By continuing to browse this site you agree to us using cookies as described in About Cookies
Wiley Online Library is migrating to a new platform powered by Atypon, the leading provider of scholarly publishing platforms. The new Wiley Online Library will be migrated over the weekend of February 24 & 25 and will be live on February 26, 2018. For more information, please visit our migration page: http://www.wileyactual.com/WOLMigration/
Inducible bacteriocin production in Lactobacillus is regulated by differential expression of the pln operons and by two antagonizing response regulators, the activity of which is enhanced upon phosphorylation
Expression of the five (pln) operons involved in the bacteriocin production of Lactobacillus plantarum C11 is regulated by a so-called pheromone-based signal-transducing network, in which the peptide pheromone (PlnA) induces bacteriocin production through the action of a histidine protein kinase (PlnB) and two antagonizing response regulators (PlnC as an activator and PlnD as a negative regulator). All pln-regulated promoters contain a conserved pair of direct repeats that serve as binding sites for PlnC and PlnD. In the present work, we show that the five PlnA-responsive operons are differentially expressed with regard to both timing and strength, and that the pheromone triggers a strong autoactivating loop of the regulatory unit (plnABCD) during an early stage of induction that gradually leads to enhanced activa-tion of the other operons. The transport operon (plnGHSTUV), which is involved in the secretion of the pheromone and bacteriocins, is also expressed relatively early upon induction, but is quickly turned off soon after peak expression. Further investigation of the various promoters revealed that, although subtle differences within the promoter regions could account for the observed differential regulation, the presence of a downstream promoter-proximal se-quence in one promoter was found to cause delayed peak activity. How phosphorylation regulates the activity of the pln response regulators was also accessed by direct mutagenesis at their phosphorylation sites. It was found that the two response regulators exert activity at two different levels: a low level when they are not phosphorylated and an elevated level when they are phosphorylated. The present data demonstrate that bacteriocin production in L. plantarum C11 is a highly regulated process, in which different regulatory mechanisms are applied to fine tune the timing and strength of expression of the five pln operons.
Lactobacillus plantarum C11 regulates its bacteriocin production through such a pheromone-based regulatory mechanism ( Diep et al., 1994 ; 1995 ; 1996 ; Anderssen et al., 1998 ). The bacteriocin locus contains five operons: two of them ( plnEFI and plnJKLR ) code for bacteriocins and immunity proteins, one operon ( plnGHSTUV ) for an ABC transport system to secrete peptides containing double-glycine N-terminal leaders ( Håvarstein et al., 1995 ), one operon ( plnABCD ) for the signal-transducing pathway, and the last operon ( plnMNOP ) containing genes with unknown functions in bacteriocin synthesis. In the regulatory operon, plnA codes for the peptide pheromone PlnA, plnB for a HPK and plnCD for two RRs. All pln operons are repressed during non-producing conditions. Upon induction by PlnA, transcription of all five operons is regulated in a co-ordinated manner: they are all activated during the exponential growth phase of the bacteria and somehow downregulated before the stationary growth phase ( Diep et al., 1996 ). This co-ordinated expression suggests a common regulatory network for all five operons. This view is supported by the fact that the promoters associated with these operons share some common features: They all contain a poorly conserved −35 region and are preceded by a conserved pair of direct repeats serving as regulatory elements ( Diep et al., 1996 ; 2001 ). We have demonstrated recently that the two pln RRs bind directly on these conserved repeats as homodimers in a co-operative manner to regulate gene expression ( Risøen et al., 1998 ; 2001 ). Interestingly, although the two RR proteins are very similar at the amino acid level (over 75% similarity) and even recognize the same genetic regulatory elements, they differ from each other in function: PlnC activates transcription and bacteriocin production, whereas PlnD somehow represses both processes ( Diep et al., 1994 ; 2001 ). The exact mechanism(s) that control their activity so that they do not antagonize each other in an uncontrolled manner are still poorly understood.
In the present work, we sought to gain further insight into the different mechanisms that control gene expression during the induction process. We report that the five promoters, despite their overall similarity, are differentially regulated in both strength and timing. We also provide evidence showing that phosphorylation enhances the activity of the two pln regulators, which otherwise have some low activity when they are in their non-phosphorylated form.
