Drug-resistant EGFR lung cancer mutations promote tumor growth by stabilizing interfaces in ligand-free signaling-competent EGFR oligomers

Abstract

Here we implement a higher-resolution FLImP version which, combined with large-scale simulations of various membrane-embedded dimer interfaces, allowed us to build an experimentally backed model of all the relevant interactions assembling signaling ligand-free EGFR oligomers.We show that WT-EGFR, T766M-EGFR, and Ex20Ins-EGFR share a ligand-free hetero conf -oligomer structure in which tetramer-scaffolds made of H2H ect /2x kin monomer and B2B ect /H2H kin dimer sub-units held by a transversal transmembrane interface 39 cantilever into position the extracellular portion of the St2St ect /Asym kin dimer under the regulation of the ectodomain tethered conformation 40 .Within these hetero conf -oligomers, St2St ect /Asym kin dimer sub-units are positively and negatively regulated intracellularly via two currently functionally-orphan kinase interfaces (PDB IDs:3VJO 41 and 5CNO 42 ).Crucially, we show that these newfound oligomer assembling interfaces, specific to the ligand-free state, are critical for EGFR mutation-dependent tumor growth in vivo.

Results
FLImP as a molecular ruler to measure oligomer size and discriminate different sub-unit conformers FLImP measures the lateral pairwise separations on the scale of 0-70 nm between fluorescent probes bound to non-monomer structures on the cell surface, and their relative abundance 33 .Mimicking the viewpoint of FLImP microscopy, we can assume the orthogonal projection onto the cell surface (xyplane) of a hetero conf -hexamer assembled by two dimer conformers (Fig. 1C).We created a dataset of synthetic FLImP separation probability distributions simulating measurements of individual separations between fluorophores bound to the hetero conf -hexamer, including spurious fluorophore localizations (clutter), and noise, which we summed (Fig. 1D; gray background).
If individual dimeric units of different conformations are present on the cell surface, FLImP will report one separation per type.If, as shown in Fig. 1C, hetero conf -oligomers are assembled by two dimer conformers, the FLImP peaks report 1 st , 2 nd , 3 rd , etc., neighbor separations in the oligomer structure (Fig. 1D; color peaks).With the help of mutations, changes in intensity and position of these peaks can be used to investigate the structural sub-units assembling an oligomer.As an example, we simulated synthetic FLImP separations for the homo-trimer that forms when one of the dimer conformers is disrupted (Fig. 1E, gray background).FLImP decomposes from these data the remaining 1 st and 2 nd -order vertical separations from the homo-trimer (Fig. 1E, color peaks), revealing which 1 st order separation belongs to the disrupted interface and which higher-order peaks are dependent on hetero conf -interactions.The mutations and treatments that we used to dissect the interactions assembling ligand-free oligomers are mapped-out in Fig. 1F.
FLImP samples a finite population of separations, and this introduces errors.We, therefore, used bootstrap-resampling 43 to estimate how this affects the decomposition.Fig. 1G illustrates that bootstrap-resampling can capture changes caused by the transition from hetero conf -hexamer to homotrimer and evaluate significance above finite sampling errors.
Given enough resolution and with the help of mutations/treatments, separation peaks from 1 st order interfaces are typically amenable to be assigned, but assigning higher order peaks can be harder.We interrogate the effect of mutations on unassigned higher order peaks by folding them into a multidimensional scaling (MDS) Wasserstein metric 44 .As shown in the example (Fig. 1H), we also include the bootstrap-estimated errors associated with finite sampling and calibrate changes to report oligomer growth direction.The B2B ect /H2H kin dimer and the St2St ect /Asym kin dimer are sub-units in hetero conf -oligomers The color peaks in Fig. 2A show the most likely positions and intensities of pairwise separations between CF640R fluorophores specifically conjugated to anti-EGFR Affibodies bound to DIII of cell surface WT-EGFR ectodomains.All FLImP separation sets have hereafter the median positions of the WT-EGFR separations superimposed (dashed lines) to facilitate comparisons.FLImP measurements require immobilizing cell surface receptors by chemical fixation via a method demonstrated not to introduce detectable artefacts 33 .Nevertheless, predictions arising from results of chemically fixed cells were validated in live cells using single particle tracking (Supplementary Fig. 1A).
Interpreting a FLImP separation set requires assigning 1 st order peaks to the structural sub-units assembling the underlying dimers/oligomers.In previous work 34 , we link the 10.8-13.5 nm peak (Fig. 2A, red dashed line) to the extracellular part of the B2B ect /H2H kin dimer, which could also exist as an oligomer sub-unit (Fig. 2B).The link between the 10.8-13.5 nm peak and the B2B ect /H2H kin dimer, either alone and/or as a protomer, is supported here by the observation that FLImP no longer indicates separation density between 8.1-14.8nm when a previously proposed double E981R/D982K (ED/RK) mutation in the C-terminus 45 that destabilizes electrostatic interactions at the heart of the intracellular portion of the B2B ect /H2H kin dimer is introduced (Fig. 2C; red dashed line).By contrast, separation density at 9.2-13.2nm increases when the H2H kin dimer is stabilized by the T766M mutation (Fig. 2D; red dashed line), further supporting the assignment of the 10.8-13.5 nm peak in the WT-EGFR separation set to the B2B ect /H2H kin dimer.
We previously linked a <9 nm separation to the extracellular portion of the St2St ect /Asym kin dimer 34 , which could also exist as an oligomer sub-unit.Our enhanced FLImP method decomposes two <9 nm components in the WT-EGFR separation set (Fig. 2A; cyan and black dashed lines).To determine which one of these components could be assigned to the St2St ect /Asym kin dimer conformer, we pre-treated WT-EGFR-expressing cells with the conformation-selective monoclonal antibodies mAb-2E9 46 or mAb-108 47 , which select for high and low-affinity Epidermal Growth Factor (EGF) binding, respectively.As the selectivity of these mAbs is bona fide against EGF binding, mAbtreated cells were next probed with an EGF-CF640R derivative.Here cells were fixed after mAb treatment but before probing with EGF-CF640R to avoid ligand-induced conformational changes.Results show that EGF-CF640R binds well to fixed cells at similar sites to Affibody-CF640R (Supplementary Fig. 1B, 1C).
