Pseudomonas aeruginosa Porin OprF

Using ion channel reconstitution in planar lipid bilayers, we examined the channel-forming activity of subfractions of Pseudomonas aeruginosa OprF, which was shown to exist in two different conformations: a minority single domain conformer and a majority two-domain conformer (Sugawara, E., Nestorovich, E. M., Bezrukov, S. M., and Nikaido, H. (2006) J. Biol. Chem. 281, 16220–16229). With the fraction depleted for the single domain conformer, we were unable to detect formation of any channels with well defined conductance levels. With the unfractionated OprF, we saw only rare channel formation. However, with the single domain-enriched fraction of OprF, we observed regular insertion of channels with highly reproducible conductances. Single OprF channels demonstrate rich kinetic behavior exhibiting spontaneous transitions between several subconformations that differ in ionic conductance and radius measured in polymer exclusion experiments. Although we showed that the effective radius of the most conductive conformation exceeds that of the general outer membrane porin of Escherichia coli, OmpF, we also found that a single OprF channel mainly exists in weakly conductive subconformations and switches to the fully open state for a short time only. Therefore, the low permeability of OprF reported earlier may be due to two factors: mainly to the paucity of the single domain conformer in the OprF population and secondly to the predominance of weakly conductive subconformations within the single domain conformer.

Using ion channel reconstitution in planar lipid bilayers, we examined the channel-forming activity of subfractions of Pseudomonas aeruginosa OprF, which was shown to exist in two different conformations: a minority single domain conformer and a majority two-domain conformer ( With the fraction depleted for the single domain conformer, we were unable to detect formation of any channels with well defined conductance levels. With the unfractionated OprF, we saw only rare channel formation. However, with the single domainenriched fraction of OprF, we observed regular insertion of channels with highly reproducible conductances. Single OprF channels demonstrate rich kinetic behavior exhibiting spontaneous transitions between several subconformations that differ in ionic conductance and radius measured in polymer exclusion experiments. Although we showed that the effective radius of the most conductive conformation exceeds that of the general outer membrane porin of Escherichia coli, OmpF, we also found that a single OprF channel mainly exists in weakly conductive subconformations and switches to the fully open state for a short time only. Therefore, the low permeability of OprF reported earlier may be due to two factors: mainly to the paucity of the single domain conformer in the OprF population and secondly to the predominance of weakly conductive subconformations within the single domain conformer. The channel properties of OprF, the major nonspecific porin of Pseudomonas aeruginosa, have been studied by several methods. Early studies (1,2), utilizing the near equilibrium redistribution of radiolabeled solutes initially trapped in reconstituted liposomes, have suggested that the channel was large. OprF allows nearly complete outward diffusion of polysaccharides of 2,000 -3,000 daltons in contrast to the Escherichia coli general porin channel that is permeable to sugars of only up to 600 daltons. Kinetic studies of solute diffusion (3,4), using osmotic swelling of proteoliposomes, have shown that OprF has much lower permeability (i.e. allows much slower permeation of the same test solute) than the classical trimeric porins of E. coli but forms channels that are wider than the channels of E. coli porins because the diffusion rates are much less influenced by the size of the oligosaccharide solutes. Intact E. coli cells expressing OprF porin from plasmid-coded gene are capable of growing on raffinose (5), which is too large (505 daltons) to serve as an effective carbon source for the wild-type E. coli.
A number of studies of OprF have been carried out with planar bilayer systems. The first study by Benz and Hancock (6) has already shown that the addition of OprF produces channels of large single channel conductance (several nanosiemens (nS) 2 in 1 M KCl or NaCl) that are also large by other criteria, such as indifference to the nature of the permeating ions and proportionality of channel conductance to the concentration of salt solutions used. It has also been reported that OprF channelforming activity is 100-fold lower than that of E. coli porins. However, the behavior of OprF channel is rather complex and probably sensitive to experimental details, so that quite different conclusions were reached in later studies. Thus Woodruff et al. (7) have reported that OprF produces mostly "small" channels whose conductances in 1 M KCl range from 0.1 to 1 nS with an average of 0.36 nS. The occurrence of rare larger channels was mentioned, but neither the actual current recordings nor conductance histograms illustrating this finding were given. In the most recent study by Brinkman et al. (8) the predominance of the small channels, with a broad distribution of conductances from 0.2 to 0.8 nS, has been confirmed. It has also been reported that channels with a similar conductance distribution are formed by the truncated OprF containing only its N-terminal domain.
