All EGF(ErbB) receptors have preformed homo- and heterodimeric structures in living cells.

The epidermal growth factor receptor (EGFR) family of receptor tyrosine kinases, also known as ErbB or HER, plays crucial roles in the development of multicellular organisms. Mutations and over-expression of the ErbB receptors have been implicated in a variety of human cancers. It is widely thought that the ErbB receptors are located in the plasma membrane, and that ligand binding to the monomeric form of the receptors induces its dimeric form for activation. However, it still remains controversial whether prior to ligand binding the receptors exist as monomers or dimers on the cell surface. Using bimolecular fluorescence complementation (BiFC) assays in the present study, we demonstrate that in the absence of bound ligand, all the ErbB family members have preformed, yet inactive, homo- and heterodimers on the cell surface, except for ErbB3 homodimers and heterodimers with cleavable ErbB4, which exist primarily in the nucleus. BiFC assays of the dimerization have also suggested that the ligand-independent dimerization of the ErbB receptors occurs in the endoplasmic reticulum (ER) before newly synthesized receptor molecules reach the cell surface. Based on BiFC and mammalian two-hybrid assays, it is apparent that the intracellular domains of the receptors are responsible for the spontaneous dimer formation. These provide new insights into an understanding of transmembrane signal transduction mediated by the ErbB family members, and are relevant to the development of anti-cancer drugs.


Introduction
Receptor tyrosine kinases (RTKs) mediate a variety of cellular responses in normal biological processes and in pathological states (Hunter, 2000;Schlessinger, 2000). The epidermal growth factor receptor family belongs to subclass I RTKs and consists of four members; EGFR (also known as ErbB1 or HER1), ErbB2 (Neu or HER2), ErbB3 (HER3) and ErbB4 (HER4) (Yarden and Sliwkowski, 2001;Olayioye et al., 2000). The ErbB receptor contains an extracellular ligand-binding domain, a single-pass transmembrane α-helix and an intracellular domain that encodes a tyrosine kinase followed by a regulatory region (Ullrich et al., 1984;Warren and Landgraf, 2006). The four ErbB family members recognize at least 11 different but structurally related growth factors including epidermal growth factor (EGF), transforming growth factor α (TGF-α) and neuregulins (NRGs) (Citri and Yarden, 2006;Linggi and Carpenter, 2006a). ErbB2 does not have a known ligand but functions as a co-receptor for the other three members. ErbB3 has an inactive kinase, and can be activated through heterodimerization by the other members of the family for signaling . ErbB4 consists of two pairs of naturally occurring isoforms differing in their juxtamembrane domain (JMa or JMb) and C termini (cyt1 or cyt2). An isoform pair is characterized by alternative splicing of exons located in the extracellular juxtamembrane region, conferring the JMa isoforms with susceptibility to proteolytic cleavage (Junttila et al., 2000;Gambarotta et al., 2004).
It is widely thought that ligand binding provokes homo-and/or heterodimerization of the ErbB receptors (Schlessinger, 2002;Burgess et al., 2003;Bublil and Yarden, 2007), resulting in the phosphorylation of tyrosine residues in the C-terminal tail, which serve as docking sites for signaling molecules that contain SH2 (Src homology region 2) and PTB (phosphotyrosine-binding) domains Pawson, 2004). The receptor activation initiates diverse arrays of downstream signaling pathways including Ras (MAP) kinase, phospholipase Cγ (PLCγ), signal transducer and activation of transcription (STATs), and phosphatidylinositol-3 kinase (PI-3K) (Schlessinger, 2000;Yarden and Sliwkowski, 2001). Furthermore, ligand binding also induces receptor clustering and endocytosis in clathrin-coated pits on the cell surface, followed by either recycling back to the cell surface or moving through a series of endosomes to lysosome for degradation (Sorkin and von Zastrow, 2002;Miaczynska et al., 2004). After endocytosis, nuclear translocation of the receptors may also occur and may regulate gene expression (Lin et al., 2001;Wang et al., 2004;Linggi and Carpenter, 2006a).
Significant insights into the molecular mechanisms underlying ligand-linduced dimerization and activation of the receptor family have been provided by recent analyses of the crystal structures of the extracellular domains of EGFR (Garrett et al., 2002;Ogiso et al., 2002;Ferguson et al., 2003), ErbB2 , ErbB3 (Cho and Leahy, 2002) and ErbB4 (Bouyain et al., 2005), and the kinase domain of EGFR (Stamos et al., 2002;Zhang et al., 2006). The unliganded EGFR, ErbB3 and ErbB4 extracellular domains adopt a 'tethered' intramolecular conformation, in which the 'dimerization arm' of the domain II (also known as CR1) is buried by interacting with the domain IV (also known as CR2) within the same molecule. By contrast, the extracellular domains of ligandbound EGFR, ErbB3, ErbB4 and unliganded ErbB2 exhibit a

