Functional isolation of activated and unilaterally phosphorylated heterodimers of ERBB2 and ERBB3 as scaffolds in ligand-dependent signaling

Most available cellular data on ERBB receptors are derived from overexpression systems. This finding reflects both technical reasons and the high importance of receptor amplification in cancer. Comparatively little is known about the rules that govern ERBB signaling at the very low receptor levels found in all but hematopoetic cells. Low endogenous ERBB levels play a critical role in the development of several organs and subsequent cardiac maintenance. To better understand both “normal” ERBB signaling and its distortion in a cancer setting, we studied ERBB2/ERBB3 signaling in MCF7 cells, a well established model system for non-ERBB2 overexpressing breast cancers with FACS-confirmed expression levels of both receptors at or below 10,000 receptors per cell. In contrast to autophosphorylation in ERBB2 overexpressing cells, ERBB2/ERBB3 signaling in MCF7 is very sensitive to nanomolar ligand concentrations. The inhibition of ligand binding or receptor heterodimerization is expected to interfere with the activation of both receptors. However, a TyrP immunoprecipitation of ligand-treated samples shows that the lower levels of remaining tyrosine phosphorylation after A30 treatment reside almost exclusively on ERBB3 (Fig. 1A). For cells with low receptor levels, mild detergent and ligand treatment can significantly influence the recovery of receptors from the membrane compartment. We confirmed the preferential inhibition of ERBB2 phosphorylation by direct SDS lysis and immunoblotting for ERBB2 pTyr1248 (Fig. 1B), pTyr1139 (e.g., Fig. 1C), and pTyr1289 for ERBB3. This approach confirms a selective inhibition of ERBB2 phosphorylation. Whereas ERBB2 phosphorylation is almost completely inhibited, ligand-induced coimmunoprecipitation of ERBB2 and ERBB3 is retained as is the phosphorylation of ERBB3 in those complexes (Fig. 1C). Compared to deliberately partial inhibition by pertuzumab (2C4), an antibody that targets the dimerization loop of ERBB2 and blocks heterodimerization (16, 17), inhibition by A30 is distinctly more asymmetric in nature (Fig. 1D). This asymmetric inhibition by A30 translates into downstream signaling (Fig. 2). The inhibition of phosphorylation at ERBB2/Y1139, one of several sites implicated in the initiation of MAPK activation, correlates with a pronounced inhibition of MAPK signaling. This is in contrast to a lack of inhibition for the ERBB3-dependent activation of AKT. Combined, these findings suggest that A30 exerts its primary inhibitory effect at a level other than ligand binding to ERBB3, heterodimerization, or the allosteric activation of ERBB2 in heterodimers. Instead, the asymmetric inhibition of ERBB2 phosphorylation creates an activated heterodimer with equally asymmetric downstream signaling properties.

A 43-nucleotide minimal aptamer segment (mA30) is sufficient for specific ERBB3 binding and targeted photo cross-linking (18). However, whereas mA30 blocks the inhibitory properties of the 78-nucleotide-long A30, it is a much less potent inhibitor of ERBB2 phosphorylation (Fig. S1). These early observations implicated steric interference with complexes beyond the heterodimer. For an initial approximation of the portion of the ERBB3 ECD involved in A30 binding, we photo crosslinked thiouracil substituted A30 and insect cell expressed extracellular domains of ERBB3 (Fig. S2). This in vitro study with purified components narrowed the putative binding region for A30 to a segment near the junction of domain III and IV that is accessible in both the extended and tethered conformation. Because no structure is currently available for the ERBB3 receptor ECD in the extended conformation, we used the available structure of tethered ERBB3 for the initial selection of residues for mutagenesis. A striking feature of the domain III–IV junction is its neutral-to-positive surface charge (Fig. 3A). Whereas charge-based interactions alone cannot account for the high specificity and nanomolar affinity of A30, electrostatics exert strong “steering” forces during a SELEX procedure. Combined, these considerations further narrowed the list of candidate residues, making an analysis of cell-surface–expressed receptors by mutagenesis feasible.