Because most of the planar bilayer studies were performed on unfractionated OprF, which is now known to be a mixture of two completely different conformers (9), we decided to examine the properties of channels formed by OprF in more detail using preparations enriched in different conformers. We also used the poly(ethylene glycol) partition assay (10 -15) to gauge the pore size.

EXPERIMENTAL PROCEDURES
Bacterial Strains and Their Cultivation-P. aeruginosa PAO1 was used as the source of OprF. The plasmid for the expression of the N-terminal domain of OprF was constructed as follows. The portion of oprF gene coding for residues 1 through 170 of the mature OprF was cloned by PCR amplification by using a forward primer containing a PstI site and a backward primer containing a stop codon and a BamHI site. The amplicon, after restriction endonuclease treatment, was ligated between the PstI and BamHI sites of the vector pBCKS(ϩ) (Stratagene) previously modified by inserting a sequence coding for the signal sequence of E. coli OmpA protein followed by the hexahistidine tag (or the tag and the enterokinase cleavage site sequence) just in front of the PstI site (see Ref. 9). After confirming the correctness of the sequence, an EcoRI-NotI fragment was excised from the recombinant plasmid and was inserted into a medium copy number vector, pKY9790 (obtained from K. Yoshida), digested with these two enzymes. This is a 5.10-kb plasmid with the pBR322 origin, chloramphenicol selection marker, lacI gene, and a ptac promoter. The recombinant plasmid, pKY-OprFN, was transformed into E. coli host BLR (Novagen). Bacteria were grown in LB medium containing 0.5-1% glucose (and 30 g/ml chloramphenicol when needed for plasmid maintenance) with aeration by rotary shaking. * 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. 1  Preparation of the N-terminal Domain of OprF-Overnight culture (30 ml) of E. coli BLR containing pKY-OprFN at 30°C was diluted into 1 liter of fresh LB medium containing 30 g/ml chloramphenicol, and the suspension was incubated at 30°C with shaking until the A 600 reached 0.6. At this time, 0.1 mM isopropyl 1-thio-␤-D-galactopyranoside was added to initiate the expression of the truncated OprF, and the incubation was continued for a further 3 h. Crude cell envelope fraction was prepared by French pressure cell disruption, and it was then extracted with octyl ␤-D-glucoside as described for intact OprF (9). The truncated OprF, with the N-terminal hexahistidine tag, was purified by using a nickel-nitrilotriacetic acid Superflow column (Qiagen) according to the manufacturer's instructions.
The fragment lacking the hexahistidine tag was made from a plasmid containing the tag in front of the enterokinase cleavage site. The protein was purified as above and treated with enterokinase, and the flowthrough fraction from a nickel-nitrilotriacetic acid column was collected. The absence of the hexahistidine tag was further confirmed by SDS-PAGE.
Preparation of Unfractionated OprF Protein-P. aeruginosa OprF protein was purified as described in the preceding article (9) by selective detergent solubilization of the outer membrane followed by ion-exchange chromatography and gel filtration.
Isolation of Open Conformer-enriched and Open Conformer-depleted Fractions of OprF through High Resolution Size Fractionation-We utilized the tendency for the open conformers to form oligomers (especially at high concentrations) as follows. OprF samples were fractionated by gel filtration with a 1.5 ϫ 90-cm column of high resolution medium with narrow particle size distribution (Toyo Pearl HW-50F, Tosoh Biosep, Montgomeryville, PA). The buffer used contained 0.1% (w/v) dodecyl maltoside, 0.4 M NaCl, 20 mM Tris-HCl, pH 8.0, 1 mM dithiothreitol, and 3 mM NaN 3 . Fractions of 0.5 ml each were collected and analyzed for protein with A 280 or BCA protein assay reagent (Pierce) and for pore-forming activity with liposome swelling assay (described below). The tailing edge of the OprF peak had little pore-forming activity and could be used as open channel-depleted fraction. The leading edge, which had a pore-forming activity that was much higher than the bulk OprF (9), was fractionated once more by gel filtration, and again the leading edge of the OprF peak was collected. In this manner, we could obtain open channel-enriched fraction that had specific poreforming activity at least 5 times higher than the unfractionated OprF. This was used for further studies.
The specific activity of various fractions in liposome swelling with L-arabinose as the solute was 4.7, 24, and 0.7 milliabsorbance unit/g of protein/min for unfractionated, open form-enriched, and open formdepleted OprF samples, respectively. Each number represents an average of five samples.