Summary
The epidermal growth factor receptor (EGFR) family of receptor tyrosine kinases, also known as ErbB or HER, plays crucial roles in the development of multicellular organisms. Mutations and over-expression of the ErbB receptors have been implicated in a variety of human cancers. It is widely thought that the ErbB receptors are located in the plasma membrane, and that ligand binding to the monomeric form of the receptors induces its dimeric form for activation. However, it still remains controversial whether prior to ligand binding the receptors exist as monomers or dimers on the cell surface. Using bimolecular fluorescence complementation (BiFC) assays in the present study, we demonstrate that in the absence of bound ligand, all the ErbB family members have preformed, yet inactive, homoand heterodimers on the cell surface, except for ErbB3 homodimers and heterodimers with cleavable ErbB4, which exist primarily in the nucleus. BiFC assays of the dimerization have also suggested that the ligand-independent dimerization of the ErbB receptors occurs in the endoplasmic reticulum (ER) before newly synthesized receptor molecules reach the cell surface. Based on BiFC and mammalian two-hybrid assays, it is apparent that the intracellular domains of the receptors are responsible for the spontaneous dimer formation. These provide new insights into an understanding of transmembrane signal transduction mediated by the ErbB family members, and are relevant to the development of anti-cancer drugs.
dimerization-competent conformation, in which the dimerization arm is exposed for the extracellular domain contacts between two monomers together with an adjacent loop Dawson et al., 2005). By crystal structural analysis of the isolated kinase domain, two distinct, symmetric and asymmetric, dimeric structures have been determined (Stamos et al., 2002;Zhang et al., 2006). The symmetric back-to-back structure of the kinase domains, which is stabilized by the carboxyl-terminal tails of the receptor molecule, may correspond to the receptor's inactive form since the substrates may not be accessible to the catalytic sites for autophosphorylation. Furthermore, based on the asymmetric structure together with site-directed mutagenesis of the contact residues in the full-length receptor, it has been proposed that the active kinase domains of the ErbB receptors form the asymmetric dimer in which the C-lobe of one kinase domain is juxtaposed to the N-lobe of the other. In this proposed mechanism, the C-lobe of one ErbB kinase domain allosterically activates the other kinase domain by binding to the latter's N-lobe and repositioning its activation loop such that catalysis is facilitated (Zhang et al., 2006;Hubbard and Miller, 2007).
However, it has been reported by several groups that EGFR exists as a preformed dimer on the cell-surface (Moriki et al., 2001;Yu et al., 2002;Martin-Fernandez et al., 2002;Clayton et al., 2005). Furthermore, it has recently been shown by fluorescence crosscorrelation spectroscopy (FCCS) and Föster resonance energy transfer (FRET) that EGFR and ErbB2 expressed on the cell surface exist as preformed homodimers and heterodimers at physiological expression levels (Liu et al., 2007). In the present study using BiFC, we further demonstrate that all the four ErbB receptors have preformed, yet inactive, homo-and/or heterodimeric structures on the cell surface in the absence of bound ligand, whereas preformed ErbB3 homodimers and heterodimers with cleavable ErbB4 are mainly localized in the nuclei. Consistent with the preformed dimeric structure, BiFC analysis has also revealed that the receptor dimerization occurs in the endoplasmic reticulum (ER), where ligand may not be available, before newly synthesized receptor molecules reach the cell surface.

