Facilitated Interaction between the Pyruvate Dehydrogenase Kinase Isoform 2 and the Dihydrolipoyl Acetyltransferase*

The dihydrolipoyl acetyltransferase (E2) has an enormous impact on pyruvate dehydrogenase kinase (PDK) phosphorylation of the pyruvate dehydrogenase (E1) component by acting as a mobile binding framework and in facilitating and mediating regulation of PDK activity. Analytical ultracentrifugation (AUC) studies established that the soluble PDK2 isoform is a stable dimer. The interaction of PDK2 with the lipoyl domains of E2 (L1, L2) and the E3-binding protein (L3) were characterized by AUC. PDK2 interacted very weakly with L2 (Kd ≃ 175 μm for 2 L2/PDK2) but much tighter with dimeric glutathione S-transferase (GST)-L2 (Kd ≃ 3 μm), supporting the importance of bifunctional binding. Reduction of lipoyl groups resulted in ∼8-fold tighter binding of PDK2 to GST-L2red, which was ∼300-fold tighter than binding of 2 L2red and also much tighter than binding by GST-L1red and GST-L3red. The E2 60-mer bound ∼18 PDK2 dimers with a Kd similar to GST-L2. E2·E1 bound more PDK2 (∼27.6) than E2 with ∼2-fold tighter affinity. Lipoate reduction fostered somewhat tighter binding at more sites by E2 and severalfold tighter binding at the majority of sites on E2·E1. ATP and ADP decreased the affinity of PDK2 for E2 by 3–5-fold and adenosine 5′-(β,γ-imino)triphosphate or phosphorylation of E1 similarly reduced PDK2 binding to E2·E1. Reversible bifunctional binding to L2 with the mandatory singly held transition fits the proposed “hand-over-hand” movement of a kinase dimer to access E1 without dissociating from the complex. The gain in binding interactions upon lipoate reduction likely aids reduction-engendered stimulation of PDK2 activity; loosening of binding as a result of adenine nucleotides and phosphorylation may instigate movement of lipoyl domain-held kinase to a new E1 substrate.

The mitochondrial pyruvate dehydrogenase complex (PDC) 1 catalyzes the irreversible conversion of pyruvate to acetyl-CoA along with the reduction of NAD ϩ . Mammalian PDC has a highly organized structure in which the dihydrolipoyl acetyltransferase (E2) component plays a central role in the organization, integrated chemical reactions, and regulation of the complex (1)(2)(3)(4). The other components of the complex, required for the overall reaction, include the pyruvate dehydrogenase (E1) component, the dihydrolipoyl dehydrogenase (E3) component, and the E3-binding protein (E3BP). The PDC reaction is regulated by a phosphorylation-dephosphorylation cycle, which is carried out by dedicated kinase and phosphatase components. The desired control in different organs is achieved by the selective expression and the novel regulation of at least four pyruvate dehydrogenase kinase (PDK) isozymes and at least two pyruvate dehydrogenase phosphatase (PDP) isoforms (Refs. 3-11 and references therein). The kinases act to inactivate PDC and the phosphatases to reactivate PDC by phosphorylation and dephosphorylation of the E1 component.
The PDKs and the related kinase regulating the branchedchain ␣-keto acid dehydrogenase complex are unrelated to cytoplasmic Ser/Thr/Tyr kinases; however, the C-terminal half of these kinases share structural motifs and fold in a similar manner (12)(13)(14)(15) to the ATP-binding regions of DNA gyrase B, heat shock protein 90, and histidine kinases such as CheA (16 -18). Indeed, within the three-dimensional structure of PDK2 (14), the C-terminal domain, excluding the last 34 residues for which structure has not been resolved, are folded in a very similar manner to these ATP-binding regions as is a C-terminal domain in the branched-chain kinase (15). PDK2 was interpreted as a dimer (14), although definitive evidence was not presented for this oligomeric state in solution. Here we establish that functional PDK2 is a stable dimer in solution; the other three PDK isoforms have a greater tendency to form larger aggregates.
Specialized interactions within the PDC assemblage operate to enhance PDK activities and are required for the selective processing of specific regulatory effects (1-4, 9, 10, 19 -27). Of particular importance is the functional interplay of PDKs with E2. E2 transforms kinase function and regulation through serving as an adaptable binding framework that directly abets efficient phosphorylation, acting as a processing unit in translating and transmitting effector signals, and modifying the sensitivity to allosteric effectors that directly bind to the kinases (reviewed in Refs. 3 and 4). Thus, it is important for understanding these roles to elucidate the nature of the dynamic interactions of the kinases with E2.
E2 and the E3BP component have remarkable structures that allow them to carry out their related general roles. These components form an adjustable framework that binds the other components and integrates the sequential reactions in the assembled complex. When expressed by itself, 60 E2 subunits associate by their C-terminal domains as 20 trimers assembled in the form of a dodecahedron; each C-terminal domain of this inner core connects by a series of mobile linker regions to an E1-binding domain and then to two lipoyl domains (1)(2)(3)(4)28). The N-terminal lipoyl domain is designated L1, then the second lipoyl domain, L2. L2 is flanked by linker regions connecting to L1 on one side and the E1-binding domain on the other. The E3BP component has a similar segmented structure with globular domains connected by linker regions (1)(2)(3)(4)29). E3BP associates with the inner E2 core via its C-terminal domain, binds E3 via a binding domain, and has a single N-terminal lipoyl domain (designated L3) (29, 30 -34). Besides serving as substrates and intermediate carriers, the mobile lipoyl domains are involved in crucial interactions with the PDKs and PDP1 (3,4).
The binding of PDK2 by lipoyl domains of E2 and binding of E1 by binding domain of E2 results in much faster phosphorylation than when free PDK2 phosphorylates free E1 (9). This difference in kinase activity is increasingly amplified when samples with equivalent E1 but including or not including E2 are evaluated under increasingly more dilute conditions. 2 Additionally, PDK2 responsiveness to regulatory effectors both requires (NADH and acetyl-CoA effects) and is altered (pyruvate and ADP) by the E2-confining interactions within the complex (3,9,10,19,26). Specifically, NADH and acetyl-CoA stimulate PDK2 activity by increasing the state of reduction and acetylation of the lipoyl domains, particularly the L2 domain (9,19,26). E2 activation of PDK2 activity transforms PDK2 from being poorly inhibited by pyruvate or dichloroacetate to being markedly inhibited (9). Synergistic inhibition by allosteric binding of pyruvate or dichloroacetate results from the binding of these inhibitors by PDK2⅐ADP (and PDK2⅐ATP) but not by free PDK2, which slows down ADP dissociation. 3 To investigate the dynamic interactions of PDK2, we utilize free monomeric lipoyl domains, dimeric fusion constructs (GST-lipoyl domains), as well as E2 60-mer and E2⅐E3BP subcomplex with their lipoyl groups in an oxidized or reduced state. The L2 domain and its lipoyl group make a critical contribution to the E2-activated function of PDK2 (9). Beyond establishing that PDK2 is a stable dimer, we have used biophysical studies to evaluate the contributions of the interactions of the PDK2 dimer with the free L2 domain, GST-L2 dimer, GST bearing the L1 and L3 domains, the E2-60-mer alone and with E2 binding E1 and phosphorylated E1. We also evaluated the effects of catalytic reduction of the lipoyl domains of these structures on PDK2 binding. Our data establish that preferential binding to a lipoyl domain-bearing structure acutely depends on the oligomeric state of the lipoyl domain source and is markedly influenced by the state of oxidation/ reduction state of the lipoyl group. Reduction of lipoyl groups alters PDK2 binding in a manner that fits a gain in cofactorkinase interactions, which is apparently needed for reduction of E2 lipoyl groups enhancing PDK activity (1-4, 9, 21-27). Having E1 bound to E2 and the state of phosphorylation of E1 also influence PDK2 binding to the complex. As considered under "Discussion," our results are consistent with a mechanism proposed based on functional studies (1)(2)(3)(4), whereby the binding of PDK2 to E2 may aid continued access of PDK2 to many bound E2⅐E1.

