Introduction

The epidermal growth factor receptor (EGFR or HER1/ErbB1) is the founding member of the human EGFR tyrosine kinase family (HER2/ErbB2/Neu, HER3/ErbB3, and HER4/ErbB4)1. EGFR plays a fundamental signalling role in cell growth and is frequently hyper-activated in human cancers via mutation and/or overexpression2. This driving role in malignancy has made EGFR a key target for anti-cancer therapy3,4.

An EGFR monomer consists of an N-terminal ligand-binding extracellular module (ECM) connected to an intracellular module (ICM) by a single-pass transmembrane (TM) helix (Fig. 1a). The ECM comprises four domains (DI–DIV) and adopts a tethered conformation via an interaction between DII and DIV5. The ICM includes a short juxtamembrane (JM) segment, a tyrosine kinase domain (TKD) and a disordered carboxy-terminal region, locus of the key tyrosine phosphorylation sites6,7. Ligand binding stabilises the extended conformation of the ECM promoting the formation of back-to-back dimers8,9 (Fig. 1a). Subsequent EGFR signalling across the plasma membrane depends on an allosteric interaction between an activator and receiver kinase effected through an asymmetric TKD (aTKD) dimer10. Signal transduction also requires ligand-bound EGFR oligomers11,12 formed by face-to-face interactions between back-to-back dimers12 (Fig. 1b).

Evidence has accumulated over the years for ligand-free EGFR dimers and oligomers (see e.g. refs. 13,14,15,16,17,18,19,20,21). However, the mechanisms by which ligand-independent activation of non-monomers is prevented remain unclear. Nonetheless, it is widely believed that autoinhibition is related to the adoption of an inactive symmetric TKD (sTKD) dimer revealed by X-ray structures of EGFR TKDs bearing the V924R (or V948R) and I682Q mutations at the C-lobe and N-lobe, which inhibit aTKD dimer formation (PDB ID 3GT8 (ref. 22), 2GS7 (ref. 10), and 5CNN (ref. 6)). The sTKD was putatively associated to a speculative side-to-side ECM tethered dimer20 (Fig. 1c), presumably because this would provide a fail-safe approach to autoinhibition. Alternatively, molecular dynamics (MD) simulations23 suggested that the sTKD dimer is coupled via a C-crossing TM domain dimer to a ligand-free back-to-back dimer analogous to the X-ray structure of the Drosophila ECM dimer24 and a model based on SAXS data from Caenorhabditis elegans EGFR16 (Fig. 1d). In this back-to-back dimer, which resembles the ligand-bound dimer, the autoinhibitory heavy lifting would be done by the sTKD dimer alone25. A flexible ECM dimer held by DIV–DIV contacts by the plasma membrane was also suggested by electron microscopy (EM) images of purified, near-full-length Δ998-EGFR26,27 (Fig. 1e). This stalk-to-stalk dimer is promoted by kinase-mediated interactions in response to the binding of type I tyrosine kinase inhibitors (TKIs), which reversibly bind the ATP-binding pocket stabilising the aTKD dimer26 inhibiting C-terminal phosphorylation. The challenge is the lack of high-resolution methods on cells, which has made it impossible so far to obtain evidence on the architecture of non-monomer ligand-free species.

Here we exploit the nanoscale resolution of fluorophore localisation imaging with photobleaching (FLImP)12,28,29,30 in a multi-technique study including two-colour single particle tracking (SPT)31, fluorescence resonance energy transfer (FRET)32,33, and MD simulations23,34 to investigate ligand-free non-monomer species on cells. The results reveal structural insights on the inactive species and on how mutations circumvent the autoinhibition of the basal state.

Results

Monomers and dimers and oligomers populate the cell surface

We used FLImP on Chinese Hamster Ovary (CHO) cells expressing ~105 wild-type EGFR (wtEGFR) copies/cell, maintained by an inducible Tet-ON promoter35 (Fig. 2a). FLImP measures pairwise lateral separations between identical fluorophores with ~5–7 nm resolution, returning normalised distributions in the 0–60 nm range, which is ideal to investigate EGFR dimers and oligomers12. We conjugated CF640R fluorophore to the single cysteine of an anti-EGFR Affibody which binds DIII of EGFR’s ECM without activating the receptor36 and accumulated a FLImP histogram of DIII–DIII separations from ligand-free wtEGFR complexes. Bayesian parameter estimation with Bayesian Information Criterion (BIC)12 objectively decomposed the FLImP histogram into a sum of five peak components at 5, 13, 22, 30, and 46 nm (Fig. 2b). An EGFR dimer has two DIIIs and must give rise to a unique DIII–DIII separation. This must be <~20 nm, given the <3.7 nm size of Affibody probes37 and the ~5–8 nm size of the EGFR monomer5,8,9. Multiple separations >20 nm and periodic DIII–DIII components therefore report ligand-free wtEGFR oligomers larger than dimers and stoichiometric extracellular interactions, respectively. These results agree with previous data from T47D cells28, which express 10-fold fewer EGFR copies/cell, indicating that the ligand-free structures formed by EGFR are independent of receptor expression.

Being a pairwise measurement, FLImP is insensitive to monomers. To estimate the relative proportions of monomers and other species, we used model-free photobleaching image correlation spectroscopy (pbICS)38 analysis, which we applied to CHO cells expressing wtEGFR labelled with an Alexa 488-Affibody fluorescent derivative (Fig. 2c). The results suggest that ~65% of receptors are incorporated into dimers and larger oligomers, consistent with previous estimates18,21,39.

A structural model of extracellular ligand-free dimers

The nanoscale proximity between receptors in oligomers translates into a local surface density of ~5000 receptors/μm2. This is above that on cells overexpressing >2 × 106 EGFR copies/cell (~2000 receptors/μm2), where significant phosphorylation occurs in the absence of ligand40. In contrast, CHO cells expressing 20-fold fewer receptors have low basal phosphorylation36 despite ~65% of receptors forming non-monomer complexes. This suggests that ligand-independent phosphorylation is inhibited in ligand-free complexes, leaving two questions: what is their architecture and how is autoinhibition achieved?

