1 μg ml⁻¹ erythromycin). All assays were performed in six parallel wells, and results are presented as means ± standard deviation.

Transmission and scanning electron microscopy

Bacteria were grown overnight in C + Y medium to an OD₆₂₀ of 0.5 and fixed in 0.1 M cacodylate buffer, pH 7.4, containing 2.5% glutaraldehyde. After this procedure, bacteria were post-fixed in 1% osmium tetroxide, dehydrated in gradients of ethanol, stained en bloc with 2% uranyl acetate and embedded in Spurr's resin. Sections were stained with 2% uranyl acetate and Reynold's lead citrate and examined under a JEOL 1200 EX II electron microscope.

Computer-assisted sequence analysis

Sequence analysis and alignments were conducted with the program DNASTAR and with the Genetics Computer Group sequence analysis software package. The computer program BLAST was used to search for amino acid sequences that were homologous to those of psaB, psaC and psaD gene products (Altschul et al., 1990).

Nucleotide sequence accession number

The DNA sequence of the 3.141 kb Psa locus has been assigned the GenBank accession number A7 055088.