Conformation of CCAAT/enhancer-binding protein alpha dimers varies with intranuclear location in living cells.

The structure of a protein defines its biochemical properties, but the impact of intracellular location and environment on protein structure remains poorly defined. CCAAT/enhancer-binding protein alpha (C/EBPalpha) is a master regulator of transcription and cellular proliferation that concentrates and is kept inactive at transcriptionally quiescent, pericentromeric regions in mouse cell nuclei. C/EBPalpha dimer structure was measured in living cells from the amounts of fluorescence energy transferred between derivatives of the green fluorescent protein attached to different C/EBPalpha domains. Comparing the levels of fluorescence resonance energy transfer at pericentromeric and nonpericentromeric regions of the nucleus indicated that the DNA binding domains of C/EBPalpha dimers were further apart and interacted more poorly at pericentromeric heterochromatin than in the more euchromatic regions of the nucleus. In contrast, the position and interactions of the transcriptional activation domains were similar throughout the nucleus. Phorbol ester treatment caused a shift in the position of the transcriptional activation domain relative to the DNA binding domain. Thus, C/EBPalpha conformation varies with intranuclear location and with cellular environment. These "fluorescence resonance energy transfer nanoscopy" techniques will be broadly applicable for associating conformational and kinetic variations to subcompartment-specific actions of C/EBPalpha or any protein in the dynamic intracellular environment.

The structure of a protein specifies its interactions with itself and with other factors. X-ray crystallography and NMR characterize the structure of concentrated, isolated proteins of static composition and uniform conformation under nonphysiologic conditions. Methods that investigate protein structure in living cells, where conformation may vary with localized protein interactions and with dynamic, subcellular microenvironments, would complement high resolution x-ray crystallographic and NMR structures.
It is becoming increasingly evident that the nucleus is composed of distinct functional subdomains at which transcription regulatory proteins dynamically associate (1)(2)(3)(4)(5). C/EBP␣ 1 is a transcription factor that regulates gene expression and cellular proliferation through distinct mechanisms (6 -9). Within mouse cell nuclei, C/EBP␣ accumulates at one of those subnuclear domains, the pericentromeric heterochromatin (10,11), by binding to ␣-satellite DNA repeats (10) 2 that concentrate around the centromere (12,13). In non-murine cells, the homologous repetitive element and C/EBP␣ are less concentrated. Although most transcription factors do not concentrate at pericentromeric heterochromatin, some factors involved in lymphocyte and erythrocyte development transiently associate with microscopically detectable, mouse pericentromeric heterochromatin at specific differentiation stages (14 -17). This suggests that regulated compartmentalization of transcription factors may be functionally important (2).
The co-concentration of ␣-satellite DNA and C/EBP␣ at microscopically detectable pericentromeric heterochromatin permits the study of pericentromeric targeting of C/EBP␣. Point and deletion mutations of C/EBP␣ that block pericentromeric targeting still block cellular proliferation (18), indicating that pericentromeric localization is not required for the antiproliferative effects of C/EBP␣. In contrast, transcriptional activity is regulated by pericentromeric location. A C/EBP␣ mutant that no longer binds ␣-satellite DNA but that retains normal binding to some promoter binding sites also no longer concentrates at the pericentromeric subdomains of the nucleus. This altered specificity mutation releases C/EBP␣ from sequestration at the transcriptionally quiescent (11) pericentromeric subdomain, which results in a substantial elevation in C/EBP␣ activation of promoter activity. 2 More naturally, in mouse pituitary cell cultures, C/EBP␣ is redistributed to the euchromatin upon expression of the pituitary-specific transcription factor Pit-1 (19,20), whereas a Pit-1 mutation identified in human patients with combined pituitary hormone deficiency (21,22) is defective in the redistribution of C/EBP␣ (19). Thus, the sequestration of C/EBP␣ at transcriptionally inactive heterochromatin is functionally significant and regulated.
The sequestration of C/EBP␣ at pericentromeric heterochromatin may be accompanied by an alteration in C/EBP␣ structure. We investigated whether C/EBP␣ structure was different when localized at pericentromeric heterochromatin and at euchromatin in mouse pituitary progenitor cells. C/EBP␣ is a member of the bZIP family of transcription factors, which are characterized by a conserved, carboxyl-terminal dimerization and DNA binding domain (23)(24)(25). The bZIP domain forms a dimeric ␣-helical coiled-coil that binds DNA (26,27). Transcrip-tion activation (TA) functions are present in more amino-terminal domains of C/EBP␣ (28 -30). However, the relative positions of the bZIP and TA domains in the C/EBP␣ dimer are unknown under any in vitro or in vivo condition.
We combined fluorescence microscopy and fluorescence resonance energy transfer (FRET) techniques to define the relative positions of the TA and bZIP domains in dimers of C/EBP␣ at different subregions of living cell nuclei. The TA and bZIP domains of full-length C/EBP␣ were labeled with the autofluorescent green and blue fluorescent proteins (GFP and BFP) and expressed in living cells. Like endogenous C/EBP␣, fluorescent protein-labeled C/EBP␣ was constitutively nuclear and accumulated at pericentromeric heterochromatin (10,11,31). Since FRET only occurs if BFP and GFP are less than 80 Å apart (32,33) and decreases to the sixth power of the distance separating the fluorophores (34), the amount of energy transferred from BFP to GFP indicated the relative interactions between the fluorophore-tagged TA and bZIP domains at each site within the cell (34 -37).