Among possibilities, we considered cyclic shapes41 and polymer chains12 (Fig. 2d). Both architectures, when probed with a 1:1 binder like Affibody, predict a decreasing “ladder-like” DIII–DIII separation intensity consistent with FLImP results. However, the longest DIII–DIII separation (46 nm) (Fig. 2b) and largest oligomer size (octamers) (Fig. 2c) are more consistent with polymer chains. To test this further, we conjectured that, when probed with a CF640R-EGF derivative, which only binds the polymer ends12, each polymer chain would contribute a single separation (Fig. 2d, right). Thus, the FLImP separation density would move from short to long separations. This effect should be more noticeable the fewer de novo complexes form after EGF binding. A screen of EGF-induced particle colocalisation events revealed that I942E-EGFR, a kinase domain mutation that allows strong EGF-induced phosphorylation10, forms very few de novo EGF-induced complexes (Supplementary Fig. 4). FLImP data from I942E-EGFR display a decreasing ladder-like distribution of DIII–DIII separations when probed with CF640R-Affibody (Fig. 2e) and a transfer of density to longer components when probed with CF640R-EGF (Fig. 2f). These results are consistent with polymer chain architectures.

To attempt to construct a structural model of a polymer chain using long-timescale MD simulations, we searched for crystal structures of the ECM monomer in which lattice contacts might reveal previously unidentified oligomer interfaces. The structures of EGFR co-crystallised with EgA1 (PDB ID 4KRO42) and 9G8 (PDB ID 4KRP42) verified this criterion. (EgA1 and 9G8 are nanobodies (NBs) that bind DIII and stabilise the DII–DIV tether42.) Figure 3a shows the monomer contacts in 4KRP. We hypothesised that the tethered conformation and lack of two-fold symmetry of this model, reminiscent of an X-ray structure of HER2 co-crystallised with a designed ankyrin repeat protein43, are reasonable initial points because both counteract the formation of signalling dimers.

We then carried out two MD simulations (13.8 and 20 µs) starting from the asymmetric dimer seen in the crystal packing of 4KRP, after removing 9G8-NB and adding the TM helix and the lipid bilayer (Fig. 3b). In both simulations the ECM dimer arrived at a head-to-head dimer conformation within 1.5 µs and remained in this conformation to the end. This dimer conformation is similar to the crystal dimer conformation in that it remains asymmetric and open-ended and maintains the trans interaction between the DIII of one protomer and the DIV of the other, but it bears a trans interaction between DI and DII. Moreover, the tethered conformation of the crystal structure was consistently unstable, as in both simulations DI and DIII in both protomers gained stable cis interaction with one another, giving rise to a conformation similar to that in the proposed back-to-back inactive dimer23 in terms of DI, DII, and DIII. Intriguingly, the ECMs of neither of the two protomers in the head-to-head conformation bear close contact with the membrane. Encouragingly, the bulk of the amino acids involved in the head-to-head interface (6–238) is missing in the constitutively active variant EGFRvIII prevalent in glioblastoma (6–273)44. If amino acids 6–273 are deleted post-simulation in the model, the monomers break apart (Fig. 3c), suggesting biological relevance.

We next added the 9G8-NBs to the starting ECM dimer and similarly simulated a 9G8-bound head-to-head dimer up to 20 µs, starting from the conformation seen in PDB 4KRP with the TM helices and the lipid bilayer added (Fig. 3d). Consistent with the two simulations of the “naked” dimer where 9G8-NB was removed, the asymmetry and the open-endedness of the dimer remained, the DIII–DIV trans interaction of the ECM dimer was stable, and the DI–DII trans interaction was developed. The DI–DIII cis interaction again developed in both protomers, reproducing the previously reported monomer conformation23.

The head-to-head dimer can be extended into a higher-order oligomer by incorporating additional protomers and repeating the head-to-head interaction (Fig. 4a). pbICS data reveal that ~25% of species are dimers, ~10% trimers, and ~5% tetramers and above (Fig. 4b). The fraction of each species as a function of oligomer size looks approximately exponential consistent with a step-wise polymerisation process. Curvature of the polymer chain in the plane of the membrane would arise because the DIII–DIII separation between nearest neighbours is larger than DI–DI (Fig. 3b). One can imagine such curved geometry would facilitate receptor packing around the mouths of vesicles like, for example, caveolae, which have a ~50 nm diameter45. The predicted DIII–DIII separations (Fig. 4a, blue) are encouragingly consistent with the experimentally measured separations (Fig. 2b; Supplementary Fig. 1). Moreover, the largest separation resolved (46 nm) corresponds to the prediction for the largest octamer species detected by pbICS.

To further assess the possibility that this model represents the architecture of non-monomer complexes, we conjectured that ΔC-EGFR, which mirrors the amino acid sequence used in the simulated head-to-head dimer model (no ICM), should form polymers consistent with the model. To test this, we probed DIII–DIII separations on CHO cells that express ΔC-EGFR. The separations resolved are consistent with the head-to-head polymerisation model predictions (Fig. 4c, Supplementary Fig. 7).

The model also predicts shorter DI–DI separations (Fig. 4a, green). We tested this using a CF640R derivative of an anti-EGFR EgB4 nanobody (EgB4-NB) that specifically binds DI of EGFR without inducing activation42,46 (Supplementary Fig. 6). The distribution of DI–DI separations is significantly compressed towards shorter values (Fig. 4d, Supplementary Figs. 8 and 9a–c). Being smaller, fewer DI–DI separations are resolved, and those resolved are consistent with model predictions.

