Direct stimulation of ERBB2 highlights a novel cytostatic signaling pathway driven by the receptor Thr701 phosphorylation

ERBB2 is a ligand-less tyrosine kinase receptor expressed at very low levels in normal tissues; when overexpressed, it is involved in malignant transformation and tumorigenesis in several carcinomas. In cancer cells, ERBB2 represents the preferred partner of other members of the ERBB receptor family, leading to stronger oncogenic signals, by promoting both ERK and AKT activation. The identification of the specific signaling downstream of ERBB2 has been impaired by the lack of a ligand and of an efficient way to selectively activate the receptor. In this paper, we found that antibodies (Abs) targeting different epitopes on the ERBB2 extracellular domain foster the activation of ERBB2 homodimers, and surprisingly induce a unique cytostatic signaling cascade promoting an ERK-dependent ERBB2 Thr701 phosphorylation, leading to AKT de-phosphorylation, via PP2A Ser/Thr phosphatases. Furthermore, the immunophilin Cyclophilin A plays a crucial role in this pathway, acting as a negative modulator of AKT de-phosphorylation, possibly by competing with Ser/Thr phosphatases for binding to AKT. Altogether, our data show that Ab recognizing ERBB2 extracellular domain function as receptor agonists, promoting ERBB2 homodimer activation, leading to an anti-proliferative signaling. Thus, the ultimate outcome of ERBB2 activity might depend on the dimerization status: pro-oncogenic in the hetero-, and anti-oncogenic in the homo-dimeric form.


Results
Antibodies against ERBB2 extracellular domain promote AKT de-phosphorylation via ERK. To identify ERBB2-specific contribution in modulating intracellular signaling pathways, we sought to stimulate the receptor with two antibodies (Abs) targeting different epitopes in the ERBB2 extracellular domain (ECD). In particular, we used Pertuzumab (PZ 27 ) recognizing domain II, and Trastuzumab (TZ 28 ) targeting domain IV of the receptor 29 .
PI3K/AKT survival and RAS/MAPK proliferation pathways are the major signaling pathways activated by ERBB2 heterodimers 30 , thus we initially investigated their phosphorylation kinetics in response to treatments with Abs in SK-Br3 human breast cancer cells. For AKT we focused on the two major phosphorylation residues phosphorylated in response to RTK activation, and necessary for AKT full enzymatic activity, i.e. Thr 308 and Ser 473 . To avoid long-term signaling adaptation mechanisms, short-term kinetics (2, 20 and 45 min) after treatment with Abs was evaluated. By western blotting (WB) upon treatment with either TZ or PZ, we observed an early persistent increase of phospho-ERK, and a dramatic reduction of the AKT Thr 308 and Ser 473 residues phosphorylation, preceded by a transient mild increase (Fig. 1a).
As Abs modulate ERK and AKT in opposite directions, we tested whether the two events are interdependent, evaluating the ERK and AKT phosphorylation state upon treatment with either one of the two Abs in the presence of the MEK1 inhibitor U0126, to prevent ERK activation (Ab/U0126). Ab/U0126 treatment significantly delayed AKT de-phosphorylation of both Thr 308 and Ser 473 residues (Fig. 1a), showing that the reduction in phospho-AKT levels in the presence of ERBB2-targeted antibodies depends on ERK activation.
To corroborate these findings, we performed TZ treatment in an additional cell line (BT474) showing ERBB2 levels comparable to SK-Br3 and in MDA-MB-468 cells, which express negligible levels of ERBB2 ( Supplementary  Fig. 1a). We found that only cells expressing high levels of ERBB2 responded to TZ, showing an ERK-dependent AKT de-phosphorylation ( Supplementary Fig. 1b).
As a proof of principle that our findings obtained in SK-Br3, and BT474 cells could be reproduced in primary human ERBB2-BrCa, we established primary cell cultures from patient-derived-xenografts (PDX) developed from one ERBB2-positive patient and one ERBB2-negative BrCa patient, as control. The fraction of cells expressing ERBB2 was assessed by immunofluorescence (IF, data not shown), and only cell cultures obtained from the ERBB2-positive PDX displayed ERBB2 expression in a sizable proportion of the cells (18%).
ERBB2-positive PDX derived cells, but not ERBB2-negative PDX derived cells, responded to TZ and TZ/ U0126, displaying an ERK-dependent reduction of the levels of phospho-AKT ( Supplementary Fig. 1c), thus supporting the data obtained using SK-Br3, and BT474 immortalized cell lines (Fig. 1a, and Supplementary Fig. 1b).
Ab-induced AKT de-phosphorylation is an exclusive ERBB2 signaling. As SK-Br3 cells express high ERBB2 levels, detectable levels of EGFR and ERBB3, and very low or no ERBB4 ( Supplementary Fig. 1a), we sought to assess whether ERBB2 requires the cooperation of other ERBB family members to respond to Abs. Thus, we silenced SK-Br3 cells for both EGFR and ERBB3 ( Supplementary Fig. 2a), and treated them with either Scientific Reports | (2020) 10:16906 | https://doi.org/10.1038/s41598-020-73835-1 www.nature.com/scientificreports/ Ab or Ab/U0126. In all conditions, cells responded as control cells transfected with a non-targeting (NT) siRNA ( Supplementary Fig. 2b), suggesting that EGFR and ERBB3 are dispensable for the response to ERBB2-targeted Abs. We then asked whether stimulation of ERBB1 or ERBB3 would mimic the downstream signaling elicited by Abs. Thus, SK-Br3 cells were treated with EGF (ligand for EGFR), or heregulin β1 (HRG, ligand for ERBB3): in contrast to Abs ( Supplementary Fig. 3a), treatment with EGF or HRG promoted both ERK and AKT phosphorylation ( Supplementary Fig. 3b or 3c, respectively), suggesting that the ERK-dependent AKT inactivation is an Ab-exclusive signaling.
