Analysis of EGFR binding hotspots for design of new EGFR inhibitory biologics

Abstract

The epidermal growth factor (EGF) receptor (EGFR) is activated by the binding of one of seven EGF‐like ligands to its ectodomain. Ligand binding results in EGFR dimerization and stabilization of the active receptor conformation subsequently leading to activation of downstream signaling. Aberrant activation of EGFR contributes to cancer progression through EGFR overexpression/amplification, modulation of its positive and negative regulators, and/or activating mutations within EGFR. EGFR targeted therapeutic antibodies prevent dimerization and interaction with endogenous ligands by binding the ectodomain of EGFR. However, these antibodies have had limited success in the clinic, partially due to EGFR ectodomain resistance mutations, and are only applicable to a subset of patients with EGFR‐driven cancers. These limitations suggest that alternative EGFR targeted biologics need to be explored for EGFR‐driven cancer therapy. To this end, we analyze the EGFR interfaces of known inhibitory biologics with determined structures in the context of endogenous ligands, using the Rosetta macromolecular modeling software to highlight the most important interactions on a per‐residue basis. We use this analysis to identify the structural determinants of EGFR targeted biologics. We suggest that commonly observed binding motifs serve as the basis for rational design of new EGFR targeted biologics, such as peptides, antibodies, and nanobodies.

1. INTRODUCTION

1.1. Role of EGFR in cancer and EGFR antibodies in anticancer therapies

1.1.1. The epidermal growth factor receptor is an oncogenic driver of cancer

The epidermal growth factor (EGF) receptor (EGFR) is a receptor tyrosine kinase crucial for epithelial and epidermal tissue development and homeostasis (Seshacharyulu et al., 2012). EGFR mutations and overexpression are known drivers of cancers, including colorectal cancer (CRC) (Therkildsen et al., 2014), non‐small‐cell lung cancer (NSCLC) (Hayashi et al., 2022), head and neck squamous cell carcinomas (HNSCC) (Johnson et al., 2020), and glioblastomas (Li, Wang, et al., 2023). NSCLC and CRC are the most common of those cancers. In 2023, an estimated 238,340 and 153,020 Americans will develop lung cancer and CRC, respectively (Siegel et al., 2023). Approximately 85% of these lung cancer cases will be NSCLC cases, with 10%–15% of European populations and up to 50% of Asian populations harboring oncogenic EGFR mutations (Melosky et al., 2022; Sholl et al., ²⁰¹⁵; Zhang et al., ²⁰¹⁶). Over 90% of these mutations are located in the intracellular kinase domain (Lynch et al., 2004; Pao et al., ²⁰⁰⁴). For CRC, EGFR is overexpressed in 60%–80% of tumors (Oh et al., 2011; Pabla et al., ²⁰¹⁵), frequently resulting from increased gene copy number (Ooi et al., 2004). Increased EGFR gene copy number in CRC positively correlates with clinical success of EGFR targeted treatments (Algars et al., 2017; Jiang et al., ²⁰¹³; Moroni et al., ²⁰⁰⁵). Beyond CRC and NSCLC, glioblastomas frequently exhibit the EGFRvIII mutation and gene amplification, occurring in up to 40% of tumors (Gan et al., 2013; Liu et al., ²⁰⁰⁵). Finally, in HNSCC, EGFR is overexpressed in 90% of tumors and is associated with lower patient survival (Grandis & Tweardy, 1993).

