Our general experimental strategy is to label two samples of CaMKII separately with two different fluorophores, followed by mixing and analysis of whether holoenzymes containing both of the fluorophores are obtained. This type of analysis is complicated in bulk solution by the difficulty in distinguishing between aggregation of holoenzymes and true subunit exchange. Gel filtration analysis of a constitutively active mutant (CaMKII^T286D) shows no evidence for aggregation, and incubation of the wild-type holoenzyme with Ca²⁺/CaM and ATP shows no changes in the UV-vis absorption spectrum (data not shown). We also used dynamic light scattering (DLS) to determine the size distribution of activated CaMKII assemblies in solution (for both wild-type, activated by Ca²⁺/CaM, and a constitutively active mutant in which Thr 286 is replaced by aspartate [CaMKII^T286D] in the absence of Ca²⁺/CaM). DLS measurements indicate that the addition of ATP to CaMKII^T286D, under the conditions used in our mixing experiments, results in a small population with sizes that are larger than that of a single holoenzyme (Figure 2—figure supplement 1). The DLS data for both wild-type CaMKII and CaMKII^T286D demonstrate that there is no time-dependent change in the particle size distribution over the time scale of our experiments.

To rule out aggregation as a complicating factor definitively, we developed a single-molecule assay in which individual CaMKII holoenzymes are visualized directly. In this assay, two samples of CaMKII are labeled separately with the donor (Alexa 488, green fluorophore) and acceptor (Alexa 594, red fluorophore) using cysteine chemistry, and then mixed and incubated together in the presence or absence of Ca²⁺/CaM and ATP. CaMKII holoenzymes are then immobilized on a glass slide such that individual holoenzyme assemblies are separated spatially from each other, and visualized by TIRF microscopy (see schematic diagram in Figure 2A). An exchange event is detected by the colocalization of red and green fluorophores, and potential aggregation is monitored by analysis of the distribution of fluorophore intensities.

Figure 2 with 5 supplements see all

Download asset Open asset

For the single-molecule experiments, wild-type CaMKII was expressed with a C-terminal Avitag (i.e., the tag is located after the hub domain) (Howarth and Ting, 2008). The purified protein was biotinylated in vitro, and labeled separately with either the green or the red fluorophore. Mass spectrometric analysis showed that the labeling occurs at Cys 280, in the C-terminal lobe of the kinase domain. Labeling efficiencies ranged between 20 and 40% for the wild-type protein (see ‘Materials and methods’), and this is sufficient for our analysis because of the direct visualization of holoenzymes. After mixing the two labeled species, a small amount of the CaMKII solution was removed at each time point, diluted, and then immobilized on the PEG-treated glass slides that were coated with streptavidin (Figure 2A). Multi-color TIRF imaging was performed and the extent of colocalization was determined using a custom particle-tracking program (see ‘Materials and methods’). We score exchange events by treating all particles that have both red and green fluorophores equally. That is, our analysis does not distinguish between holoenzymes in which a single subunit has been exchanged and those in which more than one subunit has been exchanged.

The positions of CaMKII holoenzymes were determined by fitting a two-dimensional Gaussian function to particle intensities (see ‘Materials and methods’). Importantly, the software filters out particles of insignificant brightness (i.e., noise) and aggregates (i.e., unusually bright particles) (Figure 2—figure supplement 2). On average, 6–20% of particles were eliminated from the analysis. Since samples from different time points are from the same reaction, the distribution of fluorescence intensities of each sample should not change over time in the absence of aggregation. Histograms of the fluorescence intensities were created for each image, after processing in the particle-tracking program, and monitored to check for any population of extra-bright particles (most likely aggregates). The distribution of fluorescence intensities did not change significantly over the time course of the subunit exchange reaction, indicating that the colocalization we report is not due to aggregation, which would cause the distribution to shift towards values of greater intensity (Figure 2—figure supplement 3). This does not, however, rule out that aggregation or self-association of CaMKII holoenzymes may be necessary as an intermediate step for subunit exchange to occur, since such self-association might be reversed by dilution prior to the single-molecule analysis.

We compared the results of mixing experiments in which CaMKII had either not been activated or had been activated by Ca²⁺/CaM and ATP. Protein samples were mixed and incubated together at 37°C for varying lengths of time in the absence of Ca²⁺/CaM and ATP and analyzed using the single-molecule assay. The level of colocalization detected under these conditions is very low, even at long times (∼5% at 240 min, Figure 2B). A dramatically different result is obtained when the samples are activated by incubation for 240 min at 25°C with Ca²⁺/CaM and ATP (∼40% colocalization in about 50 min, Figure 2B).

The absolute levels of colocalization observed in our experiments are likely to be underestimates of the true degree of colocalization. Incomplete labeling, a population of fluorescent molecules in a dark state and errors in image processing would lower the total percentage of colocalization detected. The maximum extent of colocalization (62%) was observed using a sample of CaMKII that has been labeled with both dyes simultaneously. We report the degree of colocalization in any particular experiment as a fraction of this maximal value. The values for colocalization are variable between experiments done on different days, due to labeling efficiencies and instrumental variability (compare Figure 2B and C). All direct comparisons made in this paper are done using data from experiments performed on the same day.

The extent of colocalization increases markedly with CaMKII concentration and with temperature, as shown in Figure 2B,C. For example, activated wild-type CaMKII reaches 40% of maximal colocalization within 6 min at 37°C, but at 25°C the time taken to achieve the same degree of colocalization is ∼250 min. If the concentration of CaMKII subunits is reduced from 8 μM to 1 μM, the extent of colocalization in 10 min at 37°C is reduced from ∼20% of maximal to ∼10% (Figure 2C). The concentration of CaMKII subunits in dendritic spines is estimated to be ∼100 μM (Otmakhov and Lisman, 2012), suggesting that in this environment the timescale of subunit exchange may be substantially shorter than seen in our experiments. We have not studied subunit exchange in vitro at concentrations higher than 8 μM.

We asked whether the activation-triggered colocalization involves a dead-end conversion in which holoenzymes that have undergone one cycle of exchange are inert to further exchange. To test this, we added saturating Ca²⁺/CaM and ATP to red-labeled and green-labeled forms of the holoenzyme and incubated these activated species separately for 15 min at 25°C. We then mixed the two samples and observed robust colocalization that is similar for samples that had not been incubated separately after activation (Figure 2—figure supplement 4). These data indicate that once CaMKII is activated, its subunits are capable of multiple cycles of exchange.