EXPERIMENTAL PROCEDURES
Materials-Highly purified human PDK2 was prepared generally as previously described (9) but with some minor changes. 3 The maximal specific activities of all human PDK2 preparations were Ͼ1100 nmol⅐min Ϫ1 ⅐mg Ϫ1 when assayed at 30°C in 20 mM potassium phosphate buffer. 3 Purified human E2, E1, E2⅐E3BP, and mutated E2⅐E3BP were prepared as described elsewhere. 4 An E2⅐E3BP mutant was substituted for all three lysines that undergo lipoylation in L1 (Lys 46 ) and L2 (Lys 173 ) domains of E2 and L3 (Lys 44 ) of E3BP. Free lipoyl domains and GST-fused lipoyl domains were prepared as previously described (35,36). AMP-PNP was purchased from Sigma.
Analytical Ultracentrifuge (AUC) Studies-Sedimentation velocity runs were conducted at 20°C using an Optima XL-I ultracentrifuge using the 4-hole An 60Ti rotor. In sedimentation velocity studies, scans of the radial profiles of protein concentration (see standardization below) were measured using UV optics or the Rayleigh interference optical system. When only UV absorbance was measured at fixed wavelengths (most often at 280 nm), quartz windows were used. When the refractive index increment was measured with the Rayleigh interference optical system or both that signal and 280 nm absorbance were measured, sapphire windows were used. After an initial scan at 3000 rpm to check the total absorbance and detect very large aggregates, sedimentation velocity runs were conducted at the indicated speed with scans continuously collected at 1-7-min intervals for the three solventmatched samples (typically 390-l samples and 410-l matched solvent) using double sector cells in charcoal-filled Epon centerpieces or, for runs above 40,000 rpm, aluminum centerpieces. In some studies using interference optics to observe low levels of protein (particularly when a high level of NADH was present; see below), minisci were precisely matched (37) using a cell with a capillary connections for transfer of a small amount of solvent from the reference solvent side to sample side by a short time of centrifugation at 10,000 rpm. After allowing the transfer, the run was then stopped and the cell gently rotated to attain even mixing of solvent and sample throughout the sample side and then the run was restarted and completed. Absorbance base lines were established by high speed sedimentation at the end of runs for sufficient time to clear all protein from the inner third (near meniscus) region of the sample column.
The majority of sedimentation velocity studies were conducted in buffer A: 50 mM potassium phosphate, pH 7.5, containing 0.5 mM EDTA, which had a density of 1.003 g/ml and viscosity of 1.015 cp. Proteins not prepared in this buffer were introduced to this buffer either by gel filtration or by exhaustive dialysis. Some studies were conducted in buffer B: 20 mM Na-Hepes, pH 8.0, 0.15 M NaCl, 1% ethylene glycol, and 0.1% Pluronics F68. Buffer B had a calculated solvent density of 1.005 g/ml and a viscosity of 1.0343 cp; all studies with buffer B used absorbance for detection of protein movement. PDK2 was initially prepared in buffer B. The calculated partial specific volume of PDK2 is 0.728 ml/g. The partial specific volumes of proteins were calculated based on the amino acid sequence, and the solvent density and viscosity as well as frictional coefficients were calculated using UltraScan Software 4.0. 5 When high levels of adenine or pyridine nucleotides were included, the interference optics system was used to observe sedimentation boundaries. In most studies in which lipoyl groups were maintained in a reduced state, this was done by inclusion of a low level of E3 (0.1 g/cell) with 2.0 -2.5 mM NADH plus 0.1-0.25 mM NAD ϩ . Some studies pro-ducing partial reduction were conducted with lower levels of pyridine nucleotides as described under "Results." Sedimentation velocity data were analyzed by the van Holde-Weischet method (38), which analyzes the rate of sedimentation of species throughout the sedimentation boundary with the effects of diffusion removed by extrapolation to infinite time. Sedimentation velocity results were analyzed by the time derivative (dc/dt) procedure (39) and for g(s*) distribution using DCDTϩ software (version 1.16) developed by John Philo (40). 5 The dc/dt analyses provides estimates of M, s 20,w , and D for homogeneous samples; the apparent sedimentation coefficient distribution function g(s*) versus s* provides useful boundary shape for evaluating associating systems (39 -41). Weak hetero-interactions of similarly sized components (PDK2 and GST-lipoyl domains; A ϩ B^C) require further analyses to estimate binding affinities. These were simulated (39,41,42) using the SedAnal program 5 as recently revised (versions 2.49 -2.70) that was provided by Walter Stafford. Specifically data were simulated using the A ϩ B ϭ C model. Simulations were conducted using the following parameters. Protein concentrations and S values of the independent species were measured. Protein concentrations were standardized by interference optics as described in the next section. Estimates of the values of f/f o and the maximum S are described under "Results." For PDK2 binding to GST-L2 and GST-L2 red , this involved using sedimentation velocity measurements on the relatively tight complex between PDK2 and GST-L2 red . For the weak binding of L2 to PDK2, it was necessary to use the estimated S of the saturated complex for the much tighter binding of two L2 to a different kinase, PDK3. 6 Corrections were made for the effects of protein concentration decreasing the S observed the saturating ranges. We used a dissociation rate constant (termed the reverse rate constant) of 0.01s Ϫ1 , which is fast relative to the transport process. Even though functional studies 2 require fairly rapid dissociation of PDK2 from GST-L2 (Ͼ 0.2 s Ϫ1 ), we confirmed that slow reverse rate constants (Ͻ0.001 s Ϫ1 ) gave poorer fits with variation in the K d . Initially in comparing SedAnal simulation g(s*) profiles and our experimental g(s*) profiles, graphs of isobars were constructed from a series of simulations. Then, for the series of simulations around the apparent optimum, we plotted the different K d versus the root mean square error ([͚([sim-data] 2 )/n] 1/2 ϭ RMSE) and standard error (RMSE/ ͌ n) and used these to determine the confidence interval around K d . The margin of error was generally considered to be the range defined by K d that were within 2 ϫ RMSE/ ͌ n of the minima (termed standard deviation); we also considered some indications of systematic error based on the fit of the data to simulations.