ErbB family members have preformed homodimeric structures in living cells
We first analyzed subcellular expression patterns of the receptors fused with a fluorescent protein (FP) at the receptor's C terminus by transfecting the constructs into Chinese hamster ovary (CHO) cells, in order to compare the patterns with those of BiFC constructs of the receptors shown below. CHO cells have the advantage of no endogenous expression of the EGFR family members except for a low background expression of ErbB2 (Tzhar et al., 1996;Zurita et al., 2004) (supplementary material Fig. S1). When plasmid constructs encoding FP-fused receptors were transfected into CHO cells, EGFR-CFP, ErbB2-YFP and ErbB4-JMa-GFP were expressed on the cell surface as expected. By contrast, ErbB3-CFP was primarily expressed in the nucleus although a minor fraction of the fusion receptor was also observed in the cytoplasm (Fig. 1A). Furthermore, co-expression of FP-fused receptors in a pair-wise manner in CHO cells revealed that all the receptors except for coexpressed ErbB3-CFP and ErbB4-JMa-YFP were produced on the cell surface ( Fig. 1B; supplementary material Fig. S2). The majority of co-expressed ErbB3-CFP and ErbB4-JMa-YFP were found in the nucleus. When ErbB3-CFP was co-expressed with EGFR-YFP or ErbB2-YFP, most of the ErbB3-CFP molecules were localized on the cell surface, unlike singly expressed ErbB3-CFP. These results suggested that ErbB3 interacts with EGFR, ErbB2 and ErbB4 in the absence of ligand by preforming heterodimers, as shown below by BiFC analysis.
For BiFC analysis of receptor dimerization in the absence of bound ligand, we fused the C terminus of the receptor molecule to the N-terminal or C-terminal fragment of the Venus FP, a derivative of YFP (Nagai et al., 2002). BiFC is based on the formation of a fluorescent complex from two separate non-fluorescent fragments, which are brought together by the association of two interacting partner proteins fused to the fragments ( Fig. 2A) (Ghosh et al., 2000;Hu et al., 2002;Kerppola, 2006), and is applicable for direct visualization of protein-protein interactions and their subcellular localization in many types of living cells and organisms under physiological conditions (Deppmann et al., 2003;Zhang et al., 2004;Fig. 1. Subcellular localization of the ErbB family members. (A) CHO cells expressing EGFR-CFP, ErbB2-YFP, ErbB3-CFP or ErbB4-JMa-GFP were serum-starved for 24 hours at 37°C, and then observed by confocal microscopy. Note that the FP-fused EGFR, ErbB2 and ErbB4-JMa were localized in the plasma membrane, whereas ErbB3-CFP was mainly in the nucleus. (B) CHO cells co-expressing EGFR-YFP and ErbB3-CFP, ErbB2-YFP and ErbB3-CFP, or ErbB4-JMa-YFP and ErbB3-CFP were serumstarved, and then observed by confocal microscopy. Note that a large fraction of ErbB3-CFP was localized in the plasma membrane when co-expressed with EGFR-YFP or ErbB2-YFP. By contrast, the majority of the ErbB4-JMa-YFP protein was localized in the nucleus when co-expressed with ErbB3-CFP. Bars, 10 μm. Hynes et al., 2004;Blondel et al., 2005;Giese et al., 2005;Shyu et al., 2006). Two complementary N-terminal (VN) and C-terminal (VC) fragments of the Venus FP were separately fused to the C-terminal end of EGFR. When these constructs were separately expressed in CHO cells, or EGFR-VN was co-expressed with EGFR-VC, the fusion constructs behaved like unfused wild-type CHO cells expressing EGFR-VN, EGFR-VC, or both of the constructs were serum-starved and then treated with (+) or without (-) 100 ng/ml EGF for 30 minutes at 4°C. Whole cell lysates were subjected to immunofluorescent staining with anti-phospho-EGFR (p-EGFR) or anti-phospho-Akt (p-Akt) antibody, or with anti-EGFR (EGFR) or anti-Akt (Akt) antibody. (C) Specific homodimerization of EGFR detected by BiFC. CHO cells co-expressing the BiFC constructs indicated were serum-starved, and then observed by confocal microscopy. Cells co-expressing EGFR-VN and -VC were also treated with 100 ng/ml EGF for 30 minutes at 37°C (+ EGF). Bars, 10 μm. (D) Examination of expression levels of BiFC constructs in the positive and negative controls by immunofluorescent staining. CHO cells co-expressing the BiFC constructs indicated were fixed with methanol-acetone, and immunostained with murine anti-FLAG primary antibody and goat anti-HA primary antibody, followed by incubation with Cy3-conjugated donkey anti-mouse secondary antibody and Alexa Fluor 633-conjugated donkey anti-goat secondary antibody. Fluorescence signals of FLAG-tagged proteins (cyan), HA-tagged proteins (red) and Venus (green) were observed by confocal microscopy. Bars, 10 μm. (E) EGF binding does not affect fluorescence intensity of cells co-expressing EGFR BiFC constructs. CHO cells co-expressing EGFR-VN and EGFR-VC were serum-starved, and then treated with ATP synthesis inhibitors for 1.0 hour as described in Materials and Methods. Fluorescence intensities of ROI (region of interest) on the cell membrane were observed by confocal microscopy. 100 ng/ml EGF was added to the cell culture at the time indicated by arrow. The relative intensities were normalized by the average intensity at 3 minutes, and the data points are the means ± s.d. (n=32). (F) Subcellular localization of BiFC construct homodimers of the ErbB receptors in the presence of bound ligand. CHO cells expressing the BiFC constructs indicated were serum-starved, and then observed by confocal microscopy. Cells expressing ErbB3-VN and -VC, or ErbB4-JMa-VN and -VC were also stained with Hoechst 33342 to visualize nuclei. Cells co-expressing ErbB3-VN and -VC, or ErbB4-VN and -VC were treated without (-NRG) or with (+NRG) 1.0 nM NRG for 30 minutes at 37°C. Bars, 10 μm. receptors as revealed by western blotting. Although the fusion receptors expressed in the cell and the downstream kinase Akt were not phosphorylated before ligand binding, upon EGF addition, the fusion receptors and the downstream kinase Akt were phosphorylated (Fig. 2B). These results indicated that the BiFC fusion receptor constructs had intact activities in terms of ligandinduced phosphorylation of the receptor and Akt.
When EGFR-VN was co-expressed with EGFR-VC in CHO cells, Venus fluorescence was detected on the cell surface in the absence of added ligand. Upon EGF binding the fluorescent EGFR dimers were internalized (Fig. 2C). No fluorescence was observed when EGFR-VN was co-expressed with EGFR(ΔICD)-VC in which the intracellular domain of EGFR was deleted, or with EpoR-VC, a BiFC fusion construct of erythropoietin receptor, in CHO cells (Fig. 2C). These results indicated that EGFR can specifically form inactive homodimers on the cell surface in the absence of bound ligand. However, the non-detectable Venus signals in the negative controls might result from no, low or separate expression of transfected plasmids in different cells. To rule out these possibilities, we performed immunostaining to examine the expression levels of the fusion proteins in the positive and negative controls (Fig. 2D). The expression levels of EGFR-VN and EGFR(ΔICD)-VC or EpoR-VC in the negative controls were comparable to and not significantly different from that of EGFR-VN and EGFR-VC in the positive control. We detected the EGFR-VN and EGFR-VC signals as well as Venus fluorescence on the cell-surface in the positive control. By contrast, Venus fluorescence was not observed in the negative controls, although we detected the EGFR-VN and EGFR(ΔICD)-VC, or EGFR-VN and EpoR-VC signals on the same cell surface. The expression levels of the BiFC fusion proteins in the positive and negative controls were also confirmed by western blotting (supplementary material Fig. S3). Furthermore, when the ligand EGF was added to the culture medium of CHO cells co-expressing EGFR-VN and -VC, relative fluorescence intensity was not increased by ligand binding under the conditions in which endocytosis was blocked by inhibiting ATP synthesis (Fig. 2E). By contrast, upon ligand binding, EGFR homodimers were internalized in the absence of the inhibitors (Fig. 2C,E). Taken together, these results demonstrate that without bound ligand, most, if not all, of the EGFR molecules on the cell surface have an inactive homodimeric structure.
Prior to ligand binding, we also detected ErbB2 and ErbB4 homodimers on the cell surface (Fig. 2F). Both isoforms of ErbB4 homodimers, cleavable ErbB4-JMa and non-cleavable ErbB4-JMb, were also observed on the cell surface in the absence of the ligand NRG. Upon ligand binding, the cleavable isoform ErbB4-JMa was internalized and localized in the cytoplasm and nucleus, whereas the non-cleavable isoform ErbB4-JMb was internalized and localized only in the cytoplasm. These results suggest that a fraction of the cleaved cytoplasmic domain of ErbB4-JMa can be relocated to the nucleus. Like ErbB3-CFP (Fig. 1A), the majority of ErbB3 homodimers were located in the nucleus, although minor fractions of the homodimers were also observed in the cytoplasm.