Our results indicate that EGF binds cell surface St2St ect /Asym kin dimer conformers with high affinity (Supplementary Fig. 1D).The separation set from mAb-2E9-treated cells suggests two components of <9 nm (4.9-6.8 nm and 7.2-8.9nm) (Fig. 2E; cyan and black dashed lines), which are consistent with those in the WT-EGFR separation set.In contrast, no separation density at 6.8-9.9 nm is apparent in the set from cells treated with mAb-108, which blocks high-affinity binding, (Fig. 2F; black dashed line).To assess the robustness of this result, because incompletely resolved components cannot be perfectly separated, we compared the evidence for separations in bootstrap-resampled datasets after pooling the individual components (marginalized probability).This analysis indicates that the absence of separation density in the vicinity of ~8 nm associated to the mAb-108 treatment is robust to finite sampling errors (Fig. 2G).We therefore assigned the 7.0-8.6nm peak in the WT-EGFR separation set to the St2St ect /Asym kin dimer conformer.Further validation of this assignment is provided below.
Despite the T766M mutation destabilizing the Asym kin dimer 34 , the separation density at ~8 nm is similar between WT-EGFR and T766M-EGFR (Fig. 2A, 2D).We hypothesized that the T766M mutation underpins the St2St ect /Asym kin dimer via interactions with the inactive B2B ect /H2H kin dimer, which is stabilized by the T766M mutation.To test this, WT-EGFR expressing cells were treated with Erlotinib 48 , a TKI that binds to EGFR's kinase ATP-binding pocket, stabilizing the St2St ect /Asym kin dimer 49 .With our previous poorer resolution, we found that Erlotinib treatment enhances a broad peak encompassing separations 1.6-10.2nm 34 .Our higher resolution method decomposes three components under this peak (Fig. 2I).Interestingly, Erlotinib treatment recapitulates effects induced by the T766M mutation, most notably on separations of <20 nm associated with Erlotinib-treated WT-EGFR and T766M-EGFR (Fig. 2I, 2D, 2A).Quantified in Fig. 2J, both induce a decrease in the 5.8-7.1 nm component, the assignment and function of which is discussed below, and an increase at ~9-12 nm.These results argue that the stabilizing the St2St ect /Asym kin dimer directly by Erlotinib binding shares characteristics of stabilizing the B2B ect /H2H kin dimer via the T766M mutation, suggesting that the B2B ect /H2H kin dimer stabilizes the St2St ect /Asym kin dimer within hetero conf -oligomers.
Further evidence that the B2B ect /H2H kin dimer and St2St ect /Asym kin dimer are structural sub-units in ligand-free hetero conf -oligomers is shown in Fig. 2H.This includes the increase (decrease) in oligomer size induced by the T766M (ED/RK) mutations and the different oligomer sizes associated with mAb-2E9 and mAb-108.Notably, Erlotinib, as previously found 34 , does not significantly increase oligomer size.These results assign the role of underpinning oligomer growth to the H2H kin dimer.
H2H ect dimer/2x kin monomers are sub-units in hetero conf -oligomers In a previous study 34 , we proposed a third ligand-free dimer conformer based on a lattice contact in an X-ray structure of the tethered ectodomain monomer (PDB ID:4KRP 50 ), in which the monomers are held by ectodomain interactions (H2H ect dimer) (Fig. 3A).This model was supported by the FLImP results associated with C-EGFR, a mutant in which the intracellular domains are deleted, here reanalyzed with the higher resolution decomposition (Fig. 3B).Results show peaks at almost a fixed interval consistent with the previously proposed homo-oligomers of repeating extracellular H2H ect dimer units 34 .The presence of H2H ect dimer/2x kin monomers is further supported by the finding that the 1 st -order separation of the truncated H2H ect dimerof C-EGFR shares position with the shortest component in the separation set for WT-EGFR (Fig. 3B; cyan dashed line).Taken together, previous and current evidence confirms the presence of H2H ect dimer/2x kin monomers as a third ligand-free dimer conformer.
Higher resolution unmasked differences beyond the 1 st order separation between the C-EGFR and WT-EGFR sets (Fig. 3B quantified below, 2A,).Given this, we speculated that H2H ect dimer/2x kin monomer conformers might participate in hetero conf -oligomerization.To investigate this, we introduced mutations that disrupt the H2H ect dimer.Our MD simulations suggest that inhibition of the tether via two well-understood DIV mutations in EGFR's ectodomain, H566F and G546P 51 , disrupts the H2H ect dimer (Supplementary Fig. 2A, 2C).These mutations induce a shift in the shortest separation component assigned to H2H ect dimer/2x kin monomer conformers from a median position of 6.5 nm to 4 nm, which implicates the ectodomain tethered conformation in the formation of H2H ect dimer/2x kin monomers (Fig. 3D,   3E; blue and orange).Results for G564P are in Supplementary Fig. 3A-3D.
Because MD simulations suggest that the WT-EGFR H2H ect dimer can explore a separation range of 3-7 nm between the center of mass of the two DIIIs (Supplementary Fig. 2B, Supplementary Method 1), the shift in the separation component annotated to H2H ect dimer/2x kin monomer conformers from a median position of 6.5 nm to 5.2 nm introduced by the ED/RK mutation (Fig. 3E; blue and green) suggests that inhibiting the B2B ect /H2H kin dimer changes H2H ect dimer/2x kin monomers conformation.The ED/RK mutations also induce a comparable oligomer size reduction to that from the H566F mutation (Fig. 3C).Together these results suggest that B2B ect /H2H kin dimer sub-units interact with H2H ect dimer/2x kin monomers, and hence that the latter is also a sub-unit in ligand-free hetero conf -oligomers.
From this, it follows that H2H ect dimer/2x kin monomers might interact with the St2St ect /Asym kin dimer.To test this, WT-EGFR expressing cells were treated with Lapatinib (Fig. 3F), a TKI that binds to EGFR's kinase ATP-binding site breaking the Asym kin dimer 52 .As expected, Lapatinib induced a significant reduction around the 8 nm position assigned to the St2St ect /Asym kin dimer conformer (Fig. 3E; blue and red).This is accompanied by a shift in the component assigned to the H2H ect dimer/2x kin monomer conformer, from a median position of 6.5 nm to 4.2 nm (Fig. 3E, 3F; cyan dashed line), arguing that inhibiting the St2St ect /Asym kin dimer changes the conformation of H2H ect dimer/2x kin monomers, and hence that H2H ect dimer/2x kin monomer and St2St ect /Asym kin dimer sub-units interact.Interestingly, Lapatinib does not decrease oligomer size (Fig. 3C).This will be discussed later with more data (Fig. 7).

