Effect of Cavβ Subunits on Structural Organization of Cav1.2 Calcium Channels

Background Voltage-gated Cav1.2 calcium channels play a crucial role in Ca2+ signaling. The pore-forming α1C subunit is regulated by accessory Cavβ subunits, cytoplasmic proteins of various size encoded by four different genes (Cavβ1 - β4) and expressed in a tissue-specific manner. Methods and Results Here we investigated the effect of three major Cavβ types, β1b, β2d and β3, on the structure of Cav1.2 in the plasma membrane of live cells. Total internal reflection fluorescence microscopy showed that the tendency of Cav1.2 to form clusters depends on the type of the Cavβ subunit present. The highest density of Cav1.2 clusters in the plasma membrane and the smallest cluster size were observed with neuronal/cardiac β1b present. Cav1.2 channels containing β3, the predominant Cavβ subunit of vascular smooth muscle cells, were organized in a significantly smaller number of larger clusters. The inter- and intramolecular distances between α1C and Cavβ in the plasma membrane of live cells were measured by three-color FRET microscopy. The results confirm that the proximity of Cav1.2 channels in the plasma membrane depends on the Cavβ type. The presence of different Cavβ subunits does not result in significant differences in the intramolecular distance between the termini of α1C, but significantly affects the distance between the termini of neighbor α1C subunits, which varies from 67 Å with β1b to 79 Å with β3. Conclusions Thus, our results show that the structural organization of Cav1.2 channels in the plasma membrane depends on the type of Cavβ subunits present.


Introduction
Voltage-gated Ca v 1.2 calcium channels react to membrane depolarization by creating a rapid and transient increase in intracellular free Ca 2+ concentration, thereby playing an essential role in initiation of calcium signaling in a wide variety of cells [1]. In order to exhibit this function, Ca v 1.2 calcium channels require association of the pore-forming a 1C subunit with accessory Ca v b and a 2 d subunits as well as calmodulin. Calcium channels are clustered rather than evenly distributed along the surface membrane of neurons [2][3][4] and cardiac myocytes [5][6][7]. Singlemolecule imaging of the functional recombinant EYFP N -a 1C /b 2a / a 2 d channels revealed clusters composed of ,40 channels [8]. In neuronal cell bodies and proximal dendrites in hippocampus and cerebral cortex, Ca v 1.2 clusters of 1.5-2 mm in diameter were observed with anti-a 1C antibody [9]. Using electron microscopy in bird and amphibian cardiac muscle [5,6] and immuno-gold labeling in mammalian ventricular myocytes [7,10] it was shown that Ca v 1.2 clusters are loosely tethered to ryanodine receptors (RyR) of the sarcoplasmic reticulum. Although association of calcium channels and ryanodine receptors appears to be weaker in cardiac myocytes than in skeletal muscle [11] and may involve different mechanisms of coupling [12], Ca v 1.2 clustering is essential for excitation-contraction coupling [13,14].
Little is known about the factors affecting the structure of Ca v 1.2 clusters or the mechanisms of their formation. Because the carboxyl-terminal ''IQ'' region of a 1C mediates the calmodulindependent Ca 2+ -induced inactivation of the channel [15][16][17][18], it is reasonable to suggest that both calmodulin and the cytoplasmic 750-amino acid C-tail of a 1C have a role in the formation and maintenance of the Ca v 1.2 clusters. Indeed, a splice variant of a 1C (a 1C,86 ) deprived of IQ does not show a distinct tendency to form clusters [19]. The role of IQ sequences in intermolecular interactions between neighboring a 1C molecules was experimentally confirmed in recent diffraction study [20]. The impact of bulky cytoplasmic Ca v b subunits on Ca v 1.2 structure and clustering is not known. Ca v b subunits are important differential modulators of the electrophysiological properties of calcium channels [21][22][23]. These peripheral proteins of variable size are encoded by four different genes (Ca v b 1 -b 4 ), some of them being subject to alternative splicing [24]. They have a common binding site in the cytoplasmic linker between repeats I and II of a 1C known as the a-interaction domain (AID) [25]. Here, we applied total internal fluorescence reflection (TIRF) and three-color FRET microscopy to assess the effects of Ca v b on cluster size and density of Ca v 1.2 as well as to measure inter-and intramolecular distances between the N-and C-termini of a 1C and the N-tails of b 1b , b 2d and b 3 . Our results demonstrated that Ca v 1.2 channels form plasma membrane clusters and revealed the effect of the type of Ca v b present on molecular distances and packing of the channels.

Results
Differential effect of Ca v b subunits on cluster organization of Ca v 1.2 channels Ca v 1.2 calcium channels retain functional activity after fusion of fluorescent proteins to the N-and C-termini of a 1C and to the Nterminus of Ca v b. In our experiments, we labeled a 1C at the N-tail with monomeric mVenus (Va 1C ) and/or at the C-tail with monomeric mCerulean (a 1C C) [26]. To investigate the effect of Ca v b subtype on size and density of Ca v 1.2 clusters, we chose three major Ca v b variants, neuronal/cardiac b 1b [27], cardiac b 2d [28,29] and neuronal/cardiac/vascular b 3 [30][31][32], which is the predominant Ca v b subunit in vascular smooth muscle cells [31,33]. The more commonly used b 2a was excluded from the experiments because its N-tail is palmytoylated and anchored to the inner leaflet of the plasma membrane.
Fluorescent microscopy is a convenient approach to detect clusters of recombinant calcium channels as fluorescent foci or groupings of labeled molecules [34]. In this study, we used TIRF microscopy to visualize Ca v 1.2 clusters on the basal plasma membrane. Wavelet transform was used for the detection of clusters (see Methods and Figure 1A) to estimate the effect of the type of Ca v b present on the Ca v 1.2 clusters size ( Figure 1B) and density (defined here as number of clusters per mm 2 of the plasma membrane, Figure 1C). The smallest Ca v 1.2 clusters were observed with b 1b present. Ca v 1.2 clusters were significantly (P,0.001) larger with b 2d (by ,20%) and b 3 present (by ,30%) ( Figure 1B). We also found that the average density of the Va 1C / b 1b clusters in the plasma membrane was 2.5 times higher (P,0.01) than Va 1C /b 3 , with b 2d again taking an intermediate value ( Figure 1C). Thus, Ca v b subunits differentially regulate the architecture of the Ca v 1.2 clusters.
In principle, the close proximity of channels within a cluster may generate intermolecular FRET between the V and C fluorophores of neighboring Va 1C C channels. This intermolecular FRET should be absent outside of clusters, where only intramolecular FRET should occur. The Va 1C C/a 2 d/b 3 channel was expressed in COS1 cells and two-color TIRF-FRET was measured inside and outside of clusters identified by wavelet transform. Based on FRET efficiency, a V-C distance (r)-frequency histogram of the total number of pixels within clusters revealed a possible bi-modal distribution, where a second (intermolecular) component of FRET is seen within clusters ( Figure 2A) but not outside of the clusters ( Figure 2B). Because TIRF microscopy captures only a small fraction of the cell plasma membrane, we used epifluorescent three-color FRET microscopy to quantitatively analyze the effect of Ca v b subtype on inter-and intra-molecular distance of Ca v 1.2 channels.
The type of Ca v b present does not affect intramolecular distance between the N-and C-termini of the a 1C subunit We investigated the effect of Ca v b subtype on molecular distances in Ca v 1.2 channels by three-color FRET between Va 1C C and tagRFP (Rb) fused to N-termini of b 1b , b 2d and b 3 . The advantage of three-color FRET cell microscopy applied to multisubunit complexes is that the method simultaneously detects the relative arrangement of the three different fluorophores (C, V and R) at a distance #26R o , where R o is the Förster radius (R o(C-V) = 53 Å ; R o(V-R) = 58 Å ; R o(C-R) = 51 Å ). Both mCerulean and mVenus are close analogs of GFP and can be approximated by a cylinder of 32648 Å [35]. However, tagRFP [36], a monomeric analog of eqFP611, is larger in size and can be approximated by a cylinder of 34654 Å [37]. Use of monomeric forms of fluorescent proteins excludes artifacts due to dimerization after expression [38]. The labeled constructs were co-expressed with a 2 d in two different combinations as shown in Figure 3, and three-color FRET was measured using a multicube system [39]. Although membrane potential was not controlled during experiments, it was found to be on average 210.063.3 mV (n = 5) indicating that channels were predominantly in an inactivated state. In each fluorescent cell image, region of interest (ROI) was determined using a standard procedure as described earlier [40]. Within this ROI, only pixels with donor/acceptor ratio from 0.2 to 5 ( Figure 4, left panels) were selected for further analysis [41]. FRET efficiency was determined according to [42] (Figure 4, middle panels) and then converted to the distance (r) between donor and acceptor (right panel) according to [43].
Results of our measurements revealed that the tested Ca v b subunits did not affect intramolecular distance between the N-and C-termini of a 1C . Measurement of FRET in the double-labeled Va 1C C co-expressed with a 2 d and Rb 1b , Rb 2d or Rb 3 showed that the intramolecular distance r C-V between V and C did not vary significantly and was on average 68-69 Å , independent of the type of co-expressed Ca v b (Figure 4, A-C; see Table 1 for statistics).
Estimation of r C-V in the absence of Ca v b was not possible because of poor plasma membrane targeting by Va 1C C/a 2 d under such conditions. To overcome this problem, we co-expressed Va 1C C and a 2 d with tagRFP-labeled b 2 CED, a 42-amino acid fragment of b 2 subunits which does not bind to AID, but interacts with the IQ region of the a 1C subunit C-terminus, facilitates voltage gating and stimulates surface expression of the channel [44]. Results of FRET measurements showed r C-V = 6861 Å (n = 22), essentially the same distance as that estimated when AID was occupied by Ca v b. Taken together, these results of our study suggest that type of Ca v b subunits present does not significantly affect the intramolecular distance between the N-and C-termini of a 1C in Ca v 1.2 calcium channels.
Intermolecular distance between the a 1C subunit N-and C-termini depends on the type of Ca v b present Fitting of FRET data obtained with b 2d and b 3 to a sum of two Gaussian distributions (Table 1) revealed a statistically significant second component of Va 1C C FRET ( Figure 5). Arising from neighboring Va 1C C molecules, this FRET provided estimates for the intermolecular distances (r C,V ) that were significantly different for b 2d (7263 Å , n = 5) and b 3 (7763 Å , n = 6). To verify our intermolecular distance measurements, we co-expressed a mixture of Va 1C and a 1C C along with Rb 1b , Rb 2d or Rb 3 (Figure 4, D-F). Any FRET between V and C in this recombinant system must be intermolecular FRET between termini of neighboring channels. Results, presented in Table 1, showed that intermolecular distances r C,V measured in these complexes with b 2d (7262 Å , Figure 2. Intramolecular vs. intermolecular FRET in Va 1C C revealed in TIRF images. Va 1C C, a 2 d and b 3 were co-expressed in COS1 cells. Two-color FRET was measured in TIRF images and converted into distances r between V and C as described in Methods. Shown are normalized cumulative histograms (n = 11) for r calculated for ROI inside clusters (A, total number of pixels m = 231) and outside clusters (B, m = 3908) identified by wavelet transform. The same intramolecular (r V-C ) distance <6.8 nm (light gray bars) was observed both inside and outside clusters, while intermolecular (r V,C ) distance <8.1 nm was observed only in clusters (dark gray). doi:10.1371/journal.pone.0005587.g002  Table 1. Intra-and intermolecular distances between the Ca v 1.2 calcium channel a 1C and b subunits measured by three-color FRET microscopy.
FRET efficiency between the indicated fluorophores fused to the a 1C and b 1b , b 2d or b 3 subunits was measured in the plasma membrane of expressing COS1 cells and fitted to a Gaussian function. In cases when the routine curve fit showed two significantly different Gaussian distributions, the corresponding dispersion coefficients c  FRET efficiency (middle column) and distance (right column). Relative frequency was calculated for total number of pixels in ROI as described in Methods. The red solid line is the best fit to a Gaussian distribution with indicated means for r V-C and r V,C . doi:10.1371/journal.pone.0005587.g004 With b 1b , the intermolecular distance r C,V measured between Va 1C and a 1C C was 6761 Å (n = 26), a value not significantly different from the estimate for intramolecular Va 1C C distance (r V-C = 6861 Å , n = 17). This explains why the data obtained in the presence of b 1b were best fitted by a single Gaussian distribution. Thus, unlike b 2d and b 3 , in the presence of b 1b the inter-and intramolecular distances appear to be similar. The measurements with a mixture of Va 1C and a 1C C confirm that Ca v 1.2 calcium channels containing b 1b , b 2d or b 3 subunits are in close proximity to each other, thus supporting their clustering in the plasma membrane. The distance r C,V between the N-and C-termini of the neighbor a 1C subunits depends on the type of Ca v b. In the presence of b 1b , the distance r C,V (6761 Å ) is 1.2 nm smaller (P,0.002) than with b 3 (7964 Å ), while r C,V estimated in the presence of b 2d (7262 Å ) is of an intermediate value. Subsequent measurements of three-color FRET between Rb and the fluorophores of the a 1C subunit added more certainty to this general picture ( Figure 6A).

FRET between tagRFP-labeled Ca v b and mCerulean/ mVenus-labeled a 1C
The three Ca v b subunits selected for our study vary in molecular mass (b 1b , 53.2 kDa; b 2d , 73.5 kDa; b 3 , 54.5 kDa) and in the size of the variable N-terminal (V1), central (HOOK) and C-terminal (V2) regions (see Figure 6B). There are large differences between the three Ca v b subunits in variable regions on both sides of the AID-binding pocket, which anchors Ca v b to the I-II linker of a 1C ( Figure 6A). In spite of that, the intramolecular distance r V-R between Rb and Va 1C estimated in all tested threecolor FRET combinations, including single-or double-labeled a 1C (Va 1C +a 1C C+Rb, Va 1C C+Rb), was not significantly different for all tested Ca v b subunits except for Rb 2d (see 1 in Table 1) Although the average distances r C-R between Rb and Va 1C C were significantly different for Rb 1b (8562 Å , n = 13) and Rb 3 (7961 Å , n = 14), they were not significantly different between Rb and a 1C C. A superposition of all three simultaneously measured arrangements between Rb and Va 1C C ( Figure 6C) illustrates differences in the positions of Rb subunits as reflected by statistically significant differences in r V-R and r C-R (Table 1).
Fitting to a sum of two Gaussian distributions did not reveal the second (intermolecular) component of FRET between Va 1C C and Rb 1b (Table 1). However in the case of Rb 3 two intermolecular FRET components were clearly observed, one corresponding to the distance r V,R = 10062 Å (in 15 of 19 cells) and the other corresponding to r C,R = 8561 Å (in 10 of 14 cells). In the presence of Rb 2d , the latter component was not observed (n = 13), suggesting that the related distance r C,R exceeded 102 Å . However, intermolecular FRET between Va 1C and Rb 2d was distinctly revealed in 3 out of 8 cells in a range close to the limits of resolution of the method with an estimate of r V,R = 10761 Å (Table 1). Taken together, FRET measurements between Rb and the labeled tails of Va 1C C corroborated data on intermolecular FRET obtained with Va 1C +a 1C C+Rb and demonstrated that (a) calcium channels are in close proximity in the plasma membrane, and (b) both the intra-and intermolecular architecture of Ca v 1.2 channels depend on the type of Ca v b present.

Discussion
Ca v 1.2 calcium channels initiate Ca 2+ signal transduction to many different downstream targets in wide variety of cells. Investigation of factors affecting structural organization of Ca v 1.2 channels is crucial for better understanding the mechanisms of Ca 2+ signaling. The tendency of Ca v 1.2 channels to form clusters in the plasma membrane of different cell types has been poorly investigated. Here we studied effects of three major Ca v b subunits on structural organization of recombinant Ca v 1.2 channels expressed in COS1 cells. Because untransfected COS1 cells do not express endogenous calcium channels, they lack natural intracellular partners (e.g., cardiac RyR2) in proximity of exogenous Ca v 1.2 channels that might promote their clustering through ''junctional'' coupling [45]. However, recombinant Ca v 1.2 channels expressed in COS1 cells establish functional coupling to CREB-dependent transcriptional activation [46], pointing to a physiologically relevant integration of recombinant Ca v 1.2 into a naturally occurring signaling cascade with Ca 2+ / calmodulin-dependent protein kinase II mediating this activity in native cells [47].
TIRF microscopy revealed clusters of recombinant Ca v 1.2 channels in the plasma membrane of COS1 cells. The size and the plasma membrane density of the clusters significantly depend on the type of Ca v b present. This important observation suggests that the type of Ca v b present determines the structure of the Ca v 1.2 clusters. The average cluster size varies from 360 (b 1b ) to 450 nm 2 (b 3 ). Corroborating reasonable dimensions of these values, a mean size of the Ca v 1.2 cluster with the major cardiac b 2d (430 nm 2 ) is within the estimated size range (250-560 nm 2 ) of rat ventricular RyR2 clusters [48].
Relative arrangement of a 1C and Ca v b was estimated with subnanometer precision using three-color FRET microscopy in live cells with calcium channels in a stable, inactivated state. Our study revealed that in spite of substantial differences in molecular structure ( Figure 6B), the intramolecular distance between the a 1C subunit tails does not significantly depend on the type of Ca v b present. Relative position of Rb 1b , Rb 2d and Rb 3 did not differ significantly. This is interesting because, unlike b 1b and b 3 , b 2d has a C-terminal b 2 CED domain, which interacts with the IQ region of the a 1C C-tail [44].
Another important observation is that N-and C-termini of a 1C and N-termini of Ca v b subunits of neighbor channels are in close (,120 Å ) proximity to each other, which corroborates with the tendency of Ca v 1.2 to form clusters. Intermolecular distance between the a 1C subunits significantly depends on the type of Ca v b and increases from 67 Å in the presence of b 1b to 79 Å with b 3 . Measurements of FRET between Rb and neighbor V/C-a 1C supported this general picture and showed a significant effect of the type of Ca v b present on the relative position of neighbor channels.
Interestingly, freeze-fracture of the surface membrane revealed that distances between Ca v 1.2 channels trapped in cardiac junctions with RyR2 is variable and, on average, are larger than those identified by FRET [49]. It is known that the cytoskeleton and RyR2 associate with Ca v 1.2 plasma membrane clusters in heart cells [50]. Thus, one can not exclude that the distance between Ca v 1.2 channels in clusters in cardiac junctions is affected by RyR2. However, it is not clear whether clustering affects the ability of Ca v 1.2 channels to initiate Ca 2+ signaling and whether every channel is responsive to depolarizing stimuli. In cardiac muscle cells, a single Ca v 1.2 opening triggers activity of 4-6 RyR2 [51]. The average size of a RyR2 cluster in ventricular myocytes plasma membrane is 250 nm 2 (,100 RyR2 molecules) [48] and interaction between Ca v 1.2 and RyR2 is weaker than that between Ca v 1.1 and RyR1 in skeletal muscle. Thus, activation of a RyR2 cluster may be mediated by random opening of few Ca v 1.2 channels in clusters located at a larger distance than that estimated by FRET.  [37] and 1t0j [62], respectively. FRET measurements with ECFP-labeled plekstrin homology domain in the inner leaflet of the plasma membrane [40,63] showed that the N terminal tags of both the a 1C and Ca v b subunits are located within the 26 Fö rster distance (,100 Å for ECFP/ EYFP) from the plasma membrane. Little is known about molecular determinants underlying physiologically important cluster organization of Ca v 1.2 channels in neurons [52]. It was shown recently that scaffolding proteins (AKAP79/150 and PDZ) participating in organizing plasma membrane signaling complexes in neurons were not responsible for organizing Ca v 1.2 channel clusters [53]. The involvement of Ca v b in Ca v 1.2 channel cluster organization, identified in our study, does not contradict the earlier report that the calmodulinbinding IQ region of a 1C has a role in Ca v 1.2 clustering [19]. Because Ca v bs interact with IQ [23,44], it is possible that both act as concerted determinants in Ca v 1.2 channel clustering.
In conclusion, our study revealed effects of Ca v b subunits on the structural organization of Ca v 1.2 calcium channel in the plasma membrane in the absence of ''junctional'' interactions. It remains to be seen whether the observed differences in the cluster packing and arrangement of Ca v 1.2 contribute to the observed differences in calcium signaling among the cell types with preferential expression of a certain type of Ca v b [54][55][56].

Materials and Methods
Labeling a 1C subunit with mVenus and/or mCerulean To avoid dimerization, only monomeric forms of fluorescent proteins were used. The C-terminus of human Ca v 1.2 calcium channel a 1C,77 subunit was amplified by PCR with sense 59-ctattgaattcgatatcTGCCAGCAGCCTGGTGGAAGCG-39 and antisense 59-gtattaccggtggCAGGCTGCTGACGTAGACCCTGC-39 primers. The PCR fragment was cleaved with ECoRI and AgeI and incorporated into an mCerulean-N1 [57] vector cleaved with the same enzymes, and the 59-ECoRV/NotI-39 fragment from the resulting plasmid was then incorporated into a 1C,77 -pCDNA3 cleaved with AleI and NotI, resulting in the mCerulean C -a 1C,77 -pCDNA3 plasmid coding for a 1C C. The 59-NdeI/KpnI-39 fragment from mVenus-C1 vector [26] was incorporated into a 1C,77 -pCDNA3 and mCerulean C -a 1C,77 -pCDNA3 cleaved with the same enzymes to yield mVenus N -a 1c,77 -pCDNA3 and mVenus N -mCerulean C -a 1c,77 -pCDNA3, respectively, coding for Va 1C and Va 1C C.

Labeling of Ca v b subunits with monomeric fluorescent tags
The cDNA of human b 1b and b 3 subunits was cloned from a human heart mRNA (Promega) by a nest RT-PCR strategy. For b 1b , 59-GACGGGCAGGGCGCCCACTAC-39 was used as primer for the reverse transcription, sense 59-GAGGCTCCTCTCCA-TGGTCCAG-39 and antisense 59-CCACTACATGGCATGT-TCCTGC-39 primers were used for the first round PCR, sense 59-GCCACCATGGTCCAGAAGACCAG-39 and antisense 59-CAC-TACATGGCATGTTCCTGCTC-39 primers were used for the second round PCR. For b 3 , primer 59-CGCCTGTGCCT-GCCAGGGTAGGGCAGCAGG-39 was used for the reverse transcription, sense 59-GACTCCCCCATGTATGACGAC-39 and antisense 59-GGCTGTCAGTAGCTATCCTTG-39 primers were used for the first round PCR, sense 59-GCCACCATGTAT-GACGACTCC-39 and antisense 59-TGTCAGTAGCTATC-CTTGGGC-39 primers were used for the second round PCR. The cDNA was cloned into a TA cloning vector pCR 2.1 (Invitrogen) and confirmed by DNA sequencing. The 59-EcoRV/ BamHI-39 fragment of a b 1b TA clone was incorporated into the pTagRFP-C vector (Evrogen, Moscow, Russia), which was cleaved with XhoI, filled in with Klenow and then cleaved with BamHI to generate RFP-b 1b (Rb 1b ). In a similar way the 59-XhoI/HindIII-39 fragment of a b 3 -TA clone was incorporated into the pTagRFP-C vector to generate monomeric Rb 3 . To prepare RFP-b 2d , b 2d was amplified by PCR using mVenus-b 2d [44] as template with sense primer 59-CGGAGATCTATGGTCCAAAGGGACATGTC-39 and antisense primer 59-GGGGTCGACTCATTGGGGGATG-TAAACATC-39, and then the PCR product was cleaved with BglII and SalI, and incorporated into the pTagRFP-C vector cleaved with the same enzymes.

Imaging
Images were recorded with a pixel size of ca. 200 nm using a 14-bit Hamamatsu C9100-12 digital camera (Hamamatsu City, Japan) mounted on a Nikon TE2000 epifluorescent microscope (Tokyo, Japan) equipped with a 6061.45 numerical aperture (n.a.) oil objective and multiple filter sets (Chroma Technology, Rockingham, VT). Excitation light was delivered by a 175 W xenon lamp. Excitation filter sets were changed by a high-speed filter wheel system (Lambda 10-2, Sutter Instrument, Novato, CA). The Dual-View system (Optical Insights, Santa Fe, NM) was used for the simultaneous acquisition of two fluorescence images (donor and FRET). Images were collected and analyzed using C-Imaging (Compix, Cranberry Township, PA) and MATLAB v.7.0.4 (The Mathworks, Natick, MA).
Clusters within TIRF images were identified using 2D continuous wavelet transform similar to [58]. Images were analyzed using a two-dimensional mexican hat wavelet over scales 0.5 through 2 to identify ROI of locally increased signal fluorescence up to 5 mm 2 in area. Similar approaches have been employed for cluster detection in clinical and cell biology imaging [46,59,60]. Corrected FRET intensity was calculated from data acquired using the three filter sets (CFP, YFP, and FRET) as described previously [40] using MATLAB. Briefly, corrected FRET values (FRET c ) were calculated according to where a and b are bleedthrough coefficients and I FRET , I d and I a are FRET, donor and acceptor intensities.
Measurement of the G factor, which relates the increase in sensitized acceptor emission to the loss of donor fluorescence (quenching), is critical for calculating FRET efficiency (E) using the three-filter cube method. G factor is a constant for a particular fluorophore pair and imaging setup [42]. This method requires preparation of cDNA constructs encoding donor-acceptor fusion fluorescent proteins differing as widely as possible in FRET efficiency. This was accomplished by varying the length and composition of the linker residues connecting mCerulean and mVenus, mCerulean and tagRFP or mVenus and tagRFP. G factor was determined as where I aa1 , I dd1 and F c1 are acceptor, donor and corrected FRET intensity of the construct with the shortest linker between donor and acceptor, and I aa2 , I dd2 and F c2 are acceptor, donor and corrected FRET intensity of the construct with the longest linker between donor and acceptor. Using this formula, we found G factors of 1.81 for the mCerulean/mVenus pair, 1.30 for the mVenus/tagRFP pair, and 0.38 for the mCerulean/tagRFP pair. These G factor values allowed us to calculate FRET efficiency according to [42] as follows: The distance between two fluorophores was calculated in accordance with Förster theory: The Förster distances (R 0 ), the characteristic distance where the FRET efficiency is 50%, was calculated according to [61]: where Q D is the donor quantum yield, e A is the maximal acceptor extinction coefficient, and J(l) is the spectral overlap integral between the normalized donor fluorescence and the acceptor excitation spectra. All these parameters were calculated based on data obtained from Evrogen for tagRFP and reported for mCerulean and mVenus in [61]. Other parameters included coefficient C = 8.786610 211 mol%L 21 %cm%nm 2 , k 2 #2/3 representing the angle between the two fluorophore dipoles assuming random orientation, and g#1.4, the typical refractive index for biomolecules in aqueous solution [43]. The Förster distance estimated for mCerulean-mVenus was 5.3 nm, while R 0 for mCerulean-tagRFP was 5.1 nm and R 0 for mVenus-tagRFP was 5.8 nm. The k factorthe ratio of donor to acceptor (D/A) fluorescence intensity for equimolar concentrations in the absence of FRET, was determined for each construct in accordance with [42]: The k factor for mCerulean/mVenus was calculated to be 0.41, while mVenus/tagRFP gave k = 1.60 and mCerulean/tagRFP gave k = 0.27. D/A ratio for arbitrary concentrations of donor and acceptor was calculated according to [42]: For corrected FRET efficiency measurements, this ratio should be in the range from 0.2 to 5.0 [41]. During analysis, the pixels with D/A ratio outside this range were eliminated from the FRET efficiency calculations.
Validation of G and k factors is presented in Figure S1 for twoand three-color FRET standards with different FRET efficiencies (linkers) and D/A stoichiometry. In our three-color FRET experiments, the major energy transfer was observed directly between mCerulean and tagRFP and not from cascade transfer through mVenus. If there would be a significant contribution of cascade FRET through mVenus, we would see a decrease in efficiency when we used two-color FRET (mCerulean/tagRFP) compared with three-color FRET, potentially including contributions from mCerulean/mVenus/tagRFP cascade. We did not observe a decrease in efficiency with two-color FRET, as experiments with Rb 3 and a 1C C gave the same efficiency of 0.05 (r = 80 nm) as three-color FRET experiments with Rb 3 , a 1C C and Va 1C . Additional control experiments showed that the third fluorophore did not have a significant effect on mCerulean-mVenus FRET: we did not observe a significant difference between the distance between mCerulean/mVenus fluorophores (7363, n = 10) measured by two-color FRET with Va 1C C and unlabeled b 2d and that obtained with three-color FRET using Rb 2d and Va 1C C (6862 nm, n = 13).
For each cell, we calculated FRET efficiency and distances (r) between fluorophores in each pixel of ROI. Gaussian fitting of the r distribution (20 bin histogram) was done in MATLAB using the fit function: where b is the position of the center of the peak (mean) and c (dispersion coefficient) reflects the width of the distribution.