Proteoliposome Swelling Assay of Pore-forming Activity-Pore-forming activity was assayed by the osmotic swelling rates of proteoliposomes containing different amounts of OprF (4). Proteoliposomes were made in 15% Dextran T-40 (originally obtained from Pharmacia Corp.), and isoosmotic solutions of various compounds were used to examine the effect of solute size on penetration rate (4).
Single Channel Reconstitution-"Solvent-free" bilayer lipid membranes were formed as described elsewhere (13). The film and the membrane capacitances were close to 25 picofarads each. The single channel measurements were carried out at room temperature of 23 Ϯ 2°C; the aqueous phase of the membrane chamber contained 1 M KCl and 5 mM Tris or HEPES at pH 7.42. Single channel insertion was achieved by adding 1-5 l of 30 g/ml open channel-enriched OprF stock solution (P. aeruginosa PAO1 was used as the source of OprF) in the buffer that contained 1 M KCl and 1% (v/v) octyl polyoxyethylene (Alexis Biochemicals Corp., Lausanne, Switzerland) to 1.5 ml of aqueous phase in the "cis" compartment of the chamber after a bilayer membrane had been formed. For the multichannel experiments (see Fig. 10A of the present study as well as Fig. 6 in the preceding article (9)), we used 0.6 -2 mg/ml OprF stock solution. Under otherwise equal conditions, the channelforming concentration of OprF exceeded that of OmpF (13,16,17) by a factor of 100 -500. The sign of potential is defined as positive when potential is higher at the side of protein addition.
Measurements were performed using an Axopatch 200B amplifier (Axon Instruments, Inc., Foster City, CA) in the voltage clamp mode (17). Data were filtered by a low pass eight-pole Butterworth filter (Model 9002, Frequency Devices, Inc., Haverhill, MA) at 15 kHz, recorded simultaneously by a video cassette recorder operated in a digital mode, and directly saved into the computer memory with a sampling frequency of 50 kHz. Amplitude analysis was done using software developed in house.
Estimation of Pore Size Using the Partitioning of Poly(ethylene Glycol)-The poly(ethylene glycol) partition assay (10 -15) was used. The design of the chamber permitted the change of membrane-bathing solution without damaging the membrane. In a typical experiment, a bilayer was first formed by raising solutions in both chamber compartments 2 mm above the opening in the partition. After a single OprF channel was inserted and its parameters were recorded, the bathing solution near the membrane was changed by slowly replacing the initial electrolyte with 15% (w/w) solutions of differently sized poly(ethylene glycol)s (Aldrich). The change of the bathing solution took 0.5 min. To keep the ion/water molar ratio constant, the polymers were added to 1 M KCl stock solutions. At the end of each experiment the contents of both compartments were taken to measure conductivity using a CDM 83 conductivity meter (Radiometer, Copenhagen, Denmark) to determine the actual polymer concentrations in the experimental cell.

RESULTS
The preceding study (9) has demonstrated that OprF protein exists in two distinct conformations: the majority two-domain conformer (similar to the major conformer of OmpA) with the separate N-terminal eight-stranded transmembrane domain and the C-terminal periplasmic domain, which interacts with the peptidoglycan (18), and the minority conformer that apparently folds as a larger single domain ␤-barrel.
The Channel-forming Activity of OprF N-terminal Domain in Liposome Swelling-The N-terminal domain of OprF has been produced by the cloning of the 5Ј-terminal part of the oprF gene earlier (8). However, it was not clear whether this domain actually functions as a small channel for nutrients because only planar bilayer studies were done. We therefore examined the channel-forming activity of this construct by using the osmotic swelling of proteoliposomes. The protein coded by the truncated oprF gene was preceded by a hexahistidine tag and the signal sequence of OmpA and under our conditions nearly exclusively found in the outer membrane fraction (not shown). Its signal sequence was cleaved off as the product had the correct size for the sum of the 1-170 N-terminal domain of the mature OprF and a hexahistidine tag by SDS-PAGE and by mass spectrometry.
The results of proteoliposome swelling assay ( Fig. 1) showed that this domain cannot facilitate transport of even the smallest of organic nutrients, including L-arabinose, L-alanine, and glycine. Although we also used urea and erythritol for the assay, these solutes traversed the membrane bilayer at significant rates so that an unequivocal conclusion could not be obtained for these solutes.