Ligand-free conformers interact via transmembrane contacts
The conformation of the three ligand-free dimer sub-units suggest hetero conf -oligomer assembly might be mediated by transmembrane interactions.Therefore, we next considered the Lzip transmembrane dimer named after its leucine zipper-like interactions 39 (Fig. 4A).In principle, the Lzip interface could mediate interactions between transmembrane monomers and dimers (Fig. 4B).To investigate this possibility, we mutated all three Lzip dimer transmembrane helix amino acids (V635, L638, L642) to either serine or alanine, named Lzip3S and Lzip3A, respectively.Based on previous literature, Lzip3S mutations would be expected to strongly inhibit the Lzip interaction, unlike the more conservative Lzip3A 53 .
To evaluate the consistency of the FLImP results, we reasoned that simultaneously inhibiting the H2H kin dimer and Asym kin dimer, and thus all intra-dimer intracellular interactions, together with Lzip contacts should recapitulate the C-EGFR results.This hypothesis was evaluated in two stages.First, by combining the ED/RK mutations with L680N, a kinase N-lobe mutation that inhibits the kinase domain from acting as receiver 45 , and thus the St2St ect /Asym kin dimer.Then the Lzip3S mutations were added.Reassuringly, the ED/RK+L680N mutations induce a shift in the peak assigned to H2H ect dimer/2x kin monomer sub-units comparable to the Lzip3S mutations alone, confirming that the conformation of the H2H ect dimer/2x kin monomer sub-units depends on hetero conf -oligomer interactions.Also reassuringly, combining ED/RK+L680N+Lzip3S mutations recapitulated the results for C-EGFR (Fig. 4E, 4G), e.g., pseudo-periodic separations, oligomer size increase, and the loss of the 2.8-4.3 nm peak (Fig. 4F), suggesting that Lzip contacts inhibit the formation of the homo-oligomers of repeating extracellular H2H ect dimer interfaces.The extracellular structure of ligand-free hetero conf -oligomers Based on the above data, we constructed a model of the orthogonal projection on the cell surface of the hetero conf -oligomer based on the known shape and dimensions of the three ligand-free dimer subunits (Fig. 1B, Supplementary Fig. 4A-4C).H2H ect dimer/2x kin monomer and B2B ect /H2H kin dimer could form a tetramer in which, remarkably, the Lzip interface aligns one ectodomain of the H2H ect dimer/2x kin monomer so it can link with another tetramer and form the extracellular portion of the St2St ect /Asym kin dimer (Fig. 5A).
To test this model and its relevance to the dysregulated ligand-free state, we implemented a 2D version of FLImP that reports triangular arrangements between probes bound to EGFR structures.The 3-fold higher probe concentration required for 2D FLImP was better suited to the less sticky EGF-CF640R derivative, as Affibody-CF640R at this higher concentration began to show signs of nonspecific binding on the glass supporting the cells 54 .
As an example, we probed cells expressing T766M-EGFR with EGF-CF640R after chemical fixation to avoid ligand-induced conformational changes, as discussed above.The resulting 2D FLImP triangle dataset was optimally grouped into seven distinct triangles (Fig. 5B-5H).Reassuringly, the 1D separations in the triangles are found as components decomposed by 1D FLImP for T766M-EGFR and WT-EGFR (Fig. 2D, 2A), arguing that T766M-EGFR and WT-EGFR share hetero conf -oligomer structure.
To evaluate whether the 2D FLImP data supports the proposed model, the model was expanded to the size required by the triangles and the positions expected for EGF-CF640R bound to EGFR's ectodomain DIII marked (Fig. 5I).We found that the smallest triangle (T1) accounts, within errors, for four triangular probe motifs in the model (Fig. 5I).Triangles T2-T7 each account for one motif (Fig. 5J).
To further validate the applicability of the hetero conf oligomer structure to T766M-EGFR, we verified that combining the T766M mutation with the tether-disrupting H566F mutation or the Lzip3S mutations also disrupts the hetero conf -oligomers (Supplementary Fig. 6A-6D).This, together with the excellent results of superimposing the triangles from 2D FLImP data collected from CHO cells expressing T766M-EGFR indicate that the model is an accurate representation of the extracellular portion of the ligand-free hetero conf -oligomers.