Fig. 3. Preformed heterodimers of the ErbB receptors in the absence of bound ligand. (A) BiFC analysis of the ErbB receptor heterodimerization in living cells. CHO cells expressing the BiFC constructs indicated
were serum-starved, stained with or without Hoechst 33342, and then observed by confocal microscopy. Bars, 10 μm. (B) CHO cells co-expressing the BiFC constructs indicated were serum-starved, treated with 100 ng/ml EGF (+EGF) or 1.0 nM NRG (+NRG) for 30 minutes at 37°C, and then observed by confocal microscopy. Bars, 10 μm. (C) Ligand-induced phosphorylation of preformed ErbB heterodimers. CHO cells co-expressing ErbB2-VN and EGFR-VC, or ErbB3-VN and ErbB2-VC were serum-starved, and then treated with (+) or without (-) 100 ng/ml EGF or 1.0 nM NRG for 30 minutes at 4°C. Whole cell lysates were subjected to immunoblotting using specific antibodies for phosphorylated forms (p-EGFR, p-ErbB2, p-ErbB3 and p-Akt) or for unphosphorylated forms of EGFR, ErbB2, ErbB3 and Akt. Preformed ErbB receptor dimers The subcellular localization of ErbB3 homodimers was not affected by the addition of the ligand NRG (Fig. 2F).

ErbB receptors can have preformed heterodimeric structures in living cells
Using BiFC, we also investigated whether the ErbB family members could form heterodimers on the cell surface in the absence of bound ligand. When ErbB2-VN was co-expressed with EGFR-VC, Venus fluorescence was detected on the cell surface ( Fig. 3A). By contrast, Venus fluorescence was not detected when ErbB2-VN was co-expressed with EpoR-VC as a negative control (Fig. 3A). These indicate that ErbB2 can specifically form heterodimers with EGFR on the plasma membrane in the absence of bound ligand. Similarly, EGFR-ErbB3, EGFR-ErbB4-JMa, ErbB2-ErbB3, ErbB2-ErbB4-JMa and ErbB3-ErbB4-JMb heterodimers were also detected on the cell surface in the absence of bound ligand (Fig. 3A,B). Like the ErbB3 homodimers, however, the majority of ErbB3-ErbB4-JMa heterodimers were observed in the nucleus although a minor fraction of the heterodimers was also found in the cytoplasm (Fig. 3A). Upon addition of EGF to the culture medium, the ErbB2-VN-EGFR-VC heterodimers were observed to be internalized (Fig. 3B). Upon NRG binding, ErbB3-VN-ErbB2-VC and ErbB3-VN-non-cleavable ErbB4-JMb herodimers were also internalized and relocated to the cytoplasm. By contrast, the majority of ErbB3-VN-cleavable ErbB4-JMa heterodimers were located in the nucleus, and did not respond to the addition of NRG (Fig. 3B). The ligand binding also induced autophosphorylation of the heterodimers as well as the phosphorylation of the downstream kinase Akt as shown by western blotting (Fig. 3C). These results indicate that the BiFC heterodimers on the cell surface are inactive forms, which can be activated and internalized by ligand binding. As shown above in the case of EGFR homodimers, the relative fluorescence intensities of cells co-expressing ErbB2-VN and EGFR-VC, or ErbB3-VN and ErbB2-VC were not significantly different before and after the addition of ligand under the conditions in which endocytosis was blocked by inhibiting ATP synthesis (supplementary material Fig. S4). These indicate again that in the absence of ligand, most, if not all, of the receptor molecules on the cell surface have inactive dimeric structures.

Intracellular domains of ErbB receptors are required for ligandindependent homo-and heterodimerization
The BiFC assay suggested that the intracellular domains of the ErbB receptors are required for their dimerization, as shown above in Fig. 2C. To test this possibility, we performed a mammalian twohybrid assay (Dang et al., 1991). When pair-wise interactions between intracellular domains of all the ErbB receptors were examined, significantly stronger interactions were observed for all the combinations between the intracellular domains of the receptors than those of the negative controls (Fig. 4). These results demonstrate that the intracellular domains of the ErbB receptors are responsible for ligand-independent homo-and heterodimer formation, although other parts of the receptor molecules may also be involved.