EGFR targeted therapy has shown efficacy in improving patient outcomes. For example, the EGFR tyrosine kinase inhibitor, osimertinib, induces high response rates and median progression free survival of ~19 months in patients with advanced/metastatic EGFR‐mutated lung cancer (Soria et al., 2018); however, acquired osimertinib resistance via EGFR mutations and activation of alternative growth signaling pathways ultimately limits therapeutic efficacy (Papadimitrakopoulou et al., 2018; Ramalingam et al., ²⁰¹⁸). For patients with RAS/RAS wild‐type metastatic CRC, chemotherapy in combination with EGFR antibody cetuximab or panitumumab extends median survival by 2–4 months compared to chemotherapy alone (Biller & Schrag, 2021). However, antibodies induce resistance mutations and activation of alternative growth signaling pathways (Graves‐Deal et al., 2019; Li et al., ²⁰¹⁷; Zhao et al., ²⁰¹⁷). For locally advanced HNSCC, cetuximab is used as a radiation sensitizer for chemotherapy ineligible patients. Treatment with cetuximab and radiation increases the median overall survival (OS) from 29.3 to 49.0 months compared radiation treatment alone (Bonner et al., 2006). For recurrent or metastatic HNSCC, cetuximab is used in combination with platinum‐based chemotherapy and fluorouracil (Johnson et al., 2020; Li, Tie, et al., ²⁰²³). Combination with platinum‐based chemotherapy and fluorouracil increases median OS from 7.4 to 10.1 months (Vermorken et al., 2008). Unfortunately, current EGFR‐targeted treatments are only effective on subsets of EGFR‐driven cancers. TKIs are ineffective for CRC and antibodies are not used as first‐line treatments for NSCLC. For EGFR antibodies, the largest survival extensions are seen in patients ineligible for more efficacious chemotherapy regimens and when effective antibodies may only extend survival for a few months. Additionally, EGFR targeted therapy suffers from toxicity of wild‐type EGFR inhibition, presented through dermatologic issues including rashes (Lacouture, 2006) and adverse gastrointestinal events (Aw et al., 2018). However, EGFR's clinical significance coupled with ongoing challenges developing effective targeted therapies suggests the need to consider alternative therapeutic modalities in EGFR‐driven cancers.

1.1.2. Ligand‐induced conformational changes activate EGFR for downstream signaling

To generate alternative therapeutic modalities, we need to understand the molecular mechanisms of EGFR activation and inhibition. Prior to activation by endogenous EGFR ligands, EGFR exists predominantly as a monomer or inactive dimer. Upon activation, EGFR forms dimers, multimers, and higher‐order oligomers (Du et al., 2021; Huang et al., ²⁰¹⁶; Needham et al., ²⁰¹⁶). Endogenous ligands activate EGFR by binding the EGFR extracellular domain between domains I, which contains site 1, and III, which contains sites 2 and 3 (Figure 1). Ligand binding causes conformational changes that promote EGFR domain II dimerization of two EGFR monomers (Ogiso et al., 2002), domain IV dimerization, and formation of an N‐terminal juxtamembrane dimer characteristic of active EGFR (Arkhipov et al., 2013; Endres et al., ²⁰¹³; Jura et al., ²⁰⁰⁹). Juxtamembrane dimerization enables the formation of an active, asymmetric kinase domain dimer (Huang et al., 2021; Zhang et al., ²⁰⁰⁶). However, EGFR ligands differ in their ability to stabilize an active EGFR dimer (discussed in Section 2.1.2). Post activation, the tyrosine kinase domain initiates downstream signaling of growth pathways including the RAS, PI3K, and JAK pathways (Yarden & Sliwkowski, 2001; Zhao et al., ²⁰¹⁷). RAS and PI3K pathway activation drives cancer proliferation (Ciardiello & Tortora, 2008; Downward, ²⁰⁰³; Mizukami et al., ²⁰¹⁹).

FIGURE 1.

(Left) Depiction of EGFR in the inactive monomeric state adopting the tethered conformation prior to EGF induced dimerization of EGFR with another EGF bound receptor. The ectodomain of EGFR includes the ligand binding domain of EGFR while the tyrosine kinase domain is intracellular. Created with biorender.com. (Right) Cartoon depiction of active EGFR domains I–IV, and sites 1–3.

1.1.3. Limited clinical success of EGFR antibodies in cancer treatment

EGFR targeted biologics inhibit EGFR activation by preventing EGFR ligands from binding or preventing the conformational changes necessary for dimerization (Figure 2) (Li et al., 2005). Cetuximab and panitumumab, for example, occlude the EGF binding site. However, those antibodies are only effective for KRAS wild‐type metastatic CRC tumors (Biller & Schrag, 2021). Altogether, EGFR antibodies are applicable to only a subset EGFR‐driven cancers and are susceptible to resistance mutations (discussed in Section 2.3.5) that emerge in treatment of cancers including CRC (Bagchi et al., 2018; Dienstmann et al., ²⁰¹⁵; Sickmier et al., ²⁰¹⁶). Additional challenges that antibodies face clinically include difficulty penetrating the tumor microenvironment (Cai et al., 2020) and the potential to stimulate an immune response (Stas & Lasters, 2009); though, some evidence suggests that immune response activation by antibodies contributes to clinical efficacy in EGFR driven tumors (MacDonald & Zaiss, 2017). Overall, these data suggest that existing biologic design strategies need optimization or alternative biologics should be considered for EGFR driven cancer treatment.