A ligand-independent mechanism of Ex20ins-induced oligomer activation
The model predicts that inhibiting the B2B ect /H2H kin dimer conformer would result in the tetramer shown in Fig. 6A.Inhibiting the Lzip interface should have the same effect, but separations found when B2B ect /H2H kin dimer sub-units are inhibited by the ED/RK mutations are inconsistent with those found when Lzip contacts are inhibited via Lzip3S mutations (Fig. 6B).This hinted at the possibility that we had not yet accounted for all hetero conf -oligomer assembling interactions.We considered a functionally orphan symmetric backbone-to-backbone kinase interface (Bb2Bb kin interface), revealed by X-ray crystallography 41 (Fig. 6C).Because a kinase monomer could dock into a Bb2Bb kin dimer to form an Asym kin dimer, we speculated that the Bb2Bb kin interface might be involved in strengthening the St2St ect /Asym kin dimer sub-units.Giving credence to this possibility, two non-naturally occurring charge-reversal R938E and K946E mutations, which compromise the Bb2Bb kin dimer, as reported by MD simulations (Supplementary Fig. 7A-7B), decrease receptor phosphorylation (Fig. 6D).
Remarkably, we found that the separation set associated with K946E-EGFR is almost indistinguishable from that of ED/RK+L680N-EGFR (Fig. 6E-6G).This reveals that disrupting the Bb2Bb kin interface via the K946E mutation recapitulates the effect of jointly inhibiting St2St ect /Asym kin dimer and B2B ect /H2H kin dimer sub-units via the combined ED/RK+L680N mutations.This is consistent with St2St ect /Asym kin dimer and B2B ect /H2H kin dimer conformers being obligate hetero conf -oligomers.Adding the K946E mutation to T766M-EGFR also reduces oligomer size and decreases receptor phosphorylation (Fig. 6G-6I).Together, results suggest that the Bb2Bb kin interface reinforces the St2St ect /Asym kin dimer subunits, and thereby their interaction with the B2B ect /H2H kin dimer in the hetero conf -oligomers.
If the above interpretation is correct, one would expect that stabilizing the Bb2Bb kin interface should mirror the effects of stabilizing the St2St ect /Asym kin dimer conformer.We conjectured that Ex20Ins might increase the number of contacts between the kinase domains of the Bb2Bb kin interface, stabilizing that interface (Fig. 6J).This was supported by MD simulations of the WT and D770-N771insNPG (insNPG), a mutant chosen because structural data is available 55 (Supplementary Fig. 7C-7D).In support of the notion that the Bb2Bb kin interface underpins the stability of St2St ect /Asym kin dimer conformers, the separation set for insNPG-EGFR recapitulates the effects of the Erlotinib treatment in WT-EGFR (Fig. 6K, 6L, Supplementary Fig. 6E), revealing that the mechanism by which insNPG-EGFR dysregulates ligand-independent phosphorylation is by strengthening the Bb2Bb kin interface in hetero conf -oligomers.

A ligand-independent mechanism of T766M-induced oligomer activation
After incorporating the Bb2Bb kin interface (Fig. 7A), another prediction is that disrupting Lzip contacts should have an analogous effect on phosphorylation to inhibiting the B2B ect /H2H kin dimer via the ED/RK mutations, but only the latter increases phosphorylation (Fig. 6D).We speculated this might be explained if the B2B ect /H2H kin dimer sequestered kinase monomers from stabilizing the Asym kin dimer via the Bb2Bb kin interface (Fig. 7B).In the crystal lattice of the activator-impaired V924R H2H kin dimer 42 we noticed a side-to-side interface (S2S kin interface) that could play such a role (Fig. 7C).
We previously reported that the non-naturally occurring I942E mutation inhibits ligand-bound oligomerisation 34 .Here we show that I942E decreases ligand-free EGFR phosphorylation (Fig. 6D).The I942 residue lies at the heart of the S2S kin interface, and modelling of the WT and I942E mutant suggested that the latter stabilizes the S2S kin interface (Supplementary Fig. 7E, Supplementary Method 1).The separation set associated with I942E-EGFR is consistent with the results obtained for WT-EGFRexpressing cells treated with Lapatinib, which inhibits the Asym kin dimer (Fig. 7D-7E).These results argue that S2S kin interface stabilization via the I942E mutation also inhibits the St2St ect /Asym kin dimer, indicating that the S2S kin interface plays an autoinhibitory role.Combining the T766M and I942E mutations support this notion (Supplementary Fig. 6F-6G).
If the B2B ect /H2H kin dimer sequestered kinase monomers via the S2S kin interface preventing these from reinforcing St2St ect /Asym kin dimer sub-units via the Bb2Bb kin interface, then inhibiting the Bb2Bb kin interface should make more kinase monomers available to stabilize the B2B ect /H2H kin dimer via the S2S kin interface, thereby growing larger oligomers as found when B2B ect /H2H kin dimer sub-units are stabilized by the T766M mutation (Fig. 2H).To test this possibility, we need to inhibit the Bb2Bb kin interface whilst preserving St2St ect /Asym kin dimer and B2B ect /H2H kin dimer sub-units so they can compete for kinase monomers.However, when the Bb2Bb kin interface-inhibitory K946E mutation is introduced in WT-EGFR, the mutation also disassembles the St2St ect /Asym kin dimer and B2B ect /H2H kin dimer conformers (Fig. 6G).In contrast, partially disassembled Lzip3S-EGFR oligomers preserve their phosphorylation (Fig. 6D) and some density around 12 nm (Fig. 6B), suggesting that Lzip3S-EGFR oligomers retain some St2St ect /Asym kin dimer and B2B ect /H2H kin dimer sub-units.Thus, we reasoned that adding K946E to Lzip3S mutations should stabilize the B2B ect /H2H kin dimer and increase oligomer size.The results are consistent with this notion (Fig. 7F).We also found that the separation sets of Lzip3S+K946E and Lzip3S+L680N are similar below ~35 nm (Fig. 7G, 7H), suggesting that inhibiting the St2St ect /Asym kin dimer and the Bb2Bb kin interface are almost equivalent, and explaining why disrupting the St2St ect /Asym kin dimer, either directly via Lapatinib or indirectly via the I942E mutation, does not decrease oligomer size, as both stabilize the S2S kin interface and thereby the H2H kin dimer counterbalancing the effect of disrupting St2St ect /Asym kin dimer sub-units (Fig. 7F).
Results, therefore, show that destabilizing the St2St ect /Asym kin dimer, either directly by the L680N mutation or indirectly by inhibiting the Bb2Bb kin interface, increases oligomer size.In WT-EGFR this depends on the S2S kin interface at the expense of phosphorylation.The T766M mutation directly stabilizes the B2B ect /H2H kin dimer, thus promoting the formation of larger oligomers without detrimental effects on phosphorylation, thereby conferring a cellular growth advantage to mutated cells.

