Quantitative detection ofin vivophysical protein–protein interactions based on DNA sequence recognition : Figure 1

Dear Editor, Understanding physical protein–protein interactions (PPIs) in vivo is a major challenge in elucidating the molecular mechanisms in complex biological systems. Although many approaches, such as the yeast two-hybrid (Y2H) assay, protein fragment complementary assay (PCA), and fluorescence resonance energy transfer (FRET) assay, have been developed to assess the intensity of interaction of proteins in living cells, they have practical limitations for quantitation (Morell et al., 2009; Padilla-Parra and Tramier, 2012; Stynen et al., 2012). Nucleic acid-based technologies, such as quantitative PCR (qPCR), DNA microarray, and deep sequencing, are probably most featured with quantification, high-throughput, and cost-effective for operation. Recently, new methods of detecting biomolecular interactions have been developed based on a strategy that labels target interactions using specific nucleic acid sequences (Söderberg et al., 2006; Yu et al., 2011; Zhu et al., 2013). Here, we describe a novel method, termed protein-dimerization footprinting (PdF), for the quantitative measurement of physical PPIs in vivo by directly transcoding the signal from physical PPIs into DNA sequences. In the PdF strategy (Figure 1A), a target protein is fused to a DNA-binding domain that can specifically recognize a defined DNA sequence but with negligible affinity. However, when the interaction between target proteins in living cells becomes strong enough to allow the formation of a stable dimer, the DNA-binding domain binds to the specific DNA sequence with substantially higher affinity, thereby protecting the DNA sequence against subsequent DNase I digestion. Hereby, a physical PPI is transcoded into a specific DNA sequence that can be detected by quantitative PCR or other nucleic acid-based techniques, thus using the copy number to quantify physical interaction intensity. The PdF strategy for quantifying physical protein–protein interactions. (A) An outline of the PdF assay. (B) A validation for the PdF assay. PCR detected much higher PdF signal (i.e. the BR signal) of CI(C)-pPIDA1 (positive control) than BR-pSP73 (blank control) and pPIDA1 (negative control), indicating the dimerization of CI(C). Here, physical PPI intensity is effectively converted to DNA sequence signal. (C) Qualitative results for homo-dimerization of proteins in the E. coli chemotaxis pathway are consistent from Y2H, FRET, and PdF measurements. ‘+’ represents strong dimerization and ‘−’ represents weak or no dimerization. (D) PdF agrees well with FRET in quantitative detection of E. coli chemotaxis protein homo-dimerization. Error bars indicate SD from five or three independent biological samples for FRET efficiencies or PdF signals, respectively. PdF signals and corresponding FRET efficiencies are compared in a scatter plot. Each point represents a chemotaxis protein homo-dimerization that was measured by both PdF and FRET. The line indicates the linear regression fit. (E) PdF agrees well with FRET in quantitative detection of PhoB dimerization after 50 min incubation in M9 broth with different phosphate concentrations. Error bars indicate SD from five or three independent biological samples for FRET efficiencies or PdF signals, respectively. An increase in PhoB dimerization intensity along with the decrease in phosphate concentration was observed. PdF and FRET measurements are compared in a scatter plot with FRET efficiency as x-axis and PdF signal as y-axis. Each point represents a different phosphate concentration at which PhoB dimerization was measured by both PdF and FRET. The line indicates the linear regression fit. (F) Dynamical detection of PhoB dimerization intensity in M9 broth with different phosphate concentrations and different incubation times. Error bars indicate SD from three independent biological samples for PdF signals. In the present study, we used protein homo-dimerization as a simple system to demonstrate the principle, performance, and utility of PdF. The DNA-binding domain used in the present study comes from the lambda repressor CI. CI is composed of a C-terminal dimerization domain, CI(C), that is responsible for CI dimer formation, and an N-terminal DNA-binding domain, CI(N), that accounts for CI binding sequence recognition. Importantly, the CI dimer has high affinity for its specific DNA sequence, whereas the CI monomer exhibits low affinity for the same sequence (Ptashne, 2004). This system was applied in designing DNA-binding domains for our PdF assay. To examine whether PdF could distinguish protein dimers from monomers by measuring the specific DNA sequence, we constructed three vectors (Figure 1B): BR-pSP73, pPIDA1, and CI(C)-pPIDA1. BR-pSP73 was a blank control that contains only CI binding region (BR). pPIDA1 was a negative control that contains BR and expresses CI(N). CI(C)-pPIDA1 served as a positive control that contains BR and expresses the dimerizable CI(N)-CI(C) fusion protein. DNase I digestion showed that the CI(N)-CI(C) fusion protein generated a much stronger PdF signal (or BR signal) than the blank control and CI(N) monomer (Figure 1B), demonstrating that physical PPI information can indeed be effectively converted to DNA sequences through specific protein–DNA binding dynamics. We next developed a mathematical model (see Supplementary material) based on the biochemical reactions involved in the PdF assay, which established the relationship between the PdF signal (Imd, the relative copy number of BR) and physical interaction intensity (Kdimer, the binding constant of target proteins during dimerization). As detailed in the Supplementary material, the mathematical model accurately captured the behavior of the PdF dynamics. According to the model, the PdF signal could directly represent the interaction intensity with a linear correlation. Taking the PdF signal of CI(C) as a reference, denoted Imdref⁠, we defined 0.1Imdref as the threshold for determining whether a physical PPI existed. That is, if the PdF signal Imd<0.1Imdref⁠, Imd is regarded as zero, or there is no significant interaction between the target proteins. To compare the performance of PdF with Y2H and FRET, we measured the intensities of four known protein homo-dimerizations in the E. coli chemotaxis pathway, CheA, CheB, CheY, and CheZ, which are of different sizes and structures (Kentner and Sourjik, 2009). As shown in Figure 1C, all three methods gave the same qualitative results. CheY and CheB had weak or undetectable homo-dimerization, whereas CheA and CheZ had stronger homo-dimerization. An interaction was defined as existed by PdF if its PdF signal was above 0.1Imdref⁠, or by FRET with the threshold adopted from a previous study (Kentner and Sourjik, 2009). Under these thresholds, CheB was determined as no homo-dimerization by both FRET (Figure 1D and Supplementary Table S1) and PdF (Figure 1D and Supplementary Table S2), consistent with undetectable CheB homo-dimerization in the Y2H assay (Supplementary Figure S7). Weak homo-dimerization of CheY was detected by both FRET and PdF, as indicated by their respective weak signal values (Figure 1D, Supplementary Tables S1 and S2), while Y2H failed to detect the presence of CheY dimer (Supplementary Figure S7). Both CheA and CheZ were detected to have stronger homo-dimerization by Y2H, FRET, and PdF (Figure 1D, Supplementary Figure S7, Tables S1 and S2). As shown in Figure 1D, PdF signals of CheA, CheB, CheZ, and CheY were quantitatively comparable with their corresponding FRET efficiencies. The high correlation between PdF and FRET signal values (R2 = 0.95) demonstrates that PdF is able to quantitatively detect physical PPIs in living cells with the accuracy similar to FRET. Indeed, all above results were consistent with the previous study systematically investigating protein interactions in chemotaxis pathways of E. coli using FRET (Kentner and Sourjik, 2009). Next, we assessed the performance of PdF on detection of protein dimerization under different conditions. In a model system of PhoB, PhoR phosphorylates PhoB under phosphate-restricted conditions, and increases dimerization intensity of PhoB, which in turn enhances the ability of PhoB to regulate the transcription of Pho regulon (Fiedler and Weiss, 1995; Gao et al., 2008; Mack et al., 2009; Gao and Stock, 2013). Dimerization affinity of phosphorylated PhoB is ∼100-fold greater than that of unphosphorylated PhoB in vitro (Gao et al., 2008). To measure PhoB dimerization by FRET or PdF, CFP-/YFP-fused PhoB or CI(N)-PhoB fusion protein was expressed in E. coli for 3.5 h or 8 h, respectively. E. coli cells were collected, incubated in M9 broth with different phosphate concentrations (1 × 10−1, 1 × 10−3, and 1 × 10−5 M) at 37°C for 50 min, and subjected to FRET (Supplementary Table S3) or PdF (Supplementary Table S4). A high linear correlation (R2 = 0.99) between PdF and FRET signals (Figure 1E) demonstrates that dimerization intensities of PhoB detected by PdF and FRET are quantitatively comparable. In addition, the observed increase in PhoB dimerization intensity along with decrease in phosphate concentration was consistent with previous studies (Fiedler and Weiss, 1995; Gao et al., 2008; Mack et al., 2009). The transcriptional activity of PhoA, one member of the Pho regulon, was lower at a high phosphate concentration (5 × 10−3 M) than that at a low phosphate concentration (1 × 10−4 M or 2 × 10−4 M) (Gao et al., 2008; Mack et al., 2009). The PhoB dimerization-induced repression of β-galactosidase expression was enhanced from 35% under phosphate excess condition (1 × 10−2 M) to 65% under phosphate limitation condition (1 × 10−4 M) (Fiedler and Weiss, 1995). Previously, phosphorylation of PhoB was shown to increase along with incubation time under phosphate limitation condition (5 × 10−5 M) (Gao and Stock, 2013). As PhoB dimerizes upon phosphorylation (Fiedler and Weiss, 1995; Gao et al., 2008; Mack et al., 2009), this result provides indirect time-course information of PhoB dimerization. Here, we sought to investigate the time behavior of PhoB dimerization directly, by applying PdF to measure PhoB dimerization intensities at different incubation times under different phosphate conditions (Figure 1F). We found that, with 1 × 10−3 and 1 × 10−5 M phosphate, PhoB dimerization intensities increased with incubation time, which was consistent with the previous study (Gao and Stock, 2013) and might result from the changing of PhoB phosphorylation degree (Gao et al., 2008; Gao and Stock, 2013). In contrast, with 1 × 10−1 M phosphate, PhoB dimerization intensities kept at a low level approaching the threshold 0.1Imdref and did not change along with the incubation time. This might result from the weak dimerization of unphosphorylated PhoB (Mack et al., 2009; Gao and Stock, 2013). Thus, the use of DNA sequence for PPI detection distinguishes PdF from contemporary protein-based PPI detection strategies. It is much simpler and better for quantification than Y2H (Sim et al., 2005). In FRET assay, the relation between the energy transfer efficiency (E) and the distance of the two proteins (R) obeys E ∼ 1/(1 + R6), and interaction intensity (K) and R are thus inversely correlated (Gordon et al., 1998). Therefore, FRET amplifies the variation of PPI intensity. In contrast, PdF is specifically designed to detect interaction intensity, and the relation between PdF signal (I) and interaction intensity (K) is I ∼ K (Supplementary material), which may explain why FRET measurements are more dispersed (Supplementary Tables S1–S4). The in situ PLA is a powerful immunohistochemical method using DNA as the reporter to detect locations and numbers of complexes of known proteins in mammalian cell or tissues (Söderberg et al., 2006). However, its application, in particular in bacteria cells, is limited due to the dependence of antibodies and small size of a bacteria cell (Clausson et al., 2011). PdF is a more universally applicable and cost-effective method with the accuracy comparable to FRET. Furthermore, PdF has the potential to largely increase the throughput of quantitative PPI detection in vivo. PdF also provides a feasible method for measuring multiple physical PPIs simultaneously in one cell, by using different orthogonal pairs of DNA-binding domains and specific DNA sequences. The limitation of PdF is that unlike FRET and in situ PLA, it does not possess the space resolution. In addition, it has a lower temporal resolution (in minutes) compared with FRET (in seconds). To improve the temporal resolution of PdF, the crosslinking kinetic (CLK) technology (Poorey et al., 2013) might be used, by which the protein–DNA binding detection can be accomplished in a time scale of seconds. Thus, it is possible for PdF to achieve a temporal resolution comparable with ordinary FRET in in vivo PPI detection. [Supplementary material is available at Journal of Molecular Cell Biology Online. We thank the critical comments from Dr Min Guo at the Scripps Research Institute. This work was supported by grants to J.W. from the National Natural Science Foundation of China (31130034), the Ministry of Science and Technology (2011CB910200), and the Science and Technology Commission of Shanghai Municipality (13DZ1940100), and by grants to Z.L. from the Ministry of Science and Technology (2012CB917200, 2014CB910600).]