The pln promoters are differentially regulated with regard to timing and strength
To study promoter activity of the five pln operons, we selected DNA fragments ranging from position −125/−126 upstream of their transcription start (referred to as +1) to the start codon of their first gene to represent the different promoters (see Table 1). The selected DNA fragments, which contain the promoter core region (−35 and −10 boxes) and the upstream regulatory repeats, are hence of about the same length with regard to the region upstream of the transcription start (+1) but differ in length in the (non-coding) region downstream of this site (see Table 1). The different promoter fragments, termed PA, PE, PG, PJ and PM, derived from the operons plnABCD, plnEFI, plnGHSTUV, plnJKLR and plnMNOP, respectively, were obtained by polymerase chain reaction (PCR) and then ligated just upstream of the promoterless cat in pGKV210, which was used as a reporter plasmid. After transformation into C11 cells, the different clones were induced with PlnA (200 ng ml−1), and CAT activity was then assayed as described in Experimental procedures.
Table 1. . PlnA-responsive promoters and their sequence.
LR RR −35 −10 TS
Left (LR) and right (RR) repeats and predicted −35 and −10 regions are underlined, whereas transcription start (TS) and the first base (A) of the start codon are indicated by bold letters. Sequences in brackets were added to include restriction sites EcoRI (GAATTC) or BamHI (GGATCC). Numbers at the end of the first five promoters indicate the length of non-coding sequence from TS to the first start codon (ATG). In the fusion promoters (PAM and PAJ), the non-coding sequences are derived from plnM and plnJ promoters (indicated by italic letters). Note that only PJ and P′J contain an extensive AT-rich sequence of 26 nucleotides (small letters) located just downstream from the right repeat (RR).
As shown in Fig. 1A, PA of the regulatory operon was strongly activated during the early stage of induction. It reached peak activity between time points 2 h and 3 h before declining gradually later in growth. The activity profiles for PG and PE somewhat resemble that for PA, albeit with some differences: their peak activity appeared to be weaker and slightly delayed and, in the hours after peak expression, their activity also declined more quickly. In contrast, the activity profiles for PM and PJ differ remarkably from the others: the first was significantly delayed, peaking about 6 h after induction, whereas the latter displayed little or no activity at all.
The five promoters examined above contain non-coding sequences located just downstream of their transcription start sites; these associated sequences are of different lengths (see Table 1). To test whether any of these elements, especially those of PM and PJ, could account for the differential regulation of their promoter activity, we designed shorter promoter fragments in which most of the downstream non-coding sequences were omitted. Thus, the new promoter fragments started at the same position as before, but all ended at position +13, giving rise to DNA fragments (termed P′E, P′G, P′J, P′M) of the same length as PA, which has a minimal non-coding sequence (see Table 1). The activity of the new promoter fragments is shown in Fig. 1B. It can be seen that the removal of the non-coding sequence from PM (resulting in P′M) gave rise to a stronger and earlier peak activity. On the other hand, the removal of the non-coding sequence from PJ (resulting in P′J) increased the activity only slightly; this result implies that the poor activity of PJ was not entirely caused by the associated non-coding sequence. For the remaining promoters, P′E and P′G, no significant difference was observed with regard to timing. However, unlike P′E, the strength of which was comparable to that of PE, the strength of P′G was significantly reduced (≈ 50%) compared with that of PG.
In view of the above results, we hypothesized that PA, which was relatively strong during the early stage of induction (Fig. 1A), would have a reduced and delayed activity when it was fused with the non-coding sequence from PM (resulting in PAM). On the other hand, the strength of PA would not be significantly reduced when it was fused with the non-coding sequence from PJ (resulting in PAJ) (see Table 1). The fusion promoters were obtained and tested. As shown in Fig. 1C, the results turned out as expected. Thus, a delayed and relatively low activity was found for PAM, whereas PAJ displayed a peak activity of a level comparable to that of PA.
Analysis of the dose dependency of PlnA concentrations
To analyse how the individual promoters respond to different inducer concentrations, we exposed the various clones described above to increasing inducer concentrations (0, 20, 200 and 1000 ng ml−1). CAT activity of cells induced for 3 h was measured, and the results are shown in Fig. 2. It can be seen that the various clones responded differently to the increasing inducer concentrations. The PA clone responded with almost full expression even at the lowest inducer concentration applied (20 ng ml−1). In contrast, the activity of PE and PG and their shorter versions P′E and P′G responded in a more marked dose-dependent manner to the increasing PlnA concentrations. On the other hand, the PJ and P′J clones responded poorly to all concentrations of PlnA applied, although the shorter version (P′J) showed a weak dose-dependent response. Similarly, the PM clone responded poorly to all inducer concentrations (after 3 h induction); however, its shorter version (P′M) showed a marked dose-dependent response to the inducer.
