Mathematical modeling and biochemical analysis support partially ordered calmodulin-myosin light chain kinase binding

Summary Activation of myosin light chain kinase (MLCK) by calcium ions (Ca2+) and calmodulin (CaM) plays an important role in numerous cellular functions including vascular smooth muscle contraction and cellular motility. Despite extensive biochemical analysis, aspects of the mechanism of activation remain controversial, and competing theoretical models have been proposed for the binding of Ca2+ and CaM to MLCK. The models are analytically solvable for an equilibrium steady state and give rise to distinct predictions that hold regardless of the numerical values assigned to parameters. These predictions form the basis of a recently proposed, multi-part experimental strategy for model discrimination. Here we implement this strategy by measuring CaM-MLCK binding using an in vitro FRET system. Interpretation of binding data in light of the mathematical models suggests a partially ordered mechanism for binding CaM to MLCK. Complementary data collected using orthogonal approaches that assess CaM-MLCK binding further support this conclusion.


INTRODUCTION
Calmodulin (CaM) is a calcium-binding protein found ubiquitously in the cytosol of eukaryotic cells. CaM contains two globular domains joined by a flexible linker. The N-terminus and C-terminus domains each have a pair of EF-hand motifs and are each capable of binding two calcium ions (Ca 2+ ). 1 CaM acts as a regulator or effector in numerous cellular processes, ranging from muscle contraction to glycogen metabolism and synaptic plasticity. [2][3][4] The Ca 2+ -mediated binding of CaM and either smooth muscle myosin light chain kinase or nonmuscle myosin light chain kinase (henceforth collectively referred to as ''MLCK'') is central to functions such as vascular smooth muscle contraction and cell motility. CaM-MLCK binding takes place as follows: Ca 2+ enters the cytosol and binds to the four Ca 2+ binding sites of CaM, leading to a dramatic change in CaM protein conformation. 5 CaM then binds to the CaM-binding domain of MLCK, triggering a conformational change in MLCK that activates the kinase by displacing an autoinhibitory sequence from the kinase's catalytic domain. 6,7 Finally, activated MLCK phosphorylates the 20-kDa regulatory light chains of myosin II, resulting in contraction caused by myosin cross-bridges moving along actin filaments.
Multiple formal models have been proposed for the activation of MLCK by Ca 2+ and CaM. [8][9][10][11] Given the cooperative nature of Ca 2+ binding at both the C-and N-terminus of CaM, each of these models treats CaM as having two Ca 2+ -binding sites. The model of Brown et al. 8 and Fajmut,Brumen et al. 9 makes no further assumptions, giving rise to an eight-state reaction network in which MLCK may bind to CaM before or after Ca 2+ ('' Model 1'' in Figure 1A). Other previously proposed models are truncations of this network, in which the binding of Ca 2+ and MLCK is either partially 10 or fully 11 ordered ('' Model 2'' and '' Model 3'' in Figure 1A, respectively). Accordingly, Model 1 corresponds to a fully random binding mechanism (Ca 2+ and MLCK can bind to CaM in any order), Model 2 to a partially ordered mechanism (MLCK can bind to CaM after Ca 2+ is bound at the C-terminus), and Model 3 to a fully ordered mechanism (MLCK can bind to CaM only after Ca 2+ is bound at both the C-terminus and N-terminus).
In recent work, Dexter et al. showed that this class of models is analytically solvable for a CaM/MLCK system at both thermodynamic equilibrium and steady state and that each model predicts distinct steady-state behavior in certain Ca 2+ concentration regimes. 12 Importantly, these predictions hold regardless of the . Each of the site-directed Asp to Ala mutations we performed has been demonstrated to prevent Ca 2+ binding, while having only a slight impact on the structure of each CaM EF hand. 20 We use a multi-step strategy to discriminate between these models. We first compare Model 1 to Models 2 and 3, and we then compare Model 3 to Models 1 and 2. During the first step, CaM-FR binding at zero [Ca 2+ ] is determined as Model 1, but not Models 2 and 3, predicts binding under these conditions. In the second step, the binding of FR to a CaM with impaired N-terminus Ca 2+ (CaM 21A,57A ) is measured at high [Ca 2+ ]. Models 1 and 2, but not Model 3,predict binding between FR and this mutant CaM at high [Ca 2+ ]. Finally, we use a CaM mutant with impaired C-terminus Ca 2+ binding (CaM 94A,130A ) to evaluate additional predictions of Model 2, including the absence of binding of FR to this mutant in high [Ca 2+ ]. The predictions about binding are based on algebraic calculations with all parameters treated symbolically; as such, they do not depend on fitting the models to experimental data.