In sedimentation velocity runs evaluating PDK2 component binding to the E2-60-mer or E2 complexes, the boundaries for precisely matched levels of PDK2 alone, kinase-binding component alone, and the combination were simultaneously recorded. The level of bound PDK2 was evaluated by different procedures with appropriate corrections, when present, for bound E1 and trace levels of free E1. When a significant portion of PDK2 was both bound and free, the level bound was estimated at various times from the difference in absorbance height of the plateau of the trailing boundaries for free PDK2 without and with E2-complex formation. This measurement was made following separation of the leading boundary of the large complexes from much slower sedimenting free PDK2. The height of the trailing plateau was estimated in the center of the trailing plateau of the E2 ϩ PDK2 sample using final base lines observed after clearing proteins. Bound PDK2 was also assessed from the increase in the height (absorbance or fringe) caused by bound PDK2 (fringe ratio technique). The absorbance or interference signal of the rapidly sedimenting complex with all components was compared with the separate components and the fraction bound calculated based on the increase above the trailing boundary of the signal for the rapidly sedimenting complex. To adjust for differences in the rates of sedimentation, the increase in absorbance or interference fringe heights for fast moving E2 with bound PDK2 was compared with that for E2 alone by measurements made at the same radial position (close to ⌬r ϭ 0.3 cm from center of gradient to the initial minisci). These results were consistent with the estimates of bound kinase made from the decrease in the trailing kinase with E2 or E2⅐E1. Finally, after thorough standardization using those measures of binding, the change in S with binding of a component, ⌬S increment, was also used to evaluate PDK2 binding using either detection technique. This involved, within experimental error, a constant increment in the S value per PDK2 bound to E2. That is also found for binding of at least 20 E1 tetramer (44) 7 to E2 and for binding of 12 E3 dimers to E2⅐E3BP. 7 With the binding of PDK2 to the larger E2⅐E1, there was a somewhat smaller increment and a slight decrease in this increment in the upper range (20 S change overall) with increasing mass of the complex. The largest E2 complexes still entrain Ͼ80% water within their observed diameters (44), including when E1 is bound at the level used (22 E1 tetramers per E2 60-mer). The standardization of the ⌬S increment in the absence of NADH was critical for evaluating binding in the presence of high levels of NADH when E2 or E2⅐E1 was held constant, because this was the parameter that could be most accurately measured. To maintain E2 constant, we did not use capillary cells. The lack of matching with high level of NADH present and the required use of interference optics made determination of the decrease in the trailing kinase too inaccurate. At intermediate PDK2 levels but not at high or low levels, the increase in the fringe ratio could be measured but with reduced accuracy (approximately Ϯ8% as compared with approximately Ϯ5% without NADH). We therefore applied to all the data the ⌬S increment estimated from the study when NADH was not present. For estimating K d for binding to n sites, , where L f and L b ϭ free and bound smaller component (PDK2 here); L f ϩ L b ϭ L t ; E t ϭ concentration of GST-L2 dimer or E2-60-mer, and n the number of binding sites (when PDK2 is assumed to have two binding sites for free L2 monomer E t ϭ PDK2 total and L f and L b indicate the levels of free and bound L2). For Klotz plot analyses of PDK2 binding to the E2-60-mer to determine K d and n, this was rearranged to E t /L b ϭ K d /nL f ϩ 1/n and E t /L b is plotted against 1/L f . As a test of the above results, we evaluated whether other experiments (both E2 or E2⅐E1 and PDK2 varied with and without NADH) were fit by the constants derived from the Klotz plots. The errors reported are beyond linear least square analyses of these plots, based on judicious estimates of the systematic errors in the estimated ⌬S increments (e.g. footnote a, Table II).
Sedimentation equilibrium experiments were conducted on PDK2, alone, using three charcoal-filled Epon centerpieces with 6-sector cells. Experiments were performed with three to nine protein samples and matched solvent alone. Normally 50-l samples were overlaid over 20 l of Fluorinert R FC-43 silicon oil (3 M Industrial Chemical Products Division); underlaying this inert oil allowed the protein absorbance to be read to within 0.02 mm of the outer edge of the aqueous phase. Each protein was evaluated at three or more concentrations, and equilibrium was attained at three to four centrifugation speeds. The concentration gradients were shown to be stable at a given speed (have indistinguishable boundaries) for at least 4 h before making a transition to a higher speed. The protein gradients and control solvents of equilibrated samples were scanned at least five times at 0.01 mm/min, and these measurements were averaged and solvent boundaries subtracted to obtain the equilibrium boundary of the solute protein. Sedimentation equilibrium data were evaluated with Beckman software (version 4), which was provided with the Optima XL-I ultracentrifuge. Both global analyses and average values for best individual fits were calculated and were within experimental error.
Component Concentrations and Component Ratios by AUC Studies-Purified components were prepared with constant 280 nm to 260 nm ratios. This required extensive efforts to remove contaminating proteinnucleic acid complexes from E2 and E2⅐E3BP preparations. Essentially homogenous components (PDK2, E1, L2, GST-L1, GST-L2, GST-L3, E2, E2⅐E3BP) were analyzed individually in sedimentation velocity experiments with simultaneous analysis of the moving boundary by absorbance at 280 nm and interference signal. Based on the independence with protein source of the refractive index increment (measured by the Rayleigh interference optics), the protein concentration of each purified component was determined generally as described in Beckman Application Information A-1815A. 5 Dialyzed bovine serum albumin was used as a standard based on its extinction coefficient of 0.614 A 280 for 1 mg/ml (45). For levels of protein giving absorbance values between 0.5 and 1.2 A 280 , simultaneous scans of the absorbance at 280 nm and interference scans were made; higher levels up to 18.5 mg/ml were observed by just interference optics. To have well matched base lines, capillary cells were used to overlay buffer lacking bovine serum albumin during the run. These estimates, which have been made on all components, 4 were used for calculating component ratios in all experiments. The ratios of absorbance at 280 nm to shorter wavelengths were determined by standard spectrophotometric analyses with various di-lutions of samples with blanking against the specific buffer used in dialysis of a component.
Preparation of Phosphorylated E1-E1 (2.25 mg) plus 0.14 mg of E2 were treated with 0.027 mg of PDK2 and 0.1 mM ATP and 0.5 mM MgCl 2 at 30°C in 0.65 ml of 50 mM potassium phosphate buffer (pH 7.5). Inactivation of reconstituted PDC activity was evaluated with time by removal of 5-l samples and quenching of PDK2 activity by removal of ATP by treatment with hexokinase/glucose (46). PDC activity (initially 19.3 mol⅐min Ϫ1 ⅐mg Ϫ1 ) was evaluated at least in duplicate in 200 l of standard PDC assay (20) by addition of 1-2 g of E1 (level increased with inactivation) to 5 g of E2⅐E3BP and 3 g of E3 followed by an incubation for at least 5-10 min at 4°C. Nearly complete inactivation was observed after 30 min, and then 0.56 ml of sample was treated with treated with 4 units of hexokinase and 5 mM glucose at 4°C and small molecules removed and equilibrated in buffer A by gel filtration on Sephacryl S-300 column (1 cm ϫ 23 cm). The PDC activity of this sample was evaluated after 48 h and after longer periods of storage and found to remain constant with approximately 7% of the original activity. Fig. 1A shows selected scans from a sedimentation velocity study on PDK2 in buffer B. From analysis of the full set of sedimentation boundaries for each time scan, panel B shows the integrated van Holde-Weischet analysis for the cumulative sedimentation profile (time extrapolated G(s)). This indicates that Ͼ90% of the protein is sedimenting with a single oligomeric state. Analysis of sedimentation boundaries by the DCDTϩ method also detects essentially a homogeneous species in buffer B and gave s 20,w ϭ 5.02 S, M r ϭ 90,220 g(s*) fit. When PDK2 was observed in buffer A, similar results were obtained and gave an s 20,w ϭ 5.13 S, M r ϭ 91,540, and D ϭ 5.41 ϫ 10 Ϫ7 cm 2 ⅐s Ϫ1 . The sedimentation velocity results indicate that PDK2 subunits (molecular mass ϭ 45,816) form primarily a dimer in solution. The frictional coefficient ratio (f/f o ϭ 1.44) fits an extended globular protein structure that possibly entrains some solvent (see "Discussion"). When high levels of PDK2 were incubated at room temperature for several hours (6 -16 h) in buffer A prior to sedimentation, a portion of the PDK2 (up to 30%) sedimented at ϳ7.7 S, suggesting formation of a tetramer (n 2/3 rule predicts 8.1 S). Even with relatively high PDK2 concentrations, this was reduced to very low levels in the presence of 0.15 M NaCl in buffer B or by the weakly interacting L2 domain (below). However, as indicated below, Na ϩ hindered binding of PDK2 to lipoyl domains. The PDK3 isoform has a much greater tend-ency to form higher oligomeric states, but PDK3 binds L2 domain much more tightly than PDK2 and is stabilized as dimer by L2 (3,4). 6 Efficient completion of sedimentation velocity experiments at 20°C minimized problems with formation of the larger PDK2 oligomer in buffer A for PDK2 Յ 17 M.

PDK2 Oligomeric State-
In buffer B, the conclusion that PDK2 is a dimer was supported by sedimentation equilibrium studies conducted at four levels of PDK2 equilibrated at three different centrifugation speeds. Fig. 2 shows the global fit for the three speeds for the lowest and highest concentrations. The global fit yielded M r of 92,240 Ϯ 630 Da. The average of the best individual fits of the 16 data sets yielded a M r of 91,900 Ϯ 2,700 Da. Both of these analyses are within experimental error of the molecular mass of 91,632 calculated for the PDK2 dimer from the amino acid sequence of our detagged construct. The data support PDK2 being a stable dimer.