Abs binding promotes ERBB2 homodimerization. EGFR and ERBB3 dispensability for the response to the Abs suggest that the ERK-dependent AKT inactivation is likely due to ERBB2 homodimerization. Based on this hypothesis, we first evaluated the role of the ERBB2 tyrosine kinase activity in the process, by treating SK-Br3 cells with Abs in presence or absence of Lapatinib, an ERBB2 and EGFR tyrosine kinase inhibitor. Under these conditions, Lapatinib completely abrogated the Ab-dependent signaling ( Supplementary Fig. 3d), confirming that an active ERBB2 tyrosine kinase is required.
To assess whether Ab binding promotes ERBB2 homodimerization, we performed a mass spectrometry analysis of the ERBB2 containing complexes immunoprecipitated after 10 min of treatment with TZ or PZ (Supplementary Fig. 4 and Supplementary Table S1), and cross-linking with membrane-impermeable DTSSP. WB analysis in not-reducing conditions showed a decrease in the abundance of the monomeric form and an increase of the ERBB2-positive high molecular weight species upon treatment with TZ or PZ ( Supplementary Fig. 4d). Peptide analysis revealed that Ab treatment induced a 20-fold (TZ, Fig. 1b) or a five-fold (PZ, Fig. 1c) increase in the abundance of ERBB2-containing dimers.
To rule out the possibility that the higher levels of ERBB2 dimerization might be due to the clustering effect promoted by the bivalent Abs, we generated TZ and PZ Fab fragments (TZ-Fab and PZ-Fab, respectively). By immunoprecipitation (IP) experiments followed by mass spectrometry, we found that, similarly to the bivalent Ab, TZ-Fab promoted a 13-fold increase in the abundance of ERBB2-containing dimers (Fig. 1b). Moreover, by quantifying the fraction of EGFR (to estimate heterodimers) and ERBB2 (to estimate all dimers) engaged in dimers (normalizing the values for their respective amounts in the input fractions), we found that in each condition heterodimers represented approximately 7% of all dimers (Fig. 1b). By contrast, IP experiments followed by mass spectrometry showed that, at odds with the bivalent antibody, PZ-Fab did not induce receptor dimerization, possibly because it engages the dimerization arm of ERBB2 (Fig. 1c).
Altogether, these results support the conclusion that Abs promote the two major events leading to ERBB activation upon ligand binding, i.e. receptor dimerization and activation of a downstream signaling, thus acting as agonists for ERBB2.
Ab-induced AKT de-phosphorylation depends on Ser/Thr phosphatase activity. There are three main alternative ways to obtain an ERK-dependent reduction of phospho-AKT levels: (i) inhibition of the PI3K/ AKT phosphorylation axis; (ii) activation of Ser/Thr phosphatases leading to AKT de-phosphorylation; (iii) a combination of the two.
Phospho-ERK may control the PI3K-dependent activation of AKT via mTORC1. We assessed the phosphorylation status of the mTORC1 downstream effector p70S6 kinase (S6kinase), as readout of mTORC1 activation 32,34,35 ( Supplementary Fig. 5a), and found no differences in samples treated with TZ or TZ/U0126 ( Supplementary  Fig. 5c), thus excluding an inhibitory effect on PI3K and subsequently of AKT phosphorylation. Accordingly, SK-Br3 cells treated with either the mTORC1 inhibitor rapamycin, or the PI3kinase inhibitor LY294002, before and during TZ treatment showed a complete inhibition of S6kinase phosphorylation, thus excluding a role for the ERK pathway in controlling mTORC1 activity in this cellular model ( Supplementary Fig. 5e).
We then evaluated the possible involvement of PTEN, a cytoplasmic phosphatidylinositol-3,4,5-trisphosphate 3-phosphatase, able to counteract the activity of PI3K, once recruited on the plasma membrane (PM). It has been reported that PTEN recruitment to the PM involves the kinase activity of ERK, the subsequent phosphorylation of MEK1 Thr 292 residue, and the formation and translocation of a tri-partite complex comprising phosphorylated MEK1, MAGI1 and PTEN to the PM 33 ( Supplementary Fig. 5b). Thus, we analysed the membrane/cytosol partitioning of the three components of the complex and found that it is similar in cells treated with TZ and TZ/U0126 ( Supplementary Fig. 5d). In particular, phospho-MEK1 was exclusively present in the cytosolic fraction, MAGI1 in the membrane fraction, and PTEN mainly in the cytosol, excluding also this pathway in the Ab-induced AKT inactivation.
Having excluded a mechanism involving the inhibition of the AKT phosphorylation process, to explain the reduced levels of phospho-AKT promoted by TZ or PZ, we then evaluated the possible contribution of Ser/ Thr phosphatase activity. Different phosphatases have been reported to function on AKT specific phosphoresidues; in particular, PP2C family members work exclusively on Ser 473 , PP2A holoenzymes on Thr 308 , and PP1 complexes on Thr 450 and/or Ser 47336,37 . Thus, we exploited well-established inhibitors of PP1 and PP2A families of Ser/Thr phosphatases, i.e. calyculin, okadaic acid, and tautomycin, in SK-Br3 cells treated with TZ, either as single agents or in combination. While the phosphorylation levels of AKT Ser 473 residue were not affected by any of the inhibitor combinations (Supplementary Fig. 6a and 6c), the phosphorylation levels of AKT Thr 308 residue increased significantly in the presence of higher concentrations of okadaic acid. As the shape of the de-phosphorylation kinetics curves in control and phosphatase inhibitors-treated samples was comparable, we concluded that inhibition of the catalytic activity of Ser/Thr phosphatases per se does not mimic the TZ/  Fig. 6a and 6b). By contrast, TZ/U0126 treatment significantly delayed AKT de-phosphorylation kinetics, leading to the possibility that ERK may regulate the association of phosphatases to phospho-AKT.