The results of the single-molecule assay are corroborated by fluorescence resonance energy transfer (FRET) measurements in solution. The site of labeling in the wild-type protein, Cys 280, is located at the periphery of the holoenzyme, and does not give a good FRET signal. We therefore labeled the hub domain at a site near the central hole in the hub assembly (residue 335), where any two such sites in the same hexameric ring would be close enough to generate a FRET signal. We replaced surface-exposed cysteines (residues 280 and 289) in the kinase domain by serine and replaced Asp 335 in the hub domain by cysteine (D335C). Labeling at this site should bring donor–acceptor pairs within ∼33 Å for adjacent subunits and ∼58 Å for non-adjacent subunits within a holoenzyme (Figure 2—figure supplement 5A). The D335C mutant CaMKII was labeled with ∼80% efficiency.

The FRET signal obtained for experiments using the D335C mutant showed a marked contrast for mixed samples that were either activated by Ca²⁺/CaM and ATP or not. There is a substantial increase in the FRET signal that occurs within minutes of mixing the two labeled species for the activated sample (Figure 2—figure supplement 5B,C). In the unactivated sample (i.e., in the absence of Ca²⁺/CaM and ATP), there is only a modest, slow increase in the FRET signal with time, indicating that any exchange that might occur in the unactivated sample does so at a very slow rate.

We interpret the results of the single-molecule experiments to mean that activation results in the weakening of inter-subunit contacts, facilitating exchange. To check whether this was happening, we sought to stabilize the holoenzyme assembly. To do this, we fused CaMKII to hemolysin-coregulated protein from Pseudomonas aeruginosa (Hcp1), a protein that forms hexameric rings with roughly the same diameter as the hub domain of CaMKII (CaMKII-Hcp1) (Mougous et al., 2006) (PDB code 1Y12). The fusion protein was generated by linking the C-terminal end of the hub domain of CaMKII to the N-terminal end of Hcp1 by a 10-residue linker with a sequence that is designed to be flexible (see ‘Materials and methods’). The Avitag used to immobilize CaMKII to the glass slide was incorporated after the Hcp1 sequence. The kinase activity of the CaMKII-Hcp1 fusion was tested using a peptide substrate (syntide) and it displayed cooperative activation by Ca²⁺/CaM, with an activation profile similar to that of wild-type CaMKII (Gaertner et al., 2004; Rosenberg et al., 2005, data not shown).

We carried out a mixing experiment using the CaMKII-Hcp1 fusion protein in which this construct was labeled separately with either red or green dye and the two samples were mixed and incubated at 37°C. Colocalization of the two fluorophores is only ∼10% even after 1 hr, compared to ∼70% for the wild-type holoenzyme (Figure 3A). In an analogous experiment, we labeled wild-type CaMKII with the red fluorophore (Alexa 594) and the CaMKII-Hcp1 fusion protein with the green fluorophore (Alexa 488) and measured colocalization after activation (Figure 3A). The level of colocalization is much below that observed with the wild-type protein in this case as well.

Figure 3

Download asset Open asset

The strong suppression of colocalization seen with the CaMKII-Hcp1 fusion protein lends further support to the idea that CaMKII holoenzymes exchange subunits upon activation. Since fusion of Hcp1 to the hub domain is unlikely to impede the separation of a holoenzyme into two hexameric rings, these data also suggest that exchange involves some other disassembly process.

The fact that activation leads to subunit exchange in the intact holoenzyme made us wonder whether the hub domain assembly might be intrinsically unstable, and that the release of stabilizing contacts between the kinase domains and the hub upon activation might allow the subunits of the hub domain to separate and exchange. To test whether the hub domain is intrinsically capable of subunit exchange, we purified the hub domain and monitored colocalization. We found that the subunits of the isolated hub domain assembly do not exchange subunits (Figure 3B).

These data suggest that some combination of the kinase domain, the regulatory segment or the linker connecting the regulatory segment to the hub domain must be required for the exchange process. To examine the role of the linker we carried out single-molecule experiments using a construct of CaMKIIα in which the linker is eliminated entirely. This short-linker construct is similar to the construct used to obtain the crystal structure of CaMKIIα (Chao et al., 2011). As shown in Figure 3C, the short-linker construct exhibits fluorophores colocalization with the same rate as full-length CaMKIIα when activated by Ca²⁺/CaM and ATP, indicating that the linker is not required for subunit exchange.

We also wondered whether the release of interactions between the kinase domain and the hub might be the trigger for subunit exchange. In the crystal structure of the autoinhibited short-linker CaMKII holoenzyme, the kinase domains dock against the hub domains (Chao et al., 2011). Mutation of lIe 321 in the hub domain to glutamate disturbs this docking and results in an opening of the holoenzyme assembly (Chao et al., 2011). Introduction of the same mutation (I321E) in the context of the short-linker construct has no effect on the rate of colocalization (Figure 3C). This suggests that the trigger for subunit exchange does not involve the disruption of the interface between the kinase domain and the hub that is seen in the structure of the autoinhibited holoenzyme.

The experiments presented so far relied on activation by Ca²⁺/CaM and ATP to trigger exchange. Since the regulatory segment binds to Ca²⁺/CaM, it is difficult to introduce mutations into this segment without disturbing the interaction with calmodulin. To identify the role played by Ca²⁺/CaM in the exchange process, we used the constitutively activated form of the enzyme (CaMKII^T286D) in which the regulatory segment is expected to be displaced from the kinase domain even in the absence of Ca²⁺/CaM.

CaMKII^T286D displays a vigorous degree of colocalization when ATP is added in the absence of Ca²⁺/CaM, comparable to that of the wild-type enzyme in the presence of Ca²⁺/CaM and ATP (Figure 4A). These data demonstrate that although activation of the catalytic module is necessary for colocalization, the physical presence of Ca²⁺/CaM is not required.

Figure 4 with 1 supplement see all

Download asset Open asset

With the elimination of a direct role for Ca²⁺/CaM in the exchange process our attention was now focused on the kinase domain and the regulatory segment. We were particularly interested in the part of the regulatory segment that spans the calmodulin-recognition motif (the R3 element), because this segment interacts with the hub domain in the crystal structure of the autoinhibited CaMKII holoenzyme (Chao et al., 2011).

Starting with the constitutively active variant (CaMKII^T286D), we replaced residues 302 to 313 in the R3 element by an arbitrary linker sequence that is not expected to form regular secondary structure. Specifically, the sequence ³⁰²A I L T T M L A T R N F³¹³ in the R3 element was replaced by ³⁰²S T G G G T G S T G S T³¹³. We monitored colocalization after mixing for this construct and found that colocalization was essentially eliminated (Figure 4A).