Interaction of PDK2 with the Monomeric Form of the Inner Lipoyl Domain (L2) of E2-As indicated in the Introduction, functional studies suggested PDK2 preferentially interacts with the L2 domain of E2. In high Na ϩ -containing buffer B, sedimentation velocity analyses of PDK2 plus the free L2 domain (calculated and measured (35) molecular mass of lipoylated L2 is 12,979) and each alone detected little or no interaction between 8.2 M PDK2 and 16 M L2. L2 alone is a well behaved monomer (s 20,w ϭ 1.54 S). However, using buffer A, which lacks 100 mM Na ϩ (contains 91 mM K ϩ ), led to detectable binding at these levels (based on a decrease in trailing L2). By increasing L2 to 40 M and PDK2 to 20.4 M in buffer A and observing sedimentation at 289 nm, an S value increase of 0.17 S and a decrease in the level of trailing L2 of ϳ24% (fits K d for 2 sites of ϳ100 M) was shown. By increasing the level of L2 to 86 M, PDK2 (14.2 M) sedimented 0.29 S faster than PDK2 and the trailing free L2 was reduced by ϳ9.5% (fits K d for two sites of ϳ190 M). Fig. 3A shows average dc/dt, and Fig. 3B shows the g(s*) versus S* plots under these latter conditions for both L2 and L2 red (i.e. L2 with its lipoyl group reduced), which is considered below. Dimeric PDK3 binds two L2 with an maximum increase from 5.2 to 6.2 S. 6 PDK3 binds L2 with at least a 30-fold tighter affinity than PDK2. Thus, the small increases in S are generally consistent with the level of PDK2 binding indicated by the decrease in trailing PDK2. Using the approximation that S increases linearly with PDK2 binding and estimated decreases in trailing PDK2, the binding data suggest L2 binds to PDK2 at one site with a K d Ͼ 50 M and less tightly if binding is at two sites (Ͼ90 M). Results consistent with this conclusion are obtained by fitting with SedAnal (best fit of the data, assuming binding of two L2 per PDK2, was with a K d of ϳ175 M; experimental error required K d Ͼ70 M). This binding interaction is considerably weaker than reported by Tuganova et al. (47) for binding of two L2 to rat PDK2 (see "Discussion"). These analyses are summarized in Table I.
We would note that the g(s*) curve for PDK2 alone extends slightly beyond those in the presence of L2 or L2 red (Fig. 3, A and B); this may be caused by very low levels of PDK2 tetramer in the absence of L2. The results suggest that interaction with L2 prevents formation of low levels of larger oligomer. This effect is not nearly as marked as with PDK3, for which L2 In panels A (average dc/dt) and B (g(s*)), analyses are shown for L2 alone, PDK2 alone, L2 ϩ PDK2, and L2 red ϩ PDK2. Panel C shows the g(s*) profile for L2 red ϩ PDK2 and the fitting of this profile by SedAnal simulations using K d with the indicated values. To generate L2 red , 2.5 mM NADH plus 0.1 mM NAD ϩ and then, just prior to the run, 0.1 g of E3 were added. Other conditions are described under "Experimental Procedures" and "Results." prevents a much greater tendency to form higher oligomers. 6 Fig. 3 also shows the effect of reduction of the lipoyl group of free L2 with E3 and high NADH (cf. Fig. 3 legend and "Experimental Procedures") to produce dihydrolipoyl-L2 (L2 red ) had no or very small effect in reducing the affinity of PDK2. To avoid small concentration gradients given the high protein and high NADH and required use of interference optics, precisely matched sample and reference cell minisci were matched using the capillary cell approach described under "Experimental Procedures." Upon reduction of L2, ⌬S was estimated to be 0.37 S as a result of L2 red interacting with PDK2. This and SedAnal simulations (selected ones in Fig. 3C, K d and range estimates in Table I) gave a lower limit Ͼ70 M for binding of 2 L2 red , which is within experimental error of L2 binding to PDK2.
Interaction of PDK2 with Dimeric GST-L2 and GST-L2 red -Binding of GST-L2 dimer (calculated molecular mass of the lipoylated subunit ϭ 39,117 Da) containing oxidized lipoyl groups was evaluated at a nearly constant ratio of ϳ2.3 GST-L2 to PDK2 at four concentrations in the range of 1.93-6.4 M. Figs. 4 and 5 show g(s*) plots from these experiments. Fig. 4A shows the highest level evaluated and includes the profiles for PDK2 and GST-L2 alone; Fig. 4B also shows patterns for complex formation for the highest and lowest levels evaluated and selected SedAnal simulations. Fig. 5A includes data at intermediate levels of these components along with comparative data for GST-L2 red . Binding of these dimer structures to PDK2 was much tighter than by L2, indicating that bifunctional binding of the dimeric PDK2 by this dimeric L2 structure is greatly favored. For the simulations shown, the maximum S value and f/f o of the complex were estimated based on GST-L2 red binding to PDK2, below. Based on well demarcated minima from the RMSE analyses of differences between the data and a series of simulations for the range from 1 to 9 S, GST-L2 was estimated to form 1:1 complex with PDK2 with a K d of ϳ3.0 M. Each of the data sets at the four concentrations gave minima in RMSE analyses that were within Ϯ0.7 M of this value; the margin of error (2 ϫ standard error at minima ϭ standard deviation) gave ranges supporting K d being between 1.1 and 5.0 M. These data indicate that GST-L2, which is capable of bifunctional binding, is bound by PDK2 with Ͼ10fold tighter affinity (ϳ60-fold based on the closest best approximations of K d ) than PDK2 binds two L2 monomers.
Reduction of the lipoyl group of GST-L2 with E3 and NADH led to tighter binding to PDK2 (Fig. 5). The g(s*) profiles for binding by GST-L2 red were from sedimentation patterns of samples treated with 2.0 mM NADH and 0.1 mM NAD ϩ and were obtained using interference optics and a centerpiece with the capillary connection for precise matching of base lines. Fig.  5A directly compares the profiles for binding of 4.7 M PDK2 and 10.4 M GST-L2 and GST-L2 red and reveals a marked change in the sedimentation pattern because of the higher affinity of GST-L2 red for PDK2. Fig. 5C shows data using 1.09 M PDK2 (0.1 mg/ml) binding 2.5 M GST-L2 red (0.195 mg/ml); these levels are above but approaching the lower limit for which accurate data can be obtained using interference optics with a high level of NADH. To achieve a saturated complex, 18.9 M PDK2 was sedimented with 47.7 M GST-L2 red (Fig.  5B). An S value of 7.0 was observed, and the proportion of trailing GST-L2 was little changed from panel A and in both cases fit formation of a 1:1 complex. Assuming the observed S for essentially complete complex formation in Fig. 5B is reduced from the unimpeded S o of the complex as a result of The K d range is based on simulations that gave margins of error within 2 ϫ standard error at minima; invariably the range of K d within this margin of error was nearly the same for 2 ϫ the RMSE minima and 2 ϫ the mean absolute deviation.
b Stoichiometry assumed in data fitting and simulations. c For L2 and L2 red , these simulations gave error ranges with the lower limits shown in the table but the upper within 2 ϫ standard error encompassed no binding. Sedimentation velocity studies were conducted that overlapped the range used in Fig. 5C but that evaluated binding with lower levels of PDK2 and GST-L2 red . These studies used absorbance detection with 0.1 mM NADH, which meant that lipoyl groups were only partially reduced. With 1.14 GST-L2 per PDK2 and 0.48, 0.97, and 1.93 M PDK2, the centers of the leading peaks in g(s*) plot had positions on s* axis of 6.6, 6.95, and 7.0 S, respectively. These positions reveal rates of sedimentation that were faster, even at the lowest concentration, than that observed (bottom curve in Fig. 4B) with the highest level of PDK2 and a 2-fold higher ratio GST-L2 that had fully oxidized lipoyl groups. Therefore, partial cofactor reduction elicits a marked increase in complex formation. It seems likely that only GST-L2 dimers with both lipoyl groups reduced would bind PDK2 tightly. Although the complexity of these data with the many competing binders (oxidized, mono-, and di-reduced GST-L2) do not allow a quantitative analysis, the results firmly support tighter binding of PDK2 by GST-L2 red than GST-L2.