Dimerization of ErbB receptors occurs in ER
All the homo-and heterodimers of the receptors, except for the ErbB3 homodimer and ErbB3-ErbB4-JMa heterodimer, were found in the plasma membrane in the BiFC assay. To find where the dimerization initially occurs, we treated NIH3T3 cells expressing the BiFC homo-or heterodimers with brefeldin A (BFA), a lactone antibiotic that disassembles the Golgi apparatus and blocks anterograde transport of membrane proteins from ER to Golgi (Tamura et al., 1968;Fujiwara et al., 1988). Fluorescence due to homo-and heterodimerization of all the pair-wise combinations of the receptor-BiFC fusion constructs examined was found in the ER, where calnexin was stained as an ER marker with specific antibody ( These results suggest that the ErbB receptor dimers are initially formed in the ER. However, it cannot be ruled out that the receptors form homodimers and heterodimers at later stages of the transport, because Golgi disassembly caused by the BFA treatment may accumulate the BiFC constructs in the ER.

Subcellular localization of endogenous ErbB receptors in human carcinoma cells
It is generally considered that all the members of the ErbB family are localized to and function in the plasma membrane. However, as shown above using BiFC, the majority of ErbB3 homodimers and ErbB3-ErbB4-JMa heterodimers were found to be located in the nucleus (Fig. 1A,B; Fig. 2F; Fig. 3A,B). To rule out the possibility that the nuclear localization of ErbB3 homodimer and ErbB3-ErbB4-JMa heterodimer was due to the BiFC fusion proteins, subcellular localization of the four ErbB members were examined by immunostaining of human carcinoma cells using specific antibodies (Fig. 6). In contrast to the plasma membrane localization of EGFR in A431 cells, or ErbB2 and ErbB4 in MDA-MB-453 cells, ErbB3 was mainly localized in the nucleus of MDA-MB-468 cells although a minor fraction was also observed in the cytoplasm. These data demonstrate that ErbB3 homodimers are primarily localized in the nucleus, unlike the other three family members. Indeed, there are reports in the literature of the nuclear localization of ErbB3, although whether it is monomeric or dimeric is not known (Offterdinger et al., 2002;Koumakpayi et al., 2006).

Nuclear ErbB3 is a full-length molecule
It is known that upon ligand binding, ErbB4-JMa in the plasma membrane is cleaved at its transmembrane region and translocated into the nucleus (Ni et al., 2001;Williams et al., 2004;Linggi and Carpenter, 2006b;Sundvall et al., 2007). To investigate whether the nuclear ErbB3 was intact or cleaved, MDA-MB-468 cells were separated into nuclear and cytoplasm-membrane fractions, followed by immunoblotting for ErbB3 with specific antibody (Fig. 7A). The molecular mass of the nuclear ErbB3 was indistinguishable from that of the cell membrane-cytoplasm fractions, indicating that the nuclear ErbB3 is a full-length molecule. Intensity analysis of the fluorescence signals from CHO cells expressing ErbB3-CFP showed that about 93% of ErbB3 was located in the nucleus and approximately 7% of ErbB3 existed in the cytoplasm-membrane fractions (Fig. 7B). Similar results were also obtained from immunostaining of MDA-MB-468 cells (data not shown). The nuclear localization of FP-fused ErbB3 was not dependent on the expression levels (supplementary material Fig. S6A). By western blotting using anti-GFP antibody, it was also shown that the nuclear fluorescence observed in cells expressing FP-fused ErbB3 was not due to cleaved FP (supplementary material Fig. S6B). These data indicate that ErbB3 homodimers found in the nucleus are full-length molecules, not cleaved fragments.

