Quantitative live-cell imaging and 3D modeling reveal critical functional features in the cytosolic complex of phagocyte NADPH oxidase

Phagocyte NADPH oxidase produces superoxide anions, a precursor of reactive oxygen species (ROS) critical for host responses to microbial infections. However, uncontrolled ROS production contributes to inflammation, making NADPH oxidase a major drug target. It consists of two membranous (Nox2 and p22phox) and three cytosolic subunits (p40phox, p47phox, and p67phox) that undergo structural changes during enzyme activation. Unraveling the interactions between these subunits and the resulting conformation of the complex could shed light on NADPH oxidase regulation and help identify inhibition sites. However, the structures and the interactions of flexible proteins comprising several well-structured domains connected by intrinsically disordered protein segments are difficult to investigate by conventional techniques such as X-ray crystallography, NMR, or cryo-EM. Here, we developed an analytical strategy based on FRET–fluorescence lifetime imaging (FLIM) and fluorescence cross-correlation spectroscopy (FCCS) to structurally and quantitatively characterize NADPH oxidase in live cells. We characterized the inter- and intramolecular interactions of its cytosolic subunits by elucidating their conformation, stoichiometry, interacting fraction, and affinities in live cells. Our results revealed that the three subunits have a 1:1:1 stoichiometry and that nearly 100% of them are present in complexes in living cells. Furthermore, combining FRET data with small-angle X-ray scattering (SAXS) models and published crystal structures of isolated domains and subunits, we built a 3D model of the entire cytosolic complex. The model disclosed an elongated complex containing a flexible hinge separating two domains ideally positioned at one end of the complex and critical for oxidase activation and interactions with membrane components.

Numerous proteins contain intrinsically disordered regions (1,2). Their growing number challenges the paradigm of a strong relationship between structure and function. Those proteins require new strategies to study their shape and their interactions. Here we focus on proteins composed of structured domains linked by disordered segments. Those segments and thus the whole protein can adopt a broad ensemble of conformations. These ensembles are highly influenced by their environment. In particular, the molecular crowding, the local ion concentrations, pH, or viscosity present in live cells may change the set of accessible conformations. Nevertheless, when they are associated with diseases, knowledge about both their structure and their affinity with partners is essential to identify potential drug targets in medicinal chemistry. Their structural analysis calls for new in situ analytical strategies (1,2). Here, we combined live cell FRET-FLIM 5 and FCCS approaches to propose an integrated analytical workflow for structural and quantitative studies of a protein complex composed of such multidomain proteins in their native environment. We applied this workflow to the intra-and intermolecular interactions of the cytosolic factors of the phagocyte NADPH oxidase complex.
The phagocyte NADPH oxidase is one of the seven isoforms of the Nox family and a major enzyme of the immune system because of its pronounced production of microbicidal reactive oxygen species (ROS). A pathological hyperactivity of this oxidase leads to chronic inflammation, which is associated with cardiovascular diseases, stroke, and chronic obstructive pulmonary disease (3). The activity is regulated by the spatiotemporal organization of the intra-and intermolecular interactions of its cytosolic subunits, p40 phox , p47 phox , and p67 phox , forming a soluble complex before activation. Upon activation, these subunits translocate together with the small GTPase Rac to the membrane subunits Nox2 and p22 phox to form the active oxidase complex. The protein interactions within the cytosolic complex were extensively studied in vitro ( Fig. 1) (3). In addition to the interactions between p40 phox and p67 phox or p47 phox and p67 phox , an interaction between the SH3 domain of p40 phox and the PRR domain of p47 phox has been observed, but its physiological relevance is not clear (4). The structure of p40 phox was entirely solved (5). The presence of intrinsically disordered regions in p47 phox and p67 phox , predicted from their sequence (6), prevented the crystallization of the whole subunits. Only domains either isolated or in interaction were solved (3). To date, neither the spatial organization nor the stoichiometric composition of the entire complex in live cells was clarified. Using live cell FRET-FLIM and FCCS approaches with fluorescent protein-tagged subunits, we demonstrated a 1:1:1 stoichiometry of the three subunits, estimated their affinity, and analyzed their spatial organization. Finally, we used these findings to elaborate a new 3D in silico model of the entire cytosolic complex in the live cell situation. This model shows an elongated complex and a flexible hinge. It is fully compatible with the multiple steps of oxidase activation. In addition, it can guide the identification of potential sites for anti-inflammatory drug targets to regulate the NADPH oxidase activity.

FP-tagged subunits are correctly expressed and are functional
The cytosolic subunits p40 phox , p47 phox , and p67 phox were tagged with fluorescent proteins (FPs; Table S1), either cyan (CFP: mTurquoise or Aquamarine) (7), yellow (YFP: Citrine) (8), or red (RFP: mCherry) (9). The size of the fusion proteins expressed by COS7 cells was verified by Western blotting (Fig.  S1A). To assess whether the FP-labeled subunits are able to reconstitute the active oxidase complex, we used COS Nox2/p22 cells stably expressing the membranous subunits Nox2/p22 phox and Rac, but no endogenous cytosolic subunits (10). COS Nox2/p22 were transiently transfected with the three cytosolic subunits with or without a FP tag ( Fig. 2A). The production of superoxide anions was monitored by a luminometry assay sensitive to extracellular ROS (Fig. 2B). The ROS production started upon activation with phorbol myristate acetate (PMA) and stopped immediately after addition of diphenyleneiodonium (DPI), a NADPH oxidase inhibitor. All constructs allowed a pronounced ROS production (Fig.  2C). The production with N-terminal tagged p47 phox and p67 phox was lower compared with the C-terminal tagged variants but still much higher than the nontransfected cells. The presence of the FP tags slowed down the ROS production and consequently delayed the time point at which the maximal signal was reached (Fig. S1, B and C). Taken together, these observations show that all our FPtagged subunits are able to reconstitute an active NADPH oxidase complex and are fully suitable to explore their spatial organization and affinity.
FRET is observed between fluorescent proteins at both ends of individual subunits and is not strongly modified in presence of their partners. We then tagged each of the cytosolic subunits p40 phox , p47 phox and p67 phox simultaneously at both termini with a donor (D) and an acceptor (A) FP for FRET, resulting in so-called tandems ( Fig. 3 and Table S1). We used as a positive control a simple D/A tandem in which the donor and acceptor FPs were linked by a flexible 27-amino acid-long peptide (11). Tandems were expressed in COS7 cells, and FRET was monitored by FLIM. The apparent FRET efficiencies, E app , were derived from the average fluorescence lifetime of the donor FP measured in individual cells (Equations 1-4 under "Experimental procedures"). E app values for the p40 phox , p47 phox , and  A, images of triple-transfected COS Nox2/p22 cells expressing p47 phox -CFP, p67 phox -YFP, and RFP-p40 phox subunits (left to right). Scale bar, 20 m. Five conditions of p40 phox , p47 phox , and p67 phox with no tag, two, or three tags on either N or C terminus were tested. B, time course of the luminescence signal (L-012 ϩ horseradish peroxidase) from triple-transfected COS Nox2/p22 cells stimulated by PMA and stopped by the oxidase inhibitor DPI at the indicated times (condition 4, green). In orange, the signal obtained with nontransfected cells as reference. C, integrated PMA-stimulated luminescence signal over 30 min (n ϭ 3; means Ϯ S.E.). **, p Ͻ 0.01, Tukey's multiple comparison test.
A new 3D model of the NADPH oxidase cytosolic complex p67 phox tandems were significant, although lower than in the simple D/A tandem (E app ϭ 8.2 Ϯ 1.2%, 12.3 Ϯ 1.2%, 7.7 Ϯ 1.1%, and 32.7 Ϯ 1.8%, respectively; Fig. 3B). The low E app values for the subunit tandems are consistent with the much larger size of the central subunits as compared with a 27-amino acid peptide linker.
Next, we studied the influence of the p67 phox partner on the p47 phox tandem and of p40 phox on the p67 phox tandem. Either p47 phox or p67 phox tandems were co-expressed with a plasmid coding for the partner subunit connected to RFP via the viral P2A peptide (12,13). The P2A sequence prevents the peptide bond formation and leads to the separate expression of the subunit and the RFP, the presence of the RFP proving the correct expression of the partner subunit. The co-expression of tandems with RFP alone was taken as reference (Fig. 3C). The co-expression of RFP-2A-p67 phox did not change the apparent FRET efficiency in the p47 phox tandem, and similarly, the presence of p40 phox did not modify the FRET level of the p67 phox tandem ( Fig. 3C and Fig. S2), showing that in both cases, the overall geometry of the tandem is not significantly modified upon interaction with their partners.

FRET signals indicate specific interaction between p67 phox and p47 phox and provide information on their spatial organization
The bimolecular interaction between FP-tagged cytosolic subunits was then investigated using similar FLIM-FRET methodologies (Fig. 4). First, we studied the interaction of p67 phox and p47 phox (Fig. 4, A-C and F). Upon co-expression of p47 phox -CFP/p67 phox -YFP, the lifetimes of the CFP donor were on average significantly shorter than the reference value for p47 phox -CFP alone (Fig. 4, A and B). However, as usual in dual expression systems, FRET efficiencies E app varied strongly from cell to cell, depending on the amount of expressed acceptor. We thus determined E app as a function of the acceptor quantity, as estimated from its average fluorescence intensity (Fig. 4C), or as a function of the ratio [A]/[D] (Fig. 4F), using a custom calibration procedure (see supporting information), as shown by others (14,15). In all cases, the FRET efficiency increases with the absolute or relative amount of acceptor.
For p47 phox and p67 phox subunits labeled at their C termini (CC labeling), E app reaches a maximum value ϳ12% at high acceptor levels (Fig. 4, C and F). This is significantly higher than the negative control, consisting of the co-expression of p47 phox -CFP with YFP alone ( Fig. 4C and Fig. S3A). As a second negative control, we used a truncated version of p47 phox missing the PRR domain (p47 phox ⌬Cter ) and leading in vitro to a complete abrogation of any interaction with p67 phox (16). The coexpression of p47 phox ⌬Cter-CFP with p67 phox -YFP gives clearly different FLIM images as compared with p47 phox -CFP, with well-separated average lifetime distributions (Fig. 4, A and B), and measured E app values in the range of the first negative control ( Fig. 4C and Fig. S3A). The observation of a plateau value for E app , well above the values of the controls, provides strong evidence for a specific interaction (17). The absence of significant FRET in the case of p47 phox ⌬Cter shows in addition that the PRR domain of p47 phox is required for this interaction in live cells.
We then modified the labeling sites of our FP tags, either by switching them to the N termini or by exchanging FP colors between p47 phox and p67 phox (Fig. 4, C and F). For the CN tags, the maximum E app of 8% was lower than for the CC labeling ( Fig. 4F), yet significantly above the negative controls (Fig. S3A). For the NN tags, E app values remained very low and close to the range of the negative controls ( Fig. 4C and Fig. S3A).
To evaluate independently the occurrence of an interaction between p47 phox and p67 phox tagged at their N termini, we used FCCS. In this technique, the fluctuations of the fluorescence intensities of FP-tagged p47 phox and p67 phox diffusing in and out a confocal volume in the cytosol of COS7 cells are analyzed by auto-and cross-correlation functions (see "Experimental procedures"). A cross-correlation function with a non-null amplitude was observed for p47 phox and p67 phox tagged either at their CC and NN termini (Fig. 5A), showing in both cases a co-diffusion of the fluorophores indicative of complex formation.
We thus observe a specific interaction between p47 phox and p67 phox cytosolic subunits for all tested positions and types of FP tags, with in most cases E app well above negative controls. Apparent FRET efficiencies measured in living cells are very The upper panels show the residual difference plot. For p40 phox , a YFP/RFP FRET pair was used, whereas p47 phox , p67 phox , and the simple D/A tandem were labeled with a CFP/YFP FRET pair. The control shown on the left is p40 phox -YFP ( donor ϭ 3.18 ns, 2 ϭ 1.12), whereas the tandem shown on the right is RFP-p40 phox -YFP ( long ϭ 3.17 ns, short ϭ 1.56 ns, accounting for 77 and 23% of the decay amplitude, respectively, 2 ϭ 1.09). B, apparent FRET efficiencies for p40 phox , p47 phox , and p67 phox tandems and the positive control (simple D/A tandem). C, effect of p67 phox co-expression on the p47 phox tandem and of p40 phox on the p67 phox tandem. A co-transfection of the tandem with RFP alone was used as control (n ϭ 10 -30 cells; means Ϯ S.E., raw data Fig. S2). ***, p Ͻ 0.001, Tukey's multiple comparison test.

A new 3D model of the NADPH oxidase cytosolic complex
complex average quantities. In addition, the efficiency of energy transfer depends on both the distance and relative orientation between the donor and the acceptor, as predicted by Förster theory (see "Experimental procedures"). Because the fluores-cent proteins are attached to the subunits through variable flexible linkers (Table S1), a large range of relative FP orientations are likely allowed, leading to some averaging of the orientation factor. In the frame of this paper, we will thus assume that major A new 3D model of the NADPH oxidase cytosolic complex differences in FRET efficiencies are chiefly governed by distance. This is supported by the similar FRET efficiencies observed when different D/A pairs are used or when the anchoring sites of donor and acceptor are swapped (see for example Fig. 4, C and F). In this frame, the relative FRET efficiencies observed for donor and acceptor located at different termini provide interesting topological information. The maximum apparent FRET efficiency is higher for the CC labeling than for the CN labeling (Fig. 4F). Indeed, the two C termini are known to bind to each other and should be closer than the N terminus of p67 phox and the C terminus of p47 phox , which are separated by p67 phox itself ( Fig. 1). For the NN labeling, the distance between FPs is most likely too large to observe FRET, because we have evidence for complex formation through FCCS. The upper distance limit for a FRET-positive situation is approximately twice the Förster radius of the FRET pair (see "Experimental procedures"), ϳ100 Å. To fulfill this distance condition, the N termini of both subunits should point to opposite directions in the complex.

FRET signals indicate specific interactions between p40 phox and p67 phox and between p40 phox and p47 phox
We studied similarly the interactions between FP-tagged p40 phox and p67 phox (Fig. 4, D and G, and Fig. S3B). For all tag positions (NC, NN, or CC), specific FRET was observed with the same maximum efficiency, ϳ8% at the plateau (Fig. 4, D and G). The similar maximum FRET efficiencies indicate comparable distances between the different termini of p40 phox and p67 phox .
The direct p40 phox -p47 phox interaction was questioned to be of physiological relevance or to be rather an artificial phenomenon resulting from test tube experiments in the absence of p67 phox (4). The possible interaction sites were identified in vitro to be the SH3 domain of p40 phox and the C-terminal PRR domain of p47 phox , which is also the binding site for p67 phox ( Fig. 1). We co-expressed p40 phox and p47 phox or its truncated version p47 phox ⌬Cter with NC or CC labeling (Fig. 4, E and H). We found specific FRET in both cases with equivalent E app , which indicates that the FP tags at both termini of p40 phox have a similar average geometry relative to the C terminus of p47 phox . When p47 phox ⌬Cter was used instead of full-length p47 phox , no significant FRET was observed ( Fig. 4E and Fig.  S3C). This confirms the requirement of the PRR domain of p47 phox for the interaction with p40 phox in live cells.

The cytosolic complex has a 1:1:1 composition with a high affinity between the subunits
FRET depends strongly on the number of acceptors in the direct vicinity of the donor and thus allows exploration of the stoichiometry of the interaction (18). An uneven stoichiometry of the subunits in the complex (X:Y) will result in different maximum E app when the donor and acceptor are swapped between subunits (Fig. S4). In contrast, even subunit ratios (X:X) will give the same maximum E app . On heterodimers formed by p47 phox /p67 phox , as well as by p40 phox /p67 phox , swapping donor and acceptor did not change the maximum apparent FRET efficiency (Fig. 4, F and G). We also investigated possible homodimerization of the cytosolic subunits by co-expressing each subunit tagged with donor and acceptor FPs in the same cell. In all cases, the E app scatters in the range of the negative controls (Fig. S5), indicating that there are no detectable homodimers. Taken together, our findings support a 1:1:1 stoichiometry for the cytosolic complex in the living cell, in agreement with in vitro experiments (19,20).
The amplitudes of the auto-and cross-correlation functions obtained by FCCS provide an estimate of the relative expression levels of the FP-tagged subunits and of their fraction in interaction ("Experimental procedures" and supporting information). The FCCS analysis gives qualitatively similar results with FRET-positive CC labeling or FRET-negative NN labeling ( Fig.   Figure 5. FCCS demonstrates that all p67 phox is in complex with p47 phox . A, example of auto-correlation and cross-correlation functions and their fits from individual cells. Shown are FCCS data from COS7 cells transfected with the p67 phox tandem (left panel) or p47 phox together with p67 phox (middle and right panels). We use the p67 phox tandem as a positive control for co-diffusion. B, fraction of protein in interaction, as a function of YFP:CFP ratio, as obtained from the correlation functions. At 2-fold excess of p47-YFP, all p67-CFP is in complex, and 50% of p47-YFP is in complex. Each symbol represents the means Ϯ S.E. of five to ten cells from three independent experiments. We chose cells expressing YFP-tagged p47 phox in varying quantities while keeping the CFP-tagged p67 phox concentration at a nearly constant low level. This simulates the situation in neutrophils, where the expression of p47 phox is higher than p67 phox (32). 5B). In both cases, the fraction of interacting protein is correlated to the relative expression level of the proteins. Although all p67 phox is bound in complex in the presence of a 2-fold excess of p47 phox , the fraction of p47 phox interacting with p67 phox decreases concomitantly. The apparent fraction of molecules in interaction is somewhat lower for the FRET-positive CC labeling than for the FRET-negative NN labeling, which may be ascribed to a FRET-induced decrease of the amplitude of the cross-correlation function (21).
The concentration of bound and free diffusing p47 phox and p67 phox obtained from the FCCS measurement can also be used to estimate their apparent dissociation constant, K D app (see supporting information). The median K D app value is in the range of a few hundred nanomolar, showing a high affinity between p47 phox and p67 phox in live cells. Such a high affinity, with a K D well below 1 M, is consistent with previous values obtained in vitro ranging from 4 to 32 nM (22).
In conclusion, FRET imaging and FCCS experiments in live cells clearly demonstrate specific interactions between the three cytosolic subunits of the NADPH oxidase and give new insights into their spatial organization in the complexes. In the next paragraphs, we will use this information, together with SAXS, NMR, and X-ray crystallography data, to build a 3D model of the heterotrimer. The crystallographic structure of p40 phox was solved (5), but p47 phox and p67 phox are highly flexible, and only some domains were crystallized. First, we will describe how several putative conformations of p67 phox and p47 phox were selected from SAXS experiments. Second, we will assemble the individual subunits and choose along this process the most appropriate conformation of p47 phox and p67 phox , with the help of the topological information obtained from FRET, to finally propose a 3D model of the heterotrimer compatible with our live cell experiments, as well as structural and biochemical results available in the literature.

Sets of models for p67 phox and p47 phox based on SAXS experiments
SAXS analysis consists in producing a set of atomic models compatible with the experimental SAXS curves using existing high-resolution structures of the crystallized domains, as explained under "Experimental procedures." For p67 phox , we re-examined previous experimental SAXS results (23), whereas new experiments were performed for the truncated p47 phox ⌬Cter [1-342] and for p47 phox to improve the model previously proposed by Durand et al. (24). For each of the three proteins, we retained a representative selection of possible models, which are discussed in the following steps.
The representative models of p47 phox ⌬Cter [1-342] were first examined to find a model adopting a relative orientation of its PX domain with the SH3 domains compatible with all other previous experimental observations (Fig. S6). First, in the resting state, p47 phox adopts an autoinhibited conformation in which the residues of the PX domain that interact with membrane phospholipids during the active phase are poorly accessible (25). Most of the models were rejected using this criterion (Fig. S6, A and B). Second, the PX domain interacts with the lateral surface of N-terminal SH3 domain including Arg-162 and Asp-166 (16,25). Therefore, the latter two residues have to be located at the interaction surface between PX and SH3 domains (Fig. S6C). Third, previous kinetic analysis of hydrogen/deuterium exchange coupled to MS identified the residues involved in the intramolecular interaction in the resting state of p47 phox , which should also be masked (16).
We finally retained one model of p47 phox ⌬Cter that corresponds to all these criteria (Fig. 6C). The SAXS curve calculated on this model using the program CRYSOL is shown in Fig. 6A together with the experimental data (26). This model of truncated p47 phox was then used as starting point for the fit of the experimental SAXS pattern of the full-length p47 phox protein (Fig. 6B and Table S4), resulting in a set of

A new 3D model of the NADPH oxidase cytosolic complex
full-length models carrying a mainly unstructured C-terminal part. Representative models of the full-length p47 phox protein were selected to build the whole 3D model of the heterotrimer in the next section.

Step-by-step assembly of a 3D model of the NADPH oxidase cytosolic complex
Assuming a 1:1:1 stoichiometry, the representative SAXS models of full-length p47 phox and p67 phox selected above and the crystal structure of p40 phox were then used to build a model of the heterotrimer, following a step-by-step workflow presented in Fig. S7. We know that p40 phox and p67 phox interact via their PB1 domains (Fig. 1), and the crystal structure of the p40 phox -PB1-p67 phox -PB1 interaction has been solved (27). This was used as a template to add first the whole crystal structure of p40 phox onto p67 phox -PB1 (5) (step I in Fig. S7). Then we added on the p67 phox -PB1 domain the full-length p67 phox structural models previously selected. We selected the most appropriate model based on the following criteria, illustrated in Fig. S8. First, we kept the models that had neither clashes nor sterical hindrance with p40 phox . Second, considering the maximum dimension of p67 phox determined by SAXS (D max ϭ ϳ160 Å; Table S4) (23,28) and moreover the observation of significant FRET between the two termini of p67 phox in the p67 phox tandem, the N terminus of p67 phox has to be bent toward its C terminus. Third, for the p40 phox -p67 phox interaction, the significant and comparable FRET efficiencies for all tag positions (Fig. 4G) indicates that the termini of p40 phox and of p67 phox are at similar distances and significantly below the maximum FRET-compatible distance of 100 Å. The later condition is only fulfilled when p40 phox and p67 phox adopt a crosslike spatial arrangement (Fig. S8). A representative p67 phox SAXS model compatible with all three criteria was selected (step II in Fig. S7).
The resulting model of the p40 phox -p67 phox complex became the starting point to add p47 phox . The full-length p47 phox SAXS models were added with the help of the NMR structure of the interaction between the C termini of p47 phox and p67 phox , which are bound via their PRR (p47 phox ) and SH3 (p67 phox ) domains (29) (step III in Fig. S7). Most of our fulllength p47 phox SAXS models lead to steric clashes with p40 phox and/or a distance between the N termini of p47 phox and p67 phox that is too short to account for the absence of clear FRET observed for NN labeling of these two subunits (Fig. S9). This leads to the selection of a p47 phox SAXS model where p47 phox points in the opposite direction of p67 phox in a tail-to-tail orientation. This last step results in the complete 3D model of the p40 phox -p47 phox -p67 phox heterotrimer (step IV in Fig. S7 and Fig. 7).

Evaluation of our proposed 3D model: significant features and consistency with experimental data
Our assembled model of the NADPH oxidase cytosolic complex reveals an elongated structure where functional domains are connected by flexible peptide linkers (Fig. 7). As already stated, the proposed cross-like arrangement of p40 phox and p67 phox leads to distances between their termini compatible with significant and equivalent FRET for all combinations (Fig.  7A). The interaction of p47 phox with p67 phox also brings their C termini in close vicinity, coherent with the high E app measured for CC labeling, whereas the intermediate C-N distances are consistent with a lower E app for the CN labeling (Fig. 7B). Interestingly, the FRET efficiency for the CN-tagged subunits is close to the one of the p67 phox tandem. Indeed, the C termini of p47 phox and p67 phox are very close and separated from the N terminus of p67 phox by a portion of p67 phox itself. On the other hand, p47 phox is expected to be elongated alone in solution (D max by SAXS is in the order of 125 Å; Table S4), but the presence of intramolecular FRET in p47 phox tandems in live cells suggests that the subunit can adopt a more compact conformation. Indeed, the flexibility of its unstructured C terminus clearly allows movements of the N-terminal domain (featured by an arrow in Fig. 7C). The absence of FRET between NNlabeled p47 phox and p67 phox suggests that p47 phox cannot flip

A new 3D model of the NADPH oxidase cytosolic complex
completely toward p67 phox but might adopt either a more elongated or more condensed conformations, as well as different angles with respect to the long axis of p67 phox . Our model should thus be viewed as a relatively flexible structure with significant variability around its multiple hinges. The D max values of p67 phox and the p67 phox -p40 phox complex are close (Table S2). The D max values of p67 phox -p47 phox and p67 phox -p47 phox -p40 phox are similar as well (Table S2). This means that the presence of p40 phox does not change the overall dimensions of the complex. This is again consistent with the cross-like spatial arrangement of p40 phox with p67 phox and with the observation of the same E app of the p67 phox tandem with or without p40 phox . Finally, we find that the calculated SAXS curves using the program CRYSOL on the p67 phox -p47 phox -p40 phox model is in reasonable agreement with the experimental curves published by Yuzawa et al. (Fig. S10) (28).

Discussion
Here we presented an integrated workflow to analyze the intermolecular interactions and the conformation of a complex of proteins formed by structured domains separated by unstructured segments in living cells. This analytical strategy may be adapted to any cytosolic protein complex composed of partially disordered subunits.
We first demonstrate the complementarity of FRET-FLIM and FCCS to characterize live cell interactions, estimate binding affinity, and obtain topological information using the same FP labeling for both techniques. This was made possible because of the new cyan variants mTurquoise and Aquamarine in the classical CFP/YFP FRET pair (7). Their increased brightness and photostability allowed both improved performances as donors for FLIM and full suitability for two-color FCCS experiments.
The dynamic NADPH oxidase complex depends on protein assembly for activation, and thus the development of inhibitors of this assembly is an attractive concept to regulate its activity (30,31). For the p40 phox -p47 phox -p67 phox complex, there is a discrepancy between the 1:1:1 stoichiometry found in in vitro experiments (19,20) and the 2-3 times higher expression level of p47 phox compared with p67 phox in neutrophils (32). In addition, this higher amount of p47 phox compared with p67 phox implies that p47 phox is present as both as free and a complexbound subunit (32). However, it is not known whether free p67 phox exists. Our results revealed two pairwise interactions with a one to one stoichiometry, p47 phox -p67 phox and p40 phox -p67 phox , that strongly suggest that the ternary complex is assembled in a 1:1:1 stoichiometry inside a live cell. Furthermore, in the presence of an excess of p47 phox , all p67 phox were bound in complex. In solution, ϳ10 -20% of recombinant p47 phox and p67 phox proteins have been detected as dimers (22). We did not observe any dimers by FRET-FLIM. Indeed, in live cells, the unbound cytosolic subunits may encounter a large diversity of potential partners that likely limits dimer formation.
Our live cell FRET results also provide topological information that nicely complements structural data from X-ray crystallography, NMR, and SAXS experiments obtained on purified recombinant proteins. The combination of SAXS data and sin-gle-molecule FRET in vitro is very useful to investigate protein folding by integrating complementary information (33)(34)(35). The combination of SAXS with FRET measurements of FPlabeled proteins in cellulo provides lower spatial and temporal resolution; however, it affords access to the conditions in living cells. To our knowledge, such an approach has not been described before. These pieces of information were integrated to build a new model of the cytosolic NADPH oxidase complex with several noticeable features: (i) p47 phox and p67 phox interact via their C termini in a tail-to-tail configuration, and their N termini are over 100 Å apart; (ii) p67 phox and p47 phox are not fully elongated; and (iii) p40 phox and p67 phox adopt a cross-like conformation (Fig. 7). The model is consistent with the idea that p47 phox is required for the initial assembly of the functional oxidase. Indeed, the PX domain and the SH3 domains of p47 phox are located on one end of the complex, ideally positioned to interact with phospholipids (PX domain) and with p22 phox (SH3 domains) to initiate oxidase assembly. This may then bring p40 phox and p67 phox closer to the membrane. Indeed, the PX domain of p40 phox also interacts with membrane lipids, and the N terminus of p67 phox requires several activation steps mandatory for superoxide production. The flexible hinge between the N-terminal half of p47 phox and the rest of the trimeric complex may be required to establish the initial contact with phospholipids and p22 phox and then bend the complex to bring p67 phox closer to Nox2. In addition, the proximity of the C terminus of p47 phox with the SH3 domain of p40 phox in the model raises the question of whether p40 phox contributes to stabilize the C terminus of p47 phox at the SH3 domain of p67 phox . The position of the p40 phox PX domain in the middle of the elongated complex is unfavorable for a role of this domain in the initiation of oxidase assembly because it would require a lateral attachment of the complex on the membrane. However, if p47 phox leaves the active complex as recently suggested (36), the p40 phox PX domain would be in a favorable position to bind to phosphoinositides in the membrane after its dissociation from PB1 (5), and this would help keeping p67 phox in place (37).
In p67 phox , the TPR and the activation domain are essential for the assembly with Rac and Nox2 and, ultimately, for activation of the oxidase. In the proposed model of the heterotrimer, presented here in the resting state (Fig. 7), the N-terminal region of p67 phox is well-exposed and accessible. We assume that the global orientation and the internal flexibility between its domains is fully compatible with the binding to Rac or to Nox2 through the activation domain (38). At this stage of characterization of the ternary cytosolic complex, it seems that the main limiting step toward activation is p47 phox that needs to be activated to promote assembly of the whole cytosolic complex at the membrane. Preventing the conformational changes of the N terminus of p47 phox would probably block the subsequent interactions with the membrane and p22 phox and prevent oxidase activation. Therefore, our approach provides critical information for the design of inhibitors that would interfere with key steps in the activation process. Furthermore, the data we obtained and the model are a starting point to investigate the changes that may occur during activation of the oxidase. They may also serve to compare WT subunits with mutations found A new 3D model of the NADPH oxidase cytosolic complex in patients with chronic granulomatous disease to understand the phenotype of this oxidase deficiency.

Plasmid library and transfection
Plasmids encoding full-length human p47 phox (NCF1) and p67 phox (NCF2) both embedded in a pEGFP-N1 vector (Clontech) (36) and p40 phox (NCF4) embedded in a CMD8 vector (gift from Marie Claire Dagher) were used as a starting point to build a full library of N-and C-terminal tagged fusion proteins. The cDNA of the subunits were cut out with restriction enzymes or amplified by PCR and inserted in either cyan mTurquoise or Aquamarine-N1 or -C1 vectors (variants of the pECFP vectors: mTurquoise: T65S/S72A/H148D/S175G/ A206K (39), Aquamarine: T65S, H148G (40)), yellow Citrine-N1 or -C1 vectors (the mutation Q69M was introduced into EYFP) (8), or the red mCherry-C1 vector (Clontech, Takara Biotechnology, Co., Ltd.). The internal start codon (ATG) was removed by PCR from the FP-N1 vectors or from the subunits embedded in FP-C1 vectors. Table S1 displays an overview of the constructs. To build the plasmids coding for RFP-2A-p67 phox or p40 phox , we ordered a synthetic gene coding for mCherry-P2A (P2A ϭ ATNFSLLKQAGDVEENPGP) (12) framed by AgeI and HindIII restriction sites (Eurofins). The two enzymes were used to cut out the FP and insert mCherry-2A in pECitrine-p67-C1 or pEmCherry-p40-C1. Primers were purchased from Eurogentec (Kaneka Corp., Tokyo, Japan). Plasmids were amplified in DH5␣ Escherichia coli, DNA was purified with E.N.Z.A. mini kit 2 (Omega Bio.Tek). For transfection, cells (COS7 or COS Nox2/p22 ) were seeded either on glass coverslips (Ø 25 mm, thickness of 0.13-0.16 mm) in 6-well plates for microscopy or in 24-well plates for luminometry 1 day before the experiment. The cells were transiently transfected with XtremeGene HP (Roche Diagnostics) following the supplier's instructions and used 24 -48 h after transfection. Transfection efficiency, monitored by flow cytometry, was constantly between 20 and 30% for triple transfection. Based on the high reproducibility of transfection efficiencies, we assumed that the subunits without FP tag were present in the cells at the same average level as the FP-tagged ones.

Cell culture
COS7 cells were purchased from ATCC and cultured following the supplier's instructions. COS7 cells stably expressing Nox2 and p22 phox (COS Nox2/p22 ) were kindly provided by M. Dinauer (Washington University School of Medicine, St. Louis, MO) and cultured in medium containing selecting antibiotics (10).

Fluorescence lifetime imaging microscopy
Time-resolved laser scanning TCSPC microscopy was performed on a custom made microscope as described previously (40). Briefly, the setup is based on a TE2000 microscope with a 60ϫ, 1.2NA water immersion objective (Nikon). The epifluorescence pathway is equipped with an mercury lamp, a set of filter cubes for the different FPs, and a CCD camera (ORCA-AG, Hamamastu Photonics; Table S3). The TCSPC path is equipped with pulsed laser diodes (440 nm for CFPs; 466 nm for YFPs, PicoQuant) driven by a PDL800 driver (20 MHz, Pico-Quant). The C1 scanning head (Nikon) probes a 100 ϫ 100-m maximum field of view. To select the FP fluorescence, dichroic mirrors and filter sets were used before the detection by a MCP-PMT detector (Hamamatsu Photonics). The signals were amplified by a fast pulse preamplifier (Phillips Scientific) before reaching the PicoHarp300 TCSPC module (PicoQuant). Counting rates were routinely between 50,000 and 100,000 counts/s. Transfected cells were kept in PBS at 20°C and studied for 2 h maximum in an Attofluor cell chamber (Thermo Fisher Scientific).
The lifetime of a fluorophore is an intensive property, independent of its concentration, which can be precisely monitored even in live cells. A precision of a few percentages on lifetime is common (17,(41)(42)(43)(44). The TCSPC fluorescence decay of all the pixels of the cytosol was computed by the SymPhoTime software (Fig. 3A). The decays were fitted with a monoexponential fit function (Equation 1) for the control cells expressing a donor fusion protein (without acceptor), the donor being Aquamarine, mTurquoise, and Citrine (Fig. 3A, left panel), where C is the constant background. Because the donor has a monoexponential lifetime, its fluorescence decay in the presence of FRET to an acceptor can be fitted with a biexponential function (Equation 2, and Fig. 3A, right panel), where C is the constant background. ␣ long and ␣ short are the proportions of a long ( long ) and a short ( short ) lifetime component, respectively. For fitting, we used a custom-made procedure in IGOR Pro (WaveMetrics). The quality of the fits is evaluated by the weighted residual distribution and Pearson's 2 test (Fig. 3A). The apparent FRET efficiency E app is calculated using Ͻ DA Ͼ and donor (Equation 4).

A new 3D model of the NADPH oxidase cytosolic complex
It was calculated on a cell by cell basis from the fluorescence decays obtained for each cell using for donor the lifetime of the same donor/subunit fusion protein alone.
The apparent FRET efficiencies measured in each cell are complex averages over highly heterogeneous populations of donor molecules engaged in different FRET interactions. Fundamentally, these FRET efficiencies depend on both the distance and relative orientation between donor and acceptor. The determination of distances between FRET partners thus requires some assumption on the orientation factor 2 (45). Unfortunately, the dynamic isotropic regime leading to the popular value 2 ϭ 2/3 cannot be applied to fluorescent proteins, because their rotational correlation time in water is ϳ14 ns i.e. three to four times longer than the lifetime of the donor excited state (46,47). In the viscous environment of living cells, FPs must be considered as static on the time scale of the FRET interaction. On the other hand, because FPs are fused to the subunits through flexible linkers, they are expected to adopt a whole set of conformations, leading to broadly distributed, but probably nonisotropic and nonhomogeneous relative orientations and distances. Therefore, the precise calculation of distances from our FRET data, as proposed for example by Vogel et al. (48), is mostly out of reach. Instead, we used the apparent FRET efficiencies to establish distance limits, to guide topological reasoning, or to provide comparative information. Indeed, whatever the effective value of 2 , the observation of a significant FRET indicates some proximity within the range of the critical Förster distance R°. Typical Förster distances of most fluorescent protein FRET pairs, calculated using 2 ϭ 2/3, lie in the range of 50 -60 Å (48). R°evolves as ( 2 ) 1/6 and thus varies only slowly with the orientation factor. Using 2 ϭ 0.476, a value applicable to fully isotropic and homogeneous static averaging (17,45,49), we obtain quite similar values of R°ϭ 57 Å for Aquamarine/Citrine and 55 Å for Citrine/mCherry. For the maximum but highly unlikely value 2 ϭ 4, these distances would at most extend to 81 and 78 Å, respectively. In cases of unfavorable relative orientations, R°could on the contrary decrease significantly below 50 Å. Conversely, as the D/A distance approaches twice R°, the FRET efficiency will tend virtually to 0. As a rule of thumb, assuming some degree of orientational and conformational averaging, and considering our practical detection limits, we took 100 Å as the upper distance limit to observe FRET in this study.
The FP-tagged subunits are transiently expressed, and the level of expression may strongly vary between cells. We recorded intensity images before any FLIM experiments on the same field of view. The intracellular concentrations are proportional to the average fluorescence intensities. E app was commonly plotted as a function of the fluorescence intensity of the acceptor in the cell, and I(A) was plotted to observe a FRETpositive situation (17, 50 -53). As a negative control, the donor/ subunit fusion protein was co-expressed with the free acceptor. This control is useful to (i) determine whether E app is significant and (ii) to set the threshold for the maximum value of I(A) beyond which unspecific FRET caused by molecular crowding starts to contribute to the signals. Indeed, when the cells are crowded with the acceptors, I(A) becomes very high, and the donors show unspecific FRET with the acceptors nearby just because of the spatial proximity (17). We discarded such cells.
The plotting of E app against [A]/[D] required the calibration of fluorescence intensities measured in both donor and acceptor channel. The ratio of fluorescence intensities, I(A)/A(D), was transformed in the concentration ratio of fluorescent proteins, [A]/[D], using a custom calibration procedure of the microscopy setup (see supporting information and Figs. S11 and S12).

Fluorescence cross-correlation spectroscopy
A confocal microscope Leica TCS SP8 SMD (Leica Microsystems) was used. It is equipped with a DMI 6000 CS stand and a 63ϫ/1.2 HC PL APO water immersion objective, a continuous argon laser (514 nm from Leica TCS SP8), and a diode pulsed laser (440 nm, PicoQuant). The signal was selected by 505-nm dichroic mirror and two bandpass filters (BP 478/22, BP 540/ 30) and detected by two APD detectors (PicoQuant). For the detection, the SMD module is constituted of a PicoHarp3000 system for TTTR mode of single photon counting (PicoQuant). The cells were kept in an air-conditioned chamber at 30°C. To correct for the cross-talk between the CFP and YFP detection channels on the amplitudes of the correlation functions, we used a variant of fluorescence correlation spectroscopy called fluorescence lifetime correlation spectroscopy (54). The fluorescence signal of each channel was corrected from spectral bleed-through using fluorescence lifetime correlation spectroscopy filters as described elsewhere (55,56). In this method combining TCSPC and FCS, the different FCS contributions are separated by using the different fluorescence decay pattern of both fluorophores. The fluctuations of fluorescence intensity were auto-and cross-correlated, and the resulting curves were analyzed using a standard pure diffusion model with Sympho-Time (PicoQuant).
The auto-correlation function of fluorescence fluctuations returns the time needed for the fluorophore to cross-the confocal observation volume, V FP , and the number of molecules, N FP , present in V FP . N FP is the inverse of the amplitude to the auto-correlation function, G FP (0) (21). The concentrations of FPs in the observation volume are calculated as following where N A is the Avogadro number (Equation 5).
The amplitude of the cross-correlation function, G cross (0), allows the estimation of the concentration of the stoichiometric p47 phox -p67 phox complex. p47 phox and p67 phox are respectively labeled with YFP and CFP (Equation 6).
In practice, the observation volumes in the two channels, V YFP and V CFP , and V cross have always different sizes and are not perfectly coincident. In addition, a fraction of FPs might be nonfluorescent because of nonmatured protein, dark state, or photobleaching (21). The later was kept here below 10%. To minimize the impact of these limits on the detection of inter-A new 3D model of the NADPH oxidase cytosolic complex actions, the fractions, X, of co-diffusing fluorophores that are also the fractions of protein in interaction, were normalized to the average fractions obtained for the p67 phox tandem, whose two tags diffuse together (Equations 7 and 8).
In addition, the relative amount of both partners was estimated as follows (Equation 9).
[YFP] cplx [CFP] cplx ϭ The details for the computation of K D app can be found in the supporting information.

Protein preparation, SAXS data acquisition, and processing
p47 phox constructs were expressed as a GST fusion protein using a pGex-6P-1-derived vector containing the cDNA corresponding to the full amino acids sequence (from 1 to 397) for the full-length p47 phox or to a sequence coding for amino acids 1-342 for the p47 phox ⌬Cter construct lacking the C-terminal end of the protein after the AIR motif. For both constructs, the production protocol was the same. The proteins were expressed in E. coli, strain BL21(DE3). Expression was induced with 0.5 mM isopropyl ␤-D-thiogalactopyranoside, when the cell culture reached an OD of 0.6 at 600 nm. Temperature was then shifted to 20°C for an overnight induction. The cells were harvested and resuspended in chilled lysis buffer (50 mM Tris, pH 7.5, 0.3 M NaCl, 1 mM EDTA, 2 mM DTT, and Complete EDTA-free protease inhibitor (Roche Diagnostics)). All following operations were carried out at 4°C. The cells were disrupted by sonication and then centrifuged at 40 000 rpm for 40 min in a Beckman 45 Ti rotor. The supernatant was loaded onto a 4-ml gluthathione-Sepharose 4B column (GE Healthcare) equilibrated in the lysis buffer. Proteins were eluted at 1 ml/min with 50 ml of elution buffer (50 mM Tris, pH 7.5, 50 mM NaCl, and 10 mM glutathione). Fractions containing GST fusion of the corresponding p47 phox construct were pooled and digested overnight at 4°C with PreScission protease (70 enzyme units per 40 mg of protein). Because of the genetic construction, digested p47 phox possesses 10 and 7 additional residues at its N and C terminus, respectively, and p47 phox ⌬Cter possesses only 10 additional residues at its N terminus. Digestion products were loaded at a flow rate of 1 ml/min onto a MonoS column (GE Healthcare) equilibrated in 50 mM Hepes, pH 7.5, 50 mM NaCl, 1 mM EDTA, and 2 mM DTT. The proteins were eluted with a 40-ml linear gradient of NaCl (50 -500 mM). The resulting protein fractions containing the corresponding p47 phox construct were concentrated on a centrifugal concentration device with a 10-kDa cutoff. The protein was then diluted 20 times in 50 mM Hepes, pH 7.5, 100 mM NaCl, 1 mM EDTA, 2 mM DTT, and 5% glycerol and finally reconcentrated at different concentrations at which aliquots of p47 phox , or p47 phox ⌬Cter, were taken and stored for SAXS measurements. For p67 phox , p47 phox , and its truncated version, SAXS data processing led to a molecular mass close to that derived from the sequence showing that the protein solutions were devoid of oligomers (Table S4).
SAXS measurements on p47 phox ⌬Cter were performed on a laboratory instrument Nanostar (Bruker Nanostar, ϭ 1.54 Å). SAXS data on full-length p47 phox were collected on the SWING beamline (SOLEIL synchrotron, St. Aubin, ϭ 1.0Å). Both proteins were studied at different concentrations (Table S4). SAXS data were normalized to the intensity of the incident beam, averaged, and background-subtracted using the program package PRIMUS (57); intensities were put on an absolute scale using water scattering. The final pattern used for fitting was obtained by extrapolating to infinite dilution of the set of curves recorded at the different concentrations.
p47 phox ⌬Cter was modeled using the program BUNCH (58), which combines rigid-body and ab initio modeling approaches. The program starts from known structures of the two domains PX and SH3 tandem (Protein Data Bank code 1KQ6 and 1NG2, respectively) and finds the optimal positions and orientations of domains and probable conformations of the linkers by fitting the SAXS curve calculated on the model to the experimental curve. In a final step, we substituted the dummy residues of the linkers with all atom descriptions using the programs PD2 (59) and SCWRL4 (60). An ultimate adjustment was performed using the program CRYSOL. The modeling was repeated 100 times. Finally, models giving the best agreement with the experimental curve (lowest 2 values) were selected. The same approach was used for full-length p47 phox using the most probable model describing p47 phox ⌬Cter and the PRR domain (extracted from the Protein Data Bank code 1K4U) as a starting point. Models of p67 phox were built in the same way starting from the known structures (23). This modeling has shown that p67 phox is able to adopt a great variety of very different conformations. Not surprisingly the SAXS data can be fitted using an ensemble of models (EOM approach) (61). All SAXS data are shown in SASBDB (https://www.sasbdb.org) 6 under the SAS codes SASDEJ3, SASDEK3, and SASDEL3 (62).

Molecular structure alignment
Protein Data Bank structures and SAXS-based models were imported into the PyMOL Molecular Graphics System, version 1.8 Schrödinger, LLC.

Statistical analysis
The data are represented as means of at least three independent experiments Ϯ S.E. Significance was tested with one-way ANOVA followed by a Tukey's multiple comparison test using GraphPad Prism version 5.