Cyclophilin A interacts with phospho-AKT in an ERK-dependent manner.
To unravel the machinery involved in the regulation of ERBB2-dependent AKT de-phosphorylation, we analysed phospho-AKT interacting proteins by mass spectrometry. Lysates of cells treated for 20 min (time point displaying the greatest difference in phospho-AKT levels between Ab and Ab/U0126, Fig. 1 and Supplementary Fig. 6b) with TZ, or TZ/U0126, were immunoprecipitated using an anti-phospho-Thr 308 AKT antibody. Proteomic analysis showed that, in the TZ/U0126-treated cells, phospho-AKT interacted, among other proteins, with the Ser/Thr-protein phosphatase 2A (PP2A)-catalytic subunit beta isoform (PPP2CB, UniProtKB accession number: P62714, 5% sequence coverage), and the peptidyl-prolyl cis-trans isomerase A (Cyclophilin A, CyPA) (PPIA, UniProtKB accession number: P62937, 23% sequence coverage) ( Fig. 2a and Supplementary Table S2). In the TZ-treated cells, instead, we could not detect any PP2A or CyPA bound to the residual phospho-AKT.
CyPA silencing promotes AKT de-phosphorylation in the absence of ERK signaling. CyPA is a member of the immunophilin family, which comprises cyclophilins, FK506-binding proteins and parvulins 38 .
The best-known member of the parvulin group is Pin1, which has been reported to control AKT phosphorylation and stability 36 . We found that CyPA is associated to phospho-AKT, suggesting two alternative scenarios, i.e. CyPA exerts a regulatory role on AKT de-phosphorylation or CyPA is a target of AKT activity.
Since it has been described that AKT can phosphorylate CyPA 39 , we tested whether this event occurred in our experimental setting. Thus, we immunoprecipitated CyPA from cells treated with TZ or TZ/U0126, and probed with two antibodies specific for phospho-AKT substrates (RXXS/T* and RXRXXS/T*) 40 by WB. In neither case we could detect any labelling ( Supplementary Fig. 7a), suggesting that in our system phospho-AKT does not act upstream of CyPA, but it is rather a downstream effector of CyPA.
To identify the role of CyPA in the TZ-induced AKT de-phosphorylation, we silenced CyPA expression by transducing SK-Br3 cells with a short hairpin RNA (shRNA) targeting CyPA, and evaluated the AKT dephosphorylation kinetics upon TZ/U0126 administration (condition in which we found CyPA associated to phospho-AKT, Fig. 2b,c and Supplementary Fig. 7b). In five independent experiments, CyPA silencing consistently impaired the AKT Thr 308 de-phosphorylation delay observed upon TZ/U0126 treatment (Fig. 2b,c), but not the AKT Thr 308 de-phosphorylation promoted by TZ ( Supplementary Fig. 7c), suggesting that ERK may stimulate AKT de-phosphorylation by promoting the detachment of CyPA from phospho-AKT.
CyPA is recruited by ERBB2 upon Ab treatment. To better understand the role of CyPA in regulating AKT de-phosphorylation, we explored the landscape of molecular interactions between the members of ERBB2 signaling pathway and CyPA by performing direct and reverse IP experiments. Lysates from SK-Br3 cells either untreated, or treated with TZ or TZ/U0126 for 20 min, were immunoprecipitated with an antibody against phospho-Thr 308 AKT. We found that in TZ-treated, compared to untreated cells, phospho-AKT co-immunoprecipitated about ten folds higher amounts of the phosphorylated form of ERBB2 (phospho-ERBB2), and about two folds the amount of CyPA, but almost no ERK or PP2A catalytic subunit (possibly because of the low amount of AKT still phosphorylated in this condition); conversely, when TZ/U0126-treated cells were compared to untreated cells, phospho-AKT co-immunoprecipitated equal amounts of ERK, higher amounts of CyPA and PP2A catalytic subunit, but not phospho-ERBB2 (Fig. 3a,c).
The IP using a CyPA-specific antibody confirmed these data, with two noteworthy exceptions: a more efficient precipitation of ERK in TZ-treated as compared to untreated cells (Fig. 3b,c), and the lack of PP2A precipitation in both TZ and TZ/U0126 conditions. Strikingly, we also found that in TZ treated, but not in TZ/U0126 treated cells, CyPA co-immunoprecipitated phospho-ERBB2.
To further support this observation, we investigated the subcellular distribution of CyPA (an abundant cytosolic protein), with respect to the PM-associated ERBB2 receptor, in SK-Br3 cells upon Ab or Ab/U0126 treatment. Cells were analysed by means of direct stochastic optical reconstruction super-resolution microscopy (dSTORM), and co-localization assessed by pair cross-correlation 41 . We found that in Ab, compared to Ab/U0126 condition, the co-localization between CyPA and ERBB2 was significantly higher (Fig. 3d,e), further supporting the finding that ERBB2-targeted Abs induced the recruitment of CyPA onto ERBB2 receptor.
Altogether, these results suggest a model in which Abs induce a phospho-ERK-dependent recruitment of a signaling complex comprising phospho-ERK, CyPA, and phospho-AKT, onto the C-terminal domain of ERBB2. Inhibition of ERK phosphorylation, in turn, leads to an increased interaction between CyPA and phospho-AKT and impairs the recruitment of the complex to the receptor. Because phospho-AKT levels are dramatically reduced upon Ab, but not upon Ab/U0126 treatment, and CyPA appears to play a role in protecting AKT from de-phosphorylation, we hypothesize that CyPA recruitment onto ERBB2 C-terminal domain might release AKT from CyPA, leaving it accessible to Ser/Thr phosphatases (see below ).