Given that the calmodulin-recognition element plays a crucial role in subunit exchange, how might activation control the ability of this segment to mediate exchange? Activation by Ca²⁺/CaM releases the R3 element from the kinase domain, but it remains hidden within calmodulin. Thr 286 is, however, outside the footprint of calmodulin, and it is phosphorylated upon Ca²⁺/CaM binding to the holoenzyme. The release of Ca²⁺/CaM from a subunit after Thr 286 is phosphorylated facilitates phosphorylation at Thr 305 and Thr 306, which prevents Ca²⁺/CaM rebinding (Hashimoto et al., 1987; Hanson and Schulman, 1992; Colbran, 1993). Displacement of Ca²⁺/CaM and phosphorylation of the R3 element on Thr 305 and/or Thr 306 may therefore facilitate subunit exchange. Using this reasoning, the kinase activity of CaMKII is expected to be crucial for the exchange process.

We carried out two mixing experiments using wild-type CaMKII and Ca²⁺/CaM in parallel (Figure 4B). In one experiment, as described previously, the fluorescently labeled samples were incubated separately with ATP and Ca²⁺/CaM, then mixed together and monitored for CaMKII colocalization using the single-molecule assay. In the other experiment, the ATP concentration was lowered to 1 μM, which is below the value of the K_M for ATP (Colbran, 1993). In addition, the kinase inhibitor bosutinib was added during the incubation phase. Although bosutinib was developed as an inhibitor of Abl tyrosine kinase, it has off-target activity against CaMKII (Remsing Rix, 2009) and was used in the crystallization of autoinhibited CaMKII (Chao et al., 2011). The fluorescently labeled samples were then mixed, and monitored for colocalization, as before.

As shown in Figure 4B, the reduction in ATP concentration and the presence of bosutinib result in suppression of colocalization. For example, after 30 min of mixing, the wild-type enzyme exhibits 70% of the maximal degree of colocalization, whereas under conditions where phosphorylation is inhibited the level of colocalization is only ∼20% after 30 min. We confirmed the importance of kinase activity by using the FRET assay (Figure 4—figure supplement 1). In this case, we added Ca²⁺/CaM and the kinase inhibitor in the absence of ATP, and found that there is no FRET signal under these conditions.

We compared the colocalization of the constitutively active variant CaMKII^T286D in the presence and absence of the kinase inhibitor and low ATP. As shown in Figure 4C, inhibition of kinase activity suppresses colocalization for CaMKII^T286D, although to a lesser extent than for the wild-type protein. These data indicate that phosphorylation at sites other than Thr 286 is also important for exchange, and drew attention to a possible role for Thr 305 and Thr 306 in the exchange process. We replaced these residues with alanine separately to generate two constructs (CaMKII^T286D,T305A and CaMKII^T286D,T306A). Attempts to purify a construct in which both residues were mutated to alanine were not successful because this variant did not express well. Mixing experiments demonstrated substantial reduction in colocalization for both individual mutations, although somewhat less than the reduction obtained with treatment of CaMKII^T286D with bosutinib (Figure 4C). We conclude that phosphorylation of both Thr 305 and Thr 306, in addition to Thr 286, is important for subunit exchange.

An important question is whether subunits that are activated can exchange with holoenzymes that have not been activated, and further, whether such exchange could result in the phosphorylation of Thr 286 in the unactivated subunits. To study this, we used mixtures of wild-type CaMKII and the constitutively active CaMKII^T286D variant in mixing experiments. Because CaMKII^T286D lacks Thr 286, this allows us to detect the phosphorylation of Thr 286 in the unactivated holoenzymes by using an antibody that is specific for the phosphorylated form of Thr 286 (pT286, see ‘Materials and methods’). The antibody is labeled with Alexa 647, and it does not detect CaMKII^T286D (Figure 5—figure supplement 1).

The first important result is that subunits in constitutively active CaMKII^T286D can exchange with subunits from holoenzymes that have not been activated, when ATP is added (Figure 5A). Fusion of the hexameric protein Hcp1 to the wild-type protein reduces the rate of colocalization, indicating that the unactivated CaMKII holoenzyme has to open up in some way for the exchange process to happen.

Figure 5 with 1 supplement see all

Download asset Open asset

We also examined the ability of the activated subunits to phosphorylate subunits from holoenzymes that had not been activated. To do this we devised a three-color single-molecule experiment, in which the Alexa 488 and Alexa 594 channels monitor the locations of the labeled CaMKII variants, and the Alexa 647 channel detects phosphorylated Thr 286 in wild-type CaMKII. To determine the extent of antibody labeling that can be detected in a fully activated sample of CaMKII, we first activated CaMKII by Ca²⁺/CaM and ATP for 10 min. This is expected to result in robust phosphorylation of each holoenzyme assembly because of the direct activation of each subunit by Ca²⁺/CaM. This resulted in ∼75% colocalization of fluorescent signals from CaMKII and the pT286 antibody (Figure 5—figure supplement 1).

We then studied the extent of Thr 286 phosphorylation that is detected when the constitutively active CaMKII^T286D is mixed with unactivated wild-type CaMKII. As shown in Figure 5B, there is an increase in the degree of phosphorylation detected with time, demonstrating that CaMKII^T286D is able to phosphorylate subunits in unactivated wild-type CaMKII.

It has been shown previously that trans-phosphorylation of Thr 286 occurs principally within a holoenzyme (Hanson et al., 1994; Rich and Schulman, 1998). In order to check that the observed phosphorylation is due to the simultaneous presence of activated (CaMKII^T286D) and unactivated (wild type) subunits within the same holoenzyme, we carried out two analyses. First, we employed the Hcp1 fusion construct, for which subunit exchange was suppressed (Figure 3A). Mixing CaMKII^T286D-Hcp1 with wild-type CaMKII resulted in decreased Thr 286 phosphorylation (Figure 5B), indicating that stabilizing the activated CaMKII holoenzyme reduces the observed phosphorylation. In the second analysis, we showed that antibody binding occurs principally in those species in which the red and green fluorophores are colocalized (Figure 5C), consistent with subunit exchange being required for the trans-phosphorylation reaction.