FRET, normalized for the amounts of BFP and GFP present, was much higher when BFP and GFP were attached to the bZIP domains than to the TA domains. This indicated that the bZIP domains were closer to each other than were the TA domains in the C/EBP␣ dimer or that the interactions between the bZIP domains were kinetically more favorable (36). Bimolecular interaction plots of the amount of FRET measured against the amounts of BFP and GFP-tagged C/EBP␣ present indicated identical kinetics for interactions between the bZIP and between the TA domains. FRET, measured at each of thousands of subregions throughout each nucleus, showed that the bZIP domains were further apart or interacted less well in C/EBP␣ dimers concentrated at the pericentromeric subregions than in dimers at the remaining euchromatic regions of the nucleus. In contrast, the spatial relationship between the TA and bZIP domains was identical at the pericentromeric and noncentromeric locations but was altered by incubating the cells with a phorbol ester. Thus, we provide the first characterization of the conformational state of a gene-and cell cycleregulatory factor at localized positions within living nuclei.

Expression of C/EBP␣ Fusions with GFP and BFP in GHFT1-5 Cells
All C/EBP␣ fusion proteins were expressed under the control of the human cytomegalovirus after transfection by electroporation, as previously described, into pituitary progenitor GHFT1-5 cells (11). The transfected cells were plated on number 1 borosilicate glass coverslips and grown 40 -48 h post-transfection before imaging. Transfected cells grown for 24 h were treated with 10 Ϫ8 M PMA (Sigma) or control Me 2 SO drug vehicle for 1 day prior to imaging or to extract preparation for promoter activation studies (38). The growth hormone promoter and the C/EBP␣-sensitive promoter containing the growth hormone TATA box (Ϫ33 to ϩ8 relative to the transcription start site) and a single growth hormone C/EBP␣ binding site (Ϫ239 to Ϫ219) have been previously characterized (11,38,39).

Immunostaining of C/EBP␣
For the quantification of C/EBP␣ co-localization with Hoechst 33342stained DNA in 3T3-L1 cells, differentiated 3T3-L1 cells, and transfected GHFT1-5 cells, those cells were washed with phosphate-buffered saline (PBS), fixed for 5 min with methanol, treated for 5 min with 0.05% Triton X-100 in PBS, blocked by incubation with 5% horse serum in PBS, and then incubated for 1 h with a 1:500 dilution of the C-18, anti-C/EBP␣ primary antibody (goat) from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA) (sc-9314) in 0.5% horse serum in PBS. Slides were washed three times with PBS and incubated for 1 h with a 1:500 dilution of donkey, anti-goat, rhodamine-conjugated secondary antibody from Santa Cruz Biotechnology (sc-2094). Following washing with PBS, slides were incubated for 5 min with 0.2 g/ml Hoechst 33342 and then washed three times with PBS. For the 3T3-L1 cell images shown in Fig. 1B, the 14AA anti-C/EBP␣ primary antibody (rabbit) from Santa Cruz Biotechnology (sc-61) was used in conjunction with the donkey, anti-rabbit rhodamine-conjugated secondary antibody from Santa Cruz Biotechnology (sc-2095).

Image Collection
All quantified FRET and co-localization data were collected on an Olympus IX-70 using Olympus ϫ40 Plan Apochromat objective (0.95 numerical aperture). The images shown in Fig. 1A were collected using an Olympus ϫ100 Plan Apochromat oil immersion objective (1.40 numerical aperture). Chroma Corp. (Brattleboro, VT) filters and multiband pass dichroic mirror (61000v2bs) were used together with Sutter Instruments (Novato, CA) -10 excitation and emission filter wheels, controlled by Universal Imaging Corp. (Downingtown, PA) Metamorph data acquisition software. The single, immobile, multi-band pass dichroic mirror in combination with mobile excitation and emission filters and the use of chromatically corrected objectives maximized the image registration required for pixel-by-pixel co-localization analysis and FRET nanoscopy. An Opti-Quip (Highland Hills, NY) model 1962 long term stabilizer was used to keep light intensity constant for accurate quantitative data collection.
Antibody-stained slides were imaged by 1) exciting rhodamine with 550 -560-nm light and collecting emissions from 580 -630 nm, 2) exciting GFP with 480 -495 nm light and collecting emissions at 500 -530 nm, and 3) exciting Hoechst 33342 with 365-395-nm light and collecting emissions from 435-465 nm. Controls showed no fluorescence bleed-through of rhodamine, GFP, or Hoechst between these channels. For quantitative co-localization analysis (Figs. 1, C and D), focusing was done using the Hoechst channel so as to blind data collection to the presence and amounts of expressed C/EBP␣. Once data collection parameters were established that ensured no saturation in any pixels, all integration times, camera gain, and pixel binning were kept constant to permit accurate quantitative comparisons of fluorescence amounts between different cell nuclei.
Live cell imaging was used to analyze FRET between BFP and GFP-linked C/EBP␣. Images were always collected in the following order: 1) acceptor (excitation filter 480 -495 nm/emission filter 500 -530 nm, 2) donor (365-395 nm/435-465 nm), 3) FRET (365-395 nm/500 -530 nm). The acceptor and donor channels yielded no bleed-through between BFP and GFP fluorescence (Fig. 2). It is particularly important to collect the FRET channel after the donor channel to eliminate any possibility of false FRET signals arising from decreased donor emissions that would occur if BFP were photobleached during an initial collection in the FRET channel. Images were collected by focusing only for the GFP-labeled C/EBP␣ using the acceptor filter combination. This prevented, until data collection, the excitation of BFP, which is more prone than other GFP derivatives to photobleaching (31). Practically, we had little difficulty with BFP photobleaching as was indicated by the reproducibility of our constants obtained from the control cells expressing only C/EBP␣ fusions with BFP ( Fig. 2).

Image Analysis
Co-localization-All image analysis was done using Metamorph software (Universal Imaging Corp., Downingtown, PA). For the analysis of C/EBP␣ co-localization with Hoechst, a region of no fluorescence adjacent to the cell was used to determine the average background level of fluorescence in each of the rhodamine, GFP, and Hoechst channels. The background amount was then subtracted from each pixel in each channel. Nuclei were identified as regions of contiguous pixels containing higher than background levels of blue fluorescence in the Hoechst channel. The selected region was transferred to the matched images collected in the rhodamine and GFP channels. For co-localization analysis of cells expressing C/EBP␣ fused to either BFP or GFP, nuclei were similarly identified as regions of contiguous pixels containing significantly higher than background levels of GFP fluorescence. Correlation coefficients were calculated by comparing the amounts of fluorescence measured in each matched pixel of two different channels using the Metamorph "correlation plot" application.
