Clathrin complexes with the inhibitor kappa B kinase signalosome: imaging the interactome

Abstract Many receptors involved with innate immunity activate the inhibitor kappa B kinase signalosome (IKK). The active complex appears to be assembled from the two kinase units, IKKα and IKKβ with the regulatory protein NEMO. Because we previously found that RNA silencing of clathrin heavy chains (CHC), in transformed human lung pneumocytes (A549), decreased TNFα‐induced signaling and phosphorylation of inhibitor kappa B (IκB), we hypothesized that CHC forms cytoplasmic complexes with members of the IKK signalosome. Widely available antibodies were used to immunoprecipitate IKKα and NEMO interactomes. Analysis of the affinity interactomes by mass spectrometry detected clathrin with both baits with high confidence. Using the same antibodies for indirect digital immunofluorescence microscopy and FRET, the CHC–IKK complexes were visualized together with NEMO or HSP90. The natural variability of protein amounts in unsynchronized A549 cells was used to obtain statistical correlation for several complexes, at natural levels and without invasive labeling. Analyses of voxel numbers indicated that: (i) CHC–IKK complexes are not part of the IKK signalosome itself but, likely, precursors of IKK–NEMO complexes. (ii) CHC–IKKβ complexes may arise from IKKβ–HSP90 complexes.

The 180-kD clathrin heavy chains (CHC) are present in abundance in all cells and can assemble into an extended variety of shapes and sizes to interact with proteins during endocytosis and exocytosis. However, CHC complexes can influence the signal trajectory for several receptors toward apoptosis and MAPK recruitment (Pierce et al. 2000;Rakhit et al. 2001;Schneider-Brachert et al. 2004;McLaughlin et al. 2006McLaughlin et al. , 2008. We have previously reported that silencing CHC with specific small interfering RNAs (siRNA) significantly attenuated TNFa-induced phosphorylation of IjBa (Escobar et al. 2006). Depleting CHC also decreased the production of NF-kB-regulated MCP1 (monocyte chemoattractant protein 1), ICAM1, and phosphorylation of p65 Rel A (another recognized IKK activity; Escobar et al. 2006). Interestingly, CHC silencing also appeared to affect levels of phosphorylated I-jB and p65 NF-jB compared to unstimulated control cells (Escobar et al. 2006). Recently, Kim et al. (2011) have found that CHC regulates basal IKKa activity in unstimulated (resting) epithelial cells. The interaction appeared independent of the clathrin light chain or endocytosis. We hypothesized that clathrin forms complexes with both IKKa and IKKb in resting cells and sought (i) to detect the complexes by affinity pull-down and MS proteomics and (ii) to localize their intracellular distribution by fluorescent immunostains and FRET.

Flow cytometry
Isolated A549 cells, in suspension (1 9 10 6 cells), were fixed and permeabilized with acetone/methanol (70/30) at À20°C for 10 min. Cells were washed three times with cold PBS. The cells were then incubated with 10% normal donkey serum in PBS for 1 h at room temperature to block nonspecific serum binding sites. After removal of the blocking serum, cells were incubated with primary antibodies overnight at 4°C. Excess antibody was removed by three washes with cold PBS. Fluorophore-conjugated, species-specific donkey secondary antibodies were added for 1 h at room temperature. After removal of excess antibodies by three washes with cold PBS, cells were resuspended in a 4% PFA-PBS solution and analyzed on a Beckman FC500 flow cytometer within 24 h.

Coimmunoprecipitation (IP)
Cells were grown in 100-mm dishes until reaching 90-100% confluence and rinsed twice with 1 mL PBS per well at room temperature. Cells were lysed by adding 1 mL/ dish of M-PER lysis solution (Pierce Biotechnology, Inc., Rockford, IL) at room temperature for 5 min and then scraped with a rubber policeman into 2-mL Eppendorf tubes and centrifuged at 14,000 RPM for 10 min to eliminate particulates. Each IP tube received 400 lL of cell lysate with 2 lg of each IKK antibody conjugated to agarose. The corresponding conjugated isotype or PBS (40 lL) was used as controls. The incubations were carried at room temperature for 1 h and the pellets underwent five cycles of washing with 1 mL PBS. The final bead pellets were resuspended in 150 lL PBS with 50 lL 4 9 Laemmli digestion buffer (Pierce Biotechnology, Inc.), boiled for 5 min, and spun at 5000 g for 10 min. The bead pellets were discarded and the supernatants were processed for mass spectrometry and proteomic analysis.
The IP antibodies used were as follows: agarose-conjugated mouse anti-IKKa and rabbit anti-NEMO (Santa Cruz Biotechnology, Inc, Santa Cruz, CA); agarose-conjugated mouse IgG and rabbit IgG (Santa Cruz Biotechnology, Inc) were used as nonspecific controls.

Liquid chromatography-tandem mass spectrometry
The SDS-polyacrylamide gel electrophoresis was performed using 4À12% Bis Tris gel system (Invitrogen-Novex, Carlsbad, CA) according to the manufacturer's protocol. The gel was stained with Coomassie brilliant blue stain (Invitrogen) and slices of equal size were excised from each sample lane, reduced using 10 mmol/L DTT at 65°C for 45 min, alkylated with 55 mmol/L iodoacetamide for 0.5 h at ambient temperature in the dark and digested in-gel with sequencing grade porcine trypsin (Promega, Madison, WI) overnight at 37°C. Peptides were extracted three times from the gel using 50% ACN, 1% FA, concentrated by SpeedVac to a desired volume (~16 lL), and subjected to LC-MS/MS analysis. Nano-flow reverse phase LC-MS/MS was performed using a capillary HPLC system (Agilent 1200, Palo Alto, CA) coupled with a linear ion trap mass spectrometer LTQ-FT Ultra Hybrid ion cyclotron resonance mass spectrometer (Thermo Fisher; San Jose, CA) through an inhouse built nanoelectrospray ionization source. Tryptic peptides were preconcentrated and desalted onto a C 18 trap column ZORBAX 300SB-C 18 , (5 lm i.d. 9 5 mm; Agilent Technologies, Santa Clara, CA) with 5% ACN, 0.1% FA at a flow rate of 15 lL/min for 5 min. The separation of the tryptic peptides was performed on a C 18 reverse phase column (75 lm ID 9 360 lm OD 9 100 mm length) packed in-house with 4 lm 100 A pore size C 18 reversed-phase stationary phase (Synergy; Phenomenex, Torrance, CA) kept at a constant 40°C using an in-house built column heater at a flow rate of 380 nL/min. The mobile phases consisted of 5% acetonitrile with 0.1% formic acid (A) and 95% acetonitrile with 0.1% formic acid (B), respectively. A 90-min linear gradient from 5 to 50% B was typically used. Data acquisition was performed using the instrument supplied Xcalibur (version 2.0.6, Thermo Scientific, San Jose, CA) software. The LC runs were monitored in positive ion mode by sequentially recording survey MS scans (m/z 400-2000), in the ICR cell, while three MS2 were obtained in the ion trap via CID for the most intense ions.

Database searching, protein identification
Peptide identification was carried out using MASCOT server (Version 2.2, Matrix Sciences Ltd, London, U.K.) for MS/MS spectra assignment to the Homo sapiens subset of the SwissProt database. Peptide tolerance was set at AE 10 ppm with MS/MS tolerance set at AE0.6 Da. Trypsin specificity was used allowing for 1 missed cleavage. The modifications of Met oxidation, protein N-terminal acetylation, and peptide N-terminal pyroglutamic acid formation were allowed for, and Cys carbamindomethylation was set as a fixed modification. The results were exported into Scaffold (Proteome Software, Proteome Software, Inc, Portland, OR) that was used to filter and compare MS/MSbased peptide and protein identifications. Peptide identifications were accepted if they could be established at >95.0% probability as specified by the Peptide Prophet algorithm. Protein identifications were accepted if they could be established at >99.0% probability and contained at least two identified unique peptides. The results were converted into txt files for Fold Change calculation (FC score) and Significance Analysis of INTeractome (SAINT) scoring http://www.ncbi.nlm.nih.gov/pubmed/22948729 and http://www.ncbi.nlm.nih.gov/pubmed/21131968 to generate a ranked list of putative interactors using the tool on the CRAPome.org site (Mellacheruvu et al. 2013).
Affinity controls deposited in CRAPome.org used were all entries in the database that used agarose as the affinity support, were from HEK293, HeLa and Jurkat cells, and used anti-GFP or IgG as the affinity reagent. These parameters resulted in 11 controls (CC42,44,45,46,47,48,195,196,197,198,and 199) used for this analysis. FC-A scoring parameters: user controls, default background estimation, and average combining of replicates. FC-B scoring parameters: all controls (user & CC), stringent background estimation, and geometric combining of replicates. SAINT scoring parameters: User controls, n-burn 2000, n-iter 4000, LowMode 0, MinFold 0, Normalize 1, and geometric combining of replicates. As shown in Table 1, protein rank was generated by summing the two spectral counting methods (FC_A, FC_B) with the SAINT score (9100).

Immunofluorescent microscopy and FRET analysis
Cells were grown to 70-80% confluence on glass slides and were washed with phosphate-buffered saline solution (PBS) prior to being fixed and permeabilized with 70/30 acetone/methanol solution. They were then treated with 10% normal donkey serum (Jackson Immunologicals, Westgrove, PA) in PBS for 1 h, followed by primary antibodies to IKKa, IKKb, NEMO, CHC, HSP-90 (Santa Cruz Biotechnology, Inc) or isomolar, species-specific IgG (Pierce Biotechnology, Inc.) and left overnight at 4°C in a humid slide chamber. The slides were washed three times with PBS and the cells were incubated for 1 h at room temperature with the following fluorochrome-conjugated secondary antibodies: donkey anti-rabbit-Cy5, donkey anti-goat-Cy5and donkey anti-mouse Cy3 (Jackson Immunologicals) or donkey anti-rabbit Alexa 488, and donkey anti-goat-Alexa 488 (Invitrogen-Molecular Probes, Carlsbad, CA) as appropriate (Mandal et al. 2008). Nuclei were stained with bis-Benzamide (Sigma, St. Louis, MO). Fluorescent resonance energy transfer (FRET) images were acquired using a Marianas imaging station (Intelligent Imaging Innovations, Denver CO) based on a Zeiss 100 m Axiovert microscope, using a Zeiss 639 Plan-Apochromat objective (1.4 N/A), a Sutter Xenon light source and a Cooke SensiCam (1376 9 1040 pixel resolution, The Cooke Corporation, Romulus, MI, USA). Chroma Sedat filter sets with single emission and excitation filter, and a multiband pass dichroic, were used for emission detection. To obtain FRET images, a Z-plane stack with >20 planes at 0.2 lm was acquired and images were processed using a constrained iterative deconvolution algorithm based upon acquisition-specific point-spread functions. For each Z-stack, six channels were captured with the same exposure times (except the nuclear stain).
Images were masked to select the voxel intensities that were above nonspecific binding (obtained with isotypes at the same concentrations as the primary antibodies). Masks were created for individual fluorophores (Alexa 488, Cy3 and Cy5) and for the fluorescence energy transfer detected between the FRET pairs (Alexa 488:Cy3, Cy3: Cy5). Mask operations calculated voxels containing donor, acceptor and positive transfer channels. The corrected FRET (cFRET) was calculated as cFRET = Transfer-Fd/D donor-Fa/Aa acceptor as reported (Gordon et al. 1998;Berney and Danuser 2003). The intensity of the positive voxels in any given cell area is reported in linear pseudocolor where black is "cold" (no cFRET) and red is "hot" (high cFRET voxels). A legend demonstrating these results is displayed on each image.
To further control for false-positive FRET, each result was tested with two different secondary antibodies to allow us to determine significant cFRET signal for each protein pair when the labeling secondary was either a donor, or an acceptor. Thus, if cFRET is reported herein, it has been found to be present regardless of the fluorescently labeled secondary antibody used; likewise, protein pairs reported to be negative for cFRET did not have a positive result in either combination in areas of overlap.

FRET resolution measure
Under ideal conditions of appropriate spectral overlap and averaged dipole orientation, FRET between fluorescent dyes in aqueous media is 50% at 5-6 nm distance (Berney and Danuser 2003;Lakowicz 2006). In select cases, FRET measurements can establish distance estimates of molecular proximity. Unfortunately, a lack of FRET signal does not necessarily reflect lack of proximity between epitopes In order to span larger molecular complexes, we and others have used antibodies (mean epitope to fluorophore spacing of about 6 nm) instead of direct labeling with dyes or fluorescent protein constructs (<2 nm; Konig et al. 2006;McLaughlin et al. 2006McLaughlin et al. , 2008Wei et al. 2006;Mandal et al. 2008). Although the theoretical resolution of neighboring epitopes could be degraded to~30 nm (2 9 12 nm the maximum possible length of two primary/secondary antibody pairs, plus the interfluorophore distance), this is an improvement over the best possible deconvolved optical resolution by 5-to 10-fold (where z resolution is 130 nm). Longer spacer arms also provide conformational flexibility purveying a much wider range of favorable fluorophore orientations, even as the degraded distance resolution is still acceptable for detecting large oligomeric conglomerates (Konig et al. 2006;Wei et al. 2006). Voxel counts have been used to determine equilibration kinetics across compartments in cells. The calculated resolution of xyz dimensions of voxels obtained in the current conditions (based on objective, NA, point spread function) are around 10 2 9 10 2 9 (2 9 10 2 ) nm. Therefore, the density of individual 5-20 nm sized protein complexes could be 10-50 units, along the diagonal of the voxel. Thus, the overlap of FRET pairs in a voxel suggests congestion of different binary complexes that are perhaps no more than a few dozen molecular diameters away.
The number of voxels satisfying a threshold signal can reflect the localized amount of the species of interest across any cell, while the intensity remains a more complex function. As in conventional immunofluorescent microscopy, the recorded intensity of a voxel in FRET images is the summed emission from fluorophores. Thus, the voxel's FRET intensity depends on the number of molecules of the dye in that volume, plus the distribution of FRET efficiencies for individual excitation pairs. Other physical factors, such as quantum yields of fluorophores and refractive indices, are presumed constant for group comparisons (assuming uniform labeling across cells and in replicate experiments).

Data analysis
All experiments were repeated at least three times. Data are presented as mean AE SEM. Comparisons between two groups were assessed by t-test, and those among three or more groups were assessed by analysis of the variance using JMP 5.0 software (SAS Institute, Inc. Cary, NC USA), and presented with Microsoft Excel 2003 graphs. We accepted statistical significance for values of P < 0.01. Mean fluorescent intensities were determined with the Intelligent Imaging Innovations Slidebook 4.1 software (Intelligent Imaging Innovations, Denver, CO).

CHC immunoprecipitates with IKKa and NEMO
Using commercially available agarose-conjugated antibodies for affinity pull down, MS analysis revealed at least 91 proteins that produced at least two identifiable fragments with the anti-IKKa antibody and over 200 with the anti-NEMO antibody in at least one replicate run. The top three proteins in both pull downs were the canonical members of the IKK signalosome. The isotype antibody pull down produced no hits for these in any run, leading to high scores for both FCa and FCb (Table 1). We did not use commercially available agarose-conjugated IKKb antibody for pull downs because it is directed to an epitope at the C-terminus of the protein, which could be sterically obstructed by bound NEMO (Fontan et al. 2007;Hayden and Ghosh 2008). The clathrin heavy chain 1 (also known CLH1, CLTC, and CHC), the focus of this study, emerged as one of the top 20 partners of IKKa and NEMO (Table 1). Three other proteins (ILF2, SUB1, and TFG) appear on both pull-down interactomes. The functions of these proteins and their putative partners can be found on the curated NCBI website. TFG, a TRK-fused gene, has been directly linked to activation of the NF-kB pathway. ILF2 is one of the proteins making up the heterodimeric NFAT transcription factor. The complex has been shown to repair DNA breaks and may also negatively regulate micro-RNA processing. SUB1 is also involved with DNA repair. SUB1 may play complex roles in the steps of gene expression affecting initiation, elongation, termination, and re-initiation by RNA polymerase. Curiously, many proteins in both interactomes appear to be involved with RNA processing. For example, interacting with IKKa are DHX9, an RNA helicase, and RS16 a component of the 40S subunit of the ribosome. With NEMO, PABPC1 is a polyA-binding protein, while RBM14 is a ribonucleoprotein that acts as an RNA splicing modulator. Over six members of the heteromolecular human ribonuclear protein complex were detected with high confidence in both interactomes. This despite the fact that these proteins appear regularly in the CRAPome with agarose-conjugated antibodies and are discriminated severely by the FCb score. In our experiments, we found no peptide fragments when using agarose-conjugated irrelevant antibodies, suggesting novel RNA linked processes, unrelated to NF-kB activation or clathrin. Remarkably, none of the well-known partners of clathrin involved in endocytosis, including the stoichiometric clathrin light chain, were ever encountered. However, two myosin motor isoforms appear in the top 20 interactome s of IKKa and NEMO.
Of previously described partners, a few fragments of HSP90 beta were found but did not achieve significance over isotype controls, whereas Cdc37 (Chen et al. 2002;Hinz et al. 2007) or ELKS (Ducut Sigala et al. 2004) were never detected. This illustrates that affinity pull down may miss well-known complexes such as the one with HSP90 (Chen et al. 2002;Broemer et al. 2004;Pittet et al. 2005;Hinz et al. 2007; see figs below).

Cellular contents by flow analysis and distributions of IKKa and IKKb by digital microscopy and FRET
As a preliminary, cells were analyzed by flow cytometry to examine the relative intracellular content of IKKa, IKKb, IKKc, Hsp90, and clathrin. The cellular content of IKKa and IKKb correlated poorly with cell size (forward scatter), disparate from the intracellular distribution of clathrin or Hsp90. A systematic multichannel flow cytometric analysis of cells simultaneously stained with two or three fluorophores showed variations of almost 100-fold for IKKa and IKKb in individual cells and less so for clathrin, NEMO or HSP90 (Fig. 1A). Furthermore, correlation of the cellular content of IKKs with each other and the other members of the canonical complex showed that IKKa and IKKb levels were not correlated with one another, but were somewhat better related to CHC or Hsp90 content (Fig. 1B).
Therefore, to better understand how complexes are formed in cells from proteins with varying individual amounts, FRET analyses of intact cells were completed with special attention to detailing the percentage of voxels of FRET for discrete cellular locations. We anticipated that a large percentage of the FRET interactions among proteins would likely dictate that such interactions are important, whereas those interactions which comprise only a minority of the proteins of interest are likely to be of little intrinsic value.
A few studies have spatially located the principal IKK proteins (ab) individually (Birbach et al. 2002;Verma et al. 2004;Ear et al. 2005;Harhaj et al. 2007) but the intracellular locations of the IKK complex, together and with clathrin, have not been studied. Figures 3-8 show deconvolved z-projections of representative single cells. The panels depict combinations of CHC localized with the two IKK proteins, NEMO and Hsp90 as detected by indirect fluorescence with simultaneous, four-channel, three-dimensional imaging (Cy5, Cy3, ALEXA 488-labeled anti-species Ig; nuclear DNA in blue) of intact cells. The corresponding cFRET observed between channel pairs is always shown at right, on a linear colored scale. The spots represent pairs of epitopes clustered within 30 nm and could include from binary up to larger heteromeric complexes. In comparison, the extended CHC is itself about 47.5 nm long (Brodsky et al. 2001;Fotin et al. 2004), well beyond the 5-6 nm range of interdye FRET (Lakowicz 2006). The superimposition of cFRET pairs (red, green), and the dependence of overlap on each cFRET pair (regressions, n = 27), is shown throughout.
The magnitude of cFRET depends on the ratios of donor-acceptor pairs, ensuring a comparable fluorescent intensity at the CCD. By optimizing labeling titers, we can obtain high FRET efficiencies (Table 2) and reproducible results among replicate runs and with different batches of cells. Also, cFRET between labeled antibodies exploits pairs of modern fluorophore dyes that give minimal donor emission at the acceptor's emission wavelength, while retaining sufficient spectral overlap at the acceptor's excitation wavelength. Nonspecific fluorescence is controlled by prescripted fluorophore combinations controls, required for cFRET calculation (Gordon et al. 1998;Berney and Danuser 2003;Lakowicz 2006). This means corrections of 20% or less in the cFRET.
The overlap and dependence of IKKa and b with their chaperone Hsp90 appears as a cFRET+ signal within the nucleus and perinuclear arcs ( Fig. 2; Broemer et al. 2004;Pittet et al. 2005;Qing et al. 2006;Hinz et al. 2007). IKKa fluorescence is more abundant in the nucleus than IKKb. Although Hsp90 is present in the nucleus, there is little IKKb-Hsp90 FRET within the nucleus. Both FRET pairs overlap at the perinuclear ring. A significant fraction of IKKb-Hsp90 cFRET voxels overlap IKKa-IKKb (0.55 AE 0.090) but considerably less IKKa-IKKb appears associated with Hsp90 (0.17 AE 0.03, Table 1). The amount of IKKb-Hsp90 cFRET signal strongly influences the degree of overlap (regression slopes = 0.55) despite the linear fit being poorer in this case (r 2 = 0.66). The slope of the overlap against the IKKa-IKKb signal is weaker (0.26), but the linear fit is very high (r 2 = 0.99). Importantly, the intercepts are negligible for both linear fits.
The cFRET interactions of IKKa with IKKb are visualized at the broken perinuclear arcs (Fig. 3). Unlike Hsp90, CHC is essentially absent from the nucleus, though abundant in the perinuclear/Golgi area (Fig. 3). The cFRET+ interaction of the IKKb-CHC pair appears to be dictated by the absence of CHC in the nucleus; thus, the overlap of IKKa-IKKb with IKKb-CHC bears minimal overlap in the perinuclear and in the cytoplasmic compartments. As with Hsp90, a significant fraction of IKKb-CHC cFRET signal (0.5 AE 0.06) overlaps a small fraction of IKKa-IKKb, (0.26 AE 0.04). The slope of the overlap against the IKKa-IKKb signal is the weakest found (0.06), and the poor fit (r 2 = 0.62) appears to be due to a complex distribution.
In contrast, the CHC-IKKb and IKKb-Hsp90 cFRET signals are found in the perinuclear ring, but with substantial signal of the latter pair in the cytoplasm (Fig. 4). As with previous sets, there is little FRET positivity detected within the nucleus, while the CHC rich peri-Golgi area is not as prominent here. A very significant fraction of the CHC-IKKb signal overlaps IKKb-Hsp90 FRET pixels (0.74 AE 0.07) and the linear dependence is almost a perfect one, the best correlation among the six triads (slope 0.96, r 2 = 0.98). The dependence of the overlap on the amount of CHC-IKKb cFRET signal is also significant (0.41, r 2 = 0.78) but not as high.
The cFRET association of IKKa-CHC and IKKa-Hsp90 is strongest in the perinuclear ring but in different zones, such that overlap is minimal (Fig. 5). There are minimal IKKa-Hsp90 FRET pairs detected within the nucleus. While these features of IKKa FRETs are similar to IKKb, the dependence of the overlap of CHC-IKKa on IKKa-Hsp90 complexes is not as dramatic (slope 0.54, r 2 = 0.97). The significant and negative intercept (À2163, P = 0.0013) suggest that local concentrations of IKKa-Hsp90 appear before FRET pixels overlap with clusters of CHC-IKKa. In addition, the IKKa-CHC FRET pairs are less abundant than NEMO-IKKa FRET+ complexes, which show high efficiencies especially around the nucleus and cytoplasm (Fig. 6). Although both IKKa and NEMO are abundant in the nucleus, cFRET between these is rare in this compartment. Only about 8.1% AE 1.2% of NEMO-IKKa containing complexes overlap with IKKa-CHC voxels. The amount of voxels where both FRET signals overlap depends more on increasing numbers of voxels containing IKKa-CHC cFRET (slope = 0.48) than on the number of voxels containing NEMO-IKKa cFRET (slope = 0.15) plotted from all examined cells.
Finally, the binary NEMO-IKKb pairs show high cFRET signal in the cytoplasm and in particular perinuclear areas (Fig. 7). The NEMO-IKKb pairs overlap IKKb-CHC in similar cellular locales; however, although there is an abundance of NEMO, in the nucleus there is little cFRET with IKKb. The amount of voxels where both cFRET signals overlap depends more on increasing numbers of voxels containing IKKb-CHC cFRET (slope = 0.64) than on the number of voxels containing NEMO-IKKb cFRET (slope = 0.296).
Taken together, Figures 2-7 demonstrate that IKKa is more abundant within the nucleus than IKKb, although the IKKa-IKKb cFRET is clearly detected within nuclear voxels. Although NEMO is also found within the nucleus and cFRETs between NEMO and either IKK are detected elsewhere, there is scant IKKa-NEMO FRET and negligible IKKb-NEMO detected within the nucleus. These data support the hypothesis that NEMO plays different roles in the nucleus than in the cytoplasm Ear et al. 2005;Gloire et al. 2006;Scheidereit 2006). Similarly, while Hsp90 is present within the nucleus, its complexes with IKKa or IKKb are restricted to arcs on the nuclear periphery. CHC is rare in the nucleus of these cells but indirect immunofluorescence with spacer antibodies detects cFRET with IKKs in the cytoplasm and perinuclear areas.
To summarize these data, the FRETn for all pairs examined was calculated and is shown in Table 2. The high values indicate that the ratio of donor and acceptors fluorescence intensity was close to 1 and that the concentration of FRET complex to free dye was satisfactorily high. Furthermore, the fraction of voxels containing two cFRETs and the regression of the overlap against the abundance of each pair has been complied for six pairs examined (Table 3). Unlike the mean fraction, which has high error, the regression uses the broad range of protein distribution in resting cells to define evidence of linear relationships with high accuracy. The greater slope dependence indicates the preferred direction of ternary overlap. These data indicate that neither IKKb-Hsp90 complexes nor IKKb-CHC complexes contribute significantly to the amount of IKKa-IKKb complexes. Instead, IKKb-CHC complexes seem proportionate to the quantity of IKKb-Hsp90 complexes and IKK-CHC complexes may precede IKK-NEMO complexes.

Discussion
Previous data demonstrated that silencing clathrin alters I-kB phosphorylation and NF-kB activation in both unstimulated and stimulated A549 lung epithelial cells and that clathrin must be involved in either the activatory signaling pathway or in housekeeping cycles of the resting state (Escobar et al. 2006). In the current study, quantitative evidence (proteomics and first-order regression of bimolecular complexes against cytometric abundance) showed that clathrin is involved with elements of the IKK signalosome operating in the unstimulated state of a human Type II alveolar cell line. The data from these two reports provide independent support for recent discoveries concerning the multifaceted roles of the dynamic IKK complex. In addition to regulating the NF-kB pathway, the IKK proteins play other roles in immunity, growth, and cancer and although the structural components of the active IKK signalosome (a, b, NEMO) are known, the maturation and posttranslational processing are just being explored (Hacker and Karin 2006;Scheidereit 2006;Hayden and Ghosh 2008). IKKa-IKKb and NEMO are known to be multifunctional proteins in both cell cytoplasm and nucleus (Gloire et al. 2006;Scheidereit 2006;Hinz et al. 2007) and flow cytometric analysis showed variation of almost 100-fold for IKKa and IKKb in individual cells, less so for clathrin, NEMO or HSP90 (Fig. 1A).
Cross-correlating the cellular contents of these five proteins with one another revealed distributions approximating linear relationships between clathrin-NEMO, NEMO-IKKb or IKKa-IKKb and more restricted ranges for the total cellular FRET between IKKb-Hsp90 and IKKb-CHC. Scheidereit (2006) reviewed a large number of proteins that could potentially interact with IKKa-IKKb and NEMO and concluded that the presence of partner proteins probably depended on the cell type studied and on the stimulus used, if any (Scheidereit 2006).
Many of the studies on the IKKs, though, used labeled proteins and none have investigated the human lung epithelium-derived cell types.
In this study, no fragments of clathrin vesicle associated proteins were found with the IKKs. This finding supports previous data, which contended that the involvement of clathrin with IKKa signaling in resting epithelial cells is independent of the endocytic pathways (Kim et al. 2011).
The contaminant repository shows that affinity pull down with agarose-conjugated beads (chiefly in HeLa, Jurkat, T-cells) often finds components of the cytoskeleton, ribosomal proteins, histones, and keratins (Mellacheruvu et al. 2013). Thus, it is surprising that the SAINT analysis finds human ribonucleoprotein C (HNRNPC) in IKKa pull downs with high confidence; this protein could be a likely IKKa partner in these cells, since the isotype IgG pull downs found very few peptide fragments of the protein. Furthermore, other components of this large ribonuclear complex, such as HNRNPK, HNRNH3, HNRNPF, HNRNPU, were found with significance >0.5 (data uploaded to CRAPOME).
The roles of IKKa and NEMO in the nucleus have been scrutinized (Gloire et al. 2006) using cell disruption approaches. Birbach et al. (2002) were the first to image IKKa (and less IKKb) by immunostain within the nucleus, with a larger nuclear population of IKKa in an endothelial cell line than in HELA cells (which required blockade of nucleocytoplasmic transport for detection). Interestingly, they noted that direct YFP, GFP-labeled IKKa constructs translocated a little slower than the native proteins, perhaps because of the presence of the tag increasing their size. In addition, Sil et al. (2004) reported a nuclear localization signal (NLS) in IKKa (but not IKKb) that was essential for epidermal development. The reported data demonstrate abundant presence of nuclear IKKa, similar to pulmonary epithelium. Harhaj et al. found IKKa and NEMO (but not IKKb) in the nucleus at perinuclear "hot spots", especially after HIV Tax expression, in modified Jurkat cells. Such "hot spots" could represent highly concentrated, oligomerized forms of IKK complexes, some of which were shown to overlap the Golgi complex; IKKa was detected in the nuclear fractions of resting human neutrophils (Ear et al. 2005).
The normal cellular distributions of IKK complexes are not defined. Kim inferred that CHC is linked to IKKa but the interactions in the NF-kB interactome are not clear (Bouwmeester et al. 2004). CHC is a large cytoplasmic protein lacking a NLS, thus, it is typically excluded from the nucleus of interphase cells, although it can be perinuclear and even bind mitotic structures (Fotin et al. 2004;Sutherland et al. 2001). Our findings indicate that both IKKa and IKKb form complexes with CHC (Table 3, Figs 2-7). Analysis of the cellular overlap of IKKa-IKKb complexes with either CHC or Hsp90 shows that a very small amount of the IKKa-IKKb colocalize with about half of the complexes made by IKKb with either protein. The number of voxels of IKKa-IKKb FRET complexes overlapping either IKKb-Hsp90 complexes (Fig. 2) or IKKb-CHC complexes (Fig. 3) is a small fraction of the total IKKa-IKKb FRET complexes (17% and 26%, respectively), but it constitutes a large portion of the IKKb-Hsp90 complexes (55%, Fig. 2) or IKKb-CHC complexes (50%, Fig. 3). This suggests that the complexes of IKKb with either Hsp90 or CHC may be precursors to the final IKK complex.
Analysis of the complexes of IKKb with either Hsp90 or CHC (Fig. 4) shows that about 74% of the voxels positive for IKKb-Hsp90 FRET overlap with 41% of voxels containing CHC-IKKb FRET and are restricted to perinuclear arcs. Figure 5 shows that there is only modest overlap (about 39 and 33%) of the individual complexes of IKKa with either Hsp90 or CHC, which are both typically perinuclear. In contrast, very few voxels (about 8%) showing NEMO-IKKa FRET overlap with 50% of IKKa-CHC FRET (Fig. 6) in the perinuclear compartment. Instead (Fig. 7), 22% of NEMO-IKKb FRET voxels are involved in overlap with 41% of IKKb-CHC FRET-positive voxels in perinuclear and cytoplasmic compartments.
Although suggestive, the 99% confidence intervals for the overlapped fractions show a wide variation that indicates that the dependence of overlap on any one FRET pair may be treated better by regression analysis.
We therefore plotted numbers of voxels containing two FRET pairs against numbers of voxels containing each pair and found linear relationships with robust P values.
ABC BC¢A þ BC Either A, B, C may contain other subunits that are not visualized. We can assume that the voxels containing colocalized cFRETs may be related to "ternary" complexes (either preceding or following its dissociation). If the abundance of voxels containing colocalized cFRETs is linearly related to AB or BC complexes, the slope of the dependence suggests the direction of assembly.
In this first-order model, we anticipated that, if the number of voxels containing colocalized cFRETs relate strongly to AB or BC complexes, the slopes of the dependence would indicate the overall direction of ternary assembly (allowing for disassembly rates). This regression exploits the variable amounts of individual FRET pairs in cells. The number of voxels containing colocalized cFRET pairs (denoted by Y) could indicate "ternary" complexes (either preceding or following dissociation). For this study, we focused only on complexes that can be confirmed by IP-MS.
In most cases r 2 values are above 0.76, showing excellent fit. P values for the calculated slopes are <1E-6, except for two cases. Furthermore, the intercepts are mostly insignificant, supporting the simple linear dependence of overlap on FRET pair amounts.
Specifically, throughout the dynamic range, in unstimulated cells (i) the amount of IKKa-IKKb complexes has little effect on the overlap with IKKb-Hsp90 complexes suggesting that the two kinases together tend to not contain Hsp90. (ii) IKKa-IKKb clusters do not overlap with IKKb-CHC. (iii) the same is true for NEMO complexes with either IKK and CHC. (iv) In contrast, the number of complexes made by Hsp90 with either IKK influences overlap with the complexes made by the IKKs with CHC Thus, IKK-CHC complexes appear to lie downstream of the HSP90 complexes and precede the formation of IKK-NEMO complexes, but not as components of the IKK signalosome. The antibodies used in this study did not reveal whether the binary complexes involve CHC monomers, trimers, or even larger lattices. Similarly, we cannot discern the stoichiometry of the IKK components in the complexes.
The reported data demonstrate that labeling cellular proteins with antibodies after fixation offers complementary advantages to the more conventional tagging with GFP or other constructs. First, postfixation FRET in situ avoids problems due to the fractionation of fragile complexes. Moreover, it avoids potential problems associated with protein overexpression that can skew cellular contents, and to steric effects in fluorescent protein constructs (which can almost double the size of the protein) (Barken et al. 2005). Furthermore, indirect immunofluorescence with labeled secondary antibodies (spacers) addresses the challenge presented by the extended 47.5-nm-long CHC molecule.
These data agree with the recent findings of Hinz et al. (2007), Hayden and Ghosh (2008) that IKKs may exist as distinct complexes during maturation and that Hsp90 serves active catalytic IKK complexes at some point of the activation/inactivation cycle and/or stabilizes nascent complexes. The slopes ( the IKKb-HSP90 FRET. By those standards, the dependence on CHC is even less for IKKa-IKKb complex. Yet, we find a very strong dependence on IKKb-Hsp90 complexes that begins to overlap IKKb-CHC conglomerates at perinuclear arcs. Moreover, both IKK-CHC complexes expand into IKK-NEMO-rich areas, suggesting shows IKKc (green), IKKb (red), and nuclear stain in blue. Panel B shows the cFRET signal intensity pseudocolored from green (lowest) to red (highest). NEMO-IKKb cFRET is abundant around the perinuclear border and cytoplasm, but not within the nucleus. The middle left panel (C) shows IKKb (red) and CHC (in Cy5, shown green) while panel D shows the corresponding cFRET signal intensities ranging from green to red. cFRET from IKKb-CHC pairs is also abundant in cytoplasmic and perinuclear areas, but virtually absent within the nucleus. Panel E shows both cFRET signals, from panel A (in green) and panel B (in red). (F) The graph shows that the number of voxels where both cFRET signals overlap depends more on increasing numbers of voxels containing IKKb-CHC cFRET (slope = 0.64), and less so on the number of voxels containing NEMO-IKKb cFRET (slope = 0.296), plotted from all examined cells. these CHC complexes somehow precede ternary IKKa-IKKb-NEMO assembly. The proteomics and quantitative localization methodology outlined in this report could be utilized for studying other ternary interactomes of IKKa at natural levels or after perturbations such as knockdowns or receptor stimulation.