We note that the signal from pT286 rises much more slowly than the rate of colocalization. Colocalization reaches a plateau value of ∼60% colocalization in ∼50 min, whereas the level of three-color colocalization at a comparable time point is only ∼5%. The replacement of Thr 286 by aspartate in CaMKII^T286D may not be a good mimic of the phosphorylated form of Thr 286 in terms of its ability to promote phosphorylation. We found that the activity of CaMKII^T286D towards a peptide substrate (syntide, see ‘Materials and methods’) is only ∼20% that of activated wild-type CaMKII (data not shown). Also, the ability of CaMKII^T286D to transphosphorylate an adjacent subunit within an unactivated holoenzyme may be inherently slow without calmodulin bound to the subunit being phosphorylated (Rich and Schulman, 1998). The activation of CaMKII is highly cooperative (Gaertner et al., 2004; Rosenberg et al., 2005), so low levels of Ca²⁺/CaM that are insufficient for autophosphorylation of an unactivated holoenzyme may be sufficient for binding the target subunit and enhancing its autophosphorylation in holoenzymes that contain some activated subunits through exchange.

To get a better sense of the extent to which activated subunits can potentiate the activity of holoenzymes that have not been activated, we turned to a solution assay using two different samples of wild-type CaMKII, without fluorescent labeling. One sample, the activated sample, was prepared by incubation with ATP and Ca²⁺/CaM for 15 min, followed by removal of Ca²⁺/CaM (see ‘Materials and methods’). The second sample, the unactivated sample, was not exposed to Ca²⁺/CaM. We then measured the activity of these samples using a peptide substrate (syntide) and a continuous enzyme-coupled assay (ADP Quest, see ‘Materials and methods’). This assay relies on the conversion of ADP to a fluorescent product, resorufin, the accumulation of which is monitored over time (Charter et al., 2006).

A sample containing 1 μl of activated CaMKII at 6 μM concentration exhibits a peptide phosphorylation rate of 132 units, where the units are the change in the fluorescence of resorufin in unit time (Figure 5D). A second sample containing 3 μl of unactivated CaMKII at the same concentration exhibits a rate of 49 units (this activity represents the futile ATP hydrolysis rate of CaMKII, also observed without peptide substrate, data not shown). A third sample consisting of 1 μl of activated CaMKII at 6 μM mixed with 3 μl of unactivated CaMKII at 6 μM exhibits a rate of 592 units. In the absence of any crosstalk between the unactivated and activated holoenzymes, the reaction rate exhibited by the mixture is expected to be the sum of the reaction rates of the first two samples, that is, 181 units. The actual observed rate, 592 units, is about three times as high. This suggests that activated CaMKII holoenzymes have increased the activity of the holoenzymes that had not been activated previously.

One difficulty in developing a conceptual model for the exchange process in CaMKII is that there appears to be little precedent in the biochemical literature, as far as we are aware, for oligomeric enzymes that spread their state of activation by exchanging subunits. We envisage two possible kinds of mechanisms for the exchange of subunits between CaMKII holoenzymes. The first mechanism is suggested by the unusual allosteric activation of porphobilinogen synthase (Selwood and Jaffe, 2012). This enzyme converts between an inactive hexamer and an active octamer, with the transition requiring dissociation of the assembly into dimers (Selwood et al., 2008). We have developed a speculative structural model for how activation might promote the release of subunits, most likely dimers, from a CaMKII holoenzyme, discussed below. An alternative mechanism involves the transient formation of higher order aggregates of two or more holoenzymes, with exchange occurring within the aggregate followed by separation of holoenzymes with mixed subunits. The data available to us at present provide no clues as to how such a process might occur, so we do not discuss this further although it remains an important alternative mechanism for future investigation.

Regardless of whether subunit exchange proceeds through the release of subunits or through transient aggregation, it seems likely that the regulatory segment is able to destabilize the hub assembly when this segment is released from the kinase domain. We developed a model for how the regulatory segment might weaken the hub assembly by analyzing molecular dynamics simulations of CaMKII (see ‘Appendix’), and by considering several characteristic features of the structure of the holoenzyme and the exchange process. These include (i) the ability of the hub assembly to interconvert between dodecameric and tetradecameric forms, (ii) the importance of phosphorylation of the regulatory segment, (iii) the presence within the hub domain of a deep cavity containing three conserved arginine residues and (iv) the known ability of the hub domain to serve as a docking site for peptide segments by virtue of its exposing an uncapped β sheet, with no steric hindrance to extending the sheet by an additional strand provided by the regulatory segment. For ease of discussion, the details of the molecular dynamics simulations and the reasoning that went into the development of this model are presented in an ‘Appendix’ that follows immediately after the main body of the text. The main features of the model are summarized schematically in Figure 6. Note that the schematic in Figure 6 emphasizes the release of dimers, but we are uncertain as to whether the actual process involves a transient aggregation step that we do not yet understand.

Figure 6

Download asset Open asset

The CaMKII holoenzyme can be thought of as being assembled from six ‘vertical dimers’. In a view that has the mid-plane of the dodecamer horizontal, these vertical dimers each contribute one hub domain to the upper ring and one to the lower ring of the hub assembly. We propose that the hub assembly within a holoenzyme, whether activated or not, normally undergoes fluctuations that convert it from the closed double-ring form seen in crystal structures to an open C-shaped form with a gap between two adjacent vertical dimers. Molecular dynamics simulations suggest that this gap can easily be large enough to capture a vertical dimer provided by another holoenzyme. This could, for example, convert a dodecameric assembly to a tetradecameric one. In the absence of activation, however, the transient openings in the hub assembly of the holoenzyme simply reanneal, due to the stability of the lateral interactions between vertical dimers.

The situation changes, however, when CaMKII is activated and Thr 305 and Thr 306 are phosphorylated. We propose that the phosphorylated regulatory segment, which is freed from interaction with the kinase domain and Ca²⁺/CaM, is now able to dock on the hub domain and form an additional strand of the β sheet that is the core scaffold of the hub domain. We speculate that when this happens, one or more of the phosphorylated sidechains in the regulatory segment reach into the interior of the hub and interact with the conserved arginine residues located in the hub cavity. This interaction might distort the structure of the hub domain, slightly weakening the lateral interactions in the hub assembly. A fluctuation in the hub assembly can now result in the release of an activated vertical dimer. A released vertical dimer can either be recaptured or, if it encounters an unactivated holoenzyme that is open transiently, it can be incorporated into that holoenzyme.