A poor fit in the best simulations of GST-L2⅐PDK2 complexes (Figs. 4B and 5C) is observed at the leading (high S) side of the g(s*) profiles, which suggests reversible formation of low levels of larger complexes. In contrast, the L2 domain seems to hinder PDK2 oligomer formation (above), and the interactions of the weakly binding GST-L1 and GST-L3 ( Fig. 6; see below) fit simulations without evidence for larger oligomers. Low amounts of larger complexes may be produced by the tighter binding GST-L2 as a result of reversible intermolecular bifunctional binding (i.e. two PDK2 bound by one GST-L2 and/or two GST-L2 bound by one PDK2). Table I summarizes the estimated dissociation constants of PDK2 for binding to oxidized and reduced GST-L2 and L2. Based on the ϳK d estimated, PDK2 binds the dimer roughly 50-fold tighter than it binds two L2 and reduction of the lipoyl group of GST-L2 produces almost a 10-fold stronger interaction. Even if there is tighter binding by L2 red at a single site on PDK2, an ϳ75-fold increase in the binding affinity by GST-L2 red is indicated by the best estimates of dissociation constants. However, the substantial experimental error in measuring the weak binding of the L2 red monomer permits a range from Ͻ30 to Ͼ3000 M. Independent of such error considerations, our data support bifunctional binding and reduction of the lipoyl groups of GST-L2 substantially strengthening binding to PDK2.
Interaction of PDK2 with GST-L1 and GST-L3, and a GST-L2 Structure with a Minimally Sized Connecting Region between GST and L2-Binding by monomeric L1 and L3 was not readily detected even when the lipoyl groups were reduced. Fig. 6 shows sedimentation patterns and SedAnal fits using GST-L1 red (panel A) and GST-L3 red (panel B). For GST-L1 (subunit molecular mass of the lipoylated subunit ϭ 37,382) and GST-L3 (subunit molecular mass of the lipoylated subunit ϭ 36,707), an S 0 of 7.15 S for their complexes with PDK2 was estimate by repeated simulations to obtain the best fit. The estimated binding affinities of PDK2 by GST-L1 red (K d Ӎ 22 M) and GST-L3 red (K d Ӎ 35 M) were not only much weaker than for GST-L2 red but also weaker than the binding by GST-L2 with oxidized lipoyl group. Therefore, preferential binding of PDK2 to the L2 domain is supported. Weak but detectable binding was observed with GST-L1 with an oxidized lipoyl group. This binding was roughly in the range of the L2 red monomer and involved significantly weaker PDK2 binding than that by GST-L1 red . In contrast to the above simulations of PDK2 binding by GST-L2 structures, faster sedimenting material than predicted by the optimal fits was not observed in the g(s*) profiles with GST-L1 red and GST-L3 red (Fig. 6). Table I includes the estimated K d values and deviation ranges estimated based on comparisons of SedAnal simulations to these profiles.
Because the regions that connect the lipoyl domain to GST in the GST-L1 and GST-L3 structures were shorter than the segments connecting to the GST-L2(120 -233) structure used above, we also tested a GST-L2(128 -229) structure that lacked 8 amino acids (120 -127) in the linker region preceding the L2 domain. Furthermore, this structure had 2 fewer amino acid residues in the preceding region that completes the connection (includes a thrombin cleavage) than in the GST-held L1, L2(120 -233), and L3. The GST-L2(128 -229) with both an oxidized and a reduced lipoyl group gave, within experimental error, equivalent binding to PDK2 as the more extended GST-L2(120 -233) used above. Thus, the reduced binding by GST-L1 and GST-L3 was not a consequence of different linker lengths.
Interaction of PDK2 with the E2 60-mer and Wild Type and Nonlipoylated E2⅐E3BP Subcomplex and Effects of Adenine Nucleotides and Lipoyl Group Reduction or Removal-Binding of PDK2 to 3,600,000-Da E2 60-mer (subunit molecular mass 59,996, including two lipoyl groups) was estimated with simultaneous data collection using interference and absorbance optics. With 0.12 M E2 60-mer, using a level of components 3-fold higher than that required for maximally stimulating PDK2 activity (40 nM complex), only a portion of PDK2 (added at 5-30 dimers/complex) was bound to E2 60-mers. With 20 -120 PDK2 dimers per E260-mer (134 nM), Յ12 PDK2 dimers were bound per E2 60-mer in sedimentation velocity runs conducted in which the total Na ϩ ion increased with increasing addition of PDK2 (stored in buffer B, which contains 150 mM NaCl). These studies gave the impression that there was a maximum of 12 PDK2 binding sites per E2 60-mer. When the binding was evaluated in sedimentation velocity experiments with samples equilibrated in potassium phosphate (buffer A) and the concentration of PDK2 dimer varied from 0.64 to 13 M (6 -122 PDK2 per E2 60-mer) with constant, 106 nM E2 60-mer, more PDK2 were bound and with a tighter affinity. Fig. 7 shows a typical sedimentation study and indicates how the decrease in trailing PDK2 was estimated. The sloped plateaus where E2 is present are the result of low levels of E2 aggregate (44). A Klotz plot analysis of the results from a set of sedimentation runs is shown in Fig. 8A. The analysis indicates that PDK2 was bound at up to 17 Ϯ 2.5 sites per E2 60-mer with an affinity of 3.6 Ϯ 0.6 M (for error basis, see Table II, footnote a). Consistent with a minimal impact on the frictional coefficient, our sedimentation studies correlating the incremental increase in the amount of PDK2 bound (based on the level of the trailing, unbound PDK2 as well as the increase in protein over E2 in the rapidly sedimenting E2⅐PDK2 complex) with the incremental increase in S as a result of PDK2 binding to E2 gave essentially a constant value for the entire range of PDK2 levels (see Fig. 8 legend). The increment with E2⅐E1 binding of PDK2 (below) was estimated to be somewhat smaller and deceased in the upper range to a small extent.
The native E2⅐E3BP subcomplex bound PDK2 with a similar affinity to E2 and the extent of binding at the levels tested. Although this series of data were fit well by the constants derived from studies on E2, this set of sedimentation velocity studies involved varying both the E2⅐E3BP and PDK2 levels, so that an independent extrapolated estimate of the maximum number of binding sites was not derived. PDK2 binding was evaluated with an E2⅐E3BP construct in which all three sites of lipoylation were mutated to express alanines (K46A/K173A-E2, K44A-E3BP). With 60 PDK2 dimers per E2⅐E3BP (134 nM), no PDK2 binding was detected with this nonlipoylated subcomplex. This confirms the critical contribution of the lipoyl group for PDK2 binding. Table II summarizes the estimates of binding constants and number of binding sites for these E2 structures.
In the presence of high NADH, the same ⌬S increment was used as the primary means for evaluating PDK2 binding to E2 red . Based on the Klotz analysis (Fig. 8B), enzymatic reduction of the lipoyl groups led to PDK2 binding at 22 Ϯ 1.5 sites per E2 60-mer with a binding dissociation constant to 1.6 Ϯ 0.4 M. Although neither parameter is greatly changed, the combination of ϳ2-fold tighter binding and a modest gain in the number of binding sites (17)(18)(19)(20)(21)(22) following reduction of lipoyl groups supports ϳ2-fold higher portions of PDK2 being bound with Յ0.5 M PDK2 (46 g/ml) and Յ0.05 M E2 (180 g/ml). The apparent 5-fold weaker affinity of E2 red than GST-L2 red for PDK2 would seem to indicate either that there is less access to the L2 domains of E2 (not supported below) or that competition by the weaker binding L1 domain of E2 contributes to this difference.