FIGURE 2.

Monomeric EGFR exits in the tethered conformation. When EGF binds EGFR, EGFR undergoes conformational changes enabling the formation of an active EGFR dimer. Antibodies such as cetuximab and matuzumab bind tethered EGFR and prevent EGFR dimerization. This figure was created with PDB files 1YY9, 3C09, and 1IVO.

Development of EGFR targeted nanobodies presents an alternative strategy for EGFR inhibition. Nanobodies are derived from camelid antibodies by removing the Fc domain that links the two heavy chains. Unlike monoclonal antibodies that contain both light and heavy chains, camelid antibodies only have heavy chains and interact with antigens using three variable regions. Consequently, smaller nanobody size enables better tumor penetrance (Bannas et al., 2017). Additionally, nanobodies are faster and cheaper to generate than antibodies (Schmitz et al., 2013). However, nanobodies still share antibodies' limitations. While no EGFR nanobodies have undergone clinical trials, 7D12, EgA1, and 9G8 have received interest as EGFR targeted biologics (Hofman et al., 2008; Roovers et al., ²⁰⁰⁷; Roovers et al., ²⁰¹¹; Tintelnot et al., ²⁰¹⁹). Additionally scaffolding molecules including repebodies, leucine rich repeat scaffold proteins (Lee et al., 2015), and adnectins, human fibronectin type III domain derived scaffold biologics (Ramamurthy et al., 2012), have been explored for EGFR targeting. These biologics, like nanobodies, are scaffolding molecules smaller and with greater tumor penetrance potential than antibodies.

Finally, peptides are another class of biomolecule that has been explored for EGFR targeting (Williams et al., 2018). Small peptide size contributes to both reduced immune response risk (Ackaert et al., 2021; Harmsen & De Haard, ²⁰⁰⁷; Stas & Lasters, ²⁰⁰⁹; Wang et al., ²⁰²²) and better tumor penetrance ability than antibodies and nanobodies. Additionally, ease of synthesis and mutability of peptides can enable rapid testing. However, peptides are limited by lower interaction surface area and reduced half‐life times compared to antibodies and nanobodies. Though, cyclization and backbone modifications including D‐amino acids increase peptide plasma half‐life (Wang et al., 2022). Overall, multiple biologics with different properties have EGFR targeting potential.

1.1.4. EGFR binding interface analysis provides insight for design of new EGFR targeted biologics

Knowledge of key interactions across existing EGFR inhibitory binding interfaces will aid design of new EGFR targeted biologics. Here, we analyze biologics in the context of clinical efficacy, global interaction metrics, and residue level interactions. We apply the Rosetta macromolecular modeling software package to highlight per‐residue level interactions, using Rosetta energy units (REU) to quantify per‐residue binding contributions. Rosetta energy is meant to correlate with the folding free energy of proteins and favorable Rosetta interactions will have negative energy values (Alford et al., 2017; Vu et al., ²⁰²²). Rosetta calculations have previously been used to calculate the ΔΔG of a mutation (Barlow et al., 2018; Frenz et al., ²⁰²⁰). Here, we use Rosetta to calculate the free energy of folding for EGFR–biologic pairwise interactions observed in previously published crystal structures. This analysis enables us to parse EGFR–biologic interactions in a qualitative manner and examine similarities of known EGFR–biologic interfaces from an energetic basis. While this analysis is not as rigorous as an in silico mutagenesis study, we find that Rosetta energy has a correlation of 0.16 with experimental alanine scanning K _D values (Li et al., 2008; Schaefer et al., ²⁰¹¹; Schmiedel et al., ²⁰⁰⁸) and this correlation improves to 0.84 when lysine residues are removed (Figure 3), indicating that this analysis is useful for parsing the structure activity relationship of EGFR targeted biologics. Based on these data, this method may overestimate the importance of charged interactions, but is unlikely to miss important interactions. We hypothesize that the structure activity relationship of these biologics can inform the design of novel EGFR targeted biologics such as peptides, antibodies, and nanobodies.

FIGURE 3.

Plots of EGFR single residue alanine scan change in K _D data against the Rosetta interaction energy for the given residue in the binding complex. (a) Includes all single alanine mutations for duligotuzumab (Schaefer et al., 2011), matuzumab (Schmiedel et al., 2008), cetuximab and necitumumab (Li et al., 2008). (b) Excludes lysine residues from the analysis.