While we believe that this model provides a reasonable framework for thinking about the exchange process, the schematic shown in Figure 6 is far from definitive. For example, close encounters between two holoenzymes may be necessary for an activated dimer to ‘hop’ directly from one holoenzyme to another, without being ever released completely. In such a situation, the interaction between the phosphorylated regulatory segment and the hub domain may occur in trans, between two holoenzymes, rather than within one holoenzyme. We anticipate that future studies will establish whether the model we are proposing is correct in essence, or whether the subunit exchange mechanism relies on features that we have failed to anticipate.

The strengthening of synaptic connections between neurons that occurs following brief repetitive stimuli that induce LTP is likely to be important for the formation and maintenance of cognitive memory. Much attention has been focused on the role of CaMKII in this process, particularly its ability to maintain an activated state, through intra-holoenzyme phosphorylation of Thr 286, after cessation of LTP-inducing stimuli. But can the active state be maintained beyond the lifetime of the kinase molecules present during the initial stimulus?

Lisman introduced the idea that information about previous activating events could be transmitted to newly synthesized CaMKII molecules if the subunits of the activated holoenzyme could exchange with unactivated ones (Lisman, 1994). We now demonstrate that the CaMKII holoenzyme has precisely this property when studied in vitro. Activated CaMKII holoenzymes can indeed instruct previously unactivated holoenzymes as to their phosphorylation state through subunit exchange, and this ability is switched on by activation and autophosphorylation. In the time since Lisman’s earlier speculation about subunit exchange, there has been no experimental precedent that has pointed to such a mechanism being operative for CaMKII. Translocation of CaMKII to the NMDA receptor requires active subunits and has been postulated to serve as a molecular memory (Bayer et al., 2001; Jiao et al., 2011; Lemieux et al., 2012; Neant-Fery et al., 2012). This population of CaMKII bound to the NMDA receptor may maintain its activity for longer times (Otmakhov et al., 2004). Subunit exchange could facilitate the recruitment of additional holoenzymes to the receptor by spreading activation to unactivated holoenzymes.

Recent experiments using a CaMKII construct engineered to be a FRET reporter of its activation state indicate that the bulk of CaMKII holoenzymes in dendrites maintain their activation state for not much longer than a minute after the stimulation is withdrawn (Takao et al., 2005). Interestingly, another study demonstrated that localized stimulation of a neuron resulted in the translocation of CaMKII from the main body of the axon to synapses throughout the dendritic arbor for much longer times (15–40 min) after the local stimulation is stopped; the authors of this report commented that subunit exchange may play a role in this process (Rose et al., 2009). Given the high concentrations of CaMKII in synapses and the fact that calmodulin is limiting (Liu and Storm, 1990; MacNicol and Schulman, 1992; Persechini and Stemmer, 2002), subunit exchange provides a potential mechanism for increasing the spread of activation caused by an initial activating pulse.

Now that we have demonstrated unambiguously that activation triggers the exchange of subunits in CaMKII in vitro, future experiments will address several important questions. We have focused on human CaMKIIα in this study. Is subunit exchange common to the other isoforms of the human enzyme? Did the property of subunit exchange arise early in evolution, or did it evolve with the specialization of CaMKII to neuronal function? Finally, and most importantly, how has nature exploited subunit exchange in activated CaMKII to determine the actual outcomes of neuronal signaling as well as in heart and other systems? It will be challenging, but ultimately most informative, to devise experiments to address this last question.

The CaMKII dodecamer can be described in terms of the lateral association of six vertical dimeric units, as shown in Figure 7. We imagine that a critical step in the exchange process is a lateral opening of the ring formed by the hub domains (Figure 7A,B). The exchange process might involve the release of vertical dimers. Such a vertical dimer could not be released easily from the Hcp1 fusion construct and indeed subunit exchange is suppressed between these variants (Figure 3A). The release and capture of vertical dimers also provides a route for the interconversion of dodecameric and tetradecameric holoenzymes without requiring a more complete disassembly. As noted in the main text, we are uncertain about whether the transient aggregation of holoenzymes might be necessary for exchange to occur. In this Appendix, we focus on potential interactions between the regulatory segment and a single hub domain assembly.

Figure 7 with 1 supplement see all

Download asset Open asset

That the unit of exchange might be a vertical dimer is suggested by the fact that the hub domain is very closely related in structure, although not in sequence, to members of a large family of enzymes and binding proteins, such as nuclear transport factor 2 (NTF2) and ketosteroid isomerase like proteins, which we refer to as the NTF2 family (Figure 7C; Hoelz et al., 2003). A common quaternary structural unit within the NTF2 family corresponds to the vertical dimer in the CaMKII hub. Of the top 50 unique hits in a Dali search (Holm and Rosenstrom, 2010) for structures similar to the mouse CaMKIIα hub domain (PDB code: 1HKX), 29 form dimers in the crystal lattice that are similar to the vertical CaMKII hub dimer (Figure 7C).

Molecular dynamics simulations provide a clear indication that the lateral interfaces in the hub assembly are more likely to be disrupted than the equatorial ones (Figure 7—figure supplement 1). The hub domain of human CaMKIIγ has been crystallized as a dodecameric assembly (Rellos et al., 2010) (PDB code: 2UX0), and we generated a 100 nanosecond (ns) trajectory of this assembly. The internal motions of the hub can be described as rigid-body motions of individual vertical-dimer units, with alterations in the lateral dimer–dimer interfaces. This is demonstrated by aligning each instantaneous structure from the trajectory on each subunit in turn, and graphing the root mean square deviations in the positions of the Cα atoms in the two subunits across the lateral interfaces and the one subunit across the equatorial interface (Figure 7D, Figure 7—figure supplement 1). For comparisons across the equatorial interfaces, within a vertical dimer, the rms displacements in the neighboring subunits are ∼2 Å across the ring. In contrast, for comparison across the lateral interfaces, the rms displacements of the neighboring subunits are as much as 6 to 10 Å for several of the interfaces, consistent with the lateral interfaces being much more flexible than the equatorial ones (Figure 7—figure supplement 1).

Further analysis of the simulation suggests that there is strain associated with the closed ring formed by the dodecameric hub assembly of CaMKII. Although the hub assembly is symmetric initially, the sixfold symmetry of the ring breaks down during the course of the simulation. This is due to the tightening of some of the lateral interfaces, for which the buried surface area increases. For the CaMKIIγ structure used in the simulation, the closer packing involves the sidechains of Phe 364, Phe 367, Tyr 368, Asn 371, Leu 372 on helix αD in one subunit and Gln 406, Pro 414, Thr 416 and Ile 418 (Gln 418 in the α isoform) from strands β4 and β5, as well as Pro 379 in strand β3 in the other subunit (see Figure 7C for the notation). With the exception of Ile 418, these residues are all conserved in CaMKIIα.