Remarkably, unlike the wild-type PA, which displayed a marginal dose dependency on the inducer (Fig. 2), the fusion promoter PAJ responded to PlnA in a more marked dose-dependent manner. However, the activity of PAJ was in general much lower than that of PA. For PAM, poor activity was found at all inducer concentrations. The latter result correlates well with the results from Fig. 1, which show that this fusion clone (PAM) and the wild-type PM are poorly active during the first 4 h after induction.
Taken together, these findings imply strongly that PA of the regulatory operon is the most active promoter that is turned on with almost full expression by small concentrations of the inducer. On the other hand, the promoters of the other operons become highly active only at a much higher inducer concentration, suggesting that these promoters reach their peak activity at a later stage of induction when the inducer has accumulated to higher levels.
Transcription analysis of the various pln operons in C11
Given that the five PlnA-responsive promoters are differentially regulated with regard to both timing and strength (Figs 1 and 2), we wanted to examine how these results correlate with the transcription of their respective operons in C11. RNAs obtained from induced and non-induced C11 cells were analysed with DNA probes specific to the individual pln transcripts. As shown in Fig. 3, transcription of all five operons could only be detected in induced cells, but not in non-induced cells, and the amounts of the transcripts and the timing of gene expression varied between the different operons. For the regulatory operon (plnABCD), its transcript appeared to be highly expressed at all time points investigated, with peak expression extending over a prolonged period of time, i.e. from time point 2 h to time point 5 h (Fig. 3A). For the plnMNOP operon, its transcription seemed to follow a stepwise fashion: a very low but detectable level of expression was observed during the first 2 h after induction; in the next 3 h (time points 3–5 h), the expression was increased to a significantly higher level before it diminished abruptly at time point 6 h (Fig. 3B). For the two bacteriocin operons plnJKLR and plnEFI, their expression profiles seemed to be quite alike: the amounts of their transcripts increased gradually after the addition of the inducer, and they both reached peak expression at time point 3 h before decreasing gradually during subsequent growth (Fig. 3C and D). Lastly, the fifth operon (plnGHSTUV) resembles the regulatory operon (plnABCD), in that it also reached peak expression relatively early, i.e. only 2 h after induction; however, from thereon, it differs from the others in that it was downregulated earlier and also more quickly (Fig. 3E). It is noteworthy that the transport transcript (plnGHSTUV) was expressed at very low levels throughout the course of the experiment. The relative amounts of the different transcripts with regard to their individual peak expression are shown in Fig. 3F.
To summarize, the transcription patterns of plnABCD (strong and early), plnMNOP (strong but delayed) and plnEFI (strong and intermediate in timing) seem to correlate well with the activity profile of their respective promoters (Figs 1 and 2). On the other hand, the transcription patterns of plnJKLR (relatively strong) and plnGHSTUV (poor) are not in accordance with the poor activity of PJ and the moderate activity of PG (Fig. 1). Possible mechanisms responsible for this discrepancy in gene expression are discussed below.
Analysis of non-phosphorylatable regulators
Most response regulators become activated upon phosphorylation; this modification takes place at a conserved aspartate (D) residue located within their N-terminal half (Sanders et al., 1989; Stock et al., 1989). A sequence alignment of PlnC and PlnD with some known RRs (such as CheY, NtrC, NodW and OmpR), the phosphorylation sites of which have been characterized, identified the D residue at position 59 (D59) as the potential phosphorylation site for both PlnC and PlnD (see Fig. 4A). To gain insight into how phosphorylation regulates the functionality/activity of PlnC and PlnD, we changed, by direct mutagenesis, their D59 to an asparagine (N), which is similar to D but cannot be phosphorylated. To study these recombinant genes, we took advantage of a reporter system that has been used in our recent study on the interplay between the kinase PlnB and its RRs in the heterologous host Lactobacillus sakei Lb790 (Diep et al., 2001). The reporter system consists of two plasmids: pJB-GB, which contains the reporter gene gusA and the kinase gene plnB, each being preceded by its own plnA promoter; and, pJB37-C or pJB37-D, which contains either plnC or plnD driven by a constitutive promoter P32 (see Table 3 and Fig. 4B). In such a reporter system, the heterologous cells expressing plnC respond to the exogenously added inducer with a strong GUS expression, whereas cells expressing the other regulator gene (plnD) somehow respond to the inducer with a low but detectable GUS expression despite the fact that overexpression of plnD in its homologous host L. plantarum C11 represses both bacteriocin production and transcription of the endogenous plnABCD operon (Diep et al., 2001). Some of these results can also be seen in Figs 4 and 5. When the wild-type regulator genes were replaced with their corresponding D59N mutant genes (resulting in pJB37-Cmut and pJB37-Dmut), we could no longer observe any increase in GUS expression in response to the inducer (Fig. 4C). These findings indicate that non-phosphorylated PlnC and PlnD have little or no ability to activate the reporter gene.