CyPA and PP2A bind to the same AKT residues. In order to identify the putative AKT moieties involved in CyPA binding, we generated FLAG-tagged wt and biomimetic mutants of the Thr 308 residue of AKT1 (i.e. AKT T308A , not phosphorylatable, and AKT T308D , phosphomimetic) and performed IP experiments on cells expressing either one of the FLAG-AKT constructs. Unexpectedly, we found that upon 20 min of TZ/U0126 treatment, both biomimetic AKT mutants, with putative different outcomes on AKT activation, immunoprecipitated approximately 50% of CyPA and 60% of PP2A, as compared to wt AKT (Fig. 4a,b). These findings suggest and on their levels in control TZ-untreated SK-Br3 cells. One experiment is shown, as representative of two. Signal in the control IgG lanes in the TZ and TZ + U0126 samples might be due to the direct binding of TZ antibody to Protein G dynabeads, and was subtracted before normalization. The higher molecular weight of bands in the IP, as compared to the input, lanes might be due to the crosslinker. (d,e) SK-Br3 cells were serumstarved for 20 min, in the presence of 10 μM U0126 or vehicle (DMSO), and treated with 10 μg/ml TZ (d) or 20 μg/ml PZ (e) at 37 °C for 20 min. On the day of acquisition, cells were processed for immunofluorescence and analysed by dSTORM super-resolution microscopy. Co-localization was assessed by pair cross-correlation analysis. For TZ, ten regions/field were chosen and eleven fields were analysed, out of three independent experiments that were pooled together. Data are presented as mean ± s.e.m. (p < 5 × 10 -9 upon Kolmogorov-Smirnov test on cross-correlation signal amplitude). For PZ, ten regions/field were chosen and eight fields were analysed, out of two independent experiments that were pooled together. Data are presented as mean ± s.e.m. (p < 0.05 upon Kolmogorov-Smirnov test on cross-correlation signal amplitude). Scale bar: 4 μm. www.nature.com/scientificreports/ that any substitution of Thr 308 affects CyPA binding. Furthermore, as CyPA and PP2A do not directly interact (Fig. 3b,c), it is feasible that they may compete for the same binding site on AKT.
Since CyPA is a peptidyl-prolyl isomerase, we tested whether the proline residues of AKT close to Thr 308 were relevant for the binding to CyPA. Thus, we mutagenized Pro 313 and Pro 318 of AKT to generate single and double mutants. We then treated cells expressing FLAG-tagged wt, AKT P313A , AKT P318A or AKT P313A/P318A constructs, with TZ/U0126 for 20 min, and performed IP experiments using an anti-FLAG antibody. Under these conditions, www.nature.com/scientificreports/ Pro 318 substitution (either alone or in combination with P313A mutation) reduced the interaction between AKT and CyPA by about 60% (Fig. 4a,b). Similarly to Thr 308 , Pro 318 substitution impaired also PP2A binding to AKT. Thus, we can envisage that CyPA inhibits AKT de-phosphorylation, by competing with PP2A for the binding to AKT Thr 308 residue. Co-IP experiments (Fig. 3) suggest that the recruitment of CyPA to the ERBB2 cytoplasmic tail could be necessary to displace CyPA from phospho-AKT, thus making the latter available to PP2A phosphatase activity.
Since CyPA is a cytosolic protein and it binds to phospho-AKT, which is phosphorylated on Thr 308 by P-PDK1 on the PM, we hypothesized that CyPA is recruited onto ERBB2 upon Ab treatment by means of its interaction with phospho-AKT. Therefore, we immunoprecipitated CyPA in cells expressing FLAG-AKT T308A treated with TZ for 20 min, and found that it binds approximately 40% less phospho-ERBB2 as compared to wt AKT (Fig. 4c,d). Thus, we conclude that TZ-induced ERK-dependent phospho-AKT recruitment to ERBB2 leads also to CyPA relocalization to the receptor.
ERK and CyPA positively regulate ERBB2-Tyr 1248 activation via a feedback loop involving ERBB2-Thr 701 phosphorylation. Since the Ab-induced recruitment of the complex comprising ERK, CyPA and phospho-AKT onto ERBB2 depends on phospho-ERK, we reasoned that active ERK might either function as a scaffold protein, docking the complex onto the receptor cytoplasmic domain, or exert an enzymatic activity on ERBB2, rendering it accessible to the complex. Indeed, while both AKT and CyPA are not ERK substrates, as they do not comprise consensus sequences for ERK-mediated phosphorylation 42 , evidences from the literature suggest that ERBB2 can be a target of phospho-ERK 43 .
In particular, ERBB2 intracellular domain comprises four target consensus sequences PX[S/T]P for ERK kinase 42 . To evaluate whether ERBB2 was a putative target for ERK activity upon Ab binding, we immunoprecipitated ERBB2 at different time-points following Abs treatment, and revealed ERK kinase consensus target sites with an antibody recognising anti-phospho-Threonine-Proline (P-TP) by WB 44 . The results showed that both TZ and PZ treatment induced a transient increase in P-TP signal, as compared to starved cells, with a peak around 2 min for TZ and 20 min for PZ (Fig. 5a). Noteworthy, under these conditions we observed that phospho-ERK co-immunoprecipitated with ERBB2 (Fig. 5a).