Our binding measurements using the FR falsify key predictions of both Model 1 (random binding) and Model 3 (fully ordered binding) but are consistent with multiple distinct predictions of Model 2, which strongly suggests that CaM-MLCK binding follows a partially ordered mechanism. We further validate our key findings and the utility of the FR system using orthogonal experimental techniques, as well as the measurement of binding between full-length MLCK-FLAG protein and CaM.

RESULTS
For all binding experiments, we purified either FLAG-tagged FR or MLCK from HEK 293T cells after transient transfection, and purified His-tagged CaM WT and CaM mutants (CaM 21A,57A CaM 94A,130A and CaM 21A,57A,94A,130A ) after bacterial expression. Gel electrophoresis and SYPRO Ruby staining of purified proteins ( Figure 2B), as well as mass spectrometry analysis ( Figure S1), established that the protein preparations were free of major contaminants. It is well documented that the electrophoretic mobility of CaM in SDS-PAGE changes following Ca 2+ binding-induced conformational change; this gel mobility shift is commonly used as a readout of Ca 2+ -CaM binding. [21][22][23][24] We observed an expected Ca 2+ -dependent gel mobility shift in our CaM WT , CaM 21A,57A and CaM 94A,130A protein preparations ( Figure 2B), 25 but observed no gel mobility shift for FR or CaM 21A,57A,94A,130A in the presence of Ca 2+ . These data suggest that the purified CaM proteins each retain their expected Ca 2+ binding affinity.
Having validated our experimental tools, we collected FRET-based CaM-FR binding data to use in our model discrimination strategy. The previous theoretical analysis identified a two-part strategy for distinguishing between the three models of CaM-MLCK binding, which is described in detail in Dexter et al. 12 The analysis involves deriving algebraic expressions for the fraction of total MLCK (F) that is predicted  iScience Article to bind to CaM as a function of free [Ca 2+ ] for each of the models. Plots of F are shown in Figure 3A for CaM WT , CaM 21A,57A CaM 94A,130A and CaM 21A,57A,94A,130A (assuming the reference values for the equilibrium constants compiled by Fajmut, Brumen et al. 9 and the concentrations of CaM and MLCK used in our experiments). The model discrimination strategy rests on two predictions that differ between the three models and that hold true regardless of the specific numerical values assigned to the model parameters, such as the equilibrium constants in Figure 1. The first is that only Model 1 predicts non-zero binding of MLCK in zero [Ca 2+ ], with the fraction bound given by the following expression: where CaM tot and MLCK tot denote total CaM and total MLCK, respectively. The second is that only Model 3 predicts zero binding of MLCK to a CaM mutant with nonfunctional N-terminal EF hands, such as CaM 21A,57A at any free [Ca 2+ ]. In contrast, both Models 1 and 2 predict non-zero binding, with the fraction in the high- [Ca 2+ ] limit given by lim The algebraic structure of Equations 1 and 2 is identical; the only difference between the two expressions is which equilibrium constant appears (K 9 in Equation 1, K 4 in Equation 2). As discussed in Dexter et al., 12 this similarity is a consequence of the fact that the reaction network reduces to a bimolecular reaction in both the low-and high- [Ca 2+ ] limits.   Figure 3B. We confirmed the presence of excess CaM compared to FR, and equal protein amount across experiments, by SDS-PAGE and SYPRO Ruby staining of the samples after FRET measurement ( Figure S3).