The tighter packing at some of the interfaces in the simulation of the dodecameric hub assembly is coupled to looser packing at some of the other interfaces. At one of the interfaces, the residues on helix αD in one subunit and the β4 and β5 strands on the other are splayed apart so that the sidechains of Phe 364 and Phe 367, which are buried in the more stable interfaces, are now partially solvent exposed (Figure 8). These results suggest that the constraint of ring closure prevents the simultaneous optimization of all of the lateral interfaces.

Figure 8

Download asset Open asset

To further explore the extent to which the closed-ring form of the dodecamer is strained, we removed one vertical dimer from the crystal structure of the dodecamer, generating a C-shaped decameric model (Figure 9A). This allows us to use molecular dynamics to study the effect of relaxing the ring-closure constraint, following a strategy that was used effectively to analyze strain in circular sliding DNA clamps (Jeruzalmi et al., 2001; Kazmirski et al., 2005).

Figure 9 with 2 supplements see all

Download asset Open asset

This decameric structure, derived from the crystal structure of the dodecameric CaMKIIγ hub assembly, was used to initiate two independent molecular dynamics trajectories (100 ns and 50 ns, respectively). There is a rapid relaxation of the curvature of the ring in both trajectories. The decameric structure opens up with respect to the starting structure in both trajectories within ∼10 ns (Figure 9B). The principal outward displacement occurs in the plane of the ring so that the structures move closer to the tetradecameric form of the hub assembly (PDB code: 1HKX; Figure 9B, Figure 9—figure supplement 1).

The constraint of ring closure undoubtedly imposes an entropic strain on the ring, and removal of the constraint is expected to yield a more floppy and open structure. What is notable, however, is that in both molecular dynamics trajectories we observe relaxation to a specific interfacial arrangement that is different from the interfacial arrangement seen in the crystal structure of the dodecamer. There are two internal vertical interfaces in the decamer, between the G:H and E:F vertical dimers and between the G:H and I:J vertical dimers. The relative orientations of subunits across these two interfaces converge to a similar arrangement, with low rms displacements, both within one trajectory and between two independent trajectories (Figure 9C, bottom left, Figure 9—figure supplement 2A), and this arrangement is different from that seen in the crystal structure of the dodecamer (Figure 9C, bottom right, Figure 9—figure supplement 2B).

This relaxation towards a specific structure, as seen in two independent trajectories, suggests that there is an enthalpic contribution to the strain in the closed ring. It is difficult, however, to assign this strain to any particular sets of interaction. The structural relaxation preserves the set of residues that interact at each interface, and PISA analysis (Krissinel and Henrick, 2007) of instantaneous structures extracted from the trajectories does not reveal any systematic trend in the estimated interfacial free energy.

The experimental data suggest that the release of the regulatory segment from the body of the kinase domain upon activation allows the regulatory segment and, possibly the kinase domain, to interact with the hub domain in a way that weakens the inter-subunit interfaces in the hub assembly. Given that the calmodulin-binding R3 element is particularly important for exchange, we looked for plausible ways in which the phosphorylated R3 element could interact with the hub domain.

There is one very obvious binding site in the hub domain for negatively charged groups, and that is the deep cavity in the hub domain, which corresponds to the active site or peptide-binding site in other structurally related proteins (Hoelz et al., 2003). We shall refer to this as the ‘hub cavity’ (Figure 10A,B). CaMKIIα contains three arginine residues that are located deep within the hub cavity (residues 403, 423, and 439). These residues are highly conserved either as arginine or lysine, with three positively charged residues present in CaMKII from C. elegans, mouse, and drosophila. No function has been ascribed to the hub cavity in CaMKII nor to the arginine residues within it.

Figure 10

Download asset Open asset

A feature of the hub assembly that might allow interaction between Thr 305 and Thr 306 in the regulatory segment and the arginine residues in the interior of the hub cavity is suggested by an interesting variability in the structures of individual subunits of the hub assembly in crystal structures. Although in many cases the only access to the interior of the hub cavity is through the main opening to the cavity, which is far from where the regulatory segment connects to the hub, some individual subunits in hub domain assemblies exhibit cracks in the surface at the interface between helix αA and the β sheet against which it is packed (Figure 10B). These cracks make Arg 403, located within the cavity, partially accessible from the side of the hub domain. If such cracks were to open up further then phosphorylated residues in the R3 element might gain access to the interior of the cavity from the side of the hub domain, in the vicinity of the lateral interfaces between vertical hub domain dimers in the holoenzyme.

We carried out a 1 μs molecular dynamics simulation of a vertical hub domain dimer that was extracted from the crystal structure of mouse CaMKIIα (PDB code: 1HKX, Video 1). This simulation revealed transient fluctuations in the hub domain structure that opened the interface between helix αA and the underlying β sheet substantially. The simulation suggests that if the R3 element was to dock alongside the lateral interface then it might be positioned to insert phosphorylated sidechains into the interior of the hub domain when such a transient fluctuation occurs.

Video 1

Download asset

The molecular details for such a docking mechanism are suggested by the presence of an open edge on the β sheet of the hub domain right alongside the site where the transient openings into the hub cavity occur. This open edge of the β sheet is in fact utilized for a docking interaction by an 8 residue portion of the R3 element in the structure of autoinhibited CaMKII holoenzyme, in which this portion of the R3 element extends the β sheet by one strand (Chao et al., 2011). In that structure, however, further interaction between the R3 element and the hub domain is prevented by the fact that the kinase domain sequesters the rest of the regulatory segment.

We modeled the regulatory segment, with Thr 305 phosphorylated, as an additional strand that extends the β sheet of the hub domain, as shown in Figure 10C (see ‘Materials and methods’ for details of the molecular docking). This model was used to generate three independent molecular dynamics trajectories, each extending for 50 ns. In each of these trajectories the R3 element retains the hydrogen bonding pattern that is consistent with its incorporation into the β sheet of the hub domain and the phosphate group maintains its proximity to the sidechain of Arg 403, suggesting that this interaction is plausible (Figure 10C).