Table 3. . Plasmids used in this study.
. pMG36e and its derivatives were used in strain C11, whereas pMS37c and its derivatives were used in strain Lb790.
Next, we examined whether any of the mutated regulators can act as a negative regulator or an activator in bacteriocin production by overexpressing their genes in their native bacteriocin producer C11. Such an experiment has been performed previously with the wild-type genes, and that was how we established different functions of PlnC and PlnD, the first being an activator and the second acting as a negative regulator (Diep et al., 2001). The effects on bacteriocin production of the mutant genes and wild-type genes are compared in Fig. 5. It can be seen that the overexpression of plnDmut also repressed the induction of bacteriocin production because cells overexpressing this gene required a relatively high inducer concentration (50 ng ml−1 or more) to establish bacteriocin production. In comparison, the control clone, which contains only the vector (V), required much less inducer, i.e. 10 ng ml−1, to induce bacteriocin production. In fact, we have shown in other experiments that as little as 1 ng ml−1 was sufficient to induce bacteriocin production in this control clone (data not shown). However, the repression by plnDmut is clearly weaker than that exerted by the wild-type PlnD, as cells overexpressing the wild-type gene needed at least fivefold more inducer (250 ng ml−1) to establish bacteriocin production. Similarly, like that observed with the wild-type gene plnC, overexpression of the mutant plnCmut, which has poor activity in Lb790 (Fig. 4), also triggered bacteriocin production in the absence of PlnA (Fig. 5); this finding suggests that PlnCmut could have some trace activity that could have been amplified by the intact autoregulatory network in C11 (see Discussion).
Taken together, these studies strongly imply that both PlnC and PlnD in their non-phosphorylated form possess some low activities in gene regulation, but these activities are significantly enhanced upon phosphorylation by the kinase PlnB.
In the present study, we have shown that the five pln operons are differentially regulated with regard to both strength and timing. Most notably, the promoter of the regulatory operon (plnABCD) appeared as the most active during the early stage of induction, whereas the promoters of the other operons became highly active only at later stages of induction. Based on the transcription analysis (Fig. 3), the early expression of the regulatory operon seems to be closely followed by the expression of the operon plnGHSTUV, which codes for a transport system dedicated to maturation and secretion of not only the bacteriocin peptides but also the inducer (Håvarstein et al., 1995). Hence, it is reasonable to believe that the concerted expression of these two operons is required to bring about a strong autoactivating loop of the regulatory unit during the early stage of induction. In effect, this would lead to a rapid increase in the inducer concentration in the culture, so that the expression of the other operons that require high inducer concentrations could be enhanced.
The two operons plnMNOP and plnJKLR differ significantly from the others in promoter activity; the promoter (PM) of the first operon regulates a much delayed activity, whereas the latter (PJ) hardly shows any activity at all (Fig. 1). In addition, these two promoters are flanked at their 3′ ends by relatively long non-coding sequences. Further characterization of these non-coding sequences revealed that only the one associated with PM could have a strong regulatory effect, whereas the other did not (Fig. 1). One possible explanation for the poor activity of P′J or PJ is the presence of a remarkably long AT-rich stretch (26 nucleotides) located in the core region of the promoter (see Table 1). It is possible that this AT-rich element might cause reduced transcription by destabilizing or preventing transcription initiation. However, further investigation is required for a more definitive answer.