To identify the specific receptor residues specifically phosphorylated by ERK we immunoprecipitated ERBB2 after 5 min of TZ or TZ/U0126 and after 20 min of PZ or PZ/U0126 treatment and performed label-free quantitative mass spectrometry. Peptide analysis revealed that Thr 701 was specifically phosphorylated upon both TZ (Fig. 5b, upper table) and PZ (Fig. 5b, lower table) treatments, but not in the corresponding Ab/U0126 conditions. Noteworthy, PZ treatment induced the ERK-dependent phosphorylation of additional sites, namely Ser 998 and Ser 1073 (Fig. 5b, lower table).
As it has been reported that the ERK-mediated phosphorylation of a Thr residue in the juxtamembrane region of the EGFR cytoplasmic domain modulates the EGF-induced receptor downstream signaling 44,45 , we tested whether ERK would control the activation of ERBB2 as well. To this end, we assessed the phosphorylation state of the ERBB2 Tyr 1248 residue. Tyr 1248 is an autophosphorylation site 46 in the receptor C-terminal tail and its phosphorylation is considered a pre-requisite for ERBB2 signaling, and transforming activity 2 . By WB, we observed that both TZ and PZ treatments induced a strong Tyr 1248 phosphorylation in SK-Br3 cells (Fig. 5a,c), and that this response was lowered in the corresponding Ab/U0126 condition (Fig. 5c). These data suggest that the Ab-induced ERK-dependent phosphorylation of ERBB2 juxtamembrane region (Thr 701 ) may function as a positive feedback mechanism to promote ERBB2 C-terminal (Tyr 1248 ) activation, and downstream signaling.
Since ERK ability to phosphorylate its substrates has been shown to be sensitive to the conformation of the substrate backbone [(Ser/Thr)-Pro] 47 , in the ERBB2 context, we hypothesized that CyPA could be involved in the positive feedback loop controlling ERBB2 C-terminal tail phosphorylation. To test this hypothesis, we analyzed the levels of ERBB2 Tyr 1248 phosphorylation in SK-Br3 cells silenced for CyPA, and found significantly reduced levels of phospho-ERBB2 [50% as compared to control nt shRNA-transduced cells, Fig. 6a,b], supporting the conclusion that CyPA positively modulates ERBB2 activation state.
Thus, as we have shown that the immunophilin recruitment onto ERBB2 depends on phospho-ERK (Fig. 3), we mutagenized the Pro 702 residue of the receptor into Ala, expressed the recombinant mutant GFP-tagged proteins in SK-Br3 cells, and treated them with TZ or PZ. We found that ERBB2 P702A -GFP recruits less CyPA as compared to wt ERBB2-GFP upon treatment with either Ab (Fig. 6c-f).
Altogether, these results suggest that upon Ab treatment: (i) ERK takes part to a positive feedback loop keeping ERBB2 active through the phosphorylation of ERBB2 Thr 701 residue; (ii) CyPA participate to the regulation of ERBB2 activation and (iii) Pro 702 represents the ERBB2 binding site for CyPA.

Discussion
While almost undetectable or expressed at very low levels in normal tissues 1 , ERBB2 is overexpressed in several carcinomas, including breast cancers that display a more aggressive clinical course and a worst outcome compared to non ERBB2-positive BrCa 4,5 . ERBB2 represents the preferred partner of the other ERBB receptors 14 , and ERBB2-containing heterodimers are more oncogenic than other ERBB combinations [15][16][17][18][19] . Thus, it is widely accepted that ERBB2 has a pro-oncogenic role. However, its precise function and downstream signaling pathway have not been assessed: the absence of a known ligand for ERBB2 has been the major difficulty to address this issue, and most studies have relied on the stimulation of the partner receptor in ERBB2-containing heterodimers.
Here, we show that, at variance to ligand activation of ERBB2-ERBB3 heterodimers by HRG or of EGFR-ERBB2 heterodimers by EGF, triggering both ERK and AKT signaling pathways, treatment with ERBB2-targeting Abs leads mainly to receptor homodimerization and to an ERK-dependent AKT inactivation. This signaling is ERBB2-specific and appears not to depend on ERBB2 heterodimers as in the context of the breast cancer cell www.nature.com/scientificreports/ line used here (expressing negligible ERBB4 levels), silencing of EGFR and ERBB3 does not impair the ERKdependent AKT de-phosphorylation elicited by Abs. On the other hand, the very early transient peak in AKT phosphorylation, preceding its de-phosphorylation, is likely due to ERBB2 activity in a heterodimeric context. Indeed, at variance to EGFR and ERBB3, ERBB2 has no binding sites for PI3K in its cytosolic domain 7 , thus it cannot directly stimulate AKT phosphorylation. Hence, Ab binding to ERBB2 leads to two independent events: in the heterodimeric context, ERBB2 cooperates on the phosphorylation of AKT induced by the partner; in the homodimeric form, the receptor triggers the specific negative modulation of the AKT pathway, thus counteracting the pro-survival activity promoted by ERBB heterodimers. This agonist role played by the Abs on ERBB2 is further supported by structural studies/simulations showing that TZ or PZ binding to ERBB2 promotes conformational changes of the receptor, similar to those reported for EGFR upon binding to EGF (reviewed in 48 ), in particular: (1) fluctuation in domain II of the receptor upon TZ binding and in domain IV upon PZ binding 49 , (2) displacement of the transmembrane domain, which loses its interaction with the inner leaflet of the membrane bilayer, upon TZ binding 50 and (3) interference with the antiparallel alignment of the juxtamembrane domains upon TZ binding 50 .