We analyzed these ‘docked’ trajectories for fluctuations that would disturb the contacts made within the hub assembly. The peptide stays docked onto the β3 strand of the hub domain throughout the trajectories, which, in turn, prevents the crack between helix αA and the β sheet from closing (Figure 10D). Notably, the widening of the crack is coupled to changes in the disposition of the loops that make lateral contacts with the adjacent vertical dimers (Figure 10D). This disruption may be sufficient to allow the release of a vertical dimer from the hub assembly (see Figure 6 for a schematic representation of such a process).

Experimental validation of this model will require an extensive set of studies that are beyond the scope of this paper. For example, we mutated the arginine residues in the hub cavity that we have implicated in the exchange process, but this resulted in a substantial loss of fluorophore labeling at residue 335, indicating that the structure of the hub domain might be compromised. A systematic test of the predictions of this model will be carried out in future studies.

Full-length constructs of human CaMKII were cloned in a pSMT-3 vector containing an N-terminal sumo expression tag (LifeSensors, Malvern, PA). Mutants were generated using a Quikchange protocol (Agilent Technologies, Santa Clara, CA). The linker connecting Hcp1 to CaMKII (GGC GCG TCT GGC GCG TCT GGC GCG TCT) and the hub domain construct were made using standard PCR techniques.

All CaMKII variants were expressed using E. coli and prepared as described previously (Chao et al., 2010). Briefly, protein expression was done in Tuner (DE-3)pLysS cells that contained an additional plasmid for λ phosphatase production. Cells were induced by addition of 1 mM isopropyl β-D-1-thiogalactopyranoside and grown overnight at 18°C. Cell pellets were resuspended in Buffer A (25 mM Tris, pH 8.5, 150 mM potassium chloride (KCl), 1 mM DTT, 50 mM imidazole, and 10% glycerol) and lysed using a cell disrupter. All purification steps were carried out at 4°C and all columns were purchased from GE Healthcare (Piscataway, NJ). Cleared lysate was loaded on 5 mL Ni-NTA column, eluted with 0.5 M imidazole, desalted using a HiPrep 26/10 desalting column into Buffer A with 0 mM imidazole, and cleaved with Ulp1 protease (overnight at 4°C). The cleaved samples were loaded onto the Ni-NTA column and the flow through was loaded onto a Q-FF 5 ml column, and then eluted with a KCl gradient. Eluted proteins were concentrated and then buffer-exchanged using a Superpose 6 gel-filtration column equilibrated in 25 mM Tris, pH 8.0, 150 mM KCl, 2 mM tris(2-carboxyethyl)phosphine [TCEP] and 10% glycerol. Fractions with pure protein were frozen at −80°C.

Calmodulin (from Gallus gallus) was expressed using a pET-15b vector (generous gift of Angus Nairn), and purified as described previously (Putkey and Waxham, 1996).

Samples for mass spectrometry were prepared by the addition of 80% acetonitrile and 250 ng trypsin to ∼10 μM CaMKII. Reactions were carried out at 25°C for 2–3 hr then put on a Speed-Vac for an additional 2–3 hr to remove residual acetonitrile. Samples were analyzed using LC/MS (QB3/Chemistry Mass Spec Facility at UC Berkeley).

CaMKII is expressed with a C-terminal biotinylation sequence (Avitag) (Howarth and Ting, 2008), and after purification, labeled with biotin using the biotin ligase BirA. This reaction mixture contained 1 mM ATP, 10 μM biotin (dissolved in DMSO), 0.6 μM BirA ligase in a final volume of 0.5–1 ml. The reaction was carried out on ice for 1 hr and desalted into labeling buffer using a PD-25 column (GE Healthcare). Mass spectrometry indicates complete biotinylation.

Following biotinylation, CaMKII was labeled with Alexa fluor dyes. Purified CaMKII was desalted into a buffer containing 25 mM Tris, pH 8.0, 150 mM potassium chloride, 1 mM tris(2-carboxyethyl)phosphine (TCEP) and 10% glycerol just prior to labeling with Alexa Fluor C₅-maleimide dyes (488 and 594, Life Technologies). Alexa dye was resuspended in 25 mM Tris (pH 8.0) at a final concentration of 10–15 mM. Labeling was completed by addition of threefold to fivefold molar excess dye over CaMKII subunit concentration, and incubated for 2–3 hr at 25°C. Samples were desalted (1 or 2X depending on efficiency and amount of protein) using PD-25 columns (GE healthcare) into an imaging buffer (25 mM Tris, pH 8.0, 150 mM potassium chloride, 1 mM TCEP). Samples were concentrated using Amicon filters to a final subunit concentration of 6 μM. Dye incorporation was estimated using spectrophotometric analysis (Nanodrop, Thermo Scientific, DE). The absorbance of each dye at 280 nm (A₂₈₀) was estimated using free dye resuspended in buffer. The dye-labeled protein samples were then scanned and the A₂₈₀ was corrected for dye contribution. The concentration of protein and dye was calculated using the corresponding extinction coefficients and the ratio of protein labeled is estimated as: [dye]/[protein] (Alexa 488: ε = 71,000 cm⁻¹M⁻¹, Alexa 594: ε = 71,000 cm⁻¹M⁻¹, and extinction coefficients for CaMKII variants were estimated using an online protein calculator tool: http://web.expasy.org/protparam/).

CaMKII labeled with Alexa 488 (6 μM) was mixed with CaMKII labeled with Alexa 594 (6 μM) with or without ATP (250 μM), MgCl₂ (8 mM), calcium (500 μM) and calmodulin (6 μM). The final concentration of CaMKII in these mixtures is 4.8 μM. In the main text we refer to the stock concentration of CaMKII used to make the mixtures (6 μM). At each time point, 2 μl of the CaMKII mixture was diluted into 1000 μl of desalting buffer and spread onto the chamber housing the functionalized cover slip. CaMKII was incubated on the cover slip for 1 min and then the surface was washed 3X with 1 ml desalting buffer. These slides were then imaged using TIRF microscopy.

In experiments where the inhibitor, bosutinib, was added, 6 μM CaMKII (after labeling with Alexa fluor) was incubated with 50 μM bosutinib for 15 min at 25°C. Proteins were then mixed as above, except with less ATP (0 or 1 μM).

In the 3-color colocalization experiments, the pThr286 antibody (Pierce antibodies #A-20186, Thermo Scientific) was labeled with Alexa 647 prior to the experiment, following the protocol for the monoclonal antibody labeling kit (Life Technologies). At each time point in these experiments, antibody was added to the CaMKII mixture at 35-fold molar excess over CaMKII concentration and binding was carried out at 25°C for 3 min. The sample was then diluted 500-fold and spread onto the functionalized cover slip, as above.