To more fully assess the relative binding by GST-L2 red and E2 red , the capacity of GST-L2 red to competitively reduce the increase in the sedimentation of E2 red upon binding PDK2 was FIG. 7. Decrease in trailing PDK2 caused by binding by E2. A set of absorbance scans are shown after E2, PDK2, or the combination (59.9 PDK2 per E2 60-mer, 0.112 M) had undergone sedimentation at 25,000 rpm for 50 min. Appropriate base-line corrections were made after the components were cleared from the outer radius, which included sedimentation for 2 h at 25,000 rpm and then 1.5 h at 40,000 rpm. evaluated using 12 PDK2 per E2 60-mer (0.107 M; 12.84 M lipoyl groups). In most of these experiments, much lower initial levels of NADH (100 M) and NAD ϩ (10 M) were used so that there was only ϳ45-65% of lipoyl groups (total: 16 -29 M) were reduced. These conditions were selected so that UV optics (rather than interference) could be used to monitor sedimentation with relatively low protein concentrations. Competitive binding of PDK2 by 1.64 M GST-L2 (ϳ0.7 M with 2 L2 red ) gave Ն50% decrease, and 8.2 M (ϳ3.4 M with 2 L2 red ) caused a 78% fall-off in the incremental increase in the sedimentation of E2. The binding of PDK2 by E2, alone, was fit by the K d values in Table II and n ϭ 15 sites (reduced from 22 because of partial reduction). The observed competitive reduction in PDK2 binding to E2 by 1.64 M GST-L2 (from 0.48 to 0.22) could be fit using these K d values and the same number sites on E2, and assuming that only GST-L2 with both of its lipoyl groups reduced could bind PDK2. These assumptions predicted a greater reduction in binding of PDK2 to E2 with the higher GST-L2 level (8.6 M); the predicted fraction of PDK2 bound to E2 was 0.075, and the observed fraction was 0.105. Therefore,  Fig. 6) as described under "Experimental Procedures." The five most accurate estimates of free PDK2 (PDK2 f ) (i.e. estimated where differences were large enough to maintain an error less than Ϯ10%) were used to estimate the average increases in s 20,w for complex beyond E2 alone per amount of bound kinase (⌬s 20,w /PDK2 b ). This gave a constant value of 1.1 Ϯ 0.06 S per PDK2 bound, with the variation not having a trend as more or less PDK2 were bound. This value was then divided into the accurately estimated ⌬s 20,w for all experimental conditions to obtain an estimate of PDK2 bound under all conditions. The increase in the height of the leading fringe (or 280 nm plateau) above trailing fringe for E2⅐PDK2 complex compared with E2 was also used to estimate PDK2 bound (fringe ratio) by the steps described under "Experimental Procedures." The reciprocal of levels of bound PDK2 (PDK2 b ) to total E2 60-mer were then plotted against the reciprocal of the level of free PDK2 (PDK2 f ). Using the same estimate of increase in ⌬s 20,w per PDK2 bound, levels of bound PDK2 were estimated from the change in S for the series of sedimentation velocity experiments analyzed in panel B. In these studies the lipoyl groups of the E2 60-mer (0.109 M) were reduced by 0.1 g of E3 added just before the sample was introduced into the cell to solution, which contained 2.5 mM NADH and 0.25 mM NAD ϩ . Sedimentation for panel B experiments was monitored with interference optics only because of the high level of NADH. In this set of experiments, capillary cells were not used to match the reference and sample solution (so fixed level of E2 60-mer did not vary). The two fringe ratio data points shown were from runs in which the increase in the leading fringe is substantial and the base lines formed by the trailing fringes were well delineated.

TABLE II
Equilibrium binding dissociation constants and number of binding sites for binding of PDK2 to E2 structures Binding was analyzed by the procedures and using the equations described under "Experimental Procedures." Binding constants, the number of binding sites, and deviations for E2, E2 red , and E2⅐E1 were determined from the Klotz plots (e.g. Figs. 8 and 9) of data from sedimentation velocity data using 5-8 PDK2 levels at a fixed E2 or E2 complex level. The basis for other estimates are described in Error shown is 1.5 ϫ error indicated from linear least square fit of Fig. 8 data. The use of 1.5 is based on evaluating the error consequences due points based ⌬S increment of 1.1 S per PDK2 bound for E2 and E2 red . The derived increment is judged to not have an error Ͼ5%. For the analysis E2-E1 data (subset of points shown in Fig. 9), the error from the least square analysis was multiplied by 2 because of greater uncertainty in the sliding ⌬S increment.
b The lipoylated lysines of Lys-46 and Lys-173 of E2 and Lys-44 of E3BP were all mutated to alanines in this construct as described in detail along with other mutant E2-E3BP elsewhere (see Footnote 4 in text). The equilibrium binding constant is a lower limit reflecting experimental error; no binding was in fact detected.
c These values are based on the segmental line fit and the independent site fit, which, for two classes of sites, should constitute limiting fits. The smaller number of weak binding sites goes with the larger number of tight binding sites and vice versa so that total binding sites is between 28  e The number of sites is assumed to be 22 rather than 27.6 based on the inclusion of 18 rather than 22 E1 per E2 60-mer in this experiment; with this assumption, the control experiment gave a K d ϭ 0.95 M (rounded to nearest 0.05). If the number of sites is increased to 27.6, the estimated K d derived from the control and ϩ AMP-PNP experiments increase to 1.4 M and 8.2 M, respectively. binding was generally in the range expected by this simple analysis that did not account for competitive binding by L2 with oxidized lipoyl groups. For calculated estimates of PDK2 binding to E2 to approximate the observed data, corrections were required for variation in the reduction of GST-L2 lipoyl groups but not for the decrease in the portion of the reduced lipoyl groups of E2 when GST-L2 served as an alternate substrate. This retention of binding capacity by E2 fits a capacity of a multiple L2 in a regional area of the surface of E2 being able to pair up in bifunctional binding of PDK2. When high NADH (2 mM) was used, the capacity for GST-L2 red to competitively reduce binding of PDK2 to E2 was increased based on a larger decrease in S.
Using the interference mode, the effects of 0.5 mM ATP and ADP on PDK2 binding to E2 were evaluated by measuring the change in S. With 0.12 M E2 60-mer and 48 or 28.4 PDK2 dimers per E2, ϳ12.5 and ϳ8.7 PDK2 were estimated to be bound to E2; these results fit a K d of ϳ2.5 M assuming 18 binding sites. With this level of E2 and 28.4 PDK2 per E2, the number of PDK2 bound per core was reduced to ϳ5.5 by 0.5 mM ATP and to ϳ3.7 by 0.5 mM ADP. Assuming that there was no change in the maximum number of binding sites, this indicates that adenine nucleotides caused an ϳ2-3-fold decrease in binding affinity; ADP elicited a larger change corresponding to an increase in this apparent K d from ϳ3.0 to ϳ11 M, whereas the increase was only to Ϫ7 M with ATP (Table II). The ATP analog, AMP-PNP, gave results that were almost identical to ATP. Thus, the affinity is not enhanced but decreased by inclusion of adenine nucleotides at least in the absence of the E1 substrate.