For live cell binding studies, we focused on four locations where mutations were not likely to significantly alter the underlying backbone conformation yet have a significant impact on the immediate surrounding surface. Binding was measured on large numbers of individual live cells using a triple color fluorescent tag assay that quantifies receptor levels (C-terminal GFP fusion, green), surface expression (N-terminal FLAG epitope tag, blue), and aptamer binding (mA30-biotin and streptavidin-Texas Red) on MCF7 cells that stably express the mutant receptors in a large excess over endogenous ERBB3. Fig. 3B shows the ratio of aptamer binding to surface receptors for 200–300 individual cells. Whereas the removal of the positive surface charge at lysine 453 and arginine 456 results in a modest (10%) increase in A30 binding, the removal of two negative charges at glutamic acid 460 and 461 moderately diminishes binding (7%). Both differences are statistically significant at P < 0.01. More extensive and statistically highly significant inhibition of binding (P < 0.001) was observed after mutating histidines 446/447 (20%) or arginines 471/472 (22%) to alanines. Those four residues form a contiguous surface patch that is spatially close to glutamic acids 460 and 461. H446/H447 represent the C-terminal “cap” of domain III and R471/R472 is located directly in the loop region between domains III and IV. The R471/472 site was selected for charge reversal, resulting in a nearly complete loss of A30 binding (90% inhibition). Inhibition is independent of receptor density (Fig. S3). The impact of mutagenesis on fully adherent cells is shown in Fig. 3C, demonstrating the loss of A30 binding to the R471/472E mutant at comparable expression and cell surface presentation levels. The stably expressed R471/472E mutant displays wild-type equivalent responsiveness to ligand, both in terms of ERBB2 tyrosine phosphorylation and downstream activation of MAPK (Fig. 4A) but is resistant to inhibition by A30. This response pattern suggests that the mutagenesis impacts A30 binding but is alone insufficient to disrupt proxy phosphorylation.

To evaluate whether the mechanism of signaling and inhibition by A30 is sensitive to receptor levels, we transiently overexpressed wild-type ERBB3 in MCF7 cells (Fig. 4B). Ligand-independent activation is minimally enhanced (measurable only at longer exposures), whereas the ligand-induced phosphorylation of ERBB3 shows both a pronounced increase and enhanced A30 sensitivity. While the A30 inhibition of ERBB3 phosphorylation at endogenous receptor levels plateaus at 20% (Fig. 2), the ligand-dependent phosphorylation of overexpressed ERBB3 is suppressed by ∼60%. By contrast, the level of A30-insensitive phosphorylation is almost constant, regardless of ERBB3 levels. Combined, these inhibitor and ligand responses suggest that upon overexpression, a shift in the mechanism of ERBB3 phosphorylation occurs toward a more A30-sensitive interaction.

Whereas levels of ERBB3 that are orders of magnitude higher than ERBB2 are unnatural, the inverse scenario is a hallmark of many ERBB2 overexpressing cancers. We therefore evaluated the impact of A30 and pertuzumab on the ERBB2-overexpressing breast cancer cell line BT474 (Fig. 4C) and SKBr3 (Fig. S4), each harboring well over one million ERBB2 receptors in large excess over ERBB3. The phosphorylation of ERBB3 but not ERBB2 is stimulated by ligand, and a pertuzumab concentration that achieves the maximal achievable inhibition (Fig. S5) is more effective than A30 in suppressing the ligand-induced phosphorylation of ERBB3. Arguably more relevant to the cancer cell scenario, each inhibitor suppresses constitutive ERBB3 phosphorylation only partially, but both act synergistically, resulting in almost complete suppression of constitutive ERBB3 phosphorylation. Whereas largely ineffective in the inhibition of ERBB2 autophosphorylation, A30 is also surprisingly effective in further enhancing the ability of pertuzumab to suppress the constitutive and lapatinib-sensitive (Fig. S6) phosphorylation of ERBB2.