2. STRUCTURAL ANALYSIS

2.1. EGF–EGFR binding interface hotspots are conserved across endogenous EGFR ligands

2.1.1. Insight from the EGF–EGFR binding interface inspires EGF competitive inhibitors

Multiple EGFR targeted biologics have been studied pre‐clinically (Golam Kibria et al., 2021; Gong et al., ²⁰²¹; Oliveira et al., ²⁰¹⁰; Tintelnot et al., ²⁰¹⁹) and clinically (Ang et al., 2021; Clement et al., ²⁰²¹; Fayette et al., ²⁰¹⁶; Oh et al., ²⁰¹⁹; Patil et al., ²⁰¹⁹; Seiden et al., ²⁰⁰⁷; Syed, ²⁰²¹). Here, we review EGFR targeted biologics with available structural data and investigate their structure–activity relationships to identify potential EGFR binding hotspots. To enhance this understanding, we first identify EGF–EGFR interface interaction hotspots.

2.1.2. Critical EGF–EGFR interactions

The three sidechain–sidechain interactions conserved among other endogenous EGFR ligands have been experimentally characterized as critical for EGF–EGFR binding (Engler et al., 1992; Matsunami et al., ¹⁹⁹¹; Tadaki & Niyogi, ¹⁹⁹³). These interactions occur on domain III of EGFR across sites 2 and 3 (Figure 4). The interactions are the pi stacking interaction between Y13 of EGF and F357 of EGFR (−6.3 REU), the salt bridge between R41 of EGF and D355 of EGFR (−4.2 REU), and hydrophobic interactions between L47 of EGF an EGFR hydrophobic patch (L382, F412, A415, V417, and I438; −12.7 REU). The EGFR amino acid numbering in this manuscript is consistent with starting at the beginning of the mature polypeptide. Unsurprisingly, since high‐affinity EGFR ligands EGF and transforming growth factor‐α (TGFα) as well as low‐affinity ligands epigen and epiregulin (Freed et al., 2017; Garrett et al., ²⁰⁰²; Singh et al., ²⁰¹⁶) bind EGFR in similar manners (Figure 5), the interaction energies are similar as well (Table 1). Finally, we observe conserved hydrogen bonds between the EGFR Q384 sidechain and the backbone of endogenous EGFR ligands (−6.1 to −6.5 REU) (Figure 5). Mimicking any of these four conserved EGFR–ligand interactions may prove beneficial in design of novel EGF competitive biologics.

FIGURE 4.

The crystal structure of EGF bound to EGFR (PDB ID: 1IVO). Residues that make important binding interactions are shown. Dashed lines indicate hydrogen bonding and salt bridge interactions.

FIGURE 5.

Conserved interactions between EGFR and EGF (PDB 1IVO), TGF‐α (PDB 1MOX), Epigen (PDB 5WB8), and Epiregulin (PDB 5WB7). Site 3 interactions left. Site 2 interactions right. Residues that make important binding interactions are shown. Dashed lines indicate hydrogen bonding and salt bridge interactions.

TABLE 1.

The Rosetta energy breakdown of the energy contribution between EGF, TGF‐ α, epigen, and epiregulin with the EGFR residues that they interact with.

EGFR residue	EGF		TGF‐α		Epigen		Epiregulin
	Mean	SD	Mean	SD	Mean	SD	Mean	SD
L325	−1.75	6E−2	−3.34	2E−2	−3.47	2E−2	−2.8	5E−1
L348	−1.8	1E−1	−1.95	2E−2	−3.98	3E−2	−2.7	1E−1
V350	−2.27	2E−2	−1.881	7E−3	−1.797	9E−3	−2.6	3E−1
R353	‐	‐	−0.19	1E−2	−0.127	9E−4	−2.4	6E−1
D355	−4.16	2E−2	−5.75	2E−2	−4.992	8E−3	−4.2	2E−1
S356	0.097	3E−3	−3.71	2E−2	−0.129	2E−3	0.005	2E−3
F357	−6.29	1E−2	−5.04	1E−2	−3.48	2E−2	−6.37	5E−2
L382	−1.66	9E−2	−1.59	2E−2	−1.4	2E−1	−1.5	3E−1
Q384	− 6.08	6E−2	−6.34	1E−2	− 6.48	5E−2	−6.12	3E−2
H409	−2.56	8E−2	−4.45	8E−2	−1.22	5E−2	−4.36	6E−2
F412	−3.82	3E−2	−4.29	1E−2	−2.87	8E−2	−4.7	7E−2
A415	−1.32	2E−2	−1.543	3E−3	−1.15	1E−2	−1.34	4E−3
V417	−2.26	3E−2	−1.871	9E−3	−1.99	6E−2	−2.63	7E−2
I438	−3.62	6E−2	−3.575	6E−3	−3.07	3E−2	−3.31	4E−2
K465	−1.57	4E−2	−3.72	1E−2	−3.29	1E−2	−0.9	5E−1
I467	−3	1E+0	−0.16	2E−2	‐	‐	−0.002	2E−3