For the preparation of phosphorylated CaMKIIα, the same procedure as above was followed, except that the His tag was not removed by Ulp1 cleavage, and a subtractive Ni-NTA purification step was not done. This protein was then labeled with Alexa 594 as previously described. His-tagged CaMKII (6 μM) was activated at 25°C by addition of Ca^2+/CaM (40 μM/6 μM), ATP (250 μM) and MgCl₂ (8 mM) for 10 min. The reaction was quenched by addition of EDTA (50 mM) and EGTA (20 mM). This reaction volume was mixed with free Ni-NTA resin (1 ml), equilibrated with Buffer A (5% glycerol) and allowed to bind for 1–2 hr (4°C). After washing with Buffer A, bound CaMKII-His was eluted with 0.5 M imidazole (5% glycerol), desalted into imaging buffer using a PD-25 column and concentrated to at least 6 μM.

A mutant of CaMKIIα was used for these experiments, in which all surface exposed cysteine residues (280, 289) were mutated to serine, and aspartate 335 was mutated to cysteine (D335C). The same labeling and mixing experiments for the single-molecule assay were done. Samples were incubated at 25°C or 37°C. At each time point, 25 μl from the mixed sample was removed and diluted to a final volume of 150 μl. An emission spectrum (500–700 nm) was acquired for each diluted sample excited at 490 nm using a Fluoromax-3 fluorometer (Horiba Scientific, Edison, NJ). Data were analyzed by calculating the FRET ratio (acceptor emission at 610 nm divided by donor emission at 515 nm). Error bars are calculated from the standard deviation between separate experiments on different days.

Kinase activity was monitored using an ADP quest assay (Charter et al., 2006). Phosphorylated CaMKII (activated sample) was prepared as above. Activated and unactivated samples were diluted to 1 μM. For the mixing experiments, these samples were mixed in a 3:1 ratio of unactivated:activated protein (15 μl unactivated + 5 μl activated). For controls, each sample was measured individually at these same concentrations, and brought to the same final volume by addition of buffer. The kinase assay was carried out in a 384-well plate format at 30°C in a 27.5 μl reaction volume. The reaction mixture contained the following components (listed at final concentration): 10 mM MgCl₂, 20 μM DTT, 0.3–0.5 mM peptide substrate syntide (PLARTLSVAGLPGKK), and Tris. For each reaction, 5 μl of the protein solution was mixed with 4 μl concentrated reaction mixture, and 1 μl of either water or Ca²⁺/CaM (200 μM/4 μM) was added. The protocol for the ADP quest assay was followed for the remaining steps and reactions were initiated by the addition of 250 μM ATP to the mix. The increase in fluorescence was monitored at 590 nm (excitation at 530 nm) in a fluorescent microplate spectrophotometer (Synergy H4, Biotek) with sampling every 5 min. Slopes (Δfluorescence_{590 nm}/time) were calculated from the early data points where an initial linear rate is observed.

The molecular dynamics trajectories were generated using the Gromacs 4.6.2 package (Berendsen et al., 1995; Pronk et al., 2013). For all the hub dimer molecular dynamics simulations, with the docked pThr containing peptide, Amber12 (Case et al., 2012) was used for ease of handling of modified backbone residues. The ff99SB-ILDN force field was used for all the calculations (Lindorff-Larsen et al., 2010). All simulations were carried out in aqueous medium using the TIP3P water model and appropriate counterions (Na⁺ and Cl⁻) were added to neutralize the net charges. After initial energy minimization, the systems were subjected to 30–100 ps of constant number, volume and temperature (NVT) equilibration, during which the system was heated to 300K. This was followed by a short equilibration at constant number, pressure and temperature (NPT, 20–100 ps). The equilibration steps were performed with harmonic positional restraints on the protein atoms. Finally, the production simulations were performed under NPT conditions, with the Berendsen and v-rescale thermostats in Amber12 and Gromacs 4.6.2 respectively, in the absence of positional restraints. Periodic boundary conditions were imposed, and particle-mesh Ewald summations were used for long-range electrostatics and the van der Waals cut-off is set at 1 nm. A time step of 2 fs was employed and the structures were stored every 2 ps. SHAKE and LINCS constraint algorithms were used with Amber12 and Gromacs 4.6.2, respectively, to fix covalent bonds.

We modeled a 16-residue stretch of the R3 element as an additional strand of the β sheet in the hub domain. We created the model by first docking a β strand taken from an arbitrary protein so that the last 9 residues of the strand were aligned roughly with the 8 residue strand formed by R3 element in the autoinhibited CaMKII holoenzyme (PDB code: 3SOA). This model does not have the strand aligned optimally with the rest of the β sheet, and to improve the model we searched the protein databank, using the PDBeFOLD server (Krissinel and Henrick, 2004), for structures that had β strands arranged in the same topology and with close spatial overlap to the three strands from the hub domain and the roughly modeled fourth β strand, yielding a match in the human core Snrnp domain (PDB code: 1D3B). We then created a new model in which the β sheet from the crystal structure of the hub domain was retained and the fourth strand was taken from Snrnp, with appropriate changes to the sequence.

Using this fourth strand as a template we modeled in the sequence of the CaMKII R3 element in more than one sequence register, and carried out a series of short molecular dynamics simulations (∼100 ns each). In one of these simulations a conformational change that is similar to the transient fluctuations noted in the Appendix happened to occur, opening up access to the hub cavity. The sidechain of Met 307 from the R3 element was seen to approach Arg 403 closely in this simulation. We then changed the sequence register of the peptide so that Thr 305 occupied the position of the methionine sidechain, and we added a phosphate group to the threonine sidechain.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank members of the Kuriyan and Groves labs, especially Jon Winger, for helpful discussion. We thank Jeff Iwig, Julie Zorn, Brian Kelch (UMass Worcester), and Charles Morgan (UCSF) for helpful comments on the manuscript, Tiago Barros for help with figure design, and Qi Wang for help with DLS experiments. We thank David King (HHMI, Mass Spectrometry) for generous assistance with synthesis of peptides. We thank Tony Iavarone (QB3) for mass spectrometry support and discussions, Stewart Loh and Anna Elleman for support, Giulio Superti-Furga for providing us with bosutinib, Joseph Mougous (University of Washington) for the Hcp1 vector, and Alice Ting (M.I.T) for the BirA vector. We thank Eileen Jaffe for making us aware of the allosteric mechanism of porphobilinogen synthase.

© 2014, Stratton et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.