Tumor growth depends on hetero conf -oligomerisation
To test the relevance of the proposed ligand-free oligomer structure and assembling mechanisms in vivo, we carried out cellular transformation assays using the IL3-dependent murine lymphoid Ba/F3 cell system with the aim of using these cells to establish tumor xenografts.Ba/F3 cells fail to survive and multiply in the absence of IL3 56,57 but this phenotype can be rescued by the ectopic expression of a constitutively active receptor tyrosine kinase, like for example T766M-EGFR, which allows survival signaling in the absence of IL3 56 .
We generated Ba/F3 cell lines stably expressing WT-EGFR (Ba/F3+WT), T766M-EGFR (Ba/F3+T766M), and EGFR mutants at near equal levels using the PiggyBac system and tested the ability of the transformed cells to grow in the absence of IL3 (Supplementary Fig. 8A).To minimize mice number, from mutations that only disrupt hetero conf -oligomer structure we focused on the T766M+H566F and omitted the equivalent T766M+Lzip3S (Supplementary Fig. 6A-6D).We also chose T766M+K946E over T766M+I942E because unlike the I942E mutation, K946E does not interfere with the ligand-bound state (Supplementary Fig. 8B), hence potential drugs should carry fewer side effects.
Next, we compared the ability of these Ba/F3 cell lines to establish and grow tumors.Notably, all these cell lines also stably expressed GFP, exploited to image growing tumors.We found a lag-time prior to the onset of palpable tumors of around 20 days in all cohorts, after which the tumors formed in animals that had received Ba/F3+T766M and Ba/F3+T766M+H566F cells and, with an additional delay, in the Ba/F3+T766M+K946E cohort (Fig. 8A).Importantly, mirroring the growth pattern observed in vitro, Ba/F3+T766M tumors grew best throughout, followed by the double mutant tumors (Ba/F3+T766M+H566F > Ba/F3+T766M+K946E), while WT-EGFR tumors did not establish or grow.The lack of tumor cell survival in the Ba/F3+WT cohort was already strongly suggested by day 21 when we observed no GFP reporter signals at the injection site (Supplementary Fig. 8C, 8D).The experimental endpoint was on day 41 when we imaged all animals by IVIS (Fig. 8B-8C) to determine tumor fluorescence as an independent measure of tumor growth between cohorts.We found significant differences in fluorescence signals between groups in line with caliper measurements (Fig. 8A) except for Ba/F3+T766M vs. Ba/F3+T766M+H566F tumors.The latter highlights the limitations of epifluorescence imaging, which did not reflect accurately the signal differences between larger tumors due to limited tissue penetration and absorption within thicker tissues.Harvested tumors were first qualitatively imaged under daylight and by fluorescence (Fig. 8D), and then weighed (Fig. 8E) upon which the significant growth differences between the Ba/F3+T766M vs. Ba/F3+T766M+H566F tumors were evident as were the other differences already seen in vivo in Fig. 8A.Moreover, we subjected harvested tumor tissues to histology to demonstrate tumor morphology by H&E (Fig. 8F) and pan-EGFR expression by anti-EGFR staining (Fig. 8G vs. staining control/Supplementary Fig. 8F).All tumors were positive for EGFR with plasma membrane localization clearly visible in stained tissues (Fig. 8G).
Notably, the T766M+H566F mutant, which does not reduce basal phosphorylation compared with T766M alone or with T766M+K946E (Fig. 6I), also handicaps tumor growth, suggesting that oligomerization per se is conveying a growth advantage.An explanation is that T766M+H566F expressing Ba/F3 cells are unable to phosphorylate and activate AKT signaling to the same extent as T766M (Supplementary Fig. 8G).

Discussion
We combined super-resolution FLImP imaging with in silico modelling and mutagenesis to identify the interfaces that assemble signaling-competent ligand-free oligomers.The <3 nm resolution achieved allowed us not only to fingerprint previously proposed ligand-free dimer conformers but, crucially, show how inter-dimer interactions assemble ligand-free hetero conf -oligomers.From this knowledge and the known shapes of the ligand-free dimer conformers we derived a structural model of ligandfree hetero conf -oligomerization.In addition, the extracellular part of this model is validated by implementing a 2D FLImP version.
The tethered conformation of EGFR's ectodomain appeared during the evolution of vertebrates 58 , and was suggested to have evolved to prevent crosstalk between the different EGFR homologs in the vertebrate EGFR family 59 .Our MD simulations and FLImP results propose a new biological role for the tethered conformation, which is to regulate the formation of the H2H ect dimer, and through it, hetero conf -oligomer size, receptor activation, and, in the pathological context of T766M-EGFR, tumor formation.
Comparing models and data revealed that a previously orphan Bb2Bb kin interface reported by X-ray crystallography 41 plays a regulatory role in ligand-free hetero conf -oligomerization and activation.The role of the Bb2Bb kin interface is to buttress the Asym kin dimer, leading to activation once the B2B ect /H2H kin dimer cantilevers H2H ect dimer/2x kin monomers into position to form the St2St ect /Asym kin dimer.This mechanism explains how the Asym kin dimer can form in the absence of ligand-induced conformational changes that are typically required to overcome the activation barrier associated with the formation of the asymmetric interface between activator and receiver kinases 36 .Importantly, our work shows that Ex20Ins mutations stabilize the Bb2Bb kin interface, providing a breakthrough in our understanding of these mutations, which so far have been studied only at the monomeric level.
Comparing models and data also suggested a second S2S kin interface that we identified from a crystal contact in a structure of the H2H kin dimer 45 .This interface allows the H2H kin dimer to hijack kinase monomers, preventing formation of the stimulatory Bb2Bb kin interface, and down-regulating activation.The S2S kin interface stabilizes the H2H kin dimer, providing an explanation for its autoinhibitory role.We propose that the regulation of the activation of ligand-free hetero conf -oligomers rests on balancing the interactions between the Asym kin dimer and the H2H kin dimer with the two newfound kinase interfaces.
Beyond its autoinhibitory role, the H2H kin dimer also acts as a scaffold promoting the formation of larger hetero conf -oligomers.Thus, by stabilizing the H2H kin dimer, the T766M mutation indirectly underpins the formation of Asym kin dimer units without the need to hijack monomers from Bb2Bb kin interface, which can thereby still buttress the Asym kin dimer and amplify cell growth.This conveys a significant growth signaling advantage to the T766M receptor mutant.
As an example, we used T766M-EGFR-dependent cell growth to evaluate in vivo the effect of breaking the ligand-free hetero conf -oligomers by disrupting the tethered conformation and inhibiting the Bb2Bb kin interface.Disrupting the ectodomain tether associated with the H2H ect , which decreases oligomer size, halves the size of T766M-EGFR driven tumors.Excitingly, disrupting the Bb2Bb kin interface, which has a deleterious effect on both oligomer size and phosphorylation, almost abolishes tumor growth.This reveals an Achilles heel in drug-resistant NSCLC tumors that could be therapeutically targeted, counterintuitively, with drugs that interfere with the tether, and/or drugs that can inhibit the Bb2Bb kin interface.Furthermore, these interfaces are away from the ATP-binding pocket mutational hotspot and inhibiting the Bb2Bb kin interface has no detectable effect on ligand-bound dimerization, so one could in principle expect no adverse side effects derived from interfering with the ligand-bound state.
Targeting protein-protein interactions is a new direction in treating diseases and an essential strategy for the development of novel drugs 60 .Given the poor long-term efficacy of current treatments for NSCLC, the new structural understanding from this work of how T766M-EGFR and Ex20Ins-EGFR amplify cell growth suggests a possible route for more effective therapies.Notably, mutant and WT-EGFR share hetero conf -oligomer structure.It would be interesting to find out whether these principles apply to cancers driven by EGFR overexpression which currently have very limited therapeutic options.