Predicted distances from DI and DIII to the membrane were tested by FRET32 (Fig. 4e). In parallel experiments we labelled DI or DIII of wtEGFR with Alexa 488 derivatives of EgB4-NB or Affibody (FRET-donors) in the presence and absence of bound 9G8-NB. The plasma membrane was treated with dialkylcarbocyanine probe DiIC18(5) (DiI) (FRET-acceptor). As shown in Fig. 4e, DI is on average closer to the bilayer than DIII when 9G8-NB is bound, while in its absence DI and DIII separations are similar. The FRET measurement reports the average of monomers and non-monomers; however, because these are predicted to have similar extracellular conformations23, these FRET results are consistent with the model predictions. In conclusion, our FLImP and FRET data strongly favour the head-to-head interface as the building block of ligand-free polymer chains.

We reasoned that extracellular and intracellular interactions should together assemble more stable complexes than those assembled by extracellular interactions alone. Given this, the contribution of the ICMs to oligomerisation could be ascertained by comparing the stability of the complexes formed by ΔC-EGFR (no ICM) and the full-length wtEGFR and I942E-EGFR. This comparison is possible because ΔC-EGFR, wtEGFR, and I942E-EGFR form indistinguishable polymer chains (Figs. 2b, e, 4c, Supplementary Fig. 9d–f). For this investigation we used two-colour SPT, a method previously used to report the incidence of pairwise receptor particle interactions (colocalisation fraction) and their duration (τON)12,14. Receptors were labelled with a 1:1 mixture of Alexa 488-Affibody and CF640R-Affibody. Results show that ΔC-EGFR forms the most complexes (Fig. 4f), which have an indistinguishable duration to those formed by wtEGFR and I942E-EGFR (Fig. 4g). These results argue against a significant contribution from intracellular interactions to oligomerisation (Fig. 4h).

The sizeable separation between the N-termini of the TM helices in the head-to-head dimer predicts that the stabilisation of the aTKD dimer, by enforcing the N-crossing TM dimer40, would break the head-to-head interaction. To test this, we stabilised the aTKD dimer by treating wtEGFR-expressing CHO cells with erlotinib26, a type I TKI47. The four DIII–DIII separation components resolved (Fig. 5a) are at positions consistent with four of the five resolved in the absence of erlotinib (Fig. 2b). We also determined the proportion of measurements consistent with each of the mean positions, an analysis that revealed a dominant 5 nm component and a significant decrease in separation density at longer positions when compared with naked wtEGFR (Fig. 5b). This result correlates aTKD dimer formation with oligomer loss.

To examine if oligomers split when the aTKD dimer is stabilised in the absence of TKI, we treated CHO cells expressing wtEGFR with methyl-β-cyclodextrin (MβCD), which depletes cholesterol from the plasma membrane and activates EGFR in a ligand-independent manner48 (Fig. 5c). The associated increase in aTKD dimers returned results remarkably similar to those found upon erlotinib treatment, namely DIII–DIII separations with a dominant peak at ~4 nm and significantly fewer oligomers (Fig. 5b, d). Together these results support the notion that the stabilisation of the aTKD dimer, either in the presence or absence of erlotinib, disrupts oligomer formation.

Kinase-mediated stalk-to-stalk and back-to-back dimers

Besides disrupting oligomers, the dominance of one DIII–DIII separation component in the FLImP distributions of Fig. 5a, d indicates that the stabilisation of the aTKD dimer promotes an ECM dimer. This is supported by the increased receptor–receptor colocalisation following erlotinib binding, as shown by SPT (p < 0.01, Kolmogorov–Smirnov test, two-tailed) (Fig. 5e). The formation of erlotinib-bound aTKD dimers is also consistent with the increased duration of the interactions between receptors when compared with naked wtEGFR where only extracellular contacts are present (p < 0.01, Kolmogorov–Smirnov test, two-tailed) (Fig. 5f). The resolution of FLImP (~5–7 nm) is, however, insufficient to distinguish between the head-to-head dimer (~7 nm DIII–DIII separation) and the ECM dimer induced by erlotinib (~5 nm DIII–DIII separation). We nevertheless reasoned that these dimers are likely to be different because N-crossing TM dimers would disrupt head-to-head interfaces.

To investigate the ECM dimer linked to the aTKD dimer, we considered the stalk-to-stalk structure detected by EM26. This dimer is upright, not-extended, and held by close C-terminal DIV–DIV contacts that afford linking with the N-crossing TM dimer and conformational flexibility (Fig. 1e). To investigate the presence of DIV–DIV contacts, we examined IIIV/KKRE-EGFR, in which mutations I545K, I556K, I562R, and V592E (IIIV/KKRE-EGFR) replace a hydrophophic surface of DIV with charged amino acids11 (Fig. 5g). As shown in Fig. 5h, the IIIV/KKRE-EGFR mutations should not disrupt head-to-head interactions. Consistent with this, FLImP showed IIIV/KKRE-EGFR forms as many head-to-head dimers and oligomers as wtEGFR (Fig. 5b, i). The IIIV/KKRE-EGFR mutations are however poised to disrupt DIV–DIV contacts. Thus, if formation of the aTKD dimer is facilitated by DIV–DIV contacts in the ECM, then cholesterol depletion should induce less phosphorylation in IIIV/KKRE-EGFR. The absence of increased phosphorylation in cholesterol-depleted IIIV/KKRE-EGFR-expressing cells is consistent with the requirement for DIV–DIV contacts for stabilisation of a functional aTKD dimer (Fig. 5c).

We next used FRET to measure the separations from DI and DIII to the membrane in erlotinib-bound wtEGFR. Assuming an upright orientation, the similar DI- and DIII-membrane separations found are consistent with a non-extended conformation, as that predicted by the stalk-to-stalk dimer (Fig. 5j). In turn, error analysis of the FRET data confirms a conformation different from that of naked wtEGFR (Supplementary Fig. 10a, c, Supplementary Table 4). To investigate conformational flexibility, we compared the binding of 9G8-NB to erlotinib-bound wtEGFR and naked wtEGFR. We found that 9G8-NB, which stabilises the tether42, displays greater binding to erlotinib-bound wtEGFR (Fig. 5k), consistent with a greater ability to adopt a tethered conformation. This would be afforded by the increased flexibility of the stalk-to-stalk dimer.