FRET-The nuclei of cells expressing BFP-or GFP-tagged C/EBP␣ were identified from background-subtracted BFP or GFP fluorescent images as described above. The amounts of nuclear fluorescence collected from the acceptor (GFP) or donor (BFP) in each channel were expressed as ratios relative to the amount of fluorescence collected in the respective acceptor or donor channels (see Fig. 2). These spectral cross-talk ratios were used to calculate the contributions of BFP and GFP to each channel in cells co-expressing BFP-and GFP-labeled C/EBP␣. Briefly, from the amount of acceptor fluorescence in each co-expressing cell was subtracted the minimal fluorescence contamina-tion of C/EBP␣-BFP to the acceptor channel (0.0013ϫ donor fluorescence amount). The minimal contribution of C/EBP␣-GFP to the donor channel (0.0040 times the BFP-corrected acceptor fluorescence) was similarly subtracted. Although these contaminations were negligible in the current experiments, they can be significant with other fluorophore or filter combinations 3 and must be accounted for. For calculating FRET and FRET efficiency from the nucleus, the fluorescence contribution of the acceptor C/EBP␣-GFP then was subtracted from the backgroundsubtracted FRET channel (0.0882 times the BFP-corrected acceptor fluorescence). The amount of remaining fluorescence in the FRET channel then was divided by the amount of remaining fluorescence in the donor channel. If the resulting ratio is the same as the ratio obtained from cells containing only donor (0.4719), then there was no FRET. If higher than 0.4719, energy was transferred from the donor to the acceptor.
It is critical that the acceptor cross-talk into the FRET channel is accounted for. We have noticed, to date, a number of attempts by other laboratories to calculate FRET without taking into account the substantial contributions of the acceptor itself into the FRET channel. Simply measuring the ratio of the amount of fluorescence in the FRET and donor channels with only donor excitation incorporates acceptor bleedthrough into the FRET channel. If not accounted for, this bleed-through increases the FRET/donor ratio and results in a false conclusion of interaction. Since the ratios measured are physical parameters of the fluorophores, correctly calculated FRET measurements are highly consistent between separate experiments, provided that all parameters affecting the relative ratios of fluorescence quantification in the donor, acceptor, and FRET channels are kept constant. This includes using the same 1) objective lens, 2) dichroic mirror, 3) excitation/emission filters, 4) camera, and 5) relative integration times for the different channels.
All bleed-through-corrected measurements were downloaded into Microsoft Excel files and were expressed as a mean Ϯ S.D. from multiple nuclei collected from two or more experiments. Statistically significant differences were determined using t tests. Slopes and y intercepts were calculated using Excel. 95% confidence intervals in the slopes were calculated using GraphPad (San Diego, CA) Prism software, using only data from the linear range (acceptor/donor amounts Ͻ5) and setting the line to have the correct y intercept of 0.47. Data collected over a wider acceptor/donor range was fit into a single order interaction plot using GraphPad Prism.
FRET Nanoscopy-The amount of FRET at each pixel was determined by applying our calculations directly on each image using Metamorph software. The backgrounds were determined as above and then subtracted from each pixel within each image using Metamorph arithmetic functions. Also subtracted from each pixel were the bleed-through contributions of BFP and GFP from matched pixels to the appropriate green, blue, and FRET channels. The bleed-through-corrected images contain the fluorescence values for BFP, GFP, and FRET, from which were calculated the FRET/donor and acceptor/donor ratios at each pixel in the image. Since C/EBP␣ should assume the same relative intranuclear distribution regardless of whether it is linked to BFP-or GFP, the high correlation coefficients observed for BFP-and GFP-linked C/EBP␣ expressed in the same cell confirmed the precise image registration required for this analysis.
The formula y ϭ mx ϩ b was used to calculate the extent of FRET (m) at each pixel in every precisely registered image: y is the FRET/donor ratio at a single pixel, x is the acceptor/donor ratio at the same pixel, and b (the y intercept) is the FRET/donor ratio, where acceptor/donor ϭ 0 (i.e. the constant determined from the donor alone control). The above FRET nanoscopy analyses were done in a semiautomated fashion by linking the calculations into a Metamorph journal.

Selection of Pericentromeric Regions
In the FRET analysis of cells co-expressing BFP and GFP-labeled C/EBP␣, there was no counterstaining with the blue fluorescent Hoechst 33342, normally used to identify pericentromeric regions. Because of the high correlation of C/EBP␣ location with Hoechst 33342 staining ( Fig. 1), pericentromeric regions within the nuclei were identified as pixels containing more than 1.2 times the average amount of background-subtracted GFP-tagged C/EBP␣ fluorescence. Visually and quantitatively, this corresponded well to pericentromeric regions identified by staining with the blue fluorescent Hoechst 33342 (40). FRET and the extent of FRET measurements were compared at and away from the marked pericentromeric regions using the co-localization application of Metamorph. The proportions of pixels showing specific extents of FRET were calculated using Metamorph by measuring the number of pixels of progressively higher slope using sequential threshold values.

Estimates of Interfluorophore Distances Calculated from Relative FRET Slopes
FRET efficiency varies with distance according to the relationship first described by Förster (34) where r is the distance between two fluorophores and r o is the distance between two fluorophores at which energy transfer is 50% efficient. Other factors, including the interaction kinetics and the rotational orientation of the fluorophore, contribute to FRET efficiency and are included in more expansive equations (34,36). The relative slopes from the linear portions of the FRET/donor versus acceptor/donor graphs are a close surrogate measurement of the relative FRET efficiencies (37). Determination of the actual FRET efficiency would depend upon equipment calibrations that equate fluorescence amounts with the amounts of each molecule. However, we can compare the relative FRET efficiencies for two different interactions as the ratio of the slopes E 1 and E 2 , where r 2 and r 1 are the BFP to GFP distances under the two conditions being examined; r o for FRET from BFP to GFP is 41.4 Å (33). By assuming hypothetical values for r 2 anywhere between 40 and 100 Å, we calculated the value of r 1 . We then determined the differences between r 2 and r 1 distances over a range of values for r 2 . This indicated the range in average distances separating the fluorophores used to label the proteins under the two different experimental conditions.

C/EBP␣ Localizes to Pericentromeric
Heterochromatin-C/EBP␣, fused at either its amino (GFP-C/EBP␣) or carboxyl (C/EBP␣-GFP) terminus with GFP, was transiently expressed in mouse pituitary progenitor GHFT1-5 cells. GHFT1-5 progenitor cells contain no endogenous C/EBP␣ (11,38) and have been used extensively in our laboratories to study the effects of C/EBP␣ expression on pituitary gene transcription and cellular proliferation (11,18,38,40,41). Western blotting showed the expressed fusion proteins to be of the appropriate molecular weight (11,18,40) and were expressed, on average, at a level comparable with that of the pituitary-specific transcription factor Pit-1 (19), which is present at low levels within GHFT1-5 cells (11,38,42). The expression level of the GFP-C/EBP␣ and C/EBP␣-GFP proteins also was similar to that of endogenous C/EBP␣ in other cell types (discussed below).
Counterstaining with an antibody against the molecular chaperone Hsp70 demonstrated that GHFT1-5 cells expressing the C/EBP␣ fusions with GFP and GHFT1-5 cells not expressing C/EBP␣ both had low amounts and identical intranuclear distributions of Hsp70 (data not shown). This demonstrated that cells expressing C/EBP␣ remained healthy and did not recognize the accumulation of ectopically expressed C/EBP␣ in pericentromeric heterochromatin as a folding defect to be corrected by co-concentrating Hsp70 at those sites. Together with our data demonstrating that C/EBP␣ sequestration at pericentromeric heterochromatin is functionally significant and regulated (18 -20), 3 this strongly argued that the observed subnuclear distribution of C/EBP␣ was not a consequence of protein aggregation and precipitation into inclusion bodies, as is sometimes observed for other cellular factors (43)(44)(45).
C/EBP␣ point mutants that were expressed to the same level as C/EBP␣ failed to concentrate at the pericentromeric heterochromatin. 2 This further suggested that C/EBP␣ localization to pericentromeric heterochromatin cells was not an artifact of C/EBP␣ overexpression in GHFT1-5 cells. Indeed, endogenous C/EBP␣ expressed in mouse 3T3-L1 cells upon chemical induction of adipocyte differentiation (46,47) and detected by anti-C/EBP␣ antibody staining also targeted to pericentromeric heterochromatin ( Fig. 1B) (10). Two different anti-C/EBP␣ antibodies (see "Experimental Procedures"), counterstained with rhodamine-linked secondary antibodies, showed the same co-localization of red (rhodamine fluorescence) and blue (Hoechst 33342 fluorescence) in many induced 3T3-L1 cells. No C/EBP␣ was detected in uninduced 3T3-L1 cells (Fig. 1B). Therefore, C/EBP␣ expressed in GHFT1-5 cells takes up the same intranuclear distribution of endogenous C/EBP␣ in another murine cell type.
The similar pericentromeric targeting of endogenous 3T3-L1 cell C/EBP␣ and C/EBP␣-GFP or GFP-C/EBP␣ ectopically expressed in GHFT1-5 cells also was confirmed by parallel staining with the same anti-C/EBP␣ antibody and the subsequent imaging of differentiated 3T3-L1 cells and transfected GHFT1-5 cells under identical collection parameters. For these comparisons, co-localization was quantified by plotting the amount of background-subtracted red fluorescence (C/EBP␣, C/EBP␣-GFP, or GFP-C/EBP␣) in each of the thousands of pixels of each image against the amount of blue fluorescence in the corresponding pixel of the matched Hoechst 33342 image (Fig. 1C). Correlation coefficients were calculated that described the degree by which C/EBP␣ and Hoechst fluorescence at each pixel varied from a perfect correlation of 1.00. Overall, the correlation coefficients averaged 0.78 Ϯ 0.14 for endoge-nous C/EBP␣ expressed in differentiated 3T3-L1 cells (calculated from 205 separate nuclei), 0.69 Ϯ 0.16 for GFP-C/EBP␣ expressed in GHFT1-5 cells (n ϭ 94 nuclei), and 0.70 Ϯ 0.13 for C/EBP␣-GFP expressed in GHFT1-5 cells (n ϭ 89 nuclei). The slightly lower correlations for C/EBP␣ expressed in GHFT1-5 cells probably are due to the low levels of Pit-1 in these cells; we have observed Pit-1 to interact directly with (20) and relocate (19) C/EBP␣ from the heterochromatic to the euchromatic regions of the cell nucleus. By comparison, the correlation coefficients measured for red C/EBP␣ fluorescence and green GFP-C/EBP␣ or C/EBP␣-GFP fluorescence, which should correlate well, since the red and green signals both originate from C/EBP␣, were 0.85 Ϯ 0.07 or 0.88 Ϯ 0.06, respectively.