For the first model discrimination test, we compared the FRET ratio of the CaM WT -FR pair with baseline in zero and low [Ca 2+ ]. Contrary to the prediction of Model  A striking feature of the experimental binding curves is the significant difference in maximum binding between CaM WT and CaM 21A,57A (p = 1.16*10 À6 by a one-tailed Mann-Whitney U test). This difference is also consistent with the predictions of Model 2 ( Figure 3A). Assuming Model 2, the fraction of CaM 21A,57A bound in the high- [Ca 2+ ] limit is given by Equation 2. For CaM WT , the limiting expression is the same as Equation 2 but with K 2 instead of K 4 (for the reasons explained above), so that the model predicts equal maximum binding if K 2 = K 4 . In the reference parameter set, K 2 = 1,000 and K 4 = 16.7, corresponding to predicted binding of 99.9% for CaM WT and 92.0% for CaM 21A,57A in 39 mM [Ca 2+ ]. As such, the experimentally observed difference is also predicted by our mathematical analysis, assuming that the literature parameter values are correct within an order of magnitude.
The validity of our experimental approach for model discrimination depends on the ability of the CaM-FR interaction to accurately mirror CaM-MLCK binding with sufficient sensitivity, as well as other considerations related to the experimental perturbations. To address these potential limitations, we performed a series of control experiments and tested the robustness of our experimental design and data.
First, to confirm that the FRET ratio of FR is only altered by the binding of Ca 2+ -bound CaM, we collected 460 nm-700 mm emission spectra with Ex 430 of the FR in 0 and 39 mM [Ca 2+ ]. When FR was assayed alone, FR showed no changes in FRET ratio as a function of [Ca 2+ ] and produced fluorescence peaks at Em 480 and Em 535 ( Figure S4A).  Figure S4D).
Having seen no interactions between FR and CaM 94A,130A or CaM 21A,57A,94A,130A during our FRET-based binding assays, we wanted to confirm that the interactions we observed between CaM WT -FR and CaM 21A,57A -FR are specific, Ca 2+ -dependent, and representative of binding between CaM and MLCK.
To do so, we performed several on-bead binding assays using either FR or MLCK-FLAG. When beadbound FR was incubated in 50 mL of buffer with 505 nM CaM WT (approximately 20 ng/mL), the fraction of CaM WT bound to FR increased proportionally with free [Ca 2+ ]. At 39 mM [Ca 2+ ], the majority of the CaM WT input was bound to the FR ( Figure 4A). When the same experiment was repeated using N-terminus mutant CaM 21A,57A , more than half of CaM 21A,57A bound to FR at 39 mM [Ca 2+ ] ( Figure 4B). These results suggest that the purified CaM proteins are functional and bind to Ca 2+ and the FR, and that there is negligible non-specific binding between the FR and the purified CaM proteins. To determine We also investigated the relationship between the ''fraction bound'' calculated in the modeling analysis and the FRET ratio we use as a proxy for CaM-FR interaction. Although the Em 480/Em 535 ratio may not have a one-to-one relationship to ''fraction bound,'' the Em 480/Em 535 ratio correlates closely with CaM-FR binding at high and low [Ca 2+ ], as calculated using on-bead binding assays ( Figure S6). Moreover, the same FRET ratio has been shown to mirror MLCK phosphorylation in vivo. 14 As a result, we conclude that the ratio can be used to characterize CaM-FR interaction for our model discrimination strategy, which rests on differential binding predictions in zero and high [Ca 2+ ].
We do not observe binding between FR and C-terminus mutant CaM 94A,130A in our FRET-based binding assays. To investigate the possibility that CaM 94A,130A might bind to the FR but fail to interfere with FRET, we used a Bio-Layer Interferometry (BLI) assay as an orthogonal method to measure binding between FR and CaM. Binding detection in the BLI assay is independent of FRET measurement, which enabled us to decouple binding and FRET interference. The BLI assay showed robust binding between FR and CaM WT at high [Ca 2+ ], an intermediate degree of binding between FR and CaM 21A,57A at high [Ca 2+ ], and no detectable binding between FR and CaM 94A,130A at high [Ca 2+ ] ( Figure 5A). Performing a BLI assay using CaM WT , CaM 21A,57A , and CaM 94A,130A with purified MLCK-FLAG yielded comparable results ( Figure 5B).