We next reasoned that, if the stalk-to-stalk and aTKD dimers are not just linked across the membrane, but conformationally coupled, the stabilisation of the sTKD dimer6,10,22 might promote an ECM dimer different to the stalk-to-stalk dimer, and possibly similar to the ligand-free back-to-back architecture proposed by MD simulations23 (Fig. 1d). To test this possibility, we examined L680N-EGFR, in which a mutation in the N-lobe of the kinase domain hinders the aTKD dimer interface10. L680N is analogous to I682Q, which gave rise to the 5CNN6 crystallographic sTKD structure. As shown in Fig. 5l, the dominance of a different DIII–DIII separation in L680N-EGFR (in this case 13 nm) suggests that the L680N mutation promotes inside-out an extracellular dimer different to the stalk-to-stalk dimer. Further evidence for L680N-induced dimerisation is provided by the increased number and duration of the interactions holding L680N-EGFR dimers together (p < 0.01, Kolmogorov–Smirnov test, two-tailed) (Fig. 5e, f). The distribution of DIII–DIII separations also displays fewer separations (>20 nm) than in wtEGFR (Fig. 5b), suggesting that formation of the sTKD dimer is incompatible with the head-to-head interface.

FRET results from L680N-EGFR showed a greater separation from DI to the membrane than from DIII (Fig. 5j). These results argue that the dimers formed by L680N-EGFR are upright and adopt an extended conformation49. Given that the dimension of the Affibody is 3.7 nm37, the 13 nm DIII–DIII separation is within range of the predictions of the ligand-free back-to-back dimer, in which the distance between binding sites is ~7 nm. In turn, the lower binding of 9G8-NB to L680N-EGFR than to wtEGFR bound to erlotinib (Fig. 5k) is consistent with the smaller conformational flexibility expected in the back-to-back dimer. The data therefore argue that the aTKD (sTKD) dimer is structurally coupled across the plasma with the previously proposed stalk-to-stalk (back-to-back) dimers.

Inside-out signals regulate cancer mutant dimer geometry

To investigate changes in ligand-free EGFR structure accompanying dysregulated signalling, we examined L834R-EGFR and T766M-EGFR (or L858R and T790M in numbering that include the 24 amino acids signal peptide). These two mutations are prevalent in non-small cell lung cancer (NSCLC) and located within exons 18–21 in the vicinity of the ATP-binding site (Fig. 6a). L834R-mutated NSCLC tumours respond to first-generation TKIs (like erlotinib)50. T766M increases the affinity for ATP and confers resistance to first- and second-generation TKIs51. Western blots confirmed the constitutively phosphorylated status of these mutants (Supplementary Fig. 17). We collected FLImP data and compared the results with wtEGFR, ΔC-EGFR, wtEGFR + erlotinib and wtEGFR + MβCD, and L680N-EGFR (Fig. 6b–e). As positive control for kinase-mediated dimerisation we used Δ973-EGFR, in which truncation of the C-terminus abolishes the autoinhibitory interaction with its own kinase6. As negative control we used Δ698-EGFR, which lacks the kinase domain.

From the FLImP distributions we quantified the DIII–DIII separations consistent with 5 nm (green), 13 nm (yellow), <15 nm (pink) and >20 nm (blue) (Figs. 6b–e). Consistent with results above (Fig. 5), we found that oligomers (>20 nm) are reliably more abundant when one would not expect kinase-mediated dimerisation (Fig. 6b). Indeed, Fig. 6c shows the FLImP distributions of the three conditions displaying more oligomers. Unsurprisingly, with wtEGFR (low phosphorylation) and ΔC-EGFR (no ICM), we found Δ698-EGFR (no TKD). In contrast, oligomers were consistently depleted in conditions where one would expect kinase-mediated dimerisation. Among these conditions, Fig. 6d shows the distributions for wtEGFR upon erlotinib or MβCD treatments, in which the formation of the aTKD dimer promotes a dominant ~5 nm component that reports the formation of stalk-to-stalk ECM dimer at the expense of head-to-head interfaces, as shown in Fig. 5. In this group, we also found L834R-EGFR, in which the L834R mutation stabilises the aTKD dimer interface by suppressing the intrinsic disorder in the N-lobe kinase dimerisation interface52. According to conformational coupling, the dominant 5 nm FLImP peak and fewer oligomers found in L834R-EGFR would suggest that the stabilisation of the aTKD dimer also promotes the stalk-to-stalk dimer in this cancer mutant. We used FRET to validate this notion (Supplementary Fig. 10e). (Later we also discuss the similarity of the results for L834R-EGFR and wtEGFR under erlotinib and MβCD treatments with additional data (Fig. 7).) Lastly, Fig. 6e shows distributions with a sizeable 13 nm component and a depleted oligomer population. Consistent with conformational coupling, inhibition of the aTKD dimer in L680N-EGFR is reflected by the relatively lower 5 nm component in this group. In contrast, the sizes of the 5 and 13 nm peaks, resolved by BIC analysis (Supplementary Fig. 11c, d), are comparable in T766M-EGFR and Δ973-EGFR, whose trans activation depends on the formation of the aTKD dimer51. We were reassured to find that deletion of the C-terminus (Δ973-EGFR) resulted in oligomer loss and the promotion of dimers (Fig. 6b–e). This is consistent with the role of the C-terminus in preventing kinase-mediated dimerisation53.

T766M-EGFR can form symmetric tyrosine kinase dimers

We were surprised to find significant numbers of 13 nm separations in the constitutively phosphorylated T766M-EGFR (Fig. 6e). If, as shown by FLImP and FRET results for L680N-EGFR, this peak reports extended back-to-back dimers, and, as proposed by MD simulations23, this dimer is structurally coupled to the sTKD dimer, then, counterintuitively, our results predict that T766M-EGFR forms significant numbers of sTKD dimers. To investigate this unexpected prediction, we carried out MD simulations of the sTKD dimer and reanalysed fully converged free-energy landscapes from parallel tempering metadynamics (PTmetaD) simulations, in the well-tempered ensemble for wtEGFR and T766M-EGFR, as in Sutto et al.34.