Captions
monomers 34 .Right, a ligand-free stalk-to-stalk ectodomain dimer conformer (St2St ect dimer sub-unit) 34 coupled with the Asym kin dimer sub-unit 34 .Middle, a ligand-free B2B ect dimer sub-unit 62 coupled with a head-to-head kinase dimer conformer (H2H kin dimer sub-unit) 30 .C Cartoon of two imaginary dimer conformers assembling a hetero conf -hexamer.Individual separations between bound fluorophores (red circles) are: two 1st-order interfaces (D1, D2), a short diagonal (O1), a long diagonal (O2), and a 2 nd -order vertical (O3).D FLImP analysis of 100 synthetic pairwise separations for the hetero confhexamer (inset) (Supplementary Method 3).Sum of posteriors of individual separations between fluorophores (gray background).Probability distributions of individual components (peaks) of decomposed separation distribution (colored peaks).The area under each peak is weighted according to the estimated proportion of measurements attributed to that peak ("abundance").E As (D) for the homo-trimer formed by the Dimer 2 conformer.F Top, map of mutations and treatments superimposed on the structure of a B2B ect dimer sub-unit 62 , an Asym kin dimer sub-unit 3 , and an H2H kin dimer sub-unit 34,50 .Mutations are colored according to the effect they are expected to have on the different dimer conformers.Given the number of mutations, automating acquisition and data analysis was paramount (Supplementary Method 2).(Bottom) EGFR sequence diagram.G Comparisons between decomposed separation probability distributions between datasets.The continuous lines show the marginalized separation posterior for each condition, the sum of the abundance-weighted peaks in (D).The fluctuations around each continuous line arise from variations derived from FLImP decompositions for 20 bootstrap-resampled datasets to assess errors due to finite number of measurements.H Wasserstein MDS analysis.This measures the work needed to convert a decomposed separation set into another, thereby estimating similarities and differences between whole FLImP separation decompositions.Similarities or dissimilarities between the 21 separation sets of different conditions (one main FLImP decomposition plus 20 bootstrap-resampled decompositions) are compared; in this case for the hetero conf -hexamer (D) and homo-trimer (E).The plot axes are Component 1 (C1) and Component 2 (C2).C1 represents the dimension that captures the largest amount of variance in the data, while C2 represents the second-largest amount of variance that is orthogonal to C1.The ellipse centers (95% confidence range) mark the positions of the main FLImP decompositions.The crosses mark the positions of individual bootstrap-resampled separation sets.peaks).The median peak positions marked by dashed lines are those of WT-EGFR.C Wasserstein MDS analysis.Similarities or dissimilarities between the 21 separation sets of different conditions (one main FLImP decomposition plus 20 bootstrap-resampled decompositions) are compared.The plot axes are Component 1 (C1) and Component 2 (C2).C1 represents the dimension that captures the largest amount of variance in the data, while C2 represents the second-largest amount of variance that is orthogonal to C1.The ellipse centers (95% confidence range) mark the positions of the main FLImP decompositions.The crosses mark the positions of individual bootstrap-resampled separation sets.E Comparisons between decomposed separation probability distributions between datasets.The continuous lines show the marginalized separation posterior for each condition, the sum of the abundance-weighted peaks.The fluctuations around each continuous line arise from variations derived from FLImP decompositions for 20 bootstrap-resampled datasets to assess errors due to finite number of measurements.Note that the component that includes an 8 nm separation, corresponding to the St2St ect /Asym kin dimer sub-unit, is not significantly decreased by the H566F mutation.This is consistent with the tether-disrupting mutations not inhibiting phosphorylation 51 (Supplementary Fig. 3E).
(V 635 xxL 638 xxxL 642 ), is the only left-handed dimer.Important for our purposes, Lzip can establish contacts with Nter+ (Nter) and Cter+ (Cter) that could assemble oligomers.B Top, a speculative tetramer formed by two monomers interacting through two Lzip contacts with one Cter interface.Bottom, orthogonal projection on xy-plane.The speculative tetramer formed by two monomers interacting through two Lzip contacts with one Nter interface is shown in Supplementary Fig. 3F. C,   F, G FLImP analysis of 100 separation probability distributions between Affibody-CF640R pairs in the conditions indicated: Sum of posteriors of individual separations between fluorophores (gray background) and probability distributions of individual components (peaks) of decomposed separation distribution (Colored peaks).The median peak positions marked by dashed lines are those of WT-EGFR.D Comparisons between decomposed separation probability distributions between datasets.The continuous lines show the marginalized separation posterior for each condition, the sum of the abundance-weighted peaks.The fluctuations around each continuous line arise from variations derived from FLImP decompositions for 20 bootstrap-resampled datasets to assess errors due to finite number of measurements.E Wasserstein MDS analysis.Similarities or dissimilarities between the 21 separation sets of different conditions (one main FLImP decomposition plus 20 bootstrap-resampled decompositions) are compared.The plot axes are Component 1 (C1) and Component 2 (C2).C1 represents the dimension that captures the largest amount of variance in the data, while C2 represents the second-largest amount of variance that is orthogonal to C1.The ellipse centers (95% confidence range) mark the positions of the main FLImP decompositions.The crosses mark the positions of individual bootstrap-resampled separation sets.
clashes.More details in Supplementary Fig. 4. B-H Optimally grouped seven distinct triangles from 2D FLImP separations determined between EGF-CF640 probes bound to cell surface T766M-EGFR (Supplementary Fig. 5).The side lengths of each triangle are annotated in the colors used in 1D FLImP decompositions when found in the separation sets of either T766M-EGFR or WT-EGFR, and colored and underlined if found in both.The abundance of each triangle is also stated (Bottom right, black).I Ligand-free hetero conf -oligomer model extended as described in A. The approximate positions where EGF-CF640R would bind DIII of the ectodomains are marked (red circles).Four versions of the experimentally-optimized triangle 1 (T1a-T1d) (B) are superimposed.Table of the lengths of the superimposed triangle sides (inset).J Top, ligand-free hetero conf -oligomer model with the triangle groups T2-T4 (C-E) superimposed.Bottom, hetero conf -oligomer model with the largest triangles T5-T7 (F-H) superimposed.Inset, Table of the side lengths of the superimposed triangles.continuous line arise from variations derived from FLImP decompositions for 20 bootstrap-resampled datasets to assess errors due to finite number of measurements.C Structure and cartoon model of the Bb2Bb kin interface formed mainly through N-to-C lobe interactions (PDB ID 3VJO) 41 .A kinase monomer can dock in the Bb2Bb kin interface to form an Asym kin dimer sub-unit, as shown.The position of the A-loop in the active configuration of the Asym kin dimer sub-unit is marked.The K946E mutation breaks the Bb2Bb kin interface interface (Supplementary Fig. 7A, Supplementary Method 1) is shown.D, I Phosphorylation of the annotated mutant-EGFR in the absence of ligand in transfected CHO cells.E, H, K FLImP analysis of 100 separation probability distributions between Affibody-CF640R pairs in the conditions indicated: Sum of posteriors of individual separations between fluorophores (gray background) and probability distributions of individual components (peaks) of decomposed separation distribution (Colored peaks).The median peak positions marked by dashed lines are those of WT-EGFR.G Wasserstein MDS analysis.Similarities or dissimilarities between the 21 separation sets of different conditions (one main FLImP decomposition plus 20 bootstrap-resampled decompositions) are compared.The plot axes are Component 1 (C1) and Component 2 (C2).C1 represents the dimension that captures the largest amount of variance in the data, while C2 represents the secondlargest amount of variance that is orthogonal to C1.The ellipse centers (95% confidence range) mark the positions of the main FLImP decompositions.The crosses mark the positions of individual bootstrap-resampled separation sets.J Side and top-view of the Bb2Bb kin dimer sub-unit (described in 41 ) formed mainly through N-to-C lobe interactions (PDB ID: 3VJO) 41 .The inserted residues of the Ex20Ins mutation, which lies on the αC/β4 loop, are shown in spheres.Over the course of the simulation, one of the monomers of the insNPG-EGFR in the Bb2Bb kin interface is stabilized in the active conformation characteristic of the Asym kin dimer sub-unit, and the other samples it as well.

