Development of a Molecularly Evolved, Highly Sensitive CaMKII FRET Sensor with Improved Expression Pattern

Genetically encoded fluorescence resonance energy transfer (FRET) biosensors have been successfully used to visualize protein activity in living cells. The sensitivity and accuracy of FRET measurements directly depend on biosensor folding efficiency, expression pattern, sensitivity, and dynamic range. Here, to improve the folding efficiency of the Ca2+/calmodulin-dependent protein kinase II alpha (CaMKIIα) FRET biosensor, we amplified the association domain of the CaMKIIα gene using error-prone polymerase chain reaction (PCR) and fused it to the N-terminus of mCherry in a bacterial expression vector. We also created an Escherichia coli expression library based on a previously reported fluorescent protein folding reporter method, and found a bright red fluorescent colony that contained the association domain with four mutations (F394L, I419V, A430T, and I434T). In vitro assays using the purified mutant protein confirmed improved folding kinetics of the downstream fluorescent protein, but not of the association domain itself. Furthermore, we introduced these mutations into the previously reported CaMKIIα FRET sensor and monitored its Ca2+/calmodulin-dependent activation in HeLa cells using 2-photon fluorescence lifetime imaging microscopy (2pFLIM), and found that the expression pattern and signal reproducibility of the mutant sensor were greatly improved without affecting the autophosphorylation function and incorporation into oligomeric CaMKIIα. We believe that our improved CaMKIIα FRET sensor would be useful in various types of cells and tissues, providing data with high accuracy and reproducibility. In addition, the method described here may also be applicable for improving the performance of all currently available FRET sensors.

Introduction end) was amplified using Titanium Taq (Takara) according to the manufacturer's instructions. Subsequently, PCR fragments were digested with EcoRI/NotI and ligated into the mCherry-inserted pRSET vector (Invitrogen), thereby fusing the amplified fragment to the N-terminus of the mCherry fluorophore with a linker sequence (GSGGRTS). In order to generate a library, this construct was then transformed into electrocompetent Escherichia coli DH5α cells and grown for 18-20 hours on ampicillin-containing LB/agar plates at 35°C. We then screened the mutants with improved folding using a blue-light transilluminator. The bright red fluorescent colonies were picked from the library and grown in LB media supplemented with antibiotics. The plasmids were then purified and sequenced. These steps were repeated a few times until four unique mutations were acquired.

Plasmid construction
For construction of the Camuiα-mGsR and Camuiα-mGmC plasmids with mEGFP (monomeric enhanced Green Fluorescent Protein with the A206K mutation) and sREACh (super Resonance Energy-Accepting Chromoprotein) [22] or mCherry [23] pairs, the full-length CaMKIIα sequence was inserted into the modified pmEGFP-C1 with a linker coding the amino acid sequence SRLRSRA. Subsequently, sREACh or mCherry was subcloned into the Cterminal region of CaMKIIα with a linker coding the amino acid sequence GSGSGGSGGSG. For construction of the Camuiα mutants, a site-directed mutagenesis kit (Stratagene) was used. For construction of non-fluorescently labeled CaMKIIα mutants, calmodulin, and His-tagged association domain mutants, these DNA constructs were inserted into the modified pmEGFP-C1, replacing mEGFP. His-tagged association domain mutants for bacterial expression were constructed by inserting the domain mutants into a pRSET vector. For construction of the association domain mutants fused to super-folder GFP (sfGFP) [24], the association domain mutants were ligated with sfGFP into the pRSET vector, fusing the fragments to the Nterminus of sfGFP with the linker sequence GSG.

Protein purification from bacteria and refolding kinetics
For the purification of the His-tagged association domain mutants and their sfGFP fusion, proteins were overexpressed in E. coli DH5α cells and purified with a Ni + -nitrilotriacetate (NTA) column (HiTrap, GE Healthcare). The concentration of the purified proteins was measured by absorption spectroscopy using the extinction coefficient of sfGFP (A 489 = 83,000 cm -1 ÁM -1 ) [24] or Bradford protein assay (Biorad).
To measure the refolding kinetics of sfGFP-fused association domains, proteins were denatured and refolded as described previously [25]. Briefly, the purified proteins (3-10 μM) were dissolved in denaturation buffer (8 M urea, 1 mM dithiothreitol (DTT)), and heated at 95°C for 5 minutes to denature both the association domain and sfGFP. The refolding reactions were initiated by diluting the denatured protein with a 100-fold amount of renaturation buffer (5 mM KCl, 2 mM MgCl 2 , 50 mM Tris pH 7.5, 1 mM DTT) at room temperature. The fluorescence recovery was measured using a fluorescence spectrophotometer at 510 nm (RF-5300PC; Shimadzu).

Equilibrium unfolding measurement
To measure the tolerance of association domain for urea, His-tagged association domain mutants were purified and mixed with the various concentrations of urea in 25 mM HEPES, pH 7.3, containing 50 mM KCl and 5 mM DTT. The mixture was incubated overnight at 30°C. The samples were excited at 295 nm and the intrinsic tryptophan fluorescence spectra were recorded from 310 to 450 nm using a fluorescence spectrophotometer as described elsewhere [26].

Circular dichroism (CD) spectroscopy
To measure the CD spectra after refolding of association domain mutants, 20 μM of proteins were denatured with 8 M urea and 1 mM DTT, and were incubated for 1 hour at 37°C. The refolding reactions were initiated by diluting the denatured protein with a 5-fold dilution at room temperature. After 3 minutes of initiation, the CD measurement was carried out. CD spectra were recorded using a Jasco J-720WI spectropolarimeter (JASCO) at room temperature. All samples containing 4 μM of protein were measured in a buffer (20 mM phosphate, pH 7.3, 50 mM KCl) with 1-mm path lengths. The baseline spectra of buffer and denaturant without protein were subtracted from respective protein spectra, as described previously [26].

Dissociated culture of hippocampal neurons and transfection
Cultured hippocampal neurons were prepared as described elsewhere [27] with minor modifications. Briefly, postnatal day 1 rats were anesthetized and decapitated, followed by brain removal and hippocampal tissue dissection. Neurons triturated by papain treatment were plated at a density of 1.3 × 10 4 cells/cm 2 on polyethyleneimine-coated 35 mm culture dishes and maintained in serum-free Neurobasal medium supplemented with 2% B27, 2 mM Gluta-MAX-I, and 10 mM HEPES at 37°C and 5% CO 2 . After 12 days, neurons were transfected with 2 μg of either Camuiα-mGsR or Camuiα4m-mGsR plasmid using Lipofectamine 2000 (Invitrogen) according to the manufacture's protocol. After 18-24 hours, the neurons in the buffer (10 mM HEPES, 150 mM NaCl, 2.5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , 10 mM glucose, and Tris-base for adjusting to pH 7.4) were imaged under the epifluorescence microscope.

Two-photon fluorescence lifetime imaging and glutamate uncaging
Details of FLIM-FRET imaging using a custom-built 2pFLIM have been described previously [30]. Briefly, the mEGFP in Camuiα mutant was excited with a Ti-sapphire laser (Mai Tai; Spectra-Physics) tuned to 920 nm for HeLa cells and 960 nm for neurons, and was visualized through an objective lens (60×, 0.9 NA; Olympus). The scanning mirror (6210H; Cambridge Technology) was controlled with a PCI board (PCI-6110; National Instruments) and Scan-Image software [31]. The fluorescence photon signals were collected by a photomultiplier tube (H7422-40p; Hamamatsu) placed after the emission filter (FF01-510/84; Semrock). Fluorescence lifetime measurement was carried out using a time-correlated single photon counting board (SPC-150; Becker & Hickl) controlled with custom software [30]. For construction of the fluorescence lifetime image, the mean fluorescence lifetime at each pixel was translated to a color-coded image [32]. Analysis of the lifetime change for individual cells was carried out as described previously [16].

Protein purification from HeLa cell and Native-PAGE
For the purification of the His-tagged association domain mutants from HeLa cells, cells were transfected with the plasmids using Lipofectamine 2000 followed by a 24-hour incubation. The cells were lysed in lysis buffer (1% Triton X-100, 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10 mM MgCl 2 , 4 mM EDTA, 5% glycerol, 5 mM imidazole) and centrifuged. Supernatants were incubated with resin (Ni-Sepharose 6 Fast Flow; GE Healthcare) for 1 hour. Samples were washed three times with 25 mM imidazole and two times with 50 mM imidazole in wash buffer (25 mM HEPES, pH7.3, 50 mM KCl). Subsequently, proteins were released from resin using 500 mM imidazole in wash buffer. For native-PAGE, the purified proteins were incubated in a buffer (25 mM HEPES, pH7.3, 50 mM KCl, 1 mM DTT) for 1 hour at 37°C and mixed with 4 × sample buffer (0.2 M Tris, pH6.8, 40% glycerol, 2.5% bromophenol blue). Electrophoresis was carried out as described previously [33], and silver staining was performed using silver stain reagent kit (Cosmo Bio).

Results and Discussion
To improve the folding efficiency of the previously reported Camuiα [11,16], we decided to improve a domain of CaMKIIα, rather than the full-length protein, since CaMKIIα is over 50 kDa and is difficult to express in E. coli in general. CaMKIIα consists of a kinase domain and an association domain, and since the introduction of mutations in the kinase domain may change the enzymatic specificity of the protein, we focused on the association domain ( Fig. 1A). We used a modified version of the GFP folding reporter method in combination with error-prone PCR [21] to increase folding efficiency. As a reporter fluorescent protein, we used mCherry instead of GFP, since the chromophore maturation time of mCherry is faster [23,34], which facilitates the screening in a short time. The PCR products coding the association domain with random mutations were then fused to the N-terminus of mCherry and ligated into a bacterial expression vector, constructing a library (Fig. 1A). After screening more than 50,000 transformed colonies, we found several bright red colonies. After multiple iterations of these experiments, we found an extremely bright red colony, and the subsequent sequence analysis showed four substitutions, F394L, I419V, A430T, and I434T in the association domain ( Fig. 1B), but no mutations in mCherry. Since this method is based on the assumption that the productive folding of fluorescent protein is related to the characteristics of the upstream protein [21], the bright red colonies probably indicate the decreased chance for entanglement and aggregation between the upstream and downstream mCherry proteins [35].
To evaluate the effects of the individual mutations in the association domain on folding kinetics of the downstream fluorescent protein, we fused the respective mutants to the N-terminus of sfGFP and purified them using conventional His-tagged protein purification. The purified proteins were then denatured in urea [25], and the recovery of sfGFP fluorescence upon refolding treatment was monitored [21] (Fig. 2, Table 1). The fluorescence of the denatured wild-type association domain fused to sfGFP only recovered up to 21% with a half-time of 7 minutes. Similarly, the mutant containing three mutations (F394L, A430T, and I434T) showed only 25% recovery, but had a 2-fold improved half-time (3.5 minutes). In contrast to these results, the mutant containing four mutations exhibited a 2-fold higher recovery (46%) and a 4-fold faster half-time (about 2 minutes) compared with the wild-type, clearly indicating improved folding efficiency of the sfGFP (Fig. 2, Table 1).
Since improved sfGFP folding is expected to reflect the folding efficiency of the association domain [21], we measured the tolerance of association domain mutants for denaturing conditions and their refolding efficiency (Fig. 3). To measure their tolerance of the denaturing reagent, the unfolding of purified association domain mutants was monitored by tryptophan and acceptor fluorophores (sREACh) and full-length CaMKIIα, which contains kinase, regulatory, and association domains. For improving the folding efficiency of the sensor, the association domain was amplified by error-prone PCR and subcloned into a bacterial expression vector. Subsequently, the ligated DNA was transformed in Escherichia coli to make a library. Bright red colonies were isolated from the library and analyzed by DNA sequencing. The identified mutations were then incorporated into the FRET sensor, leading to an increase in its folding efficiency. This scheme is based on an assumption that the folding efficiency of the association domain is coupled to that of mCherry.   fluorescence in the presence of various concentrations of urea ( Fig. 3A-C). A concentration of up to 3 M urea resulted in slight change in the wavelength maxima. Beyond 3 M urea, the fluorescence of wild-type and mutant association domains started decreasing at 3.5 and 3 M urea with a concomitant red shift in spectra, respectively, suggesting that the mutant association domain has slight lower tolerance to urea (Fig. 3C). With further addition of urea (>5 M), the emission peak shifted towards 353 nm with a further decrease (up to 65%) in intensity because of the presence of an unfolded structure with a tryptophan in the association domain exposed to hydrophilic conditions (Fig. 3C). Next, to test if the refolding efficiency of the association domain mutant is improved compared with the wild-type, we measured the reversibility of the urea-induced unfolding using far-UV CD spectra measurement (Fig. 3D, E). The purified association domain was denatured and unfolded by reducing the urea concentration. The spectra of unfolding and refolding states were similar for the wild-type and mutant, suggesting that the contents of α-helix and β-sheet are similar. (Fig. 3D, E). Since the mutant association domain has slight lower tolerance to urea compared with wild-type ( Fig. 3A-C) and similar refolding characteristics (Fig. 3D, E), improved sfGFP folding may not be due to the improved folding of the association domain mutant, but rather the reduced interaction of the mutant association domain with the fluorescent protein in the intermediate state.
To test the effects of the identified mutations for the expression pattern of Camuiα, we introduced these mutations into mEGFP and sREACh-based Camuiα, Camuiα-mGsR, and then expressed the sensors in HeLa cells (Fig. 4). Interestingly, HeLa cells expressing Camuiα-mGsR consisted of two populations: one that exhibited aggregated Camuiα-mGsR fluorescence signals in a dotted pattern (designated AG cells), and another that exhibited evenly distributed Camuiα-mGsR fluorescence (Fig. 4A). The fraction of cells, expressing wild-type Camuiα-mGsR that were determined to be AG cells, was approximately 16.6%, which is not a negligible fraction (Fig. 4B). Similar results were observed for the mutants with single or double mutations (Fig. 4B). Notably, the aggregation of Camuiα-mGsR did not localize to the mitochondria, endoplasmic reticulum, Golgi body, or lysosome (data not shown). Further, the introduction of three mutations (F394L, A430T, and I434T) in the association domain slightly reduced the population of AG cells to 10.3% compared with the wild-type; this is still a large fraction. However, the introduction of all four mutations (F394L, I419V, A430T, and I434T) dramatically reduced the fraction of AG cells to 0.8%, improving expression pattern. Therefore, we have named this new Camuiα construct, Camuiα4m-mGsR. The improved expression pattern of Camuiα4m-mGsR was also observed in HeLa cells co-expressing CaMKIIα (Fig. 4B) and hippocampal neurons endogenously expressing CaMKIIα (Fig. 5). One possible explanation for the improved expression pattern of Camuiα4m-mGsR is the reduced expression level of Camuiα4m-mGsR, since AG cells tend to exhibit a high level of expression of Camuiα-mGsR (Fig. 4C). However, the similarly high expression of Camuiα4m-mGsR did not show aggregation (Fig. 4C).
Next, to evaluate the sensitivity of the sensor, we compared the FRET signals of Camuiα4m-mGsR and Camuiα-mGsR in HeLa cells using 2pFLIM [22,29]. In this experiment, Camuiα4m-mGsR or Camuiα-mGsR was co-transfected with calmodulin that is not expressed  at high enough levels in HeLa cells, but is required for CaMKII activation. First, Camuiα-mGsR activation in individual cells was monitored (Fig. 6). Because HeLa cells expressing Camuiα-mGsR comprised two populations-AG cells and the cells evenly expressing the sensor, as mentioned in the previous section-we analyzed cells from each population separately (Fig. 6F, G). The cells that exhibited uniform expression showed robust activation after ionophore stimulation (cells 3, 4, and 5 in Fig. 6C, F), whereas the AG cells showed significantly decreased levels of response signal, most likely owing to the unfavorable aggregation of Camuiα-mGsR (cells 1 and 2 in Fig. 6C, G). Next, we measured the FRET signal of Camuiα4m-mGsR after ionophore stimulation (Fig. 6H), and compared these results with the wild-type cells evenly expressing Camuiα-mGsR (Fig. 6F). Interestingly, the cells expressing Camuiα4m-mGsR showed robust FRET signals with remarkably minimized standard deviation  (mean ± SD at 7 minutes was 0.153 ± 0.016, Fig. 6H), compared with Camuiα-mGsR (mean ± SD at 7 minutes was 0.174 ± 0.056, Fig. 6F). These data indicate that Camuiα4m-mGsR has a more stable response to the stimulation compared with that of Camuiα-mGsR (Fig. 6F, H). No activation was observed in either cell population when DMSO was added (Fig. 6I, J, K), suggesting that the activation is due to the Ca 2+ influx induced by the ionophore. The decreased response variability of cells expressing Camui4mα-mGsR is due to the smaller fluorescence lifetime variability in both before and after stimulation (Fig. 6L-N). The similar results were obtained from the experiments with co-expression of CaMKIIα (Fig. 7). Unexpectedly, the average lifetime change is slightly larger for Camuiα-mGsR (0.174 ns) compared with Camuiα4m-mGsR (0.153 ns), which is due to the smaller basal fluorescence lifetime of Camuiα-mGsR (Fig. 6L, N).
We also carried out the same set of experiments with mCherry-based Camuiα mutants, and found that the response variability of Camuiα4m-mGmC was smaller than that of Camuiα-mGmC, similar to the results for the sREACh-based Camuiα (Fig. 8). However, the sREACh Among the identified four mutations, F394L, A430T, and I434T face the binding regions of an adjacent association domain (Fig. 1B) [36]. This fact raises the possibility that these mutations may inhibit proper association between the CaMKII subunits. Therefore, to test if oligomerization was hindered, we co-transfected HeLa cells with Camuiα4m-mGsR and CaMKIIα and immunoprecipitated the proteins using an anti-GFP antibody. Western blotting revealed the presence of two bands corresponding to the molecular weights of Camuiα4m-mGsR and CaMKIIα (Fig. 9). In addition, the ratio of the band intensities of Camuiα4m-mGsR and CaM-KIIα was similar to that of Camuiα-mGsR and CaMKIIα, suggesting that Camuiα4m-mGsR is incorporated into CaMKIIα in a similar way to Camuiα. However, when CaMKIIα4m and Camuiα4m-mGsR were co-expressed and immunoprecipitated, the interaction between CaM-KIIα4m and Camuiα4m-mGsR was dramatically decreased, suggesting that the association domain with four mutations does not form a homo-oligomer (Fig. 9A, B). Consistent results were obtained by monitoring the inter-molecular FRET in living HeLa cells using 2pFLIM (Fig. 10). These facts could explain that the reason for the smaller basal fluorescence lifetime of Camuiα in Figs. 6 and 8 may be due to the inter-molecular FRET between Camuiα sensors, further increasing basal FRET, because Camuiα association domain forms tight oligomer compared with association domain with four mutations (Figs. 9, 10). Next, to identify the mutation that hinders the oligomeric formation, we carried out native-PAGE analysis and found that F394L and A434T hinder oligomer formation (Fig. 11). Furthermore, we tested the Ca 2+ /calmodulin-dependent autophosphorylation capabilities of Camuiα4m-mGsR in HeLa cells by western blotting with anti-phospho antibody. After stimulation of the cells with ionophore, Camuiα4m was shown to autophosphorylate in a similar manner to the wild-type Camuiα-mGsR (Fig. 9C,  D), suggesting that Camuiα4m-mGsR is activated in a similar way.
Since CaMKIIα is specifically expressed in neurons, Camuiα4m-mGsR will be used in neurons for monitoring CaMKIIα activity. Therefore, we tested if Camuiα4m-mGsR expression alters the dendritic spine density of hippocampal neurons, compared with Camuiα-mGsR, and found that there was no significant difference between Camuiα-mGsR and Camuiα4m-mGsR (Fig. 12), showing that there is no unfavorable effect on the neuronal morphology. Furthermore, we monitored Camuiα4m-mGsR activity during spine enlargement upon local glutamate uncaging [37], and found that both activity and spine volume changes are similar to those of Camuiα-mGsR (Fig. 13). In contrast, Camuiα4m-mGsR with T286A mutation which abolishes autophosphorylation showed the decreased activation and spine volume change (Fig. 13), consistent with the previously reported result with Camuiα [15,16]. Taken together, these results indicate that Camuiα4m-mGsR may be useful for monitoring CaMKII activity.

Conclusions
We have demonstrated here that molecular evolution of the association domain in the Camuiα FRET sensor successfully improves its expression pattern and minimizes response variability (Figs. 4,5,6,7,8). While a significant fraction of cells expressing Camuiα-mGsR exhibited aggregation, the cells expressing Camuiα4m-mGsR did not (Fig. 4). The reason for the improved expression may be owing to the improved folding kinetics of the downstream fluorescent protein (Fig. 2), not owing to the improved folding of the association domain itself (Fig. 3). Unexpectedly, the introduction of four mutations (F394L, I419V, A430T, and I434T) hinders homo-oligomer formation, but not hetero-oligomer formation between wild-type and mutant association domains (Figs. 9, 10, 11). One possible explanation for the improved expression pattern of Camuiα4m-mGsR is its inability to form oligomers. However, since the introduction of a single mutation, namely F394L or I434T, disrupts oligomerization but does not improve the expression pattern (Figs. 4, 11), the inability to form oligomers may not be the reason for the improved expression pattern. Thus, we speculate that the decreased large aggregation of Camuiα4m is due to the decreased non-specific binding between inter-association domains and between association domain and GFP, sREACh, and mCherry in the folding-intermediate state [35].
FRET experiments confirmed that Camuiα4m-mGsR exhibits a more robust response with smaller response variability than Camuiα-mGsR. This minimized response is probably due to the minimized variability of both the basal fluorescence lifetime and the response signal (Figs. 6,8). Another potentially important reason for the decrease in variability may be an effect of the increased folding efficiency of the C-terminal fluorescent protein in Camuiα4m mutant (Fig. 2). Since the folding efficiency of sfGFP fused to the C-terminus of the association domain was greatly improved in terms of refolding rate and correctly refolded fraction (Fig. 2, Table 1), the folding efficiency of the C-terminal fluorescent proteins, sREACh and mCherry, may also be improved and could contribute to the minimized cell-to-cell variability of fluorescence lifetime. The Camuiα4m-mGsR developed here exhibits a robust response and can be utilized to measure the CaMKIIα activity more precisely in vitro and in vivo. Furthermore, the directed mutagenesis strategy used here may be useful for improving the quality of all FRET sensors in terms of response variability.