The Ab-induced ERK-dependent signaling cascade triggering AKT de-phosphorylation involves both PP2A Ser/Thr phosphatases and the cytosolic immunophilin CyPA. Immunophilins exert different important functions in the cell, acting either as molecular chaperones assisting and/or correcting protein folding, or as enzymes catalysing the cis-trans isomerization at Pro residues 51 . It has been reported that the immunophilin Pin1 is overexpressed in breast cancer and that its up-regulation is prevalent in ERBB2-positive tumours 52,53 ; however, its silencing or functional inhibition does not potentiate TZ activity, possibly because it accelerates receptor degradation 54 . Similarly to Pin1, CyPA is found overexpressed in different types of tumours, including breast  (upper table) or with 20 μg/ml PZ for additional 20 min (lower table) at 37 °C. Lysates were immunoprecipitated with an ERBB2-specific antibody. Samples were analysed by mass spectrometry. In the tables, mean ± st. dev. from one experiment is reported (out of two technical replicates). The results shown are representative of two independent biological replicates. (c) SK-Br3 cells were either left untreated or serum-starved for 20 min, in the presence of 10 μM U0126 or vehicle (DMSO), and treated with 10 μg/ml TZ or 20 μg/ml PZ at 37 °C for 20 min. Lysates were analysed by WB. For TZ, one experiment is shown as representative of nine; for PZ, one experiment is shown as representative of three. ERK inactivation reduces ERBB2 Tyr 1248 phosphorylation induced by TZ or PZ down to 66% and 69% (where control sample is normalized as 100 percent), respectively (p = 0.02, paired Student's t test, for either TZ or PZ treatment). The images of the full scan western blots are provided in Supplementary Fig. 12 www.nature.com/scientificreports/ cancer, and in some reports a correlation between CyPA overexpression and malignant transformation has been reported 55,56 . Our knock down experiments show that CyPA counteracts Ser/Thr phosphatase activity on phospho-AKT, possibly by competing with PP2A for binding to phospho-AKT, as suggested by IP experiments with AKT mutants. Upon Ab treatment, a complex comprising phospho-ERK, CyPA and phospho-AKT is recruited onto the cytoplasmic domain of ERBB2, and a reduction in phospho-AKT levels is observed. Altogether, these findings support the hypothesis that the recruitment of CyPA onto ERBB2, induced by Ab treatment, renders www.nature.com/scientificreports/ phospho-AKT accessible to Ser/Thr phosphatase activity, explaining the results we obtained with the phosphatase inhibitors, i.e. that TZ regulates the association of the phosphatase to its substrate rather than the catalytic activity. This unanticipated role of Ser/Thr phosphatases in TZ action reconciles published observations implicating these enzymes both in breast carcinogenesis 24 and in the onset/development of resistance to TZ treatment 57 . Furthermore, our findings are in agreement with published data showing that AKT Ser 473 de-phosphorylation is essentially mediated by PP2C phosphatases, whereas AKT Thr 308 de-phosphorylation is performed by PP2A holoenzymes 36,37 . ERK activity is pivotal in Ab-mediated ERBB2 signaling, as it plays a dual role: it promotes the recruitment of the ERK/CyPA/phospho-AKT complex onto ERBB2, leading to AKT de-phosphorylation, and the full activation of the receptor C-terminal domain, through the phosphorylation of the Thr 701 residue in the juxtamembrane domain, identified by mass spectrometry (see model in Fig. 7).
The role of phospho-ERK in the full activation of ERBB2 by Abs is highlighted by the observation that the use of MEK inhibitors reduces the phosphorylation levels of the ERBB2 Tyr 1248 residue, which is required for ERBB2 signaling, and its transforming activity 2,46 . Noteworthy, this positive feedback functions at odds with what has been previously reported. In particular, EGF has been shown to promote an ERK-dependent phosphorylation of EGFR Thr 693 , homologous to ERBB2 Thr 701 , which negatively regulates the Tyr phosphorylation of the receptor C-terminal domain 44,45,58 . Similarly, 12-O-Tetradecanoylphorbol-13-acetate (TPA)-induced ERK-dependent ERBB2 Thr 701 phosphorylation has been shown to abrogate ERBB2 phosphorylation on Tyr 1248 and Tyr 119643 . Noteworthy, EGF binding to EGFR activates both the MAPK and the AKT signaling pathways, whereas ERBB2targeted Abs have opposite effects on the two cascades, as the direct activation of ERK is propaedeutic to AKT inactivation. By contrast, TPA has been reported to stimulate ERK and inhibit AKT activity, by interfering with the heterodimerization between ERBB2 and ERBB3 43 . On the contrary, our results show that Abs function mainly on ERBB2 homodimers, thus providing a possible explanation for the discrepancies with the published observation. www.nature.com/scientificreports/ Furthermore, phospho-ERK renders ERBB2 accessible to CyPA and phospho-AKT upon Ab treatment. Here, we provide evidence for the ERK-dependent recruitment of CyPA to the membrane-localized ERBB2, by both biochemical (IP) and morphological (dSTORM super-resolution microscopy) approaches, suggesting that CyPA is a key molecular switch in ERBB2 downstream signaling. Ab-bound ERBB2 is phosphorylated by ERK and the kinase ability to phosphorylate its substrates has been shown to be sensitive to the conformation of the substrate backbone [(Ser/Thr)-Pro] 47 . Furthermore, a peptidyl-prolyl isomerase has been shown to target and isomerize the Pro residue next to the Ser/Thr residue phosphorylated by ERK 59 . Indeed, our results show that CyPA silencing significantly reduces the levels of phospho-ERBB2, thus supporting the involvement of CyPA in the regulation of ERBB2 phosphorylation and activation. Moreover, we identified ERBB2 Pro 702 as the binding site for CyPA (see model in Fig. 7). However, whether the isomerase activity of CyPA is involved in ERK-mediated receptor phosphorylation still remains an open issue.