Note: The favorability of each interaction is colored with blue being favorable and white being neutral. A dash denotes no interaction. Mean and standard deviation (SD) values were determined for 20 relaxed structures.

While domain III–ligand interactions are highly similar, domain I–ligand interactions are influenced by the ability of different ligands to stabilize different EGFR active conformations. These different EGFR dimers have physiological consequences as well. For example, compared to TGF‐α, EGF better stabilizes the tips‐separated conformation of EGFR and the flexible domain IV leg of this conformation has higher resolution in the EGF–EGFR dimer (Figure 6). Stabilization of the tips‐separated conformation is a prerequisite of the active N‐terminal juxtamembrane dimer and leads to higher phosphorylation of the intracellular signal‐regulated kinase for EGF compared to TGF‐α (Huang et al., 2021). The difference in EGFR dimers is accentuated by the slight ligand binding pocket compression observed when TGF‐α binds EGFR (Huang et al., 2021). Still, both TGF‐α and EGF stabilize a symmetric EGFR dimer with a bent domain II conformation, whereas weak EGFR ligands epiregulin and epigen trigger formation of an asymmetric EGFR dimer with domain II conformational differences and domain I–ligand interaction differences (Freed et al., 2017). The epiregulin–EGFR dimer contains a bent and an unbent EGFR molecule, while the epigen–EGFR complex is a monomer with an unbent EGFR molecule (Figure 6). The conformational changes caused by epigen and epiregulin dramatically alter the dimerization equilibrium of EGFR resulting in EGFR dimers that are weaker and shorter lived than EGF‐induced dimers (Freed et al., 2017). However, the weakness of the EGFR dimers involving epigen and epiregulin cannot be fully correlated with their low affinity for EGFR, as amphiregulin, another ligand with low affinity for EGFR promotes substantial sEGFR501 dimerization (Freed et al., 2017). Additional data shows that the effect of stabilizing different EGFR conformations could additionally affect domain IV mediated multimerization of EGFR (Figure 6), such that different ligands could potentially alter multimerization levels (Huang et al., 2016).

FIGURE 6.

EGF and TGF‐α differ in their ability to stabilize the tips separated conformation of EGFR as illustrated by cryo EM studies (PDB 7SYD, 7SYE, 7SZ5, and 7SZ7). While EGF stabilizes a symmetric dimer (PDB 3NJP) dimer with EGFR, Epiregulin (PDB 5WB7) stabilized an asymmetric dimer with bent and unbent EGFR conformations, and Epigen (PDB 5WB8) forms very weak dimer with EGFR and crystallizes as a monomer with EGFR. For EGFR multimers, EGFR domain IV is the proposed multimerization interface (Huang et al., 2016).

A grand challenge in EGFR structural biology is resolving how dimerization and multimerization regulate the conformational landscape of the kinase domain. Like other protein kinases, EGFR displays a high degree of conformational plasticity, which is critical for carrying out different regulatory mechanisms and enzymatic functions (Tong & Seeliger, 2015). Active EGFR dimers have an asymmetric kinase domain dimer in which one kinase domain serves as the activator for the second signal receiving kinase domain (Zhang et al., 2006). For EGFR multimers, one proposed model suggests that kinase domains can serve as activators and receivers, enabling all but one kinase domain in the EGFR multimer to be active (Huang et al., 2016). While the relationship between the ectodomain and the conformational plasticity of the kinase domain is not fully understood, cetuximab has been shown to be effective against activating kinase domain mutations, such as L834R, that are found in colorectal cancer (Kim et al., 2020). The conformational plasticity of EGFR indicates that design of novel EGFR targeted biologics should consider multiple EGFR active conformations, especially given structural insights that indicate strong domain I interactions are not needed by EGFR activating ligands.