We reweighted the MD trajectories and projected the free-energy surface (FES) onto the helical content of the αC helix and the contact map corresponding to the active (open) state. As expected, the FES for wtEGFR shows a single global minimum that corresponds to the intact αC helix with the A-loop in the Src-like inactive conformation (Fig. 6f, left). There is also a higher-energy minimum populated with structures with a disordered αC helix. In the case of T766M mutant, this minimum is broader and shifted towards lower values that correspond to higher disorder of the αC helix (Fig. 6f, middle). The global minimum with an intact αC helix is also present and both minima are populated with structures with an active A-loop conformation.

These results indicate that the T766M mutation has two effects. On the one hand, it is clear from the converged FES of the monomer that it destabilises the aTKD dimer interface by affecting the stability of the αC helix. On the other, the long MD simulations of the sTKD dimer in wtEGFR and T766M-EGFR show more stability for the single mutant (as indicated by root-mean-square fluctuations, Supplementary Fig. 19a) and suggest that T766M stabilises the sTKD dimer through a network of interactions (Fig. 6f (right), Supplementary Fig. 19b). In particular, as shown in Supplementary Fig. 19c, we see that Q767 engages in interactions with the interface residues (such as E981) located below the AP-2 helix that refolds in this dimer (it is disordered in the monomer). Further evidence to this observation comes from Jura et al.22 where it is shown the E981R mutation facilitates EGFR phosphorylation. This implies that the inactive sTKD cannot form.

The aTKD/sTKD dimer equilibrium depends on the C-terminus

The finding of significant numbers of sTKD dimers in the constitutively active T766M-EGFR, but not in wtEGFR (Fig. 6f), argues against a main autoinhibitory role for the sTKD dimer25. Alternatively, in the absence of structural impediments, kinase-mediated dimerisation might inherently give rise to both active aTKD and inactive sTKD dimer configurations (Fig. 7a). Recently, solid state NMR showed that EGFR structure is dynamic54, suggesting that different conformations may exchange in time. This is supported here by three observations: firstly, in L834R-EGFR, where the aTKD dimer interface is super-stable52, there are few sTKD dimers (Fig. 6d); secondly, in L680N-EGFR, in which the aTKD dimer is inhibited, the sTKD dimer population is boosted (Fig. 6e); and thirdly, in the absence of constraints, comparable numbers of aTKD and sTKD dimers form (like in T766M-EGFR) (Fig. 6e). Indeed, K721A-EGFR, in which the lysine 721 to alanine mutation in the ATP-binding pocket renders the kinase dead55, is another example in which both aTKD and sTKD dimers form at the expense of oligomers (Fig. 7b). This result highlights the critical role played by the kinase core in regulating distal dimerisation interfaces56 and the challenges associated to intervening with TKIs.

To test the equilibrium hypothesis we examined the effect of mutations on the kinetics of particle interactions. Crucially, we can interpret SPT results knowing the stoichiometry and architecture of the different conditions (Figs. 26). Results reveal that the kinase-mediated dimers induced by the L834R, L680N, and T766M mutations are more abundant (Fig. 7c) and longer lived (Fig. 7d) than the extracellular head-to-head interfaces formed by wtEGFR (p < 0.01). T766M-EGFR and L680N-EGFR dimers are similarly stable (Fig. 7c), underscoring the preference of T766M-EGFR for the sTKD dimer.

Also shown in Fig. 7c, d, the dimers induced by C-terminal truncation in Δ973-EGFR are fewer than those induced by the L834R, L680N, and T766M mutations. This is consistent with the pivotal role played by the C-terminus in the stabilisation of active and inactive kinase dimers6,22. In turn, the low stability of the dimers formed by K721A-EGFR is consistent with the key role played in stabilising dimer interfaces by the opening and closing of the ATP-binding site during ATP hydrolysis56.

Using coarse-grained metaDynamics free-energy calculations, Lelimousin et al.57 proposed that a change in bilayer thickness, such as that induced by cholesterol depletion58, promotes a transition from less stable TM domain dimers, including the C-crossing configuration associated to the sTKD23, to the most stable N-crossing TM dimer. Given this, we speculated that if receptor populations bearing sTKD and aTKD dimers are in equilibrium, we could alter the kinetics by depleting plasma membrane cholesterol. Indeed, MβCD promotes in T766M-EGFR an increase in the number of dimers48 (Fig. 7c, e) in which the duration of the interactions is consistent with the formation of active aTKD dimers. The latter is borne by the striking similarity between T766M-EGFR + MβCD with wtEGFR + MβCD and L834R-EGFR (Fig. 7f). Note these dimers are phosphorylated, and hence longer lived than erlotinib-bound wtEGFR dimers (Fig. 5f). As the predicted equilibrium would dictate, cholesterol depletion does not increase the number of dimers formed by L834R-EGFR, which is depleted of sTKD dimers (Fig. 6d), or L680N-EGFR, where the N-lobe mutation inhibits the aTKD dimer (Fig. 7d, f). These results are therefore not only consistent with a shift in the equilibrium from sTKD to aTKD dimer configuration, but also with the key role predicted for motion correlations between the TM, JM and kinase domains in the stabilisation of the aTKD dimer56. Interestingly, the shift in equilibrium from sTKD to aTKD dimer does not take place in Δ973-EGFR, but it is strongly present in K721A-EGFR, showing that the equilibrium between kinase dimer configurations depends on the C-terminus and not just on phosphorylation.

Discussion

In summary, the body of data reported here describes the architectures of three EGFR non-monomer ligand-free species populating the cellular basal state (Fig. 8). We propose a structural model of a polymer chain assembled through a previously undiscussed extracellular head-to-head interaction. Investigations on intracellular mutants and treatments found no evidence of intracellular interactions in the head-to-head polymers. The proposed role of the head-to-head interaction is therefore two-fold. On the one hand, it drives oligomer assembly, thereby bringing receptors together and priming them for a quick response to growth factor. On the other, it maintains TM helices sufficiently apart to oppose the formation of kinase-mediated dimers. Indeed, kinase-mediated dimerisation breaks the head-to-head interaction, suggesting the latter is important for autoinhibition.