OprF in the Open Channel-enriched Preparations Has the Correct
Pore Size-Comparison of swelling rates with sugars of different sizes allows the estimation of pore size by the use of the Renkin approximation (19). This approach has been used for the E. coli OmpF channel (20) and has led to an estimate that is very close to the actual size of the constriction zone of the channel as defined by x-ray crystallography (21). The same approach earlier led to the estimation of 2-nm diameter for the OprF channel (3,4). However, these studies were all performed with unfractionated OprF. Now that we have been able to obtain fractions enriched for the open channel conformers, with the specific activity much higher than that of the unfractionated protein, we compared these two preparations for the channel size. The results (Fig. 2) showed that the open channel-enriched fraction had a wide diameter very similar to that of the unfractionated OprF, thus confirming that the enrichment did not result from the purification of other contaminating poreforming proteins.
Open Channel-enriched Fraction in Bilayer Reconstitution Experiments-Of the three different OprF fractions: open channel-enriched OprF sample, unfractionated OprF sample, and open channel-depleted OprF sample, we started with the channel-forming activity of the first one.
At low concentrations of about 0.02-0.1 g/ml, the open channelenriched fraction induced well defined channels displaying several conductance substates and rich kinetic behavior. A typical current recording showing consecutive insertion of two OprF channels with amplitude of about 30 -40 picosiemens (pS) under the applied voltage of Ϫ150 mV is given in Fig. 3 at 50-ms time resolution. In addition, intensive downward flickering to much higher currents was always seen. The amplitude of these flickering events measured at higher resolution (see below) was close to 1 nS. Although this recording shows definite "channel-like" behavior, its interpretation is rather complicated due to the presence of at least two levels of different conductance. One of the main questions is whether these conductance levels represent two types of OprF-induced channels as was suggested before (8) or show two different substates of the same OprF channel.
To answer this question we performed most of the measurements at the level of minimal OprF-induced conductance. The current recordings in Fig. 4 illustrate one such experiments. Typically channel insertion was manifested by a small stepwise increase in the membrane conductance with the voltage-dependent amplitude corresponding to about 6 and 8 pA (40 and 55 pS) for Ϫ150 and 150 mV, respectively. These small current steps are marked as L low levels for the uppermost (Ϫ150 mV) and lowermost (150 mV) tracks in Fig. 4.
Insertion of OprF channels into the membrane following the cis side addition of the protein was directional (with one exception in about 50 experiments). Indeed at negative voltages the low conductance (L low ) level was always accompanied by intense flickering to a highly conductive level (L high ) with an amplitude of about 140 -150 pA (0.93-1 nS) (    Our analysis suggests that the current recordings in Fig. 4 represent only one OprF channel and not a superposition of two or more different channels. Indeed the channel always appeared as a step from zero to the low conductance level and only then reached the higher one. Moreover at negative voltages and observation intervals of several seconds, we never observed channels that would exhibit only high (L high ) or only low (L low ) conductance levels. The behavior shown in Fig. 4 was typical. OprF channels spontaneously flickered between the two conductance levels, although the flickering frequency could be different from channel to channel and, for some channels, was time-dependent. In this case, flickering to the high conductance level was clustered in bursts of several seconds separated by shorter intervals of relative silence.
On closer examination it turned out that the lower conductance level (L low ) itself was also represented by two discrete conductances. In Fig.  5A we show that the noise of this level seen in Fig. 4 actually represents fast fluctuations between two distinct sublevels (marked as L low (1) and L low (2) in the inset) with a characteristic time at the submillisecond scale. The amplitude histogram of these fluctuations is given in Fig. 5B. The bold arrows in Fig. 5, A and B, indicate the difference between the average currents of the sublevels (2.5 and 7.5 pA under the indicated conditions). The relative time the channel spends in the L low (1) versus L low (2) substate was quite reproducible in independent experiments and was close to 0.5 at Ϫ150 mV applied voltage. Again analysis of the amplitudes of the levels and their probability distributions demonstrates that these events belong to a single OprF channel. Fig. 6A illustrates conductance dependence of the two major OprF channel levels (L low and L high ) on the transmembrane voltage. Because we could not observe any fast flickering events at positive voltages, data for the high conductance level are given for the negative potentials only. It is seen that OprF in the high conductance state is ohmic, i.e. shows voltage-independent conductance, whereas the conductance of the weakly conductive subconformation is a strong function of applied voltage. Close to ohmic behavior is characteristic of highly conductive ␤-barrel channels such as OmpF at high salt concentrations (e.g. see Ref. 17). On the other hand, weakly conductive ␤-barrel channels such as LamB exhibit a pronounced non-linearity: ionic current is superlinear in voltage (22,23). The reason for this non-linearity is not clear at the moment.
Importantly, channel conductance from experiment to experiment was varying within the limits typical to those in other channel reconstitution experiments. Fig. 6B shows the histograms of channel conductances in L low and L high states collected from 60 independent channels. The solid lines through the data are Gaussian. The logarithmic scale was chosen because of an ϳ25-fold difference in average conductance of these states.
Kinetic Properties of OprF Channels-The results presented above demonstrate the dynamic behavior of single OprF channels. The chan-  nel fluctuates between its high and low conductance in a voltage-dependent manner. Fig. 4 shows that the transitions to the high conductance state, abundant at negative voltages, virtually disappear at positive voltages. In addition, the weakly conductive state itself fluctuates between two substates, L low (1) and L low (2) (Fig. 5). To characterize the kinetics of these transitions quantitatively, one needs to perform an appropriate statistical analysis of the data. The most important characteristics, such as the average lifetime in a particular state or average frequency of events, can be found by many methods. Because the characteristic times of the OprF channel fluctuations are rather short, we chose single channel noise analysis (11,13,16,17,22,23).
Examples of the power spectral densities of current noise are given in Fig. 7. First, we quantified OprF fluctuations within L low level (between L low (1) and L low (2) ) by analyzing the fragments of current recordings (the lower inset) that were free from the fast flickering to level L high (curve 2). Second, we selected the fragments (the upper inset) that contained the intense flickering events to the high conductance level (curve 3). The power spectra of both processes can be satisfactorily approximated by Lorentzians (solid lines) drawn according to Equation 1, where f is frequency, S(0) is the low frequency spectral limit, and is a characteristic time of fluctuations. This kind of spectra suggests simple Markovian character of transitions between the states. Particularly the "L low (1) Ϫ L low (2) " noise spectrum (Fig. 7, curve 2) is well described by a single Lorentzian (solid line) with the characteristic time of about 0.1 ms. The "L low Ϫ L high " noise spectra (curve 3) at frequencies below 1,000 Hz also obey a Lorentzian dependence with a characteristic time of about 1 ms (solid line). The complete curve 3 is indeed a sum of these two fast and slow Lorentzians.
Results of noise measurements illustrated by Fig. 7 allowed us to calculate the kinetic parameters of OprF fluctuations as functions of voltage. The characteristic times and rate of L low Ϫ L high flickering events are shown in Fig. 8 (A and B, filled circles). It is seen that the characteristic time of OprF fluctuations between L low (1) and L low (2) levels ( Fig. 8A, open circles) stays virtually constant. Note also a double increase of the characteristic time for the second process: the L low Ϫ L high transitions (Fig. 8A, filled circles) when voltage is shifted from Ϫ30 to Ϫ200 mV. The rate of flickering (number of events per second, n) can be calculated using the following equation (16), where ⌬i is the amplitude of the flickering events. Fig. 8B shows this rate at different negative voltages. It is seen that the rate is a function of the applied voltage with a maximum at about Ϫ150 mV and a rapid decrease toward positive voltages. An important parameter, the probability of the highly conductive state, L high , is shown in Fig.  8C. It is found as a product of the characteristic time (Fig. 8A) by the number of events per second (Fig. 8B). It is seen that this probability is at most 3⅐10 Ϫ3 at Ϫ150 mV and quickly decreases as the applied voltage approaches zero. (1) and L low (2) averaged by filtering over a time interval of 0. 1 s, F) and highly conductive L high (E) substates. The fast flickering events of OprF channel between the L low and L high states were observable only at negative voltages (see Fig. 4 for illustration), so the data for positive voltages are absent. B, reproducibility of conductance measurements from channel to channel in the weakly conductive (L low , left) and highly conductive states (L high , right). Applied voltage was Ϫ150 mV. FIGURE 7. Power spectral density of noise in the current through a single OprF channel. The background spectrum (curve 1) was measured for the membrane with a single OprF channel at 0 mV. Curve 2 represents current fluctuations within L low level, that is spontaneous transitions between sublevels L low (1) and L low (2) (see Fig. 5 for illustration). Transitions between L low and L high were excluded from this analysis. Curve 3 represents spectral analysis of a "raw" current recording that included transitions between L low and L high (see Fig. 4, the uppermost current recording).
Experiments with Other Fractions-None of our experiments showed any channel activity for the open channel-depleted OprF fraction. Addition of this fraction at high concentrations exceeding 10 g/ml to the membrane-bathing solution led to membrane instability and destruction. Usually we observed intensive irreproducible "leaky membrane" conductance noise that did not display any distinct levels even at the highest resolution of 15 s.