In conclusion, we identified and characterized for the first time an ERBB2-specific downstream anti-oncogenic signaling induced by treatment with Ab, negatively modulating the AKT pro-survival pathway, and recognized CyPA as a key regulator of this process.
These data shed new light on a dichotomous role of ERBB2: pro-oncogenic in the heterodimeric context and anti-proliferative in the homodimeric form. This previously unrecognized role of ERBB2 as a negative modulator of other ERBBs activity opens new perspectives for the treatment of ERBB2-positive carcinomas.

Methods dSTORM super-resolution microscopy and pair cross-correlation. dSTORM super-resolution
imaging was performed on a Leica SR GSD 3D TIRF microscope equipped with 300 mW 532 nm and 642 nm lasers; a 30 mW 405 nm laser; a Leica HCX PL APO 160 ×/1.43 Oil CORR GSD objective; a quad-band filter set with 417 nm, 496 nm, 544 nm, 655 nm dichroic and 421-477 nm, 497-519 nm, 547-621 nm, 666-732 nm emission bands. Images were collected with an Andor iXon Ultra 897 EM-CCD camera over an area of 18 × 18 µm. For TZ, samples were immunostained for CyPA (Alexa 647) and ERBB2 (Alexa 568) as described above. During acquisition, samples were embedded in MEA-Glucose Oxidase imaging buffer [PBS, 10 mM β-mercaptoethylamine (MEA) pH 7.4, 10% glucose (w/v), 0.5 mg/ml glucose oxidase, 40 µg/ml catalase] freshly prepared before each acquisition. In order to minimize photobleaching, Alexa 647 was imaged first, followed by Alexa 568. Fluorophores were initially pumped into dark states by illumination with 642 nm or 532 nm lights at 80% laser power until single fluorophore blinking was observed (< 15 s). Subsequently, single molecules were imaged at 60% laser power with a frame rate of 140 frames per second until a total of 10 6 (CyPA) and 10 5 (ERBB2) events was reached, using a detection threshold of 20 photons and 40 photons for event recognition, respectively. For PZ, samples were immunostained for CyPA (Alexa 647) and ERBB2 (Alexa 555) as described previously. During acquisition, samples were embedded in dSTORM Super Resolution buffer (Abbelight). In order to minimize photobleaching, Alexa 647 was imaged first, followed by Alexa 555. Fluorophores were initially pumped into dark states by illumination with 642 nm or 532 nm lights at 40% and 60% laser power until single fluorophore blinking was observed (< 15 s), respectively. Subsequently, single molecules were imaged at 40% and 60% laser power with a frame rate of 140 frames per second until a total of 10 6 (CyPA) and 2 × 10 5 (ERBB2) events was reached, using a detection threshold of 40 photons and 60 photons for event recognition, respectively. The final images were calculated using a pixel size of 20 nm in the histogram mode.
CyPA recruitment on ERBB2 was assessed by performing pair cross-correlation analysis on 1 × 1 µm regions of interest (ROIs), as previously described 41 . Fab fragment preparation. TZ and PZ were enzymatically cleaved on immobilized papain for 10 h at 37 °C, according to Thermo Scientific Pierce Fab Preparation Kit instructions. Digestion was verified by SDSpage, and Fab fragments purified using high recovery regenerated cellulose membranes (Amicon Ultra centrifugal filter units) with 100 kDa (to remove uncleaved whole Ab) and 30 kDa (to remove the Fc portion) cut-offs. Fab concentration was assessed by Bradford assay (Bio-Rad).

Immunoprecipitation (IP). p-Thr308 AKT and CyPA interactors identification. SK-Br3 cells were either
left untreated, or serum-starved for 20 min at 37 °C either in the absence or in the presence of U0126 and treated with TZ for additional 20 min at 37 °C. Before cell lysis, cells were washed twice with r.t. PBS and treated with 1 mM DSP in PBS for 10 min at r.t. The cross-linker was quenched by incubation with 20 mM Tris for 5 min at r.t. Cells were washed twice with cold PBS, scraped in lysis buffer (containing 20 mM Hepes, 150 mM NaCl, 1% Triton-X100, 10% glycerol, and protease inhibitor cocktail, phosphatase inhibitor cocktail, 1 mM NaF, 1 mM sodium orthovanadate, 1 mM beta-glycerophosphate), incubated for 10 min on ice and sonicated. Cells were incubated for 45 min on ice, and centrifuged at 16,000g for 5 min at 4 °C to remove nuclei and membranes. Protein concentrations were determined (BCA protein assay). Meanwhile, antibody binding to protein G dynabeads was performed with the protocol described in 60 with slight modifications. The antibody-crosslinked beads (50 μl) were incubated overnight at 4 °C with lysates of SK-Br3 cells. The following day, after removing the supernatant (unbound) and washing with buffer (containing 20 mM Hepes, 300 mM NaCl, 1% Triton-X100, 10% glycerol, and protease inhibitor cocktail, phosphatase inhibitor cocktail, 1 mM NaF, 1 mM sodium orthovanadate, 1 mM beta-glycerophosphate) for 5 × 5 min, the immunoprecipitates were eluted with 30 μl of 2 × Laemmli buffer without reducing agent at 37 °C for 10 min. The eluate was transferred into a new tube, and samples were subjected to SDS-PAGE separation for Coomassie and mass spectrometry analysis. Alternatively, the eluate was transferred into a new tube, 10% β-mercaptoethanol added to cleave DSP, and samples were subjected to SDS-PAGE separation for Western blotting. Mass spectrometry analysis. p-Thr308 AKT interactors identification. The entire lanes were individually cut in nine bands and in-gel digested prior to mass spectrometry analysis. In particular, the bands, once excised from the gels, were de-stained, sequentially reduced, alkylated and digested overnight with sequencinggrade trypsin, as previously described 61 . Aliquots of the sample containing tryptic peptides were desalted using StageTip C18 (Thermo Scientific) and analysed by