Plasmid construction & mutagenesis
Point mutations in EGFR plasmids were introduced using Quikchange Lightning Site-directed mutagenesis kit (Agilent Technologiescat.no.210518-5) using primer pairs listed in Supplementary Table 1.All constructs were verified by sequencing the whole coding sequence of EGFR.To generate EGFR PiggyBac plasmids, WT-EGFR was PCR amplified from WT-EGFR/pcDNA3 plasmid using primer pairs with NheI and NotI restriction enzyme sites (Supplementary Table 1) and inserted into PB513B-1 vector using standard molecular biology techniques.
Purification of mAb-108 from cell culture media mAb108 antibody was purified from mAb108 hybridoma cell culture supernatant purification using a mouse TCS antibody purification kit (Abcam, ab128749).Purified antibody was quantified using a Nanodrop and stored at 4°C.

Generation of EGFR CHO stable cell lines
CHO cells expressing one of the following were generated: WT-EGFR, ED/RK EGFR, T766M EGFR, T766M + K946E EGFR or T766M + I942E.CHO cells were seeded and transfected with a mix of 0.3 μg of Super Piggybac Transposase expression vector (PB210PA-1, System Biosciences) and 1 μg of the appropriate endotoxin-free EGFR plasmid DNA (in PB513B-1 vector, System Biosciences) using Fugene HD at 1:3 DNA:Fugene HD ratio according to manufacturer's instructions.The cells were selected in fresh media containing 4 μg/mL puromycin for 7-10 days.Surviving clones of cells were pooled and checked for EGFR expression by western blotting and confocal imaging.described above.Cells were then rinsed twice with 0.1% FBS pre-heated at 37°C and were labeled with a 1:1 mixture of 8 nM Affibody-Alexa 488 / Affibody-CF640R for 7 min at 37°C.Cells were rinsed twice with low serum medium pre-heated at 37°C and promptly imaged as described previously 34,63 .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 (for FLImP and single particle tracking) were initially analyzed using the multidimensional analysis software described previously 64 .The colocalization event duration analysis was performed in the same way as in 34 .