However, mimicking ligand–domain I interactions is a strategy for EGFR targeted biologics utilized by adnectin 1. Adnectin 1 is a biologic with 1.8 nM affinity for EGFR but has undergone no clinical trials (Ramamurthy et al., 2012). Additionally, the non‐inhibitory nanobody EgB4 binds domain I of EGFR at a location distinct from site 1 (Zeronian et al., 2022). However, its interactions do not carry insight for the design of novel inhibitory EGFR targeted biologics.

Excluding epigen, which only makes weak interactions with domain I, EGFR ligands and adnectin 1 have similar energetic binding profiles (Figure 7). They all make favorable interactions with the EGFR loop N12–G18 (Table 2). A key interaction with residues Q16–G18 is the formation of backbone–backbone hydrogen bonds with EGFR in a parallel beta sheet fashion (Figure 8). This hydrogen bonding network favors the observed sidechain–sidechain hydrophobic stacking with EGFR L17 by ligands and adnectin 1. Further this hydrogen bonding network extends to the EGFR T15 hydroxyl, which forms hydrogen bonds with both a conserved cysteine residue and the EGF E40 equivalent residue for all three EGFR ligands (−3.2 to −9.0 REU) and the sidechain of D77 for adnectin 1 (−6.8 REU) (Figure 8). Additionally, EGF, epiregulin, and adnectin 1 form favorable hydrophobic stacking interactions with L14 (−3.3 to −5.0 REU). TGF‐α has fewer hydrophobic amino acids near L14. Finally, TGF‐α and epigen position their backbones appropriately to form hydrogen bonds with the sidechain of N12 (−3.3 and −4.9 REU) (Figure 8).

FIGURE 7.

Domain I of EGFR colored by the per‐residue interaction energy with a variety of binders. Energy units are Rosetta Energy Units (REU). White represents no interaction. Red represents an unfavorable interaction. Blue represents a favorable interaction.

TABLE 2.

The Rosetta energy breakdown of the energy contribution between EGF, TGF‐α, epigen, epiregulin, and adnectin 1 (ADN1) with the EGFR residues that they interact with.

EGFR residue	EGF		TGF‐α		EPG		EPR		ADN1
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD
N12	−1.69	2E−2	−3.35	1E−2	‐	‐	−4.86	4E−2	−0.12	2E−2
L14	−5.6	2E−1	−2.56	9E−2	−2.06	1E−2	−5.43	8E−2	−3.33	2E−2
T15	−3.24	3E−2	−8.1	3E−2	−1.609	9E−3	−8.97	8E−2	−6.82	2E−2
Q16	−5.4	4E−2	−8.51	2E−2	−0.76	1E−2	−6	1E+0	−6.16	4E−2
L17	−4.68	2E−2	−4.78	1E−2	−0.57	3E−2	−3.72	4E−2	−3.49	1E−2
G18	−2.28	5E−2	−2.3	1E−2	−2	2E−2	−3.19	5E−2	−2.39	4E−2
R29	−2.3	7E−1	−4.25	3E−2	‐	‐	−0.3	3E−1	0.03	8E−3
L69	−4.93	1E−2	−4.11	2E−2	−2.25	2E−2	−1.9	1E−1	−5.06	2E−2
Y101	−0.698	4E−3	−3.64	4E−2	−0.61	4E−2	−3.09	2E−2	−3.56	4E−2

Note: The favorability of each interaction is colored with blue being favorable and white being neutral. A dash denotes no interaction. Mean and standard deviation (SD) values were determined for 20 relaxed structures.

FIGURE 8.

Interactions between EGFR domain I and EGF (PDB 1IVO), TGF‐α (PDB 1MOX), Epigen (PDB 5WB8), Epiregulin (PDB 5WB7), and Adnectin 1 (PDB 3QWQ). Residues that make important binding interactions are shown. Dashed lines indicate hydrogen bonding and salt bridge interactions.

Lastly, EGFR ligands and adnectin 1 interact with common, nearby hydrophobic residues (Figure 8). EGF, TGF‐α, epiregulin, and adnectin 1 form hydrophobic stacking interactions with L69 (−4.1 to −5.1 REU). Further, TGF‐α and epiregulin form hydrogen bonds with EGFR Y101, while adnectin 1 P51 stacks against Y101 (−3.1 to −3.6 REU).