Plotting the C/EBP␣ to Hoechst 33342 correlation coefficients calculated for each nucleus against the average rhodamine (C/EBP␣) fluorescence intensity of the same nucleus (Fig.  1D) demonstrated that C/EBP␣ co-localization with Hoechst 33342-stained DNA did not vary with the C/EBP␣ expression level. Low level rhodamine fluorescence from GHFT1-5 cells sham-transfected with the empty expression vector or from undifferentiated 3T3-L1 cells did not correlate with Hoechst 33342 fluorescence (Fig. 1D). Similarly, rhodamine fluorescence from equivalently expressed, mutant C/EBP␣ proteins that failed to target to pericentromeric heterochromatin showed correlation coefficients with Hoechst 33342 fluorescence of zero. 2 Finally, C/EBP␣ was determined to co-localize strongly with Hoechst 33342-stained DNA within the nuclei of most induced 3T3-L1 cells (94.1% of the nuclei that express C/EBP␣ have a correlation coefficient of Ͼ0.5) and of GHFT1-5 cells expressing GFP-C/EBP␣ (88.3%) or C/EBP␣-GFP (89.9%). This analysis also demonstrated that the levels of ectopically expressed GFP-C/EBP␣ and C/EBP␣-GFP studied in our experiments were variable but were globally similar to that of endogenous C/EBP␣ in 3T3-L1 cells. Therefore, C/EBP␣ concentrated at pericentromeric heterochromatin when expressed to physiologic levels in GHFT1-5 cells.
C/EBP␣ Dimerization in Living Cells-C/EBP␣ is believed to act primarily as a dimer (26,27), although the extent of dimer- ization in the physiologic environment is unknown. To measure dimerization in living cells, we fused the cDNA for BFP to the carboxyl terminus of the cDNA for C/EBP␣ and quantified if any fluorescence energy was transferred from the donor (C/EBP␣-BFP) to the acceptor (C/EBP␣-GFP) when co-expressed (31,(35)(36)(37). C/EBP␣-GFP (11,40) and C/EBP␣-BFP 3 activated a C/EBP␣-responsive promoter when expressed in GHFT1-5 cells.
Images were captured in three different fluorescence channels from each GHFT1-5 cell expressing C/EBP␣-GFP or C/EBP␣-BFP alone or in combination ( Fig. 2A): 1) the GFPspecific "acceptor" channel (excitation with light of 480 -495 nm; emission collected from 500 -530 nm), 2) the BFP-specific "donor" channel (excitation 365-395; emission 435-465), and 3) the "FRET" channel (excitation 365-395; emission 500 -530). The fluorescence contribution of C/EBP␣-GFP to the donor channel was quantified from cells expressing only C/EBP␣-GFP as a statistically insignificant 0.0040 Ϯ 0.0145 of the background-subtracted fluorescence detected in the acceptor channel (n ϭ 97 cells). Similarly, the contribution of C/EBP␣-BFP to the acceptor channel was 0.0013 Ϯ 0.0058 that detected in the donor channel (n ϭ 108 cells). To the FRET channel, C/EBP␣-GFP consistently contributed 0.0882 Ϯ 0.0151 the amount of fluorescence measured in the acceptor channel, whereas C/EBP␣-BFP contributed 0.4719 Ϯ 0.0151 the amount of fluorescence measured in the donor channel. These "spectral cross-talk" ratios are constants that reflect the spectral properties of BFP and GFP and the physical properties of the detection equipment. As such, the ratios are the same in cells expressing low and high amounts of C/EBP␣-GFP or C/EBP␣-BFP (Fig. 2B) and do not vary from experiment to experiment when using the same detection equipment.
The spectral cross-talk constants were used to determine whether there was energy transfer in cells co-expressing C/EBP␣-GFP and C/EBP␣-BFP (34,36,37). Background-subtracted donor, acceptor, and FRET signals were quantified from multiple cells co-expressing the labeled proteins. The contribution of acceptor to the FRET channel was calculated using the cross-talk ratio and subtracted from the signal in the FRET channel (see "Experimental Procedures"). The remaining fluorescence in the FRET channel contained the contribution of the donor C/EBP␣-BFP (0.4719 of the amount of corrected blue fluorescence) plus any sensitized emissions resulting from the transfer of energy from BFP to GFP. If there were no energy transfer between C/EBP␣-BFP and C/EBP␣-GFP, the FRET/donor ratios would remain at the 0.4719 Ϯ 0.0151 value determined for C/EBP␣-BFP alone. If there were energy transfer in cells co-expressing C/EBP␣-BFP and C/EBP␣-GFP, the amount of fluorescence in the FRET channel would increase, and the amount of C/EBP␣-BFP fluorescence in the blue channel would decrease, so the FRET/donor ratio would increase.
Relative Positions of bZIP and Transcription Activation Domains in C/EBP␣ Dimers-Expression vectors also were constructed in which GFP and BFP were fused to the TA domains at the amino terminus of C/EBP␣ (GFP-C/EBP␣ and BFP-C/EBP␣). The four possible pairwise combinations of C/EBP␣ tagged with BFP or GFP at its bZIP or TA domains (Fig. 3, A-D) were co-expressed in GHFT1-5 cells. All four combinations showed FRET/donor ratios greater than 0.47 (Table I), confirming an in vivo interaction between C/EBP␣ proteins. There was no indication of FRET between any combination of C/EBP␣-BFP or BFP-C/EBP␣ and the GFP-p300 or GFP-CBP (Table I). Thus, specific interactions were observed between the carboxyl terminal bZIP domains, between the amino-terminal TA domains, and between the bZIP and TA domains in pituitary progenitor GHFT1-5 cells. Although the interactions observed may occur in dimers or in higher order multimers, the term "dimer" is used hereafter for simplicity and because kinetic considerations, discussed below, indicated a bimolecular interaction.
If a transfected cell expresses more acceptor (GFP-linked C/EBP␣) than donor (BFP-linked C/EBP␣), a greater proportion of BFP-linked C/EBP␣ will dimerize with GFP-linked C/EBP␣ than with another BFP-linked C/EBP␣. Thus, FRET increases with increasing amounts of acceptor relative to donor (35,37). This was observed when the FRET/donor ratio was plotted against the acceptor/donor ratio from multiple cells for each combination of GFP-linked C/EBP␣ with BFP-linked

FIG. 2. FRET between BFP and GFP attached to the carboxylterminal bZIP domain in full-length C/EBP␣ and expressed in GHFT1-5 pituitary progenitor cells.
A, representative images of nuclei in which C/EBP␣-GFP and C/EBP␣-BFP were expressed alone or were co-expressed. The amounts of nuclear fluorescence were quantified in the green, blue, or FRET channels (shown) and then corrected (see "Experimental Procedures") for background fluorescence and acceptor, donor, and FRET from acceptor fluorescence using bleedthrough ratios determined from the control cells (see numbers in images). Energy transfer resulted in an increased FRET emission at the expense of donor emission (FRET/donor) in the co-expressing cells relative to the cells containing donor alone (see Table I). Each channel was presented with identical fluorescence scaling to facilitate the visual comparison of intensity. B, graphing the bleed-through ratios against the average intensity fluorescence in each cell showed that the ratios were consistently measured over a wide range of fluorescence amounts. Only cells containing fluorescence amounts within this region of accurate measurement were processed for FRET determination. C/EBP␣ (Fig. 4, A-D). These curves flatten toward a plateau provided enough acceptor is present to saturate interactions with donor (Fig. 5A).
The slopes of the FRET/donor versus acceptor/donor graphs, measured within the predominantly linear range for each combination of GFP and BFP-linked C/EBP␣ (Table I), reflect the relative "extent of FRET" for each combination at equivalent acceptor/donor ratios. The extent of FRET observed for C/EBP␣ tagged with BFP and GFP at their bZIP domains was significantly higher than the extents of FRET measured for interactions between the TA domains (p ϭ 1 ϫ 10 Ϫ20 ) or between the TA and bZIP domains (p ϭ 5 ϫ 10 Ϫ25 and p ϭ 5 ϫ 10 Ϫ34 ). This indicated either that the bZIP domains were closer together than were the TA domains in the C/EBP␣ dimer or that the bZIP interactions were kinetically more stable (higher on rate and/or lower off rate).
Identical Kinetics of Interaction between the bZIP and Transcription Activation Domains-To distinguish the contributions of fluorophore distance from interaction kinetics to the different levels of FRET, we examined data collected over a wide range of acceptor/donor ratios for the bZIP to bZIP and TA to TA interactions (Fig. 5A). The ability to collect data over a large range of acceptor/donor ratios is limited by the necessity of maintaining FRET data collection parameters the same for all cells (see "Experimental Procedures"). However, from the data collected, the curves clearly followed (r 2 ϭ 0.94 and 0.75, respectively) first order interaction kinetics. This single-order interaction kinetics indicated a simple bimolecular interaction between C/EBP␣ monomers or between two interacting units.
Extrapolation of the first-order curves showed that FRET between the bZIP domains saturated at a much higher FRET/ donor ratio (7.6 Ϯ 0.5) than did FRET between the TA domains (2.5 Ϯ 0.4). This showed that there was more FRET between the bZIP domains than between the TA domains when kinetic differences were minimized at saturation. In contrast, the curves reached saturation at the same rate: the acceptor/donor ratios at half the maximal FRET/donor levels were not statistically different (42 Ϯ 4 and 43 Ϯ 11). These similar k d values demonstrated that the interactions between the bZIP domains and between the TA domains were kinetically identical. Identical k d values are expected if the interaction between the fluorophores is governed only by the dimerization of C/EBP␣ itself. Thus, the different extents of FRET arose primarily from nonkinetic considerations, which would include differences in the distances separating the fluorophores and/or differences in the rotational orientation of fluorophore dipoles that do not radially emit energy (34,36).
Rotational Constraints within the C/EBP␣ Dimer-The extent of FRET between the bZIP and TA domains varied significantly (p ϭ 0.0005) if the locations of the donor BFP and acceptor GFP at the TA and bZIP domains were reversed (slopes of 0.059 -0.077 and 0.097-0.110; Table I). Since the average distance between the domains and the kinetics of their interactions would not be changed by swapping the fluorophores, the extent of FRET for the two TA and bZIP combinations should, at first glance, be the same. However, energy transfer from many fluorophores is not radial, so the extent of energy transfer also depends upon the orientation of nonradial fluorophores (34,36). Symmetry in the TA to bZIP and bZIP to TA FRET would be observed only if both the donor and acceptor fluorophores were rotating freely in space. The asymmetry in the extent of FRET detected for bZIP to TA domain interactions indicated that one or both of the bZIP or TA domains were somewhat constrained in the C/EBP␣ dimer in living cells. Thus, the different level of FRET observed between the bZIP . The relationship of FRET/donor against acceptor/donor was quasilinear at lower acceptor/ donor levels. The slope of the graphs represented the extent of FRET at normalized acceptor/donor levels for each of the four combinations of bZIP-and TA-tagged C/EBP␣. There was no FRET between BFPtagged C/EBP␣ and either CBP or p300 tagged with GFP (ϫ in the graphs). 95% confidence intervals in the slopes of these graphs, calculated for acceptor/donor levels Ͻ5 and the y intercept forced through 0.47, are presented in Table I.  and between the TA domains potentially includes some contribution from rotational constraints on the fluorophores as well as from different domain distances.
Local Variations in C/EBP␣ Dimer Conformation within the Nucleus-The relative positions and rotations of the bZIP and TA domains, determined using the total fluorescence from each nucleus, represent the average conformation of C/EBP␣ in the nucleus (Fig. 5B). However, the type of C/EBP␣ dimers formed and the ability to from dimers may vary with subnuclear location. To analyze C/EBP␣ dimerization at different sites within the nucleus, the extent of FRET was calculated at each pixel from the background-subtracted and bleed-through-corrected acceptor, donor, and FRET images (usually 5,000 -10,000 pixels/nucleus for ϫ40 magnification) (Fig. 6, A-D). This required that pixels measured in the separate acceptor, donor, and FRET channels be perfectly matched.
BFP-and GFP-linked C/EBP␣ should concentrate at the same locations in each nucleus. There was a linear correlation in the amounts of background-and cross-talk-corrected fluorescence emitted from BFP-and GFP-linked C/EBP␣ in each of thousands of pixels from each nucleus co-expressing both fusion proteins (correlation coefficients of 0.86 Ϯ 0.07, n ϭ 374 cells). This correlation demonstrated that the amount of BFP and GFP fluorescence in each channel emitted by a small number of C/EBP␣ molecules (roughly estimated to average 5-10 molecules/pixel) was the same, and accurately measured, in matched pixels.
With accurate image registration, we were able to determine local variations in C/EBP␣ conformation by calculating (see "Experimental Procedures") the extent of FRET at each pixel within an image (Fig. 6, A-D, FRET Extent). The proportion of pixels containing specific extents of FRET was determined in each cell nucleus and then averaged from multiple nuclei (Fig.  7A). For each of the four different interactions mapped, the extent of FRET distributed around a peak that was very similar to the average extent of FRET measured from all the pixels in the cell (Fig. 7B). This suggested that the conformation of C/EBP␣ within the nucleus was variable around a single pre-ferred conformation. Note the wider distribution in the extent of FRET for interactions between the bZIP domains, which suggests a broader variation in this interaction at different locations in the cell (see below).
Unique Conformations of C/EBP␣ Dimers at Pericentromeric Regions-The pixel-by-pixel analysis identified a distribution in C/EBP␣ conformation but did not associate those variations  Fig. 6) were averaged from multiple cells co-expressing full-length C/EBP␣ fusions of BFP and GFP attached to different domains in C/EBP␣ (see Fig. 3, A-D). B, extent of FRET averaged from all of the pixels of the same nuclei. The average extent of FRET was similar to the extent of FRET with the highest proportion of pixels, indicating a relatively normal variation in FRET efficiency for each of the four aspects of C/EBP␣ dimer conformation investigated. with any subnuclear structures. We therefore compared C/EBP␣ conformation at and away from pericentromeric regions by measuring the extent of FRET at and away from pericentromeric regions for each combination of BFP-and GFPtagged C/EBP␣. Since Hoechst 33342-stained pericentromeric regions corresponded to the regions of concentrated C/EBP␣ (Fig. 1), pericentromeric heterochromatin was marked in cells co-expressing BFP and GFP-tagged C/EBP␣ as pixels containing more than 1.2 times the average fluorescence intensity of GFP-linked C/EBP␣ within each nucleus (40). On average, those marked pixels had fluorescence intensities 1.74 Ϯ 0.28 times more than in the remaining pixels (n ϭ 374 cells).
Comparing the extent of FRET at, and away from, the pericentromeric regions (Table II) showed that the conformations of C/EBP␣ were different in these subnuclear domains. The extent of FRET between the bZIP domains at pericentromeric regions was a statistically significant (p ϭ 1 ϫ 10 Ϫ23 ) 0.8954 times that measured away from pericentromeric regions. This showed that the bZIP domains of C/EBP␣ dimers were in closer contact away from the pericentromeric chromatin. Comparison of k d and the amount of FRET at acceptor/donor saturation calculated at, and away from, pericentromeric chromatin indicated that the bZIP regions both interacted less well and were further apart at the pericentromeric chromatin.
The different pericentromeric conformation of the bZIP domain was confirmed by statistically decreased extents of FRET (p ϭ 0.001) between the bZIP and TA domains at the pericentromeric regions (Table II). In contrast, the extent of FRET between the fluorophore-tagged TA domains was not statistically different at and away from the pericentromeric regions. The extents of FRET between the bZIP and TA domains remained asymmetric at both the pericentromeric and remaining regions of the nucleus, indicating that the torsional constraints on the dimer were present at both locations. The simplest interpretation is that the bZIP domains were further apart and interacting less well at pericentromeric regions and that the TA domains have similar orientations and torsional constraints at and away from pericentromeric regions (Fig. 5B).
Incubation with Phorbol Ester Alters C/EBP␣ Dimer Conformation-Physiological environment also may affect C/EBP␣ dimer conformation. Previously, we found that incubation of pituitary progenitor GHFT1-5 cells with PMA and forskolin enhanced C/EBP␣ activation of the full-length rat growth hormone promoter (38). We first determined that a growth hormone promoter deleted of all sequences except those surrounding the C/EBP␣ binding site (Ϫ239 to Ϫ209) and the TATA box (Ϫ33 to ϩ8) responded to PMA, and not to forskolin, in a C/EBP␣-dependent fashion (data not shown). This suggested that C/EBP␣ or co-regulatory factors that cooperate with C/EBP␣ were a direct target of phorbol ester activation.
We investigated the effects of PMA induction on C/EBP␣ conformation by comparing the extent of FRET for all combinations of GFP-and BFP-linked C/EBP␣ in sham-treated cells and in cells treated with 10 Ϫ8 M PMA (Figs. 8, A-D). Pixel-bypixel analysis showed the interactions between the bZIP do-mains to be identical in sham-and PMA-treated cells (Fig. 8A), even when compared for the different interactions at and away from the pericentromeric regions (not shown). This suggested that the overall contact and distances between the bZIP domains were similar regardless of PMA incubation. In contrast, PMA incubation resulted in a statistically significant decrease in the extent of FRET between the BFP-tagged TA and GFPtagged bZIP domains (Fig. 8C, p ϭ 0.03) and between the BFP-tagged TA and GFP-tagged TA domains (Fig. 8D, p ϭ 0.04). For both TA to bZIP and TA to TA FRET, the decreased extent of FRET was particularly prominent at the pericentromeric regions (p ϭ 0.01 for both) but less consistent away from the pericentromeric region (p ϭ 0.06 for both). This suggested that the PMA-induced alteration in conformation was more uniform for C/EBP␣ dimers at the pericentromeric regions than for dimers away from the pericentromeric regions. The asymmetric effect of PMA incubation on the extent of FRET from TA to bZIP (Fig. 8C), but not from bZIP to TA (Fig. 8B) or from bZIP to bZIP, indicated that the change in conformation induced by PMA included a torsional rotation of the TA domains relative to each other and to the rest of the dimer (Fig. 5B).