Discussion
We have shown by BiFC that all the ErbB receptor family members form inactive homo-and heterodimers in the absence of bound ligand, consistent with the previous observations that in the absence of ligand, EGFR has a preformed dimeric structure when analyzed by chemical cross-linking, sucrose density-gradient centrifugation and cysteine disulfide bridge formation (Moriki et al., 2001), and that the majority of EGFR and ErbB2 have homo-and heterodimeric structures at physiological expression levels when analyzed by FCCS and FRET (Liu et al., 2007). The preformed, yet inactive, homo-and heterodimeric structures of the receptors are also consistent with homodimeric structures of the EGFR intracellular kinase (Stamos et al., 2002;Landau et al., 2004), transmembrane (Mendrola et al., 2002) and unactivated extracellular domains . Indeed, an intracellular segment of ErbB2 required for ligand-independent dimerization has been identified (Penuel et al., 2002). Furthermore, ligand-independent association of EGFR and ErbB2 has also previously been demonstrated (Kumagai et al., 2003). Therefore, it appears that all the ErbB receptors have inactive homo-and heterodimeric structures on the cell surface prior to ligand binding. This is further supported by the current finding that the homo-and heterodimers can be formed in ER, where ligands for the receptors may not be available ( Fig.  5; supplementary material Fig. S5A). Ligand-independent growth Journal of Cell Science 121 (19) Fig. 5. The ErbB receptor homo-and heterodimers formed in ER. NIH3T3 cells co-expressing the BiFC constructs indicated were serum-starved, and then treated with (+) or without (-) 10 μg/ml BFA for 8 hours at 37°C. The cells were fixed with methanol-acetone and immunostained with anti-calnexin antibody, followed by incubation with Cy3-conjugated secondary antibody. Nuclei were also stained with Hoechst 33342. The cells were observed by confocal microscopy for calnexin (red), Venus (green) or Hoechst 33342 (cyan) fluorescence. Bars, 10 μm. hormone receptor dimerization also seems to occur in the ER (Gent et al., 2002).
As illustrated in Fig. 8, a molecular mechanism underlying ligandinduced activation of the preformed dimers can be explained by the 'rotation/twist' model proposed previously (Moriki et al., 2001), based on the crystal structures of the extracellular ligand-binding domains and intracellular kinase domains of the receptor family. Because of the flexibility of the extracellular domains of the receptor dimers (Moriki et al., 2001;Teramura et al., 2006), all the preformed ErbB homo-and heterodimers, except for ErbB2, may take two major structures, with either low or high affinity for ligand, on the cell surface. The 'tethered/closed' and 'untethered/ open/ extended' structures of the extracellular domains (Cho and Leahy, 2002;Ferguson et al., 2003;Cho et al., 2003;Garrett et al., 2003) may be responsible for the low and high affinity structures of the receptors, respectively. Based on the crystal structure, the orphan receptor ErbB2 has been proposed to take only the untethered structure Burgess et al., 2003). Ligand-induced intermolecular interaction of the two 'dimerization arms' in the dimeric structure (Garrett et al., 2002;Ogiso et al., 2002;Ferguson et al., 2003;Dawson et al., 2005) may also induce the rotation/twist of the transmembrane domains of the receptor dimer (Moriki et al., 2001;Smith et al., 2002), and may result in the dissociation of the back-to-back symmetric structure of the intracellular domains (Stamos et al., 2002). The dissociation of the intracellular domains in its dimeric structure is likely to rearrange the dissociated kinase domains to the head-to-tail asymmetric structure for activation (Zhang et al., 2006). This asymmetric structure may not be very stable in order to activate the other kinase in its dimeric structure through rearrangement of the kinase domains for the phosphorylation of tyrosines on the other subunit and other substrates.
A similar static twist model has previously been proposed as a molecular mechanism underlying the activation of EGFR (Gadella and Jovin, 1995). Based on FRET analysis of fluorescein (donor)and Rhodamine (acceptor)-labeled EGF molecules bound to EGFR on the cell surface of A431 carcinoma cells, which overexpress the receptor at unusually high levels (>2ϫ10 6 molecules per cell), it has been found that in the absence of bound ligand, EGFR has homodimeric structures, which are temperature dependent: 13% dimer at 4°C, 36% at 20°C and 69% at 37°C. From the three-dimensional structures of the extracellular domain dimers of EGFR bound with its ligand (Garrett et al., 2002;Ogiso et al., 2002), however, the distance between the two ligands is greater than 10 nm, and longer than the detectable distance range of FRET, as pointed out by Whitson et al. (Whitson et al., 2004) and Clayton et al. (Clayton et al., 2005). It is therefore probable that the FRET detected between the two ligands was that between the two dimers because of the over-expression of the receptor cells were serum-starved, fixed with methanol-acetone and immunostained with anti-EGFR, anti-ErbB2, anti-ErbB3 or anti-ErbB4 primary antibody, respectively, followed by incubation with Alexa Fluor 635-or Cy3-conjugated secondary antibody. Nuclei were also stained with Hoechst 33342. Stained cells were observed by confocal microscopy. Note that ErbB3 fluorescence intensity was highest in the nuclei. Bars, 10 μm.