With an unfractionated OprF sample, we were able to obtain several successful single channel insertions at OprF concentrations above 5 g/ml. Such events were rare, but the corresponding recordings looked just like the recordings taken for the open channel-enriched fraction described above. The leaky membrane conductance noise was also frequent. This hindered our ability to reliably establish the minimal channel-forming concentration of this fraction. It is reasonable to believe that the low activity of this sample was due to the low content of the open OprF fraction.
Therefore, OprF reconstitution into planar lipid bilayers qualitatively supports the main conclusions of the liposome swelling assay (Ref. 9 and Fig. 2 of the present study) about the differences in the pore-forming activity of different protein fractions. However, judging by the OprF sample concentrations necessary to obtain single channels, the planar lipid bilayer technique gives much higher ratios for these activities. The origin of the discrepancy is not clear at the moment. We speculate that it is due to the difference in conditions under which protein is inserted in the membranes in these two methods. The tendency of OprF to aggregate (9) may also play a role.
In support of the proteoliposome swelling assay results (Fig. 1), our lipid bilayer experiments with OprF N-terminal domain also showed lack of any channel activity. This is not due to the presence of the N-terminal hexahistidine tag because a preparation without the hexahistidine tag, prepared as described under "Experimental Procedures," also lacked the channel activity.
Gauging Channel Size by Poly(ethylene Glycol) Partitioning-The maximum single OprF channel conductance (ϳ1 nS) (Fig. 6) is comparable with the conductance of a single monomer (ϳ1.3 nS in 1 M KCl) in the OmpF trimer (13,16,17). This is consistent with the original concept (2-6) about the large size of the open OprF pore.
Recently we have shown (13) that the characteristic polymer cutoff size of poly(ethylene glycol) (PEG) partitioning into the pores of OmpF and ␣-hemolysin correlates nicely with the effective pore radii calculated from the high resolution x-ray structures of these channels (21,24). This finding further supports polymer partitioning as a tool for sizing channel pores in their functional states (10 -15, 25, 26). To estimate the diameter of the OprF channel, here we studied partitioning of PEGs of different molecular weights into its aqueous pore.
The main results of polymer partitioning experiments are presented in Fig. 9. With 15% (w/w) PEG of different average molecular weights, we were able to measure the change of OprF conductance in the weakly conductive (L low , filled circles) and highly conductive (L high , open circles) substates. Here we show channel conductance in the presence of polymers normalized to corresponding conductances in polymer-free solutions as a function of the polymer average molecular weight. Also shown are the data for fully open OmpF channel (squares) obtained previously (13). It is seen that large polymers (with average molecular weight of 3,400 and higher) do not change OmpF conductance appreciably (thus do not penetrate into the pore), whereas smaller polymers decrease it significantly (penetrate into the pore). PEG with the characteristic average molecular weight of ϳ1000 separates these two regimes. Polymers with the average molecular weight of 400 and smaller decrease the channel conductance almost to the same extent as the bulk solution conductivity (dotted line, corresponds to the 0.6 Ϯ 0.02 drop in bulk conductivity measured by taking solution from the cell compartments after the measurements were concluded). In the case of the highly conductive  . Conductance of a single OprF channel in L low (F) and L high (E) subconformations in the presence of differently sized PEGs normalized to its conductance in the polymer-free solution. The x axis gives the average molecular weights of PEG as specified by its manufacturer. The results are compared with the OmpF data (f) obtained earlier (13). Each point for OprF represents at least three different experiments where a single channel was first reconstituted and recorded in the polymer-free solution, which was then substituted by the polymer-containing solution. The horizontal dotted line shows the PEG effect on the bulk solution conductivity. Measurements were done at Ϫ150 mV applied voltage.
OprF substate, the total curve is somewhat shifted toward polymers with higher molecular weights. Therefore, one can expect that the effective radius of state L high is larger that that of E. coli OmpF.
The radius of the weakly conductive OprF substate L low turns out to be significantly smaller that that of the highly conductive substate. It is seen (Fig. 9, filled circles) that starting with the average molecular weight of 600, polymers do not change the L low conductance and therefore do not penetrate into the pore in this subconformation.