Antibodies were obtained from Santa Cruz Biotechnology (ERBB3/C17 and ERBB2/C18), Upstate Biotechnology (pTyr/4G10, pERBB2(Tyr1248), Cell Signaling Technologies [pERBB3(Tyr1289), pMAPK(Thr202/Tyr204), MAPK, pAKT(Ser473), and AKT], Epitomics [pERBB2(Tyr1139), Biogenex (ERBB2/CB11), Invitrogen (V5-HRP and V5), and Evrogen (Dendra)]. The purification of ERBB3–ECD constructs and the Trx–NRG fusion protein of the EGF-like domain of NRG1-β1 with thioredoxin have been described previously (3). Pertuzumab was provided by Mark Sliwkowski (Genentech Inc., South San Francisco, CA).

MCF7 cells were maintained in RPMI-1640 [10% (vol/vol) FBS, 5% (vol/vol) CO₂]. GFP fusion constructs of ERBB3 were expressed in the pFLAG-MYC-CMV-19 expression vector (Sigma), and cotransfected with puromycin resistance marker. Stably expressing cell lines were sorted for receptor–GFP fusions by FACS before expansion. For transient ERBB3 overexpression, MCF7 cells were transfected with either Dendra2 (Evrogen) or pFlag-ERBB3-Dendra2.

A30 was transcribed as previously described (15) using the RiboMAX large-scale RNA production system from Promega. For inhibition studies, the minimal aptamer (mA30) was generated using a PCR template with a shortened 5′ end. Synthetic minimal aptamer with a 3′ biotin was used for fluorescence microscopy studies.

Indicated cell lines at 60–70% confluency were treated with aptamer in RNaseOUT supplemented RPMI-1640 for 30 min at 37 °C, stimulated with Trx-NRG (10 nM) for 10 min at room temperature, and lysed in SDS sample buffer (57 mM Tris-HCl, 10% (vol/vol) glycerol, 3.3% (wt/vol) SDS, 0.17 mg/mL bromophenol blue, 1.7 mg/mL DTT) before Western blot analysis with the indicated antibodies. For coimmunoprecipitation, cells were lysed in mild lysis buffer [20 mM Tris, 137 mM NaCl, 1% (vol/vol) Triton X-100, 10% (vol/vol) glycerol, 5 mM EDTA, 1 mM sodium orthovanadate, 1 mM PMSF, 1× mixture]. Lysates were passed 10 times through a fine-gauge needle and incubated for 5 min at 37 °C. Cell debris was removed by centrifugation at 10,000 × g before immunoprecipitation of supernatants at room temperature for 2 h, five washes in mild lysis buffer and denaturation in 1× SDS sample buffer containing DTT. Inhibition studies in BT474 and SKBr3 cells were carried out as described above for MCF7 cells. For ERBB2 coimmunoprecipitation studies, MCF7 cells were cotransfected with ERBB2 constructs carrying either an N-terminal V5 epitope tag or C-terminal Dendra.

Primary sequences were aligned using clustalW. Structural models were created using Swiss Model builder (24) and surface charges were calculated in Swiss viewer using the default parameters. The EGF dimer was used as a structural framework to model the ERBB2/ERBB3 heterodimer. The C terminus of domain IV of ERBB3 (obtained from the tethered conformation) was projected into the N-terminal segment of domain IV of EGFR in the dimer structure.

MCF7 cells with stably overexpressed mutants of ERBB3–GFP were established after FACS sorting for GFP. Cells were dissociated with Versene (PBS/EDTA) and incubated for 10 min with 100 nM 3′ biotinylated minimal A30 and anti-FLAG(M2) antibody (Sigma) followed by 20 min of staining with streptavidin-Texas Red conjugate and Pacific-Blue–labeled secondary. Cells were allowed to settle and images of the uniformly rounded cells were acquired on a Zeiss Axiovert 200 M fluorescence microscope at low magnification in a rapid and automated manner using Openlab command script execution. Following density slicing, cells (<5%) with GFP signal outside two SDs of the mean (mainly due to well boundary effects) were excluded from the analysis. The ratio of surface localization (Pacific Blue) and aptamer binding (Texas Red) was analyzed as discussed. The statistical significance was analyzed using an uncoupled dual distribution t test, with P < 0.01 considered significant and P < 0.001 very significant.