Binding of PDK2 to E2⅐E1 and E2⅐Phosphorylated E1-PDK2 binding to E2⅐E1 was evaluated in sedimentation velocity studies that compared E2⅐E1 and free PDK2 to the combination. Estimating the decrease in PDK2 trailing the E2⅐E1⅐PDK2 complex required a correction for very low levels of trailing E1. Data estimating trailing PDK2 and increases in the fringe ratio fit a single class of sites when E2 had an oxidized lipoyl group; however, a small decrease in the ⌬S increment was needed as described in the legend to Fig. 9 and below for the changes in S fitting a single class of sites. Analysis of these sedimentation velocity studies indicated there were approximately 27.6 binding sites with K d of ϳ1.9 M (Table II, representative data in Fig. 9). Estimated binding of PDK2 in other experiments in which the level of E2⅐E1 and PDK2 were varied were approximated well using these constants. In the above study, 22 E1 per E2 60-mer were used; inclusion of 30 E1 per E2 60-mer did not increase PDK2 binding but did increased somewhat the level of trailing E1. All these analyses assume that binding of PDK2 to E2 does not dislodge E1; results from pelleting of complex (ultraclear tubes placed in custom holder SW 28 rotor (Ref. 31)) and measuring activities in the supernatant are consistent with that assumption (data not shown).
The combination of the presence of E1 and reduction of the lipoyl groups in the complex resulted in even tighter binding for most of the bound PDK2 (Fig. 9, Table II). Using parameters that gave a linear fit with E2⅐E1, the analysis of binding to E2 red ⅐E1 did not yield binding estimates that fit a single affinity well. A closer fit resulted from use of two classes of sites particularly in fitting the estimated binding at the extreme concentrations. Critical to the estimated binding was the use of ⌬S increment that was estimated to decrease from 0.9 to 0.8 ⌬S per PDK2 bound for the transition from few to large number PDK2 bound. Fig. 9 shows a segmental fit using an overlapping data set; the extrapolated lines give estimates of ϳ21 sites with a K d of ϳ0.22 M and roughly 10 sites (estimated from the difference of 32 Ϯ 3 total sites and 21.5 Ϯ 3 higher affinity sites) with K d of ϳ0.63 M. Assuming two classes of independ-ent sites, the data are fit well by many combinations of K d and n. The best fit was K d ϭ 0.285 M and n ϭ 18 for tight sites and K d ϭ 1.3 M and n ϭ 16 for the weak sites (Fig. 9, dots). The fit by other combinations and ones at the outer range for experimental error are described in footnote c to Table II. If there are two classes of sites, it seems more likely that the weaker class of sites is propagated as a result of crowding after many PDK2 are bound to E2 red ⅐E1 rather than being independently present. There is insufficient information to support or to fit such a complex model; the segmental line analysis approximates a sequential model in which there is a sharp transition from tighter to weaker sites. Despite the limitation of the analyses, these AUC studies clearly demonstrate that, for a major portion of PDK2 binding sites, the combination of lipoate reduction and inclusion of E1 elicits severalfold tighter binding in comparison to either of these conditions alone. Even the weaker set of sites are estimated to involve somewhat tighter binding by E2 red -E1 than E2⅐E1, although the error ranges overlap (Table II).
Binding to E2⅐E1 was compared using 0.092 M E2 60-mer, 18 E1, and 20 PDK2 per core with or without inclusion of AMP-PNP, a very effective inhibitor of PDK2 (see "Discussion"). This combination of E1 and AMP-PNP was selected to mimic the situation in which PDK2 interacts with the full complement of substrates. Binding was reduced Ͼ2-fold from that by E2 containing the same level of E1. The data fit an increase in apparent binding constant from ϳ0.95 to 6.3 M with AMP-PNP included. Those estimates assumed that n ϭ 22; the apparent K d values increase somewhat (footnote e, Table II) using n ϭ 27.6 as was estimated from the larger data set in the study (Table II) in which somewhat more E1 (22 E1 per E2 60-mer) was included.
A plausible mechanisms for aiding dissociation or intracore movement to sustain PDK2 activity is that dissociation is fostered by a decrease in affinity following phosphorylation of E1. FIG. 9. Klotz plot analysis of binding of PDK2 to E2 red ⅐E1 and comparison to binding by E2⅐E1. This figure shows the full set of data for two studies involving binding to 79.6 or to 76.6 nM E2 red ⅐E1 with 10 levels of PDK2 ranging from 5.3 to 81.8 PDK2 per E2 60-mer. As a reference for the marked change, three points are shown for a similar study conducted for PDK2 binding to E2⅐E1 with oxidized lipoyl groups. Data were analyzed by the means described in Fig. 8 legend, except that with these larger and faster sedimenting complexes (22 E1 tetramers per E2 60-mer), there was a smaller ⌬s 20,w increment per PDK2 bound; this average increment was evaluated (see "Results") to decrease above 15.8 PDK2 per E2 60-mer added from 0.9 to 0.8 for the highest level of PDK2 added. The figure shows a segmental fit using overlapping sets of data points, which approximates two classes of sites developing sequentially (see text) and the best fit (small closed dots) for independent sites (K d of 0.285 M at 18 sites and K d of 1.3 M at 16 sites). Other combinations gave good fits with the independent site model (see footnote c in Table II).
Although the E1 component has multiple phosphorylation sites, previous (33) and more recent data 8 indicate that phosphorylation proceeds by sequential encounters and not in by a processive mechanism. With a preparation that contains a mixture of singly and doubly phosphorylated E1 but retained nearly 10% activity (nonphosphorylated) E1, binding of PDK2 was reduced 3-fold. A conservative correction for the nonphosphorylated E1 suggests that there is Ͼ4-fold decrease in the affinity of PDK2 when associating with complex containing phosphorylated E1. Further studies are needed to evaluate the effects of individual phosphorylation sites. DISCUSSION We have found that PDK2 is a stable dimer in solution. Its frictional coefficient ratio of 1.44 is consistent with a combination of the extended (a/b ratio 1.52) and the irregular structure ascertained by x-ray diffraction analysis (14). The PDK2 structure appears to be ideally suited for bifunctional binding of two lipoyl domains, as was previously proposed based on studies indicating that the inner lipoyl domain of E2 has a central role in greatly enhanced PDK2 function and PDK2 regulation (1,9,26,27,35,49). PDK2 has the appearance of two slightly twisted wedges in extended apposed positions with the extended openings running in opposite directions. The monomers associate by the ATP using C-terminal domains interacting near the base of the wedges; the N-terminal domains form the outside of the wedges. The wedge pockets have an extended seam that runs the length of the interface between these domains with the ␥-phosphate of bound ATP exposed at one end. On the opposite side from these pocket openings, the combined wedges produce a large horseshoe-shaped cavity with oppositely twisted ends formed by the N-terminal domains. Significant parts of the PDK2 subunit structures, including an extended loop over the active sites, a large C-terminal segment, and short loops in the N-terminal domain, were not resolved in the three-dimensional structure (14). It seems likely that lipoyl domain binding occurs within the horseshoe-shaped cavity, aided by stabilizing interactions involving one or more of the flexible termini and the N-terminal domain. Studies with PDK4 indicate the C-terminal segment is important for the stimulation of the PDK4 by reduction and acetylation of the lipoyl group. 9 The combination of pockets, mobile loops, and a flexible terminal region contributes to the frictional coefficient ratio of 1.44 by extending the asymmetry of the molecule.
We have used lipoyl domain monomers and dimers to aid understanding of the interactions of PDK2 within the complex. Binding of PDK2 by the lipoyl domain structures was found to depend on the oligomeric state and the reaction state (below) of the lipoyl groups of the domain. Dimeric GST-L2 bound PDK2 with a binding affinity roughly matching the affinity of approximately 18 sites of the E2 60-mer. This indicates that bifunctional binding is occurring in both complexes and suggests that the manner in which the lipoyl domains are held, by linker segments at opposite corners at one end of the GST dimer, does not hinder binding relative to the more flexible and extended tethering of lipoyl domains in the E2 structure and that GST, itself, does not inadvertently contribute to the interactions. Lack of a contribution of GST is also indicated by the weak binding by GST-L1 and GST-L3. Reduction of the lipoyl groups of GST-L1 and GST-L3 allowed easily observable binding of these struc-tures to PDK2 but, even then, the estimated binding affinities were weaker than found for GST-L2 with an oxidized lipoyl group and much weaker than GST-L2 red . GSTL2(128 -229) bound PDK2 as well as GST-L2(120 -233), so that the size of the linker region was not a factor in the tighter binding by GST-L2 than GST-L1 or GST-L3. Thus, preferential binding to L2 is indicated.