Confocal imaging
For all confocal experiments, cells expressing WT-EGFR, L680N-EGFR or ED/RK-EGFR were seeded, serum-starved for 2h as described above and rinsed twice in ice-cold PBS and cooled down for 10 min on ice.For anti-EGFR Affibody and EGF competition binding experiments, cells were then pre-treated with either 200 nM, 400 nM or 600 nM of ice-cold Affibody-CF640R in PBS or with mock treatment (PBS) for 1h at 4°C, rinsed with ice-cold PBS and fixed with 3% paraformaldehyde in PBS for 30 min at 4°C.After fixation, cells were rinsed again with RT PBS and labeled with 400 nM EGF-Alexa488 in PBS for 1h at RT, rinsed with RT PBS, then fixed with 3% paraformaldehyde + 0.5% glutaraldehyde for 15 min at RT.For the mAb-2E9 binding experiments, after starvation cells were then pre-treated with 200 nM of mAb-2E9-AF488 in PBS or with mock treatment (PBS) for 2h at 4°C, rinsed with ice-cold PBS and fixed with 3% paraformaldehyde in PBS for 15min at 4°C.For the EGF binding post-fixation test, cells were labelled with 200 nM EGF-CF640R for 2 h at 4°C then fixed.After fixation, cells were rinsed again with RT PBS and labeled with 200 nM EGF-CF640R in PBS for 1h at RT, then rinsed with PBS, fixed with 3% paraformaldehyde + 0.5% glutaraldehyde for 15 min at RT and rinsed with PBS.All samples were stored in PBS at 4°C until the time of acquisition, and allowed to pre-warm at RT, before loading on the microscope.Image acquisition was performed on an Elyra PS1, using Zen Black v2.3 SP1 using 633nm or 488nm laser excitation.Colocalization analyses were carried out on images of 600 μm optical slices and performed using Huygens software (Scientific Volume Imaging) For the mAb-2E9 binding experiments, pixel-wise intensity or intensity ratio distributions were extracted from the data using Huygens (SVI).The non-parametric Kruskal-Wallis statistical test was performed and T-test post-hoc analysis with Bonferroni multiple comparison correction was applied to calculate P Values in Python.

Mice tumor models
In this study, young adult male (6-7 weeks old, 24.6 ± 2.1 g) were used for all animal experiments (NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ mice mice, Charles River UK, Strain code: 614; RRID:IMSR_JAX:005557).All mice were maintained within the King's College London Biological Services Unit under specific pathogen-free conditions in a dedicated and licensed air-conditioned animal room.Male NSG mice were used to establish subcutaneous tumor models (in right flanks) with indicated stable Ba/F3 cell lines.After acclimatization, mice were randomly allocated into four cohorts with six individuals each, shaved on their flanks, and then subcutaneously received each 2 x 10 6 tumor cells suspended in 100 μL PBS.Tumor growth was followed by calipers and tumor volumes calculated.Tumor models were grown to compare tumor growth between cohorts.The experimental endpoint was defined by the time the humane endpoint was reached for the cohort with the largest tumor growth, and then all animals were sacrificed.

In vivo imaging of tumor models
In vivo GFP fluorescence imaging of superficial tumor models was performed to visualize tumor growth in some animals per group over time and to quantify tumor growth differences in all animals at the experimental endpoint.ROIs were manually drawn including the whole tumor (or the injection sites where no tumors were visible) and used to calculate the radiant efficiency.Prism software version 9 (GraphPad, La Jolla, USA) was used to calculate all statistical parameters as indicated.Generally, pvalues were calculated using significance levels of α = 0.05.In-text numbers indicate means of pooled data ± standard deviation (SD) unless otherwise stated.

Tissue staining and histologic analysis
Formaldehyde-fixed paraffin-embedded (FFPE) tissues were prepared using standard methods as described in 65 .Morphologic analysis of tumor tissues was performed on hematoxylin-and eosinstained sections.For antibody staining, sections were blocked (Dual Endogenous Enzyme Blocking Reagent, Dako, S200389-2) in 1% (w/v) BSA for 60 min at RT, incubated with primary antibodies (Anti-EGFR, D38B1, CST 4267; RRID: AB_2246311) at 4°C overnight and secondary antibody (2 µg/mL Anti-Rabbit Ig-HRP, Dako P044801 in TBS) for 60 min at RT.Samples were developed (using Liquid DAB+ Substrate Chromogen System, Dako, K3467) and counterstained with hematoxylin before mounting.Slides were scanned using a Nanozoomer (Hamamatsu, Japan) with images being analyzed and processed by ImageJ.

Fig. 2 :
Fig.2: The B2B ect /H2H kin dimer sub-unit underpins the formation of the St2St ect /Asym kin dimer sub-unit A, C-F, I FLImP analysis of 100 separation probability distributions between Affibody-CF640R pairs in the conditions indicated: Sum of posteriors of individual separations between fluorophores (gray background) and probability distributions of individual components (peaks) of decomposed separation distribution (Colored peaks).The median peak positions marked by dashed lines are hereafter also superimposed on all the FLImP separation diagrams to facilitate comparisons with WT-EGFR.B Left, cartoon of a labeled B2B ect /H2H kin dimer sub-unit showing two fluorescent anti-EGFR Affibody bound to the two DIII of the ectodomains.Right, a labelled St2St ect /Asym kin dimer sub-unit.Erlotinib binds the ATP pocket of the kinase.G, J Comparisons between decomposed separation probability distributions between datasets.The continuous lines show the marginalized separation posterior for each condition, the sum of the abundance-weighted peaks.The fluctuations around each continuous line arise from variations derived from FLImP decompositions for 20 bootstrapresampled datasets to assess errors due to finite number of measurements.H Wasserstein MDS analysis Similarities or dissimilarities between the 21 separation sets of different conditions (one main FLImP decomposition plus 20 bootstrap-resampled decompositions) are compared.The plot axes are Component 1 (C1) and Component 2 (C2).C1 represents the dimension that captures the largest amount of variance in the data, while C2 represents the second-largest amount of variance that is orthogonal to C1.The ellipse centers (95% confidence range) mark the positions of the main FLImP decompositions.The crosses mark the positions of individual bootstrap-resampled separation sets.
cohort with error bars representing standard deviation (SD).Differences were tested for significance using 2-way ANOVA (alpha=0.05and Tukey's multiple comparison correction).B IVIS imaging of all animals from A at the experimental endpoint (day 41) showing obtained epi-fluorescence signals from animal tumors.C GFP fluorescence signal quantification of tumors corresponding to animals in B. Statistical analysis by 1-way ANOVA with Tukey's multiple comparison correction; error bars are SD.D Harvested tumors from those animals with established tumors were photographed under (top) daylight and (bottom) blue light illumination to detect fluorescence from GFP.No tumors had established in the cohort that had received Ba/F3+WT cells.E Tumor weights from harvested tumors.Statistical analysis by 1-way ANOVA with Tukey's multiple comparison correction; error bars are SD.F Hematoxylin and eosin staining of tumors that did establish; scale bars are 100 μm.G Immunohistochemistry staining using a pan anti-EGFR antibody; corresponding background control staining is shown in Supplementary Fig. 8F.Scale bars are 100 μm.