Fig. 7. Nuclear ErbB3 is a full-length molecule. (A) MDA-MB-468 cells were
serum-starved, and then nuclear and cytoplasmic-membrane fractions were prepared as described in Materials and Methods. The fractions were subsequently analyzed by western blotting using anti-ErbB3 antibody. The efficiency of cell fractionation was confirmed by examination of a cytoplasmic protein, α-tubulin and a nuclear protein, histone H3, using anti-α-tubulin and anti-histone H3 antibodies, respectively. (B) CHO cells expressing ErbB3-CFP were serum-starved, and fluorescence intensities in the nucleus and cytoplasm were measured by confocal microscopy. Relative nuclear and cytoplasmic fluorescence intensities are expressed as the mean ± s.d. (n=24). molecules in A431 cells. Since the FRET efficiencies were highly dependent on temperature, higher temperatures might enhance rapid lateral mobility of the cell-surface receptor dimers for efficient FRET.
It has also been proposed that ligand-induced tetramers may be active forms of the ErbB receptors (Huang et al., 1998;Clayton et al., 2005), although there is evidence indicating that tetrameric forms of the ErbB receptors are inactive (Furuuchi et al., 2007). As described above, in the current study we failed to detect ligandinduced tetramers or higher-ordered oligomers by the BiFC assay under the conditions in which the receptor phosphorylation was inhibited ( Fig. 2E; supplementary material Fig. S4). Furthermore, we did not observe intensity peaks larger than dimers in the FCCS analysis of homo-and heterodimers of EGFR and ErbB2 expressed on the cell surface (Liu et al., 2007). These results suggest that ligands may not be able to induce tetrameric forms of the ErbB receptors unless the receptors are phosphorylated. Considering that activated ErbB receptors are trans-autophosphorylated and form clusters in clathrin-coated pits for endocytosis, it is possible that the tetrameric or higher-ordered oligomers of the receptors may be formed after the activation of the receptors.
Interestingly, unlike all the other ErbB receptor dimers, ErbB3 homodimers and ErbB3-ErbB4-JMa heterodimers were primarily localized in the nucleus. These dimers are formed in the absence of ligand as shown in the present study (Fig. 1A,B; Fig. 2F; Fig.  3A,B; Fig. 6). The nuclear localization of ErbB3 has previously been observed in mammary epithelial cells, and is mediated by the nuclear localization signal at the receptor's C terminus (Offterdinger et al., 2002). The ligand-independent nuclear localization of unphosphorylated ErbB3 has also been reported previously (Koumakpayi et al., 2006). The nuclear localization of the ErbB3 homodimer may explain the previous result in which NRG did not induce its homodimerization (Berger et al., 2004). Although ErbB3 is primarily localized in the nucleus as described above, it has often been detected on the plasma membrane in cells that co-express other ErbB members (Offterdinger et al., 2002;Koumakpayi et al., 2006). As shown above using BiFC, indeed, most of EGFR-ErbB3, ErbB2-ErbB3 and ErbB4-JMb-ErbB3 heterodimers were located in the plasma membrane (Fig. 3A,B). These results suggest that ErbB3 tends to form heterodimers with EGFR, ErbB2 or ErbB4-JMb rather than to form homodimers. As described above, upon activation ErbB4-JMa is cleaved by a metalloprotease and then by γ-secretase, and the ErbB4-JMa intracellular domain is also translocated to the nucleus (Ni et al., 2001). This translocation is also mediated by the nuclear localization signal of the intracellular juxtamembrane region (Williams et al., 2004). As shown in Fig.  1B and Fig. 3A,B, the majority of ErbB3-cleavable ErbB4-JMa, but not ErbB3-non-cleavable ErbB4-JMb, heterodimers were localized in the nucleus prior to ligand binding. This suggests that cleavage of ErbB4 may be necessary for their relocation to the nucleus. The nuclear localization of the ErbB3-cleavable ErbB4-JMa prior to ligand binding may occur through spontaneous cleavage of the ErbB4 molecules since the cleavage of ErbB4 without bound ligand has previously been observed (Williams et al., 2004;Sundvall et al., 2007). In the nucleus, the intracellular fragment of ErbB4 enhances STAT5-mediated transcriptional activation on the β-casein promoter (Williams et al., 2004), or abrogates ETO2-dependent transcriptional repression (Linggi and Carpenter, 2006b). However, the function of ErbB3 homodimers and ErbB3-ErbB4-JMa heterodimers in the nucleus remains to be explored.
The results described in the present study provide new insights into an understanding of transmembrane signal transduction mediated by the ErbB receptors, and are relevant to the development of anti-cancer drugs. For example, inhibitors of the ErbB kinases have been developed as anti-cancer drugs, but as previously demonstrated (Moriki et al., 2001), some drugs may activate the kinases under low concentrations of the drugs by inducing the dissociation of the inactive kinase dimers. However, stabilization of the inactive kinase dimers may be an alternative strategy for the development of novel anti-cancer drugs.
The construction of pBiFC-ErbB3-VN was as follows: total RNA from MDA-MB-453 cells was isolated using an RNeasy Mini kit (QIAGEN KK, Tokyo, Japan) according to the manufacturer's instructions. The first strand cDNA was synthesized using SuperScript III (Invitrogen KK, Tokyo, Japan) according to the manufacturer's instructions. A fragment encoding the full-length ErbB3, 1342 amino acids long, with an EcoRV site at the N terminus and a SalI site at the C terminus was amplified from the above cDNA by PCR using oligonucleotides, 5Ј-TGATATCGATGAGGGCG -AAC GACGCTCTG-3Ј with an EcoRV site (ErbB3-VN-N) and 5Ј-TGTCG -ACCGTTCTCTGGGCATTAGCCTTG-3Ј with a SalI site (ErbB3-VN-C), as primers. The resulting PCR fragment was subsequently digested with EcoRV and SalI, and was inserted between EcoRV and SalI sites of pBiFC-VN173. Similarly, for the construction of pBiFC-ErbB3-VC, a cDNA fragment encoding the full-length ErbB3 was amplified from the above cDNA by PCR using oligonucleotides, 5Ј-TGG -TACCATGAGGGCGAACGACGCTCTG-3Ј (ErbB3-VC-N) and 5Ј-TGGTACCC -GT TCTCTGGGCATTAGCCTTG-3Ј (ErbB3-VC-C), as primers, was digested with KpnI, and was then cloned into the KpnI site of the vector pBiFC-VC155. To construct pErbB3-CFP, a cDNA fragment encoding the full-length ErbB3 was amplified by PCR from the pBiFC-ErbB3-VN using oligonucleotides, 5Ј-CCGGGGTACCG -ATGAGGGCGAACGACGCTCTG-3Ј (ErbB3-CFP-N) and 5Ј-CCGGGGTACCG -TC GTTCTCTGGGCATTAGCCTTG-3Ј (ErbB3-CFP-C). The PCR product was digested with KpnI and was inserted into the KpnI site of pECFP-N1 (TAKARA-BIO/Clontech, Tokyo).
The construction of pBiFC-EpoR-HA-VC was carried out by the PCR amplification of a cDNA fragment encoding the full-length murine EpoR with a HA tag at the C terminus, from pMuEpo-R190 purchased from the American Type Culture Collection (ATCC No. 40546;Manassas,VA), using primers with a KpnI site, 5Ј-CCGGGGTACCATGGACAAACTCAGGGTGC-3Ј (EpoR-HA-VC-N) and 5Ј-CCGGGGTACCAGCGTAATCTGGAACATCGTATGGGTAGGAGCAGGCCACA TAGCC-3Ј (EpoR-HA-VC-C). The product was digested with KpnI, and was cloned into pBiFC-VC155 vector.
The validity of all the constructs was confirmed by DNA sequencing and western blotting.