We next assessed the intrinsic sensitivity of the FRET assay. To test if our assay can detect small changes in binding, which is necessary for many of the comparisons involved in our model discrimination analysis, we measured FRET at 39 mM [Ca 2+ ] with different FR:CaM ratios. Assuming Model 2 and the reference parameter values from Fajmut, Brumen et al., 9 if the FR:CaM molar ratio is increased from 2:1 to 2:1.2, the amount of CaM-bound FR in 39 mM [Ca 2+ ] is predicted to increase by 8.5% (from 46.2% bound to 54.7% bound). When we repeated our FRET-based CaM-FR binding assay using these ratios (22.9 nM FR and 11.45 nM or 13.75 nM CaM), we observed a significant increase in F480/F535 ( Figure S7) (p = 0.00029 by a one-tailed Mann-Whitney U test), with a calculated 8.7% difference in FRET ratios between the two conditions. We therefore conclude that, under our experimental conditions, the assay is sensitive enough to detect at least a 9% increase in the fraction of CaM-bound FR. iScience Article The predictions of zero MLCK binding in a particular [Ca 2+ ] hold for any set of numerical values assigned to the model parameters (the equilibrium constants K 1 . K 11 in Figure 1), as shown in Dexter et al. 12 For predictions of non-zero binding, however, the magnitude of binding does depend on the choice of parameters. As such, it is theoretically possible that the model could predict a binding fraction that is non-zero but too small to detect experimentally in an important concentration regime. To address this potential concern, we undertook a sensitivity analysis of the two key predictions of non-zero MLCK binding. For numerical calculations with the models, we used the set of reference parameter values compiled by Fajmut, Brumen et al. from previous biochemical studies. 9 As shown in Equation 1, the prediction of Model 1 for the fraction of MLCK bound in zero [Ca 2+ ] depends on the value of a single parameter (K 9 ). Assuming the parameter value selected by Fajmut, Brumen et al. 9 based on several prior studies (K 9 = 0.078 mM À1 ) and the concentrations of CaM WT and FR used in the main experiment, the model predicts 5.2% binding in zero [Ca 2+ ]. As is clear from the structure of Equation 1, the fraction of MLCK predicted to bind in zero [Ca 2+ ] increases with the ratio of total CaM to total MLCK. To confirm our falsification of Model 1, we therefore repeated the binding experiment in zero [Ca 2+ ] with a much higher [CaM] (35.65 mM), for which 73.5% binding is predicted with the reference parameter values and 21.7% binding is predicted with K 9 = 0.0078 mM À1 (i.e., 10-fold lower than the reference value). As in the main experiment, the FRET ratio did not increase above baseline in zero [Ca 2+ ] ( Figure S8), providing strong evidence against Model 1 even if previous parameter estimates are incorrect by an order of magnitude (p = 0.998 by a one-tailed Mann-Whitney U test).
For Model 2, the fraction of MLCK predicted to bind to the N-terminus mutant CaM 21A,57A depends on two parameters, K 4 and K 6 (Equation 2). We confirmed the robustness of the key prediction of non-zero binding in 39 mM [Ca 2+ ] by calculating F N 2 for 200,000 combinations in which values for each of the two parameters were chosen at random from the interval [0.01v, 100v], where v is the reference value from Fajmut, Brumen et al. 9 Across the combinations F N 2 was never less than 18%, a level of binding straightforward to distinguish from baseline; the full distribution is shown in Figure S9.
Finally, we confirmed that the addition of the purified proteins to the prepared Ca 2+ buffers did not significantly alter the buffers' free [Ca 2+ ]. We measured the fluorescence intensity of two Ca 2+ indicator dyes with different Ca 2+ affinities (Fluo-4 and Calcium Green), before and after the addition of the buffer in which our CaM proteins were stored. The largest volume of protein added to any of our FRET-based binding experiments was 1.3 mL per assay, so we tested this volume. We did not observe a significant change in indicator dye fluorescence intensity after CaM storage buffer addition, which suggests that any changes to free iScience Article [Ca 2+ ] that are experimentally introduced are smaller than can be detected reliably with Fluo-4, a dye that has a very high Ca 2+ affinity (17 nM) ( Figure S10) (p = 0.23-0.99 by a two-tailed Mann-Whitney U test).