Protein-Protein Interaction-Most of the findings reported in this study were obtained on single OprF channels where protein concentration in the membrane-bathing solution was kept at the minimal level compatible with channel formation. At higher concentrations channels seem to interact with each other, which is manifested by substantial changes of their dynamic properties. Fig. 10 illustrates membrane activity of the open channel-enriched fraction at its different concentrations in the membrane-bathing solution. At a relatively high concentration of 0.5 g/ml, the ionic conductance induced by OprF corresponds to a multichannel system (Fig. 10A). At this concentration the stepwise changes in the current described above are difficult to resolve; large amplitude steps (arrows 2 and 3) similar to those described or mentioned by other researchers (6 -8) are seen instead. These current steps most likely represent insertion of aggregates of OprF channels.
Using more diluted OprF stock solutions (30 g/ml instead of 0.6 -2 mg/ml) we reduced protein concentration in the membrane-bathing solution by a factor of 5, which allowed us to observe reproducible discrete conductance levels. Fig. 10B shows current tracks corresponding to only several (two or three) channels in the membrane. However, even at this relatively low concentration, the channels tend to interact with each other. Although the amplitudes of transitions between different states are very close to those described above (Fig. 4), the channel dynamics are changed. Indeed, the current tracks in Fig. 10B are not superpositions of the currents shown in Figs. 3-5. Quantitative statistical analysis of these data is difficult, but their visual examination suggests that the higher conductance states are somehow stabilized by protein aggregation.
The possibility of interaction between channels, which is seen as their changed dynamics in multichannel reconstitution experiments, is indirectly supported by the tendency of OprF to form aggregates (9). Mod-ification of channel behavior in aggregates found in the present study may explain some of the existing discrepancies in the results reported by different laboratories, for example large OprF conductances reported by Benz and Hancock (6).

DISCUSSION
In our preceding study (9) we showed that the OprF outer membrane protein of P. aeruginosa occurs as a mixture of two conformers. The two-domain conformer, which corresponds to probably more than 95% of the population, was shown to have a conformation generally thought to be the conformation of OmpA, an OprF homolog. We could split the OprF protein by tobacco etch virus protease cleavage in between the two domains and show that the N-terminal domain was essentially a ␤-barrel, whereas the C-terminal domain was a hydrophilic globular protein rich in ␣-helices. The minority conformer, which produces channels allowing the diffusion of large solutes including sugars, could be enriched in different ways. First, it could be enriched by taking advantage of the tendency of this conformer to associate loosely to form an oligomeric structure. Second, it could be enriched by introduction of a cysteine residue close to the C terminus followed by the surface labeling of intact cells with a bulky biotinylation reagent. Finally it could be enriched by fractionation of proteoliposomes based on the permeability of the channel, although this method could not be used for the isolation of proteins in amounts sufficient for biochemical study. We have prepared open form-enriched preparations of OprF by the first method, and the main object of the present study was to investigate these conformers by using planar bilayer approaches.
The idea about the presence of a small fraction of open OprF conformers with presumably different folding was confirmed by our channel reconstitution experiments. With the open channel-enriched OprF fraction, we observed successful insertion of single channels with several levels of conductance (see Figs. [3][4][5]. In contrast, with the open channel-depleted fraction coming from the tailing edge of the gel filtration peak, there was never any activity observed, and with the unfractionated OprF, formation of the channels identical to those seen for the open channel-enriched fraction was observed but only at higher protein concentrations. Channel Radius-Although single channel conductance cannot be used directly to obtain the channel size in many cases (27), our results on polymer partitioning showed that in the highly conductive state the diameter of the OprF pore is significantly larger than in the weakly conductive state (Fig. 9). Under conditions discussed previously (13), the relationship between channel conductance and polymer partition coefficient p(w) can be expressed as Equation 3, where g(w) is channel conductance in the presence of polymer with the molecular weight w; g(∞) is channel conductance in the presence of large, completely excluded polymers; and parameter describes the relative amplitude of the channel conductance change between the regimes of completely excluded and completely penetrating polymers (for more details see Refs. 12 and 15).