The curved architecture of head-to-head polymers provides an elegant solution to the autoinhibition conundrum. Achieved via homo-extracellular interactions, such curved architecture delivers the largest possible consistent separation between TM helices in multimers, while assembling polymers commensurate with the diameter around relevant plasma membrane vesicles45. The head-to-head interaction averts the two-fold symmetry advantageous to kinase-mediated interactions, and, by holding receptors via DI, also averts highly dynamic monomer fluctuations that could lead to fortuitous dimerisation54. The bulk of the amino acids involved in the head-to-head interaction are deleted in EGFRvIII44, providing a rationale for the transformation potential of this mutant59.

Conformational coupling was demonstrated in the ligand-bound state40. Here, in the ligand-free state, it is supported by the finding that kinase-mediated dimerisation and its requirement to form TM helix dimers, breaks head-to-head dimers and polymers inducing across the plasma membrane the formation of two dimer species, both distinct from the head-to-head dimer, the architectures of which were previously reported by long-timescale MD simulations23 and EM experiments26. One kinase-mediated extracellular dimer is upright and extended, and consistent with a back-to-back conformation23. Results for L680N-EGFR together with MD simulations for the T766M mutant exploiting fully converged free-energy landscapes validate the prediction that the extended dimer is structurally coupled to the sTKD dimer23. The second is a stalk-to-stalk dimer formed by DIV–DIV contacts and structurally coupled to the aTKD dimer26. Our results suggest that kinase-mediated dimer conformations coexist in equilibrium in constitutively active and in inactive receptor dimers, regulated by C-terminal interactions54,56.

One key question remains. If the inactive sTKD dimer readily forms in constitutively active receptors, but not significantly in wtEGFR, what is its role if not autoinhibitory? It is tempting to speculate that the ligand-free EGFR structures we propose may be applicable to other HER family members. Indeed, recent results60 show that, by stabilising a particular HER2 conformer, the TKI lapatinib drives the formation of a HER2–HER3 kinase domain heterodimer via a symmetric interface. The lapatinib-induced HER2–HER3 heterodimer is primed to respond to neuregulin, leading to signal amplification. It will be interesting to ascertain the possibility of an analogous role for the sTKD dimer.

Methods

Cell culture

All reagents unless otherwise stated were from Invitrogen, UK. CHO cells expressing wtEGFR or the ΔC-EGFR, Δ973-EGFR, K721A-EGFR, Δ698-EGFR, L680N-EGFR mutants under an inducible Tet-ON promoter35,61 were a gift from Prof. Linda Pike (Washington University, St. Louis). They were grown in 5% CO2 in air at 37 °C in phenol-red free Dulbecco’s modified Eagle's medium (DMEM) supplemented with 10% (v/v) foetal bovine serum, 2 mM glutamine, 100 μg/ml hygromycin B, and 100 μg/ml geneticin. Cells transiently transfected with I942E-EGFR, L834R-EGFR, IIIV/KKRE-EGFR, control wtEGFR or T766M-EGFR were allowed to grow for 48 h after transfection. All cells used were regularly tested for mycoplasma contamination.

Drug treatments

For all relevant experiments cells were treated with 1 μM erlotinib (Tocris Bioscience) in medium containing 0.1% serum during starvation and labelling, and imaging (for live cells). Cholesterol depletion was performed by incubating cells with 10 mM methyl beta-cyclodextrin (MβCD) (Sigma) in Opti-MEM (Invitrogen) for 30 min at 37 °C after the starvation step. Cells were rinsed twice with medium with 0.1% serum prior to labelling.

Fluorophore localisation imaging with photobleaching

The method was first described in ref. 28. Briefly 1 × 105 CHO cells were seeded on 1% BSA-coated 35 mm no. 1.5 (high tolerance) glass-bottomed dishes (MatTek Corporation, USA) in 2 ml of media plus 50 ng/ml of doxycycline hyclate (Sigma), resulting in expression of ~105 receptors/cell35. After 48 h the medium was changed to 0.1% serum plus 50 ng/ml doxycycline for 2 h. Cells were rinsed and cooled to 4 °C for 10 min and then labelled with 4 nM CF640R-EGF or CF640R-Affibody, or with 8 nM EgB4-NB for 1 h on ice. The N-terminus of EGF was labelled at a 1:1 ratio by Cambridge Research Biochemicals (Cleveland, UK). The EGFR Affibody was labelled at a 1:1 ratio at its single cysteine residue. EgB4-NB was conjugated to CF640R-NHS ester (Biotium) according to the manufacturer’s instructions. Cells were rinsed and fixed with 3% paraformaldehyde (Electron Microscopy Sciences, USA) plus 0.5% glutaraldehyde (Sigma) for 15 min at 4 °C, then 15 min at room temperature. If required, cells were pre-treated for 1 h on ice at 4 °C with 200 nM 9G8-NB dissolved in PBS, or with erlotinib or MβCD as described above. We used an Axiovert 200M microscope with TIRF illuminator (Zeiss, UK), with a ×100 oil-immersion objective (α-Plan-Fluar, NA = 1.45; Zeiss, UK) and an EMCCD (iXon X3; Andor, UK). The microscope is also equipped with a wrap-around incubator (Pecon XL S1). Samples were illuminated with a fibre-coupled laser combiner (Andor) with a 100 mW 640 nm diode laser (Cube, Coherent). Images were collected every 0.28 s. Typically, for each experiment, approximately 120,000 single particle image spots were obtained from at least 750 cells and at least three biological repeats. Empirical posterior FLImP distributions were then obtained based on Affibody, EGF, or EgB4-NB separation measurements.

FLImP decomposition and post-analysis

The method has been described in detail in ref. 12. Briefly, a FLImP histogram is a sum of multiple FLImP measurement empirical posterior distributions, and can be modelled as sum of Rician-distributions, one for each discrete underlying separation. We use a Bayesian parameter estimation to fit a given number of Rician components to the FLImP distribution, thus inferring the underlying distances, their uncertainties, and the number of measurements for each component. We use a BIC to choose objectively the number of components justified by the data. We incorporate bootstrap-resampling of the measurements to account for the effect of the finite number of measurements on the error bars and BIC test.

Photo-bleaching imaging correlation spectroscopy

Fluorescent images of the labelled receptor at the basolateral membrane were repetitively scanned with an Olympus FV1000 laser scanning confocal microscope to induce partial photobleaching. The resulting images were subjected to pbICS as described in ref. 38. Briefly, for the “model-free” analysis, an aggregation model with monomer, dimer, trimer, tetramer, pentamer, hexamer, septamer, and octamer was assumed, where the fraction of each oligomeric species was variable and allowed to float in the fitting of the pbICS curves. The reported histogram is the cumulative average of seven pbICS curves per treatment condition, acquired over three independent biological replicates.

The derived cluster density (CD) as a function of fractional intensity remaining after each round of photobleaching (p) is given by Eq. (1).

$${\mathrm{CD}}\left( p \right) = \frac{{\left( {\mathop {\sum }\nolimits_j jc_j} \right)^2p}}{{\mathop {\sum }\nolimits_j \left[ {jc_j + j\left( {j - 1} \right)c_jp} \right]}},$$
(1)

Where j is the oligomer size (j = 1 for monomer, j = 2 for dimer, etc.) and cj is the molar concentration of an oligomer size j. In the fitting procedure, we fixed j=1, 2, 3, 4, 5, 6, 7, 8 and allowed the concentrations C1C8 to vary62. A typical raw dataset (CD versus p) together with fit is shown in Supplementary Fig. 3.

FRET distance of closest approach

Uncoated 35 mm no. 1.5 glass-bottomed dishes were seeded with 1 × 105 CHO cells expressing wtEGFR in 2 ml of media with 250 ng/ml of doxycycline (~4 × 105 receptors/cell). After 48 h, the medium was changed to 0.1% serum with the same concentration of doxycycline for 2 h. Samples were labelled with 5 µM C18 DiI (Thermo Fisher Scientific) for 10 min at 37 °C, then with either 100 nM EGFR Affibody or 200 nM EgB4-NB both labelled with Alexa Fluor 488 (custom conjugation) at 4 °C on ice for 1 h. Samples were fixed with 3% paraformaldehyde plus 0.5% glutaraldehyde as described above. The ensemble averaged DOCA of EGFR-bound Alexa 488 probes to cell membranes loaded with the fluorescent lipid analogue C18DiI was estimated from the measured FRET efficiencies for increasing acceptor density within the membrane using the procedure described previously32. Briefly, bootstrap regression analysis was performed by randomly resampling datasets with replacement 3000 times and fitting each resampled dataset to the FRET response generated by Monte-Carlo simulation of a donor placed at increasing distances above a plane of acceptors. For each dataset a distribution of possible DOCAs is generated and the mean and standard deviation of these distributions are reported. Bootstrap hypothesis testing was used to compare differences between mean DOCAs. For each pair of datasets compared, bootstrap regression analysis was performed by resampling with replacement from the pool of the combined data points, two new datasets of equal size to the originals. The test statistic was calculated for each of 10,000 bootstrap replicates to generate a test distribution corresponding to the null hypothesis that the difference between the mean DOCA values is zero. The p-value reported is the proportion of the test distribution larger than the observed test statistic. The number of data points (each point from an individual cell) for wtEGFR + Affibody, wtEGFR + EgB4-NB, wtEGFR + 9G8-NB + Affibody, wtEGFR + 9G8-NB + EgB4-NB, wtEGFR + erlotinib + Affibody, wtEGFR + erlotinib + EgB4-NB, L680N-EGFR + Affibody, L680N-EGFR + EgB4-NB, L834R-EGFR + Affibody, and L834R-EGFR + EgB4-NB are 54, 63, 97, 30, 119, 170, 40, 47, 30, and 11, respectively. Data were taken from at least two independent biological replicates.

Western blots

CHO cells were seeded in six-well plates at a density of 1 × 105 cells/dish. 48 h later, cells were washed twice with ice-cold PBS chilled for 10 min then incubated on ice with ice-cold EGF for 1 h at 4 °C. Cells were washed twice with ice-cold PBS and scraped into PBS plus inhibitors (phosphatases and proteases) and spun down. Cells were lysed in 10× volume of M-PER lysis buffer (Pierce) + 100 mM NaF + 1 mM Na3VO4 + 1% protease inhibitors (Cell Signalling Technologies) + 150 mM NaCl + 1 mM EDTA at pH 8 and incubated for 10 min at room temperature. Cells were cleared by centrifugation and total protein measured. Sample buffer was added to 1× final concentration. Samples were run in parallel on 1.5 mm thick, 3–8% Tris-Acetate NuPAGE gels (Invitrogen) with HiMark Prestained HMW and Novex Sharp Prestained protein standards (Invitrogen) using an XCell apparatus (Invitrogen). Proteins were blotted using an iBlot system (Invitrogen) on PVDF membranes, blocked for 1 h at 4oC with 5% BSA in TBS + 0.1% Tween and probed overnight with mouse anti-EGFR pY1173 (Upstate (Millipore)—cat. no. 05–483) antibody diluted 1/1000. Gels were probed with secondary anti-mouse-HRP antibody (Jackson ImmunoResearch—cat. no. 715-035-150) diluted 1/1000 and incubated with Supersignal West Pico Chemiluminescent Substrate solution (Pierce) for 5 min, then imaged with a BioRad ChemiDoc MP system imager. Each blot was stripped with 25 ml stripping buffer (2% SDS, 0.75% β-mercaptoethanol, 62.5 mM Tris HCl pH 6.7) for 50 min at 60 °C, and re-probed with an anti-EGFR cocktail composed of anti-EGFR D38B1 (Cell Signalling Technologies—cat. no #4267), anti-EGFR N-Terminal polyclonal ab137660 (abcam), and anti-EGFR polyclonal 10005: sc-03 (Santa Cruz Biotechnology), each diluted 1/2000, all derived from rabbit. Anti-rabbit HRP (Jackson ImmunoResearch—cat. No 711-035-152) diluted 1/1000 was used for all blots and images were acquired as above. Densitometry was performed using Fiji63 to extract the intensity of the bands and then calculating the ratio between the intensity of the pY1173 EGFR band and the corresponding total EGFR band for each condition. Experiments were performed in triplicate and data were presented in a box and whisker plot, showing the inclusive median, with the 25th and 75th quartile as box limits and the first and last quartile as whiskers.