These additional experiments and sensitivity analyses show that our experimental system, which includes the well-validated FRET-based binding assay, is robust and rigorously controlled, as outlined here. We find that the FR and CaM proteins we used are free of major contaminants; SYPRO stained SDS-PAGE gels of the CaM proteins also confirm that these proteins display the predicted altered electrophoretic mobility upon binding to Ca 2+ , which strongly suggests that they bind Ca 2+ as expected. Follow-up plate reader experiments demonstrated that FRET interference is sensitive enough to allow discrimination between candidate CaM-MLCK binding models. We also show that a change in the FRET ratio is caused only by the binding of Ca 2+ -bound CaM proteins; conversely, orthogonal BLI experiments confirm that a lack of FRET interference corresponds to a lack of FR-CaM binding. We further confirmed that the [Ca 2+ ] of our experimental buffers is not altered by the addition of our experimental proteins. Finally, on-bead binding experiments with both FR and MLCK-FLAG reproduced the key findings of our FRET-based binding assays, a critical check of the robustness of our experimental system. These data, in combination with the MLCK-FLAG data from our BLI assays, suggest that FR-CaM binding accurately mirrors binding between CaM and full-length MLCK protein, and that our FR data can confidently be used for model discrimination.
In sum, our binding measurements falsify key predictions of both the fully random binding mechanism (Model 1) and the fully ordered binding mechanism (Model 3). Our data instead support Model 2, which assumes partially ordered binding between CaM and MLCK ( Figure 1). As predicted by Model

DISCUSSION
In recent work, Dexter et al. analyzed a class of previously developed theoretical models of CaM-MLCK binding and proposed a multi-part strategy for distinguishing between them. 12 Our primary contribution here is an experimental implementation of this model discrimination strategy, which suggests a partially ordered mechanism for binding. Our analysis sheds new light on a controversy that has persisted for several decades and demonstrates a productive interplay between mathematical modeling and biochemical analysis.
It is important to remember that model discrimination strategies of this kind work by process of elimination. Models can be ruled out when their predictions contradict experimental data, but the failure of some models does not guarantee the correctness of others. We present here evidence sufficient to falsify all but one of the models of CaM-MLCK binding in the literature (Table 1). Data collected using wild-type Our data are consistent with Model 2, which makes six correct predictions. Taken together, these results strongly support a partially ordered mechanism in which MLCK can bind to CaM after Ca 2+ is bound at the C-terminus (Model 2).
To investigate MCLK-CaM binding, we centered our experimental design on an FR system that uses the CaM-binding domain of smooth muscle MLCK. Prior studies have demonstrated that this reporter accurately reflects CaM-MLCK binding in a variety of in vivo and in vitro contexts. [13][14][15][16][17][18][19] Moreover, our FRETbased CaM-FR binding assay enabled the collection of robust and reproducible data using purified CaM and FR proteins, with minimal risk of interference from endogenous proteins or contaminants. This system is also amenable to complementary, non-FRET-based biochemical techniques to assess CaM-FR binding, which allowed us to confirm the FRET-based binding data we obtained. Although the FR we used has been extensively validated as a good proxy for CaM-MLCK binding, the use of an MLCK fragment still does raise concerns about the behavior of the fragment relative to the full-length protein. In HEK 293T lysates, changing myosin phosphorylation has been shown to correspond to changes in the FRET ratio produced by FR. Changes to the FRET ratio can therefore be used as a proxy for MLCK activation. 14 It is important to note that changes in the FRET ratio do not necessarily correspond 1:1 with changes to the fraction of FR bound to CaM across the full range of [Ca 2+ ] tested. Determining the proportionality constant, however, is not necessary for our model discrimination strategy, which relies primarily on FRET ratios measured at 0 mM [Ca 2+ ] and 39 mM [Ca 2+ ]. We also complement this FRET-based binding assay data with on-bead binding assays showing that in 39 mM [Ca 2+ ], the majority of CaM WT binds to either FR or full-length MLCK protein (MLCK-FLAG) expressed using human smooth muscle MLCK gene MYLK1, while a negligible amount of CaM binds to FR or MLCK-FLAG in 0 mM [Ca 2+ ]. We therefore conclude that FRET ratio data represent CaM WT -MLCK binding in 0 mM [Ca 2+ ] and 39 mM [Ca 2+ ].
It should be noted that mammalian myosin light chain kinases are a group of serine/threonine kinases encoded by at least four genes: MYLK1, MYLK2, MYLK3, and MYLK4. 26 MYLK1 has alternative initiation sites that enable the expression of at least four protein products including nonmuscle (long isoform) MLCK and smooth muscle (short isoform) MLCK. MYLK2 encodes an MLCK isoform expressed solely in skeletal muscle, MYLK3 encodes a cardiac-specific MLCK (MLCK3), and the gene product(s) of MYLK4 remain largely uncharacterized. 25,26 All MLCK proteins, except MLCK4, have a CaM-binding peptide that shows sequence homology to both the peptide used in our FR, which is derived from avian smooth muscle MLCK, and to the CaM-binding domain of our MLCK-FLAG protein. As a result, we expect that our findings here can be generalized to other MLCK proteins and their interaction with CaM and Ca 2+ .
In addition to providing a blueprint for future model discrimination efforts that integrate mathematical and biochemical approaches, our work may also prove useful in a translational or pharmaceutical context, as MLCK and its activation by CaM have been linked to the pathogenesis of human disease. For example, increased organization of the sarcomere, the contractile unit of the striated muscle, is observed during the onset of cardiomyocyte hypertrophy. CaM-activated MLCK has been shown to mediate sarcomere organization induced by a hypertrophic agonist in cultured cardiomyocytes and in vivo. 27 Fuller characterization of the Ca 2+ -CaM-MLCK interaction network may therefore prove relevant for future drug discovery efforts. iScience Article In addition, MLCK is only one of the many binding partners of CaM; future work is needed to determine the binding mechanisms of other proteins regulated by CaM-Ca 2+ .

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:  Figure 3) B Comparison of on-bead and FRET-based binding assessment between FR and CaM WT as a function of [Ca 2+ ] ( Figure S6) B Bio-Layer interferometry (BLI) binding assays ( Figure 5) B CaM-FR binding sensitivity assay ( Figure S7

DECLARATION OF INTERESTS
The authors declare no conflicts of interest. One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in their field of research or within their geographical location. One or more of the authors of this paper self-identifies as a gender minority in their field of research. One or more of the authors of this paper self-identifies as a member of the LGBTQIA+ community. Bacteria were pelleted, frozen, and stored at À80 C until protein purification. Cells were lysed using sonication for 5 min at 40% power before being centrifuged at 17,400 RPM for 40 min to separate the soluble and insoluble portions of the lysate. The clarified lysate was loaded onto a Ni-NTA gravity column and then washed with 10 column volumes of lysis buffer before elution with 5 column volumes of elution buffer. The eluted protein was loaded manually into a 1 mL anion exchange chromatography column. HPLC was performed using buffers A and B. The HPLC fractions that contained the highest concentrations of protein were collected and concentrated to their final concentrations using a centrifugal filter with a 10 kDa molecular weight cutoff. Stock protein concentrations were as follows: CaM WT : 3.98 mg/mL; CaM 21A,57A : 4.96 mg/mL; CaM 94A,130A : 1.36 mg/mL; CaM 21A,57A,94A,130A : 1.52 mg/mL. Purified protein was aliquoted, flash frozen in a liquid nitrogen dewar, and then stored at À80 C until use. To test the robustness of the falsification of Model 1 to uncertainty in parameter values, we repeated the binding assay in zero [Ca 2+ ] using 35.65 mM [CaM WT ]. CaM WT was concentrated to a stock concentration of 66.6 mg/mL using a concentrator column so that a comparable volume of protein stock (relative to previous assays) could be used. The ''FRET-based CaM-FR binding assay'' was then performed as described above, except using 35.65 mM [CaM WT ].

FRET-based CaM-FR binding assay
To determine a lower sensitivity limit for the CaM-FR binding assay, 22.9 nM FR was added to 150 mL of high [Ca 2+ ] buffer along with either 11.45 or 13.74 nM [CaM WT ], which resulted in a CaM WT :FR molar ratio of either 1:2 or 1.2:2. 150 mL of the resulting mixture was then read in a plate reader using the program employed by the ''FRET-based CaM-FR binding assay'' (see above) and analyzed accordingly. CaM and FR were added to either 0 or 39 mM [Ca 2+ ] as per ''FRET-based CaM-FR binding assay'' above.

CaM-FR spectral scanning
CaM was used at a final concentration of 673 nM (13 ng/mL); FR was used at a final concentration of 22.9 nM (1.3 ng/mL). Using an excitation wavelength of 430 nm, the wells were read using spectral scanning between wavelengths 460 nm and 700 nm, with an emission step of 10 nm. The fluorescence intensity of each wavelength was then plotted.

Binding assay protocol
Transiently transfected cells were washed once with chilled PBS, and then harvested in 1% Triton buffer using a cell scraper. Cells were triturated to ensure complete lysis, and then were spun down at maximum speed (17,000g) for 10 min using a 4 C centrifuge. 160 mL bead:glycerol M2 FLAG slurry was prepared for FLAG-tagged protein binding by washing three times with 1 mL 1% Triton buffer before being divided evenly between 8 tubes. The cleared lysate from the transiently transfected cells was divided evenly between 7 of the tubes, while cleared lysate from control WT HEK 293T was added to the eighth tube (control lysate  Figure S5). All tubes were incubated at room temperature for 15 min and were flicked occasionally. Following incubation, the tubes were spun for 1 min at 1,000 g to pellet the beads. 40 mL of the unbound portion was conserved and prepared with 10 mL 5X reducing sample buffer.
The beads for all conditions were then washed with 50 mL of the appropriate Ca 2+ buffer and all liquid was aspirated from the beads using a syringe to remove residual unbound CaM. Running the gel 30 mg of wild-type CaM and N-terminus mutant CaM were each mixed with high Ca 2+ calibration buffer to a final volume of 20 mL. A blank control consisting of CaM storage buffer and Ca 2+ calibration buffer was also prepared. All samples were reduced using 1 mM final concentration of TCEP and alkylated using 2 mM final concentration of chloroacetamide (CAM) at 37 C for 20 min on a shaker. They were then quenched using 1 mM final concentration of TCEP at 37 C for 20 min on a shaker. Next, they were mixed with 5X gel loading buffer, boiled for 5 min, and loaded into a 10% gel (see ''gel electrophoresis'' above). The samples were run until the dye front reached the bottom of the gel. The gel was then rinsed and stained using Coomassie stain for 1 h, and destained overnight.

In-gel digestion protocol
The next day, two bands from each lane were excised and cut into small cubes using a single-use scalpel; one band was comprised of the gel at $18 kDa (where the major CaM band is located) and one was comprised of the gel between $10 and 17 kDa, the area under the major band. The rest of the in-gel digestion was performed as previously described in. 28

Extraction of peptides and StageTip purification
StageTip extraction of peptides was performed as described previously in Rappsilber et al. (2007). 29 nanoLC-MS/MS analyses LC-MS analyses were performed as described previously with the following minor modifications. 30

Semi-quantitative analysis of contaminants and CaM protein
The intensity of peptides annotated as CaM, keratin, or bacterial protein was summed up and their relative intensities were calculated by dividing them to the total peptide intensity in the samples. While not loaded with ligand, the control probes were quenched with His-GST. In each experiment, ''Control'' refers to the analyte (either FR or MLCK-FLAG) binding to the ligand-free probe. This control is performed to demonstrate that the association seen in the ligand bound probes is not due to nonspecific binding but is, in fact, specific. The binding of the analyte (either FR or MLCK-FLAG) to the optical probes was measured simultaneously using instrumental defaults for 236 s. The dissociation was measured for 287 s. There was no binding of FR to the unloaded probes; however, slight binding of MLCK-FLAG to the unloaded probes was observed. The data were analyzed by the Octet data analysis software. The association and dissociation curves for FR binding were globally fit with a 1:1 ligand model, and the curves for MLCK-FLAG binding were locally fit for 80s. The data were plotted using Prism 7.