Cell culture
The human epithelial carcinoma cell line A-431, human breast carcinoma cell line MDA-MB-453, human breast adenocarcinoma cell line MDA-MB-468 and mouse fibroblast cell line NIH3T3 were purchased from ATCC, and were maintained in Dulbecco's modified Eagle's medium (DMEM; Nacalai Tesque, Kyoto, Japan). The CHO-K1 cell line purchased from ATCC was cultured in Iscove's modified Dulbecco's medium (IMDM; Invitrogen). Both of the media were supplemented with 10% fetal bovine serum (FBS; Invitrogen) and 100 U/ml of penicillin-streptomycin (Nacalai Tesque).

Cellular fractionation
Cellular fractionation was performed as described previously (Lin et al., 2001;Giri et al., 2005). MDA-MB-468 cells maintained in DMEM were serum-starved for 24 hours, washed twice with ice-cold PBS, and resuspended in buffer A [50 mM NaCl, 10 mM Hepes, pH 8.0, 500 mM sucrose, 1.0 mM EDTA, 0.2% Triton X-100, 0.5 mM 2-mercaptoethanol, 1.0 mM NaF, 1.0 mM Na 3 VO 4 , 1.0 mM phenylmethylsulfonyl fluoride (PMSF), and 2 μg/ml aprotinin] for 30 minutes on ice. Cells were homogenized using a Dounce homogenizer for 30 strokes. An aliquot of cells was checked for cell lysis under a microscope after staining with Trypan Blue to confirm that >98% of cells were lysed. The homogenate was centrifuged at 1,500 g for 5 minutes to sediment the nuclei. The supernatant was then resedimented at 15,000 g for 5 minutes, and the resulting supernatant was collected as a membrane-cytoplasmic fraction. The nuclear pellet was further washed three times with isotonic sucrose buffer B (250 mM sucrose, 6 mM MgCl 2 , 10 mM Tris-HCl, pH 7.4) containing 0.5% Triton X-100 to remove contaminants. The purity of the nuclei was evaluated under a microscope by staining the nuclei with 1% Methylene Blue. The prepared nuclei were clean, with no membrane sticking to the outside. To extract nuclear proteins, the isolated nuclei were resuspended in buffer C (350 mM NaCl, 10 mM Hepes, pH 8.0, 25% glycerol, 0.1 mM EDTA, 0.5 mM 2-mercaptoethanol, 1.0 mM PMSF, and 2 μg/ml aprotinin) with gentle rocking for 30 minutes at 4°C. The extracted material was sedimented at 15,000 g for 10 minutes and the resulting supernatant was collected as a nuclear fraction. The fractionation efficiency was also analyzed using antibodies against mouse α-tubulin (CP06; Merk/Calbiochem KK, Tokyo) and rabbit histone H3 (Upstate).

Confocal laser scanning microscopy
Confocal laser scanning microscopy was performed basically as described previously (Tao et al., 2006), CHO cells (~2ϫ10 5 cells/dish) were seeded in 35 mm glass-base dishes (Asahi Techno Glass, Chiba, Japan) and transfected with 0.15 μg/dish of expression plasmid for proteins of interest as indicated in the figures using 2.5 μl Plus Reagent and 6.25 μl Lipofectamine LTX Reagent (Invitrogen). Before observation, the cells were serum-starved for 24 hours followed by treatment with or without ATP synthesis inhibitors (10 mM NaN 3, 2 mM NaF and 5 mM 2-deoxy-D-glucose) 1.0 hour before adding 100 ng/ml EGF (PeproTech EC, London, UK) or 1.0 nM HRGβ1 (Lab Vision, Fremont, CA). The cells were observed under an Axiovert 200M inverted microscope equipped with an LSM 510 META Ver 3.5 scan head (Carl Zeiss, Jena, Germany) using a Plan-Apochromat 63ϫ/1.4 oil DIC immersion objective. Images were collected at a 12-bit depth resolution of intensities over 1024ϫ1024 pixels. For excitation of CFP, GFP, YFP, Cy3, Alexa Fluor 633 or Hoechst 33342, argon laser lines at 458 nm, 488 nm or 514 nm, HeNe laser lines at 543 nm or 633 nm, 716 nm laser line (titanium:sapphire from MaiTai) were employed, respectively. For simultaneous multi-imaging, emission signals were obtained and separated using Multi Track established by Carl Zeiss. The analytical conditions were kept constant throughout all the experiments.

Statistical analysis
Experimental data are reported as mean ± s.d. of triplicate independent samples. Oneway analysis of variance followed by Scheffé's test was used for multigroup comparisons. Statistical significance for two groups was assessed using an unpaired t-test. A P value <0.05 was considered to be statistically significant.