C/EBP␣ Forms Dimers in Living
Cells-FRET between spectral derivatives of the green fluorescent protein fused to C/EBP␣ defined that C/EBP␣ is a dimer in living cells (Fig. 2). For BFP and GFP, FRET is 50% efficient when the donor and acceptor are 41.4 Å apart (33) and falls very rapidly, to the sixth power, as the fluorophores are separated (33,34,36). Thus, energy transfer demonstrated that the fluorophores, and therefore C/EBP␣, were well within 80 Å of each other within the cell. These interactions were very specific. No FRET was observed between C/EBP␣ and the co-activators CBP and p300 (Table I) although GFP-tagged CBP and p300 co-localized with C/EBP␣ when expressed in GHFT1-5 cells (11). Both CBP and p300 enhanced transcriptional activation by C/EBP␣ (11,48) but have never been observed to directly interact with C/EBP␣ (11,48). However, C/EBP␣ expression does cause CBP (11) and p300 3 to redistribute to the intranuclear location of C/EBP␣, suggesting that C/EBP␣ forms some sort of complex with CBP and p300 within the cell.
C/EBP␣ Dimers Form throughout the Nucleus of Living Cells-We combined the angstrom level resolution of FRET with the resolution of light microscopy (250 nm for green light) to map C/EBP␣ dimer structure throughout the cell (Figs. 4 -8). By changing the positions of the fluorophore tags within C/EBP␣, this "FRET nanoscopy" technique was used to define, in living cells, the interactions of the bZIP and TA domains in C/EBP␣ dimers at localized intranuclear domains and under different cellular conditions. FRET nanoscopy demonstrated that, in virtually all regions of the nucleus, C/EBP␣ was positioned sufficiently close to another C/EBP␣ molecule to allow energy to transfer from the attached BFP to GFP. Thus, intermolecular C/EBP␣ interactions were spread throughout the cell nucleus and were not excluded from any subnuclear structure.  8. Incubation with the phorbol ester PMA alters the rotational position of the TA domain relative to the rest of the C/EBP␣ dimer. PMA incubation had no effect on the extent of FRET measured for BFP and GFP fused to the bZIP end (A) or BFP fused to the bZIP end and GFP fused to the TA end (B). In contrast, PMA incubation decreased significantly the extent of FRET between GFP fused to the bZIP end and BFP fused to the TA end (C) and between BFP and GFP fused to the TA ends (D). The asymmetric effect of PMA on TA to bZIP (C) and TA to TA (D) FRET indicated that the effect of PMA incubation on C/EBP␣ dimer conformation included an alteration in the torsional constraints of the donor fluorophore around the TA domain.
other factors that form complexes with C/EBP␣ (11), by regulating the C/EBP␣-induced alteration in histone acetylation at pericentromeric regions (40) or by regulating dimerization of C/EBP␣ with other C/EBP family members (46,51). PMA incubation also may indirectly change the environment of the pericentromeric region to affect C/EBP␣ dimer conformation. Regardless of the underlying basis, the FRET measurements show that C/EBP␣ dimer conformation was altered when the intracellular environment was changed.
FRET Nanoscopy Complements Structural Analyses-FRET nanoscopy will be very useful for determining the structural parameters of many interacting molecules at specific intracellular locations. The structural details of C/EBP␣, to date, were limited to the assumption that the last 80 amino acids (the bZIP domain) of the 358-amino acid-long C/EBP␣ were similar to coiled-coil structures identified for other bZIP domains (26,27). This predicted that the carboxyl termini of C/EBP␣ should be in very close proximity. Indeed, our FRET measurements demonstrated that the carboxyl termini were considerably closer to each other in living cells than were the other domain interactions that were measured.
The structures measured by FRET also may be affected by the requirements of packing of GFP into C/EBP␣. Given the close proximities of the carboxyl terminal ␣-helices in the predicted C/EBP␣ dimers, it was somewhat surprising that the C/EBP␣-GFP fusions remained transcriptionally active (11,40) and competent to block cellular proliferation in GHFT1-5 cells (18). The fusion proteins containing GFP fused to the aminoterminal TA domain of C/EBP␣ also remained competent to block cellular proliferation (18) but were defective in transcriptional activation (40). Packing constraints imposed by the large GFP fluorophore at the amino terminus even may have contributed to the rotational constraint necessary to detect the PMA-induced change in TA domain conformation at the pericentromeric regions. Thus, the large size of the GFP fluorophores may have some unexpected stearic advantages as well as limitations. The development of smaller fluorophore tags (52, 53) may reduce, but not eliminate, the stearic consequences of fluorophore tags.
FRET nanoscopy is a powerful tool with which to investigate the structural parameters of interacting molecules at localized sites within living cells. We have already applied FRET nanoscopy to investigate interactions between other molecules. 3 In some instances, FRET nanoscopy has permitted us to observe interactions that are limited to a small percentage of sites within the cell. Certainly, techniques such as two-hybrid interactions also may be employed to investigate such interactions, but the extent of FRET uniquely measures the degree to which seemingly similar interactions are different (37). FRET nanoscopy also correlates those interaction nuances with spatial, and potentially temporal, considerations in living cells. We envisage that FRET nanoscopy will become a central technique with which biochemical interactions and structural parameters may be measured directly in the physiologic environment with unprecedented accuracy and detail.