Tracking of EGFR complexes on CHO cells

Cells were seeded at a density of 1 × 105 cells/dish on 1% BSA-coated 35 mm no. 1.5 (high tolerance) glass-bottomed dishes (MatTek Corporation, USA) in 2 ml of media. Receptor expression in CHO cells was induced with 50 ng/ml doxycycline. Prior to imaging, cells were starved for 2 h at 37 °C in 0.1% serum supplemented with 50 ng/ml doxycycline if required. Treatments with erlotinib and MβCD were performed as described above. Cells were then rinsed twice with 0.1% serum without doxycycline pre-heated at 37 °C and were labelled with a 1:1 mixture of 8 nM Alexa 488-Affibody/CF640R-Affibody for 10 min at 37 °C. Cells were rinsed twice with low serum medium without doxycycline pre-heated at 37 °C and promptly imaged. Single-molecule images were acquired using an Axiovert 200M microscope with a TIRF illuminator (Zeiss, UK), with a ×100 oil-immersion objective (α-Plan-Fluar, NA = 1.45; Zeiss, UK) and an EMCCD (iXon X3; Andor, UK). The microscope is also equipped with a wrap-around incubator (Pecon XL S1). The 488 and 642 nm lines of a LightHub laser combiner (Omicron Laserage GmbH) were used to illuminate the sample and an Optosplit Image Splitter (Cairn Research) was used to separate the image into its spectral components as described previously64. The field of view of each channel for single-molecule imaging was 80 × 30 µm. Typically, for each condition, at least 30 field of views comprising one or more cells were acquired from a total of at least 3 independent biological replicates. All single-molecule time series data were analysed using the multidimensional analysis software described previously31. Briefly, this software performs frame-by-frame Bayesian segmentation to detect and measure features to sub-pixel precision, then links these features through time to create tracks using a simple proximity-based algorithm. The software determines cubic polynomial registration transformations from images of fluorescent beads. Feature detection and tracking was performed independently in each channel.

Calculation of colocalisation and τON

Two-colour TIRF images of the basolateral surfaces of cells were chromatically separated by a beam splitter and registered using custom made software to map the relative positions of the probes over the time course of data acquisition31. Single particle tracks were extracted as above31. The fraction of co-localised tracks was reported by tracks in which a receptor particle in one channel colocalises with another particle in the other channel, spending at least five 50 ms frames in total moving together within a pixel of each other. Coincidental colocalisation statistics were calculated for a dataset as follows. In each channel, a randomised set of tracks of the same size as the measured set was produced, where each track in the random sample was chosen (with replacement) from the measured tracks for that channel, recentred at random with uniform probability density across the field of view, rotated through a random angle with uniform probability density between 0 and 360°, and randomly flipped in x with probability 0.5. The random tracks therefore have key properties such as durations and path structures representative of the true tracks, but now randomised in distribution and orientation. The colocalisation statistics were then calculated for the randomised tracks. This was performed a total of 50 times for each dataset, and the colocalisation statistics pooled to give a final estimate coincidental colocalisation fraction for that dataset. The reported colocalisation fraction is then the fraction for the real data minus the estimated coincident colocalisation fraction. The duration (τON) of individual events in which a track in one channel moves within a pixel of a track in the other channel and then they move apart again was also calculated. To reduce the impact of localisation error on these results a temporal Gaussian smoothing filter of FWHM 4 frames (200 ms) was applied to the position traces before the colocalisation analyses.

9G8 nanobody binding affinity for different receptor conformations

Ectodomain conformation was investigated by measuring the binding of 9G8-NB, which stabilises the tethered conformation, using the ratio between the binding of CF640R-9G8-NB, a probe which is selective for the tethered state of the receptor, and Alexa 488-EgB4-NB, a probe which is not, as determined by quantitative confocal microscopy. Briefly, uncoated 35 mm no. 1.5 glass-bottomed dishes were seeded with 1 × 105 CHO cells expressing wtEGFR in 2 ml of media with 250 ng/ml of doxycycline (~4 × 105 receptors/cell). After 48 h, the medium was changed to 0.1% serum with the same concentration of doxycycline for 2 h and cells were treated with 1 μM erlotinib as described in the drug treatments section above, if required. Cells were then treated with 50 nM Alexa 488-EgB4-NB for 1 h at 4 °C and then fixed with 3% PFA for 15 min at 4 °C, followed by 15 min fixation at room temperature. This was done to prevent 9G8-NB from inducing conformational changes. After fixation, cells were washed three times with PBS and then treated for 1 h at 4 °C with 200 nM of CF640R-9G8-NB. Twenty confocal images of equatorial regions of the cells were collected from three replicates, using a Leica TCS SP8 microscope with a (NA = 1.4; Leica) and a Leica HyD hybrid detector. Fluorescence was excited with 488 or 640 nm light produced by an NKT Extreme supercontinuum light source. Using FIJI, regions of interest corresponding to the plasma membrane of individual cells were drawn manually. The mean Alexa 488 and CF640R fluorescence were measured and the ratio of CF640R:Alexa 488 fluorescence intensity calculated for each cell.