We observed that the levels of the plnGHSTUV and plnJKLR transcripts did not correlate well with the results from the promoter study (Figs 1 and 3). Although the exact mechanism responsible for this discrepancy is unclear, these results might suggest an unknown regulatory mechanism. It should be emphasized that there are important differences in the way in which these two experiments were set up. In the promoter study, the individual promoters were subcloned (in plasmids) and studied one by one. In contrast, the transcription analysis was performed to reveal transcription patterns of the individual operons in their natural genetic environment, where in cis interactions between proximal regulated promoters of the pln regulon might occur and affect gene expression. Such a regulatory mechanism, which involves interactions between proximal DNA regulatory elements belonging to the same regulatory network, has been identified in prokaryotes (exemplified by the ara regulon; Harmer et al., 2001) and even in viruses (such as the genetic switch in lambda; Ptashne, 1992). Both the regulated promoters of plnGHSTUV and plnJKLR are in fact in proximal distance from the regulated promoters of plnEFI and plnMNOP respectively (see Fig. 6) and, therefore, it is tempting to speculate that these two operons could be subject to such an in cis regulation. Another possible explanation is that the low levels of the plnGHSTUV transcript could result from an intrinsic mRNA instability. Evidence supporting this notion comes from a previous study (Diep et al., 1996), in which a detailed Northern analysis revealed that the plnGHSTUV transcript was rapidly degraded, whereas the other PlnA-induced transcripts were relatively more stable. A low expression of the ABC transport system has also been reported for the sap regulon, which regulates a similar inducible bacteriocin production in L. sakei through a pheromone-based signalling pathway (Risøen et al., 2000). As bacteriocin production is often regarded as an accessory and energy-costly process, it is therefore feasible that a limited expression of the pheromone-secreting system could tightly regulate expression of the genes involved.
The second part of this work dealt with the pho-sphorylation of the regulators. We constructed non-phosphorylatable versions of the regulator genes by direct mutagenesis to examine how phosphorylation affects their functionality. Although the mutant genes were found to be functionally active under certain conditions, they are in general less active than their wild-type counterparts. The study in the heterologous expression host clearly demonstrated that only the phosphorylatable PlnC, but not the non-phosphorylatable PlnCmut, could activate GUS expression in response to the added inducer (Fig. 4). Similarly, although PlnDmut was found to be capable of repressing the induction of bacteriocin production, the repression caused by the corresponding wild-type pro-tein was much stronger (Fig. 5). In a recent study by Risøen et al. (2000), it was demonstrated that the non-phosphorylated form of either regulator could bind specifically to DNA regulatory repeats, but the DNA-binding affinity was enhanced for both regulators when the proteins were phosphorylated. These findings together suggest that phosphorylation enhances the activities of the regulators by increasing their DNA affinity. The fact that PlnC or PlnCmut was capable of triggering bacteriocin production in C11 in the absence of the added inducer is also consistent with the idea that the non-phosphorylated PlnC possesses low activity. This is based on the following reasoning. The activity of the non-phosphorylated PlnC (or by PlnCmut) might be low but sufficient to trigger the endogenous autoregulatory network (PlnABCD) in C11, the expression of which is then instrumental in the establishment of bacteriocin production. Such a functional autoregulatory network was absent in the heterologous host Lb790 and, therefore, the expression of plnC alone (without added inducer) or plnCmut (with added inducer) was not sufficient to drive GUS expression. A schematic model summarizing the possible functions of the different components of the pln regulatory network is shown in Fig. 6. It is pertinent to mention here that a number of regulators, e.g. PhoB (Makino et al., 1988), NodW (Loh et al., 1997) and OmpR (Igo et al., 1989), have been reported to regulate genes in such a manner, i.e. low regulatory activity being converted to high activity upon phosphorylation.
In the present work, we have shown two different mechanisms applied in the pln regulon: one involving sequence differences between the promoters that lay down the basis for their differential regulation and the second involving phosphorylation of two antagonizing regulators to regulate their activities. Apparently, a combination of these two mechanisms is necessary to fine tune expression of the five pln operons in L. plantarum C11, not only during the activation of bacteriocin production but also during the downregulation process. Our study also implies that other unknown mechanisms must be involved in order to bring about a complete and dynamic regulation of the pln regulon. For instance, it is challenging to unravel the mechanism(s) that control the timing of PlnC and PlnD so that their antagonizing actions do not outdo each other in an uncontrolled manner.
Bacterial strains and growth conditions
Lactobacillus plantarum strains 965 and C11, L. sakei Lb790 (LMG collection, Norway) and their derivatives were grown in Man–Rogosa–Sharpe (MRS) medium at 30°C without shaking. Where appropriate, erythromycin and/or chloramphenicol were added at 5 µg ml −1 .
Plasmid construction and transformation
Most molecular techniques were performed as described by Sambrook et al. (1989). PCR primers and all plasmids used in this study are described in Tables 2 and 3. All primers were obtained from Gibco BRL. Primers dbd62 and dbd63 were used for PA, dbd64 and dbd65 for PM, dbd66 and dbd67 for PG, dbd68 and dbd69 for PJ, dbd70 and dbd71 for PE, dbd64 and dbd72 for P′M, dbd66 and dbd73 for P′G, dbd68 and dbd74 for P′J and dbd70 and dbd75 for P′E. For hybrid promoters, a two-step PCR protocol was used (Higuchi, 1990). In the first step, dbd62 and dbd77 were used for isolating the PA part, dbd76 and dbd65 for the non-coding sequence of PM, and dbd78 and dbd69 for the non-coding sequence of PJ; in the second step, the PA part was fused with the non-coding sequences to obtain PAM and PAJ. In all cases, genomic DNA from C11 was used as template DNA in PCR. The resulting promoter fragments, the sequences of which are shown in Table 1, were ligated into the reporter vector pGKV210 in front of the cat gene between restriction sites EcoRI and BamHI. Primers dbd60 and dbd61 were used to introduce the substitution D59N into plnC and plnD, giving rise to plnCmut and plnDmut. Correct sequence of all clones was confirmed by DNA sequencing. Transformation of Lactobacillus strains was performed as described by Aukrust et al. (1995).
Table 2. . Primers.
. Degenerate primers: S, G or C; K, G or T; Y, C or T; M, A or C; R, A or G.
Growing cell cultures (OD600 of 0.15) were induced with PlnA (200 ng ml−1). Cells were then harvested at the indicated time points after induction, and DNA-free RNA was isolated using an RNA isolation kit (Qiagen). RNA was disposed onto a cellulose membrane using a vacuum slotblotter (Bio-Rad). Briefly, RNA dilutions (100 µl) were applied into wells followed by the addition of 200 µl of 10 mM NaOH and 1 mM EDTA to each well to denature. Membranes were then washed gently with 2× SSC for 1 min before baking for 2 h at 80°C. DNA probes were obtained by PCR using primers dbd87 and dbd88 for probe J, dbd89 and dbd90 for probe A, dbd91 and dbd92 for probe E, dbd93 and dbd94 for probe G and dbd95 and dbd96 for probe M. The various primers are listed in Table 2. Each probe was of ≈ 0.5 kb and specific to the 5′ end of the detected transcript. Probes were labelled with 32P using a random labelling kit (Amersham). Hybridization was carried out as described by Church and Gilbert (1984) and Lillehaug et al. (1991).
Growing cell cultures at an OD600 of 0.15 were induced with PlnA (200 ng ml−1), and samples (10–50 ml) were then harvested at the indicated time points by centrifugation for 10 min at 4°C. Cell pellets were washed once with 8 ml of ice-cold 0.9% NaCl before cell-free extracts were obtained by disrupting cells with glass beads as described previously by van de Guchte et al. (1991). The CAT assay was based on liquid scintillation counting (LSC) of CAT reaction products, using a protocol provided by the supplier (Promega). Briefly, CAT reaction was carried out using [3H]-chloramphenicol and n-butyryl-CoA as substrates, and the resulting butylated [3H]-chloramphenicol was separated from the non-reacted substrates by xylene extraction; the extracted product was read by LSC. Samples were assayed in duplicate, which yielded standard deviations of < 10% of the value. Protein concentration was determined by the method described by Bradford (1976) using BSA (Bio-Rad) as standard.
Bacteriocin agar plate assay and induction on agar plates were performed as described by Diep et al. (1995). For GUS experiments, cells were grown and harvested, and GUS was assayed as described previously (Diep et al., 2001).
The authors wish to thank Linda Godager for excellent technical assistance. This work was supported by grants from the Norwegian Research Council (NRC).