nLC-MS/MS using an LTQ-Orbitrap (Thermo Scientific, Bremen, Germany) equipped with a nano-electrospray ion source (Proxeon Biosystems) and an nHPLC Easy LC (Proxeon Biosystems). Peptide separations occurred on a homemade (75 µm i.d., 25 cm long) reverse phase silica capillary column, packed with 3-µm ReproSil-Pur 120 C18-AQ (Dr. Maisch GmbH, Germany). The nLC-MS/MS was performed with the protocol described in 62 with slight modifications. A gradient of eluents A (distilled water with 2% v/v acetonitrile and 0.5% v/v acetic acid) and B (acetonitrile and 20% v/v distilled water with 0.5% v/v acetic acid) was used to achieve separation (150 nl/min flow rate), from 8% B to 50% B in 50 min. Full scan spectra were acquired with the lock-mass option, resolution set to 60,000 and mass range from m/z 300 to 1750 Da. The ten most intense doubly and triply charged ions were selected and fragmented in the ion trap. ERBB2 post translational modifications (PTM) assessment. The band, corresponding to ERBB2 protein, was excised from the gel, de-stained, sequentially reduced, alkylated with iodoacetamide and digested overnight with sequencing-grade trypsin, as above described. The peptides were extracted from the bands and subdigested overnight with endoproteinase Glu-C (Roche Diagnostics). The phosphopeptides were enriched on TiO 2 resin and subsequently desalted using POROS Oligo R3 reversed-phase resin, as previously reported 63 . The flow-through peptides were also desalted using a Stage Tip C18 (Thermo Scientific). All the peptide mixtures were analysed by nLC-MS/MS using a Q-Exactive mass spectrometer (Thermo Scientific, Bremen, Germany) equipped with a nano-electrospray ion source (Proxeon Biosystems) and a nUPLC Easy-nLC 1000 (Proxeon Biosystems) as described in 64 with slight modifications. Peptide separations occurred on a homemade (75 µm i.d., 12 cm long) reverse phase silica capillary column, packed with 1.9-µm ReproSil-Pur 120 C18-AQ (Dr. Maisch GmbH, Germany). A gradient of eluents A (distilled water with 0.1% v/v formic acid) and B (acetonitrile with 0.1% v/v formic acid) was used to achieve separation (300 nl/min flow rate), from 0% B to 45% B in 45 min. Full scan spectra were acquired with the lock-mass option, resolution set to 70,000 and mass range from m/z 300 to 2000 Da. The ten most intense doubly and triply charged ions were selected and fragmented in the ion trap 64 . The experiments were performed in technical duplicates and biological duplicates. In order to quantify the amount of phosphorylation on each site, the raw data were loaded into the MaxQuant software version 1.5.2.8. Searches were performed against the UniProt_Human Complete Proteome_cp_hum_20170315 (92,919 sequences; 36,868,442 residues), with trypsin + Glu-C as proteolytic enzymes, 2-missed cleavages allowed, carbamidomethylation on cysteine as fixed modification, protein N-terminus-acetylation, methionine oxidation and phosphorylation on Ser/Thr/Tyr as variable modifications. Mass tolerance was set to 5 ppm and 20 ppm for precursor and fragment ions, respectively. The intensities of precursors were used for the label-free protein quantification. Peptides and proteins were accepted with a FDR less than 1%, two minimum peptides per protein with one unique.
ERBB2 dimerization assessment. The high molecular weight bands containing ERBB2 protein, as indicated by WB analysis, were excised from the gel, de-stained, sequentially reduced, alkylated with iodoacetamide and digested overnight with sequencing-grade trypsin, as previously described 61,65 . The peptides were extracted from the bands, desalted using a Stage Tip C18 (Thermo Scientific) and analysed by nLC-MS/MS using a Q-Exactive mass spectrometer (Thermo Scientific, Bremen, Germany) equipped with a nano-electrospray ion source (Proxeon Biosystems) and an UPLC Easy-nLC 1000 (Proxeon Biosystems), as above reported. The experiments were performed in technical triplicates and biological duplicates. In order to quantify the amount of proteins in the high-molecular weight bands, the raw data were loaded into the MaxQuant software version 1.5.2.8. Searches were performed against the UniProt_Human Complete Proteome_cp_hum_20180228 (93,786 sequences; 37,179,059 residues), with the same parameters reported above plus the addition of 145.0198 Da on lysine and protein N-terminus (CAMthiopropanoyl modification) due to the cross-linking as variable modifications. Mass tolerance was set to 5 ppm and 20 ppm for precursor and fragment ions, respectively. The intensities of precursors were used for the label-free protein quantification. Peptides and proteins were accepted with an FDR less than 1%, two minimum peptides per protein with one unique. In order to normalize the amount of proteins, the rabbit anti-human antibody, which was equally added to the samples during the IP, was also quantified. In this case, when using the MaxQuant software for the label free analysis of the bands corresponding to the Ab ( Supplementary Fig. 4) searches were performed against the UniProt_Rabbit Complete Proteome_cp_rab-bit_20180228. All the other parameters were as above reported. To quantify the relative abundance of heterodimers vs homodimers, we normalized the intensities of EGFR (to estimate heterodimers) and ERBB2 (to estimate homodimers) peptides (per μg of immunoprecipitate) for the equivalent peptides obtained per μg of control SK-Br3 lysate.
Ethics statement. All experimental protocols were approved by the Institutional Ethical Board of the European Institute of Oncology (IEO, Milan, Italy); all patients recruited for this study signed informed consent under these ethics, and all methods were carried out in accordance with relevant guidelines and regulations.