Based on this bifunctional binding, we conclude that two L2 can bind to PDK2. The weak binding by L2 and L2 red do not allow us to definitively determine an upper limit for their dissociation constants or to ascertain whether equivalent intrinsic binding occurs at both sites on the PDK2 or tighter binding by the first L2. Assuming equivalent binding of 2 L2 to PDK2, the combination of analyses (the increases in S of L2-PDK2 over PDK2, decreases in trailing L2, and SedAnal simulations) favor a K d between 120 and 240 M for L2. Even if the intrinsic (macroscopic) affinity for binding of L2 is equivalent at two sites on PDK2 (K d ϭ 120 -240 M ϭ [L2 f (2 Ϫ L2 b /PDK2)]/ (L2 b /PDK2)), the explicit microscopic binding at the first site is stronger and at the second site is weaker because there is one way to bind and two to dissociate (K 1 ϭ K d /2 ϭ 60 -120 M and K 2 ϭ 2K d ϭ 240 -480 M). Consequently, to firmly establish that two L2 bind and estimate the affinity of the second L2 requires use of very high levels of L2. Sedimentation velocity analyses could not evaluate binding in this range.
Binding by GST-L2 is 40 -80-fold tighter than the above range of 120 -240 M. This indicates that if an L2 of the GST-L2 dimer binds with the same or a somewhat weaker affinity than monomeric L2, then the rate of the intramolecular binding (k intra step) must be Ն40 times faster than its reversal (k d2 step) to the singly held kinase dimer. The appreciably tighter binding of GST-L2 red than GST-L2 explicitly fits enhanced binding being caused by a gain in interactions of the reduced lipoyl group with its greater length and gain in H-bonding capacity. In contrast, there is only a small or no gain in binding affinity upon reduction of monomeric L2. This seemingly indicates that the primary change with GST-L2 red binding occurs in the reversible intramolecular step. Such an enhanced affinity might result from a faster k intra step, but it is much more likely that there is a slower rate of release of the second lipoyl domain (k d2 step) as a result of a gain in a binding interactions by the reduced prosthetic group. Assuming that only the k intra step is different in binding by GST-L2 and L2, it is interesting that the effective concentration that is then estimated for this step 10 is very close to the concentration of L2 estimated at the surface of GST-L2.
Tuganova et al. (47) reported a binding affinity of 10 M for binding of two L2 to rat PDK2 using the Hummel-Dryer gel 8 S. A. Kasten and T. E. Roche, unpublished data. The sequence in phosphorylation of sites has been followed for different PDKs with E1 lacking any mutation in the phosphorylation sites. This required development of a procedure that yields complete cleavage of the tryptic peptide containing the first and second phosphorylation site. 9 J. Dong, L. Hu, J. C. Baker, and T. E. Roche, manuscript in preparation. 10 The two C termini of GST to which L2 are connected are located at catacorner positions at one end of GST dimer with a separation that is well matched to the distance between the subunit wedge pockets in the PDK2 dimer. Many variations in rate constants could produce the apparent relationship between binding by L2 and GST-L2. However, as a reference estimate, the effective concentration in the k intra step can be calculated assuming that k d2 ϭ k d1 and that binding of GST-L2 at first site has the same K d as binding of L2 with an equal macroscopic affinity at two sites. Then, using our estimate that GST-L2 has 40 -80-fold tighter affinity for PDK2 than L2 (K d of ϳ180 M), k intra /k ia ϭ k d1 /k intra ϭ 7.2-14.4 mM ϭ effective concentration for k intra step. Interestingly, these approximated concentration fall within the range estimated for the local concentration of the tethered L2 domains at the surface of GST-L2 (2.45-19.6 mM) based on access to 0.75 volume of sphere with radius between a maximum of 6 to minimum of 3 nm. L2 is held by GST at its distal end from the part of L2 that engages in kinase binding based on mapping of the surface of L2 that interacts with PDK3 (3,4) (see Footnote 7). The interference of GST with productive encounters of PDK2 with the tethered L2 should be compensated by a greater portion of first step encounters having a favorable orientation of L2 along with having two lipoyl domain targets participating in this initial association step. filtration procedure. They equilibrated their L2 in 5 mM dithiothreitol, which would result in a significant portion the L2 lipoyl groups being in the reduced form. Our results with a more active (Ͼ2-fold) human PDK2 indicate that this reduction step alone is insufficient to allow as tight of binding as they report by monomeric L2. The L2 that they used was made and purified as a His-tagged structure, apparently without removal of the tags. A nonspecific His tag interaction or His tag-based aggregation (possibly via metal binding of His tags) could lead to enhanced PDK2 binding with tight binding by aggregates containing reduced lipoyl groups.
Reversible bifunctional binding has been proposed as a means for a PDK gaining access to E1 by "hand-over-hand" walking on the surface of the complex (1-4, 9, 49). By directly supporting bifunctional binding, our findings are consistent with this mechanism. The capacity to move rather than dissociate depends on the relative s Ϫ1 rates for k d1 step producing complete dissociation of singly held PDK2 as compared with this tethered structure interacting via the k intra step with a second lipoyl domain. In support of the PDC reaction, the great reach and flexibility of the connecting linkers allow the lipoyl domains to very rapidly move between active sites separated by Ն45 Å (2,50,51). This demonstrated mobility and reach together with the high local concentration of the lipoyl domains (ϳ1.6 mM at the exterior of E2 (Ref. 44)) constitute the requisite properties for supporting a rapid intramolecular association step. At the same time, the limited gain in affinity for bifunctional binding (above) also predicts a rapidly reversible unimolecular step (k intra /k d2 ).
The phosphorylation of free E1 proceeds with a K m for PDK2 of 4.2 M (high K ϩ buffer) to 14 M (low K ϩ buffer). 3 However, we could not detect binding between free PDK2 and free E1 in AUC studies (data not shown). We have found that E2⅐E1 binds PDK2 only slightly tighter than E2. The ϳ2-fold increase binding affinity and increase in the number of binding sites (from ϳ18 to ϳ28) suggest that E2-bound PDK2 interacts very weakly with its protein substrate and/or that the gain in that interaction is accompanied by opposing thermodynamic changes. This constitutes the first insight into the nature of the interaction between PDK2 and its protein substrate in the absence of catalytic turnover.
Previous studies demonstrated that the PDKs bind ATP in the absence of E1 (55). The prospect that ATP might enhance binding to E2⅐E1 was not supported by the further inclusion of AMP-PNP, an ATP analog, because this decreased PDK2 binding to E2. AMP-PNP is a very effective inhibitor (K i ϭ 3 M) of PDK2. 3 Both products of the PDK2 reaction, phosphorylated-E1 and ADP, decreased the affinity of PDK2 for E2. Preliminary studies indicate that the effect of phosphorylation is less when only the first site on the E1 substrate is phosphorylated. The generation of these products of PDK2 reaction may contribute to movement of PDK2 to new substrate. It seems likely that there is a particularly strong impact on dissociation when both sites 1 and 2 are phosphorylated in an E1 tetramer. In the context of the hand-over-hand mechanism, encountering a non phosphorylated or singly phosphorylated E1 might serve to short-circuit full dissociation. Such a mechanism of sustained access is required to explain sustained kinase activity upon dilution of complexes. 2 The present study direct supports bifunctional binding required for the proposed mechanism of hand-over-hand movement; furthermore, our results suggest that, coupled to product formation, a reduction in binding affinity may support kinase movement to protein substrate that is susceptible to undergoing phosphorylation. The finding that the kinase-stimulating condition of lipoate reduction also strengthens PDK2 binding to the L2 domain seemingly reinforces the requirement for such a movement precipitating feature.