Overall, adnectin 1 and EGFR ligand–domain I interactions are highly similar. Due to this similarity, it is unsurprising that adnectin 1 and EGF have similar shape complementarities of 0.71 and 0.70 respectively with domain I (Table 3). Although the interaction surface area between adnectin 1 and EGFR of 590 Ų domain I is lower than that of EGF 720 Ų, adnectin 1 retains a high affinity of 1.8 nM for EGFR (Table 3). Altogether, adnectin 1 highlights the potential to design domain I targeted biologics, particularly biologics that bind EGFR loop N12–G18.

TABLE 3.

Affinity, interaction surface area, and shape complementary for EGFR binders.

Binding domain	7D12 a	EgA1 a	9G8 a	CET b	PAN c	NEC d	DUL e	REP f	059‐152 g	MAT h	GC1118 i	ADN1 j	EGF k
	III	III	III	III	III	III	III	III	III	III	III	I	I	III
EGFR affinity (nM)	220	276	166	2.3	0.05	3.3	1.9	0.3	5.9	113	0.16	1.8	0.8
Interaction surface area (Å2)	705	701	636	882	840	932	788	1041	1149	758	649.35	590	720	730
Shape complementarity	0.68	0.65	0.67	0.71	0.68	0.69	0.72	0.63	0.62	0.62	0.803	0.71	0.70	0.71

^a Data from Schmitz et al. (2013).

^b Cetuximab data from Li et al. (2005).

^c Pantitumumab data from Sickmier et al. (2016).

^d Necitumumab affinity and surface area data from Bagchi et al. (2018). Shape complementarity calculated with Rosetta.

^e Duligotuzumab data from Schaefer et al. (2011).

^f Repebody affinity data from Lee et al. (2015). Interaction surface area and shape complementarity calculated with Rosetta.

^g Affinity data from Matsuda et al. (2018). Interaction surface area and shape complementarity calculated with Rosetta.

^h Matuzumab affinity data from Schmiedel et al. (2008). Interaction surface area and shape complementarity calculated with Rosetta.

ⁱ Data from Lim et al. (2016).

^j Adnectin 1 data from Ramamurthy et al. (2012).

^k Affinity data for cell bound EGFR from Lax et al. (1989). Interaction surface area data from Ogiso et al. (2002). Shape complementarity calculated with Rosetta.

2.1.3. Challenges to targeting EGFR based on endogenous ligand binding information alone

As highlighted by adnectin 1, analyzing EGFR binding hotspots in the context of biologics and endogenous ligands provides multiple mechanisms of targeting EGFR hotspots and overcomes challenges of targeting EGFR based on endogenous ligands information alone. For example, biologics may have difficulty mimicking certain EGF–EGFR interactions. Secondly, a biologic that is very similar to EGF may activate the receptor. Finally, EGFR binding interface analysis in conjunction with EGFR targeted biologics expands the number of hot spots to base new biologics on. In the following sections, we discuss the structure activity relationship of known EGFR domain III targeted biologics and evaluate this information in the context of the clinical relevance of those biologics and the structure activity relationship of EGF.

2.2. Site 2 binders interact with residues critical for EGF–EGFR binding

2.2.1. Structures of biologic bound EGFR provide multiple EGFR binding mechanisms

We examine known antibody/nanobody‐EGFR binding complexes to determine EGFR binding hotspots that can inform design of novel EGFR targeted biologics. These interfaces are analyzed in the context of previously characterized sites 2 and 3 of EGFR (Ogiso et al., 2002) (Figure 1). Additionally, the only crystalized site 1 inhibitory biologic, adnectin 1, is discussed above with the description of the site 1 EGFR ligand interactions (Ramamurthy et al., 2012). Site 2 is defined by EGFR residues D355 and F357. These residues protrude from a loop and neighbor the N328 glycosylation site (Figure 1). These site 2 features make identification of site 2 targeted biologics more difficult. Site 3 is defined by the hydrophobic patch that L47 of EGF interacts with and Q384. Site 3 is a flatter surface more attractive for biologic binding. While some biologics bind to neither site 2 nor site 3, many cluster around site 2 or site 3. Starting with site 2, we examine the EGFR biologic and ligand interface interaction hotspots, using an energy decomposition analysis to highlight the interface residues that contribute most strongly to complex stability. This energy decomposition is consistent with existing literature.

2.2.2. 7D12, GC1118, and 059‐152 bind EGFR residues critical for EGF binding

Nanobody 7D12, antibody GC1118, and antibody 059‐152 bind to site 2 of EGFR (Lