Challenges of the in vivo BiFC screen
We have showed that it is possible to find proteins from a random library that interact with a bait protein in vivo and in planta. This was done by observing protein-protein interactions measured by BiFC of mYFP. Although this screen is one of the few purely random , in planta screens ever established, its efficacy is not perfect. We had to come up with a method that would allow the identification of a single population of interacting proteins from a single protoplast. We used several techniques that were impossible to avoid but are difficult to control: transfection of protoplasts, recovery of plasmids from protoplasts and transformation of bacteria with plasmid. The first condition is obvious: in order to make a screen in living plant tissue, it is necessary to have a method that has the capacity of performing many interactions simultaneously and we choose protoplast transfection. However, there is no way to exclusively introduce a single plasmid into a protoplast, the nature of the method means that each protoplast takes up multiple plasmids. Thus methods that would rely on PCR (single cell PCR) for single cells or pooling of the cells (454 sequencing) would most likely not eliminate false-positives. An attempt to dilute the mixed library DNA to some statically desirable transfection rate of 1:1 of bait to prey would be dependent on the transfection rate, which varies from batch to batch, and most likely produce undetectable protein amounts. An alternative to our random library approach would be to use defined grids of prey plasmids that could be transfected into the protoplasts as binary pools; such an approach was already performed using the split-luciferase system . This method of course, requires a cloned or clonable library ORF set. While this worked for the split-luciferase as the detection of luciferase activity has good signal-to-noise ratios, the BiFC system relies on weaker mYFP emission, and this emission must exceed that of the plant autofluorescence. Thus, we found that the screening was only possible using a flow cytometer where we can compare YFP versus autofluorescence to detect those cells expressing BiFC of YFP.
Due to the chanciness of the protoplast transfection and the DNA recovery, we choose to stay with the recovery of the plasmid DNA first. At the beginning of the screen design, we had anticipated that we would be able to isolate a single bacterial colony carrying a single plasmid type that could be screened pair-wise against the bait and that plasmid’s cDNA would be available for immediate downstream cloning via GatewayTM technology. Unfortunately, although only few bacterial colonies were obtained, those that were obtained also carried multiple, independent plasmids as well. Furthermore, the majority of colony forming units derived by plasmid recovery from the protoplasts did not have interacting prey (Figure4B). This suggested to us that although the plasmid recovery from positive BiFC cells sorted by FACS did lead to a slight enrichment of the BiFC generating plasmids (Figure4E), other plasmids were still present in those protoplasts and also consequently in the bacteria. Our solution to the multiple-plasmid problem was an enrichment of the actual plasmids encoding the interacting protein by isolating all of the plasmids from the first round and transfect them against the bait once again and sort again by FACS on BiFC of YFP. This strategy worked, as the positive interaction rate went up from 2% to 27%.
In the end we utilized the protoplast transfection and the recovery of the bacterial plasmid DNA without any additional interventions to identify novel bait interacting partners and recovered the cDNA by PCR and cloning. Together these observations mean that the screen is not saturated. For example, the one positive found in the first round of the CPK3 screen was not found in the second round (Figure4). This also indicated that there were probably many more plasmids still present encoding for other CPK3-interacting partners that had been not detected. Nevertheless, we conclude that it is possible to recover interacting proteins encoded on bacterial plasmids from an in planta screen.
The majority of observed BiFC signals was surprisingly weak and only detectable in the flow cytometer. One would have expected that some strong BiFC interactions should have been detected as it was the case for APX3. The best explanation for the low BiFC-YFP signals during the screening is that, in order for enough YFP signal to be detected, it must exceed a certain concentration in the cell before the total fluorescence intensity is detectable above the autofluorescence. Reducing the YFP by using BiFC fragments already reduces the total detectable protein . For those fusion proteins whom are less abundant in a cell or have localizations that restrict their abundance (for example those in the nucleus, plasma membrane or Golgi apparatus), it means that it is not possible to screen for any type of protein. It is not only remarkable that could we recover transfected plasmids from plant protoplasts after 36 hrs, but that it was also possible to identity some interacting partners with a bait construct. According to the data presented here, the in vivo in planta BiFC screen provides a lucrative alternative to search for novel protein-protein interactions that can, according to our data, only be found in planta.
Putative interaction partners of CPK3
We choose CPK3 (AT4G23650) for demonstrating the screening method. CPK3 (and CPK6) are expressed not only in stomata but also in other tissues and are well studied for their function in guard cells and ABA signaling . CPK3 is involved in the phosphorylation of plasma membrane S-type anion channels for the Ca2+-dependent stomatal closure , senses Ca2+ directly  in addition to regulating ROS signaling , which is also needed for ABA signaling . CPK3 is further implicated in mediated ABA stomatal regulation involving phosphoinositides and differential Ca2+ mediations . Putative phosphorylation targets of CPK3 were recently published  and the results show that potential CPK3 target proteins can be seemingly varied in functional classes. Hence, CPK3 is has been shown to be involved in many processes in addition to guard cell signaling [23, 40]. CPK3 has been broadly localized to the cytoplasm and in and around the nucleus , but has recently been shown to be preferentially vacuolar and plasma membrane-associated . Our CPK3 constructs were localized to the cytoplasm and in and around the nucleus in tobacco leaves and in Arabidopsis protoplasts (mYFP, Figure1; eGFP, Figure6) very much like that observed by . Unfortunately, the BiFC CPK3-prey interactions were too weak to be observed in the microscope; thus we were not able to show where the BiFC interactions were taking place inside the cell. We used two fluorescence based methods to substantiate the protein-protein interactions found in the screen: biased BiFC and FRET-FLIM. I wanted to say that theoretically BIFC interactions could be through an unseen partner or due to trapping, but this sentence ended up a bit self-contradictory the way is it currently written. We used GFP/mCherry or GFP/mRFP donor/acceptor FRET pairs that have been shown to perform very well in vivo[35, 41]. Proteins (and their fluorophore fusion) must be within 1 to 10 nm distance for FRET to occur , which is the typical distance found for interacting proteins. Similarly, BiFC has been discussed to occur over a distance around 7 nm . Both BiFC and FRET-FLIM support four previously uncharacterized protein interactions of CPK3.
APX3 (#1, AT4G35000) showed the strongest BiFC with the CPK3 baits in both orientations, N-terminal and C-terminal SPYNE (Figure5C). APX3 is targeted to peroxisomes [24, 45], but has been shown to be retarded in the cytoplasm by AKR2A . In the FRET-FLIM studies, the N-terminal mRFP-APX3 fusion showed the strongest FRET efficiency with CPK3-eGFP (Figure6D). mRFP-APX3 was clearly non-homogenous in its sub-cellular distribution as its C-terminal transmembrane domain  was not masked. In contrast, the C-terminal APX3-mCherry was mis-localized to the cytoplasm (Figure6B; Additional file 3) and showed no interaction with CPK3 in FRET-FLIM (Figure6E). This evidence combined with the very strong BiFC makes a good argument that the screen was able to find a major interactor of CPK3. Interestingly, APXs are important for scavenging ROS (H2O2) and APX3 could provide the link proposed for CPK3 and CPK6 in regulating ROS and NADPH activation in guard cell function .
ORP2A (#5, AT4G22540) is a predicted oxysterol binding protein (OSBP). ORP2A significantly interacted with CPK3 in BiFC experiments (Figure5C). It also interacted preferentially with CPK3 in tobacco epidermal cells as shown by FRET-FLIM. OSBPs are involved in sterol trafficking  affecting membrane fluidity and permeability and influencing secretory events. OSBPs are known to bind to oxysterols, which compose minor amounts of sterols in plants , but OSBPs are known in other species to bind to different lipids including phosphoinositides, ergosterol, and cholesterol (references in ). Interestingly, there is some evidence that OSBPs are involved in the regulation of processes like Ca2+ uptake and transcriptional control, both processes which relate directly to CPK3. Mechanisms how newly synthesized sterols reach the plasma membranefrom the ER are unclear in plants and OSBPs are possible candidates. ORP2A is well expressed in many tissues and is somewhat regulated by stresses . CPK3 was shown to target ER associated proteins Calnexin and Calreticulin , the latter of which regulates Ca2+ stores and signaling from the ER.
AT2G39050 (#6) is a ricin B-related lectin domain containing protein. Ricin is a heterodimeric plant protein that is toxic to mammalian and many other eukaryotic cells by binding to membrane localized galactose-containing receptors . Ricin is composed to two subunits, ricin toxin A (RTA) and B (RTB). RTA is catalytically-active and removes a specific residue from the 28 S ribosomal RNA . During its synthesis in plant cells ricin traffics from the ER, via the Golgi complex, to the vacuole . AT2G39050 showed significant interaction with CPK3 in BiFC. The FRET-FLIM studies also support the interaction of this protein with CPK3. That CPK3 is also membrane associated and that ricin moves through the ER to the vacuole strongly supports the interaction with CPK3.
HMGB5 (AT4G35570) belongs to the class of high mobility group (HMG) proteins and are, after histones, the second most abundant type of chromosomal proteins . HMGs have an ‘AT-hook’ that binds to the minor groove of short stretches of A/T-rich B-form DNA independent of the nucleotide sequence . Unlike histones, HMG proteins are very dynamic and some even shuttle in and out of the nucleus in animal and plant cells [50, 52]. HMGB5 is predominantly found inside the nucleus  and is extremely mobile within the nucleus . HMGB5 showed a significant BiFC interaction with CPK3 in Arabidopsis protoplasts. The FRET-FLIM experiments in tobacco epidermal cells also substantiate the interaction between CPK3 with HMGB5 in the nucleus.
Among CPK3’s roles in the regulation of plasma membrane-localized ion-channels, it is known to have roles in phosphorylating nuclear transcription factors [32, 53, 54], other DNA-binding proteins [23, 55]) and many RNA associated proteins . This suggests that CPK3 has a role in regulating gene expression before, during and after transcription and that it may also have a role in chromatin regulation in conjugation with, for instance, HMGB5.
We could exclude AT2G29670 (#2), AT5G08680 (#3), GLO1 (#4, AT3G14420) and CCoAMT (#8, AT1G67980) as true interaction partners for CPK3 as they did not show interaction any via BiFC in Arabidopsis protoplasts where the screen was conducted and therefore we conclude they do not meet in Arabidopsis cells and thus do not interact with each other (see Additional file 5 and 6 for details).