To compare polymer partitioning into the OprF channel with partitioning into OmpF (13), we used a simple scaling law (12), with adjustable ␣ and w 0 . The first parameter characterizes the sharpness of transition between regimes of exclusion and penetration. Sharper transitions correspond to larger ␣ values. The second parame- ter is the characteristic polymer molecular weight ("cutoff size") that separates these two regimes. The continuous lines in Fig. 9 represent the results of fitting that yield w 0OmpF ϭ 1360, ␣ ϭ 1.65 for OmpF; w 0OprF ϭ 1519, ␣ ϭ 1.56 for the highly conductive substate of OprF; and w 0OprF ϭ 190, ␣ ϭ 1.17 for the weakly conductive substate of OprF, suggesting that the effective radius (13) of the OprF channel pore in its highly conductive substate is larger than that of the OmpF pore. This observation is in qualitative agreement with the results obtained in liposome swelling tests discussed earlier. However, quantitatively the difference in radii found in polymer exclusion experiments is quite moderate: the OprF radius was larger by some 7% (calculated as (w 0OprF /w 0OmpF ) 0.6 , see Ref. 13) compared with a factor of 2 in liposome swelling. The reason for this significant discrepancy is not clear. One possible explanation is that the polymer exclusion method described here measures the average radius; the liposome swelling approach may be more sensitive to fluctuations in the pore aperture, including the "genuine heterogeneity of channel sizes" discussed by Woodruff et al. (7) to reconcile their results with the findings of a previous study (6).
Other Single Channel Studies-Our observations that the open channel conformers represent only a few percent of the entire OprF population and, in single channel experiments, predominantly exist in the weakly conductive states (Ref. 9 and this study) explain clearly and sufficiently the low permeability (or the slow rates of permeation) caused by this porin. At the same time, the ability of the open OprF proteins to exist in several substates and fluctuate between them might explain differences in the single channel data reported by other groups. For example Saint et al. (28) report a low conductance of 0.08 nS in 1 M NaCl for the N-terminal fragment of P. fluorescens OprF. This value is similar to the amplitude of the fast fluctuations within the weakly conductive level L low found in our experiments. Brinkman et al. (8) report the most frequent OprF channel conductance of 0.36 nS. This is very close to the amplitude of fluctuations between the low and intermediate levels of the single channel substates found in the present study. However, both Saint et al. (28) and Brinkman et al. (8) have observed the small single channel conductance also with the N-terminal eight-stranded ␤-barrel fragment of OprF and interpreted it as arising from the passage of ions through this presumably very narrow channel. In contrast, we were not able to observe any channel activity in the open channel-depleted fraction of OprF. Recent, perhaps relevant observations are those of Zakharian and Reusch (29), who have reported that the unfractionated OmpA, an OprF homolog, produces essentially only small conductance channels of about 30 pS at 22°C but mostly large channels of 350 pS at 37°C. Although these authors interpret this transition as that from a twodomain conformer to a single domain conformer, it seems more likely that it is due to the transition between substates in the open conformer population rather than to the nearly complete refolding needed for transition between these two distinct conformers. In any case, we did not observe the diffusion of the smallest organic solutes we could test in the liposome swelling assay (e.g. glycine) through the N-terminal domain of OprF, and we suggest that the N-terminal eight-stranded ␤-barrel domain should be thought of as a completely closed channel despite the suggestions from other laboratories (8,28) as well as from a computer modeling study of its homolog OmpA (30). In conclusion, our results as well as results from other laboratories suggest strongly that the large channel is produced by the folding of the entire OprF sequence to produce a ␤-barrel of many strands, certainly more than eight and perhaps close to 16, that are found in classical porins (see Fig. 10 of Ref. 9).
Finally the open channel-enriched preparation of OprF was purified by using the tendency of this conformer to form oligomers. This tendency may explain the difference in OprF kinetic behavior at the single channel level compared with multichannel membranes as discussed under "Protein-Protein Interaction" under "Results." In fact, in our multichannel experiments we repeatedly observed discrete conductance steps of high amplitude (up to 10 nS), which, depending on their size, could be interpreted as either insertion of a single OprF channel stabilized in its high conductance state by protein-protein interaction or simultaneous insertion of OprF oligomers. Three such events are marked by arrows 1, 2, and 3 in Fig. 10A. Their amplitudes are about 0.9, 2.1, and 8 nS, correspondingly. At low protein concentrations, the aggregates were unstable and tended to dissociate into monomers (9); this was precisely the reason why "pure" preparations of open conformers could not be prepared by repeated runs of gel filtration. Because the single channel assay necessitated the use of dilute samples of OprF, with the final protein concentration in the membrane-bathing solution smaller than 0.05 g/ml, the oligomers were probably dissociated into monomers that produced reproducible single channels. It is not known yet whether the open channel conformers of OprF exist as oligomers in the intact cells of P. aeruginosa; however, the open channel-enriched OprF fraction isolated by gel filtration contained significant amounts of oligomers when analyzed by SDS-PAGE without the heat denaturation of the samples (see Fig. 5B of Ref. 9).