We used the In-Fusion HD Cloning kit (639649, Clontech) to generate genetic fusions in the pENTR (Life Technologies) backbone, and the Gateway LR Clonase II kit (11791, Life Technologies) to shuttle the resulting fusion constructs from the entry clones to expression clones. We have used both the pcDNA3 or a lentiviral backbone (pLenti-puro-CMV/TO, 17293, Addgene) for the expression clones. PCR fragments used for In-Fusion reactions were generated using the Phusion High-Fidelity DNA Polymerase (M0530, New England Biolabs). For all fusion constructs used in this study, a flexible (GGGGS)₂ linker was genetically inserted between the PAmCherry1 fragments and the target protein. In generating the inducible heterodimerization constructs, N-Myr signal and a single DmrA domain were subcloned from pHet-Mem1, and a DmrC domain was subcloned from pHet1, both plasmids in the iDimerize Inducible Heterodimer System (635067, Clontech). KRas G12D and CRaf RBD (residues 51–131) were both subcloned from plasmids used in a previous study[14]. The RBD R89L mutation was introduced through site-directed mutagenesis.

U2OS cells (HTB-96, ATCC) were cultured at 37°C and 5% CO₂ in DMEM supplemented with 10% FBS (11995 and 10082 respectively, Life Technologies). Cells were plated in phenol red-free DMEM (21063, Life Technologies) supplemented with 10% FBS on a #1.5 Lab-Tek chamber slide (155409, Thermo Scientific) for PALM imaging after fixation, or a 0.17 mm coverslip bottom Delta T Dish (04200417, Bioptechs) for live cell imaging. Plasmids for the artificial dimerization system were transiently transfected using X-tremeGENE HP (13873800, Roche) as described by the manufacturer. Dimerization was induced by adding 500 nM A/C Heterodimerizer (635057, Clontech) and incubating at 37°C for 2 hours or overnight as indicated. KRas G12D and CRaf RBD constructs were introduced into the cells by lentiviral infection using the ViraPower packaging system (K497500, Life Technologies). For PALM imaging, cells were fixed in fresh 3.7% PFA with 0.1% glutaraldehyde for 15 minutes at room temperature and changed to imaging buffer (100 mM Tris with 30 mM NaCl and 20 mM MgCl₂, pH 8.5) after fixation. Gold particles (100 nm, EM.GC100, BBI International) were added as fiducial markers to correct for stage drift during imaging.

PALM imaging and tracking was performed on a Nikon Ti-U inverted microscope equipped with a Nikon 60× APO TIRF objective (NA = 1.49) using µManager[15]. Static PALM images were acquired at room temperature. Total internal reflection (TIR) illumination was used in all PALM imaging experiments. PALM image reconstruction was performed using home-written scripts in MatLab (Mathworks, MA). Ripley's K-test and cluster analysis were described previously[14], [16].

To quantify BiFC signal, multiple (4–10 as indicated in the text) random fields of view each containing a few cells were imaged in epi-fluorescence mode before and immediately after a pulse (∼1 s) of high 405 nm illumination (125 W/cm²). Fluorescence from cells with clear photoactivation signals above a threshold was averaged and adjusted for background. For the comparison between wild type CRaf RBD and the R89L mutant, we averaged the fluorescence intensities across the entire field of view because most of the cells with CRaf R89L were dim and had signal levels only slightly above background.

Single molecule tracking experiments were performed at 37°C using a temperature-controlled sample stage (Delta-T, Bioptech). Trajectory analyses were performed using home-written scripts in MatLab (Mathworks, MA). Localizations of molecules in neighboring frames were joined into diffusion trajectories based on spatial proximity, similar to previously reported[17]. Only molecules that lasted at least two frames were used to reconstruct diffusion trajectories. At 50 ms exposure time, we set the maximum distance allowed for a molecule to travel per frame at 500 nm (4 pixels) to avoid falsely connecting localizations of two different molecules into a diffusion trajectory. This is equivalent to a maximum diffusion constant of 1.39 µm²/s. Trajectory analysis with variational Bayes single particle tracking (vbSPT) was performed using the MatLab scripts (http://sourceforge.net/projects/vbspt/) provided by the authors [18].

The efficiency of fluorophore reconstitution in BiFC is sensitive to the position of the split site[4]. In a previous report, Fan et al. tested different split sites for mCherry, the parent protein of PAmCherry1, and found that site 159 (i.e., split between residues 159 and 160) was optimal for BiFC[19]. In the crystal structure of PAmCherry1[20], this site is located in the loop between beta-sheets 7 and 8 (Fig.1A), a position successfully used for BiFC of many FPs[4]. PAmCherry1 and mCherry also have identical amino acid sequences around residues 159 and 160 (Fig. S1). Based on these factors, we hypothesized that site 159 would be suitable for PAmCherry1 BiFC.

To test this hypothesis, we genetically fused the two fragments, RN (residues 1–159) and RC (residues 160–236), to DmrA and DmrC, two peptide components from the Clontech iDimerize Inducible Heterodimer System (Fig. 1B). DmrA and DmrC do not spontaneously interact but will form a 1∶1 complex when a small molecule heterodimer is added. The DmrA peptide has an N-terminal myristoylation signal for membrane localization, so RN or RC fragments were only fused to its C-terminus. This left us with four possible fusion configurations for BiFC: DmrA-RN/DmrC-RC, DmrA-RN/RC-DmrC, DmrA-RC/DmrC-RN, and DmrA-RC/RN-DmrC (Fig. 1C). A flexible polypeptide linker, (GGGGS)₂, was used in all fusion constructs[21].

Two of the four configurations, DmrA-RC/DmrC-RN and DmrA-RC/RN-DmrC, yielded strong BiFC signal when expressed in cells and treated with the heterodimerizer (Fig. 1D, right panels, and Fig. S2). As a control, we observed virtually no BiFC signal in the absence of the heterodimerizer (Fig. 1D, left panels) suggesting that RN and RC fragments do not spontaneously interact and reconstitute PAmCherry1 under these conditions. Similar results were obtained when using cytosolic DmrA and DmrC (data not shown), indicating that the lack of spontaneous interaction between RN and RC was not due to the difference in subcellular localizations. Upon addition of the heterodimerizer, significant BiFC signal was developed in 1–2 hours (Fig. S3), which is typical for the time required for chromophore maturation. Similar to other BiFC systems[2], [3], [4], PAmCherry1 BiFC appeared to be irreversible; that is, once PAmCherry1 is reconstituted, it does not disassemble into RN and RC (Fig. S3). This can at times limit the use of BiFC. Interestingly, PAmCherry1 fluorophore reconstituted efficiently at 37°C without cold incubation. By contrast, many other BiFC systems require low temperatures (often at 4°C or 25°C overnight) to function properly[4], which can be detrimental to the cells.

To optimize PAmCherry1 BiFC, we also moved the split site by 1 residue upstream (to site 158) and downstream (to site 160). Sites 158 and 160 only have a partial overlap with the loop between beta-sheets 7 and 8 (Fig. 1A). Using the same DmrA-RC/RN-DmrC configuration, we observed much lower BiFC signals with both site 158 and 160 compared with 159 (Fig. 1E). Similarly, site 174 (split between amino acids 174 and 175) was also much less efficient than site 159 in both mCherry[19] and PAmCherry BiFC (Fig. 1E). These results suggest that site 159 is optimal for PAmCherry1 BiFC. We therefore used fragments RN(1–159) and RC(160–236) in all subsequent experiments.

BiFC-reconstituted PAmCherry1 possesses similar photophysical properties to the original PAmCherry1, likely because their chromophore structures are nearly identical. In particular, we observed fast photoactivation and a high contrast ratio on BiFC-positive cells (Video S1). Using low power 405 nm activation (5–10 W/cm²), single molecule images of reconstituted PAmCherry1 were easily obtained, showing comparable brightness and signal to background ratio with the original PAmCherry1 (Fig. 2). Taken side by side, both emitted ∼600 photons per localization event and achieved ∼18 nm localization precision on average, which are comparable to literature values[13]. Moreover, using a method developed by Annibale et al[22], we measured the dark state life time of BiFC-reconstituted PAmCherry1 as 0.26±0.05 s (Fig. S4), identical to that of the parent PAmCherry1[14]. These similarities allowed us to perform PALM imaging on PAmCherry1 BiFC samples to obtain nanoscale spatial maps of the DmrA/DmrC artificial complexes (Fig. S5) in fixed cells.

With the principle of BiFC-PALM demonstrated, we next used the technique to study the interactions between Ras and Raf, two important proteins in normal and tumor cell signaling. When Ras is activated by upstream stimuli or by mutations, it recruits and activates Raf at the cell membrane[23]. Recent studies indicated that nanoscale clustering is critical to Ras-mediated activation of Raf[6], [14]. Specifically, Ras has been shown to form 5–8 membered clusters on the cell membrane with typical diameters around 20 nm; this clustering behavior appeared to be driven mostly by the C-terminal CAAX motif of the Ras protein[6], [24]. Ras nanoclusters may serve as signaling platforms to recruit and activate downstream effectors such as Raf. Interestingly, recent studies showed that Raf also signals as dimers and/or multimers mediated through interactions between the kinase domain, for example in the presence of mutant KRas or when the N-terminal domain is truncated[25], [26], [27]. These observations raise the possibility that a functional Ras/Raf signaling assembly may contain multiple (e.g. two) copies of a Ras/Raf complex. Existence of higher order Ras/Raf complex structures, however, has not been directly confirmed. We addressed this question by using BiFC-PALM to reveal the nanoscale spatial organization of individual Ras/Raf complexes on the membrane of fixed cells and diffusion dynamics in living cells.

We designed a BiFC-PALM system by fusing RN to the N-terminus of KRas 4B (hereafter referred to as ‘KRas’) G12D, an active mutant of KRas, and RC to the C-terminus of the Ras-binding domain (RBD[28], amino acids 51–131) of CRaf (Fig. 3A). Consistent with the strong interactions between KRas G12D and CRaf, the BiFC system yielded a clear signal when expressed in cells (Fig. 3B; top panel). Interestingly, two other KRas G12D/CRaf RBD PAmCherry1 BiFC configurations that we tested were also positive (Fig. S6, although one of the two configurations showed much lower efficiency). As a control, we introduced the R89L point mutation to CRaf RBD to disrupt Ras/Raf interaction[29] and observed that the BiFC signal decreased by nearly 10-fold (Fig. 3C and Fig. S7). These data confirm that the BiFC signal was due to specific interactions between KRas G12D and CRaf RBD.

We imaged the KRas G12D/CRaf RBD complexes in fixed cells with PALM under total internal reflection (TIR) conditions. When reconstructing the PALM images, we corrected blinking (due to transient dark states) of individual PAmCherry1 molecules using a previously described approach[14], [22] and a dark state life time of 0.26 s. As such, each dot in the final high resolution image represents a single PAmCherry1-tagged KRas G12D/CRaf RBD complex. An example BiFC-PALM image of PAmCherry1-KRas G12D/CRaf RBD is shown in Fig. 3B (left panels). From the PALM images, it became evident that KRas G12D/CRaf RBD complexes are heterogeneously distributed on the cell membrane as both monomers and clusters (Fig. 3B; middle and bottom left panels, and insets). Spatial pattern analysis with Ripley's K-test revealed that the clusters have an apparent diameter of ∼30 nm (Fig. 3D). We note that this apparent diameter reflects mostly the actual resolution achieved in our PALM imaging experiments and not necessarily the actual size of the clusters. Analysis with simulation aided DBSCAN (SAD)[14] further showed that each cluster typically contains 2–3 KRas G12D/CRaf RBD complexes (Fig. 3E). These observations suggest that multiple Ras/Raf complexes can cluster and form a higher order assembly.

Next, we performed single molecule tracking (smt-) PALM measurements in living cells[17] expressing the RN-KRas G12D/CRaf RBD-RC BiFC pair. Under TIR illumination and low dose (2.5–5 W/cm²) 405 nm activation conditions, smt-PALM allowed us to sparsely activate and individually track a small number of PAmCherry1 BiFC tagged KRas G12D/CRaf RBD complexes at a time. We typically acquire ∼10,000 diffusion trajectories (localization events that span 2 frames or more) per cell with 50 ms time resolution (Video S2 and Fig. S8). Analysis of the diffusion trajectories from a single cell revealed that the KRas G12D/CRaf RBD complexes exist in at least two states: a mobile state and an immobile state, which is evident in both the positional trajectories (Fig. 4A) and the histogram of displacement per frame (Fig.4B). Here, the displacement per frame represents the instantaneous diffusion rate of the molecules per unit time (50 ms). The existence of multiple diffusion states of the KRas G12D/CRaf RBD complexes was also confirmed by variational Bayes single particle tracking (vbSPT), a new algorithm developed for analyzing single molecule diffusion trajectories[18]. Specifically, vbSPT showed that the molecules have three diffusion states with diffusion constants of 0.44, 0.08, and 0.02 µm²/s (Fig. 4C). This observed heterogeneous diffusion behavior is consistent with the co-existence of KRas G12D/CRaf RBD monomers and multimers (Fig. 3B) as well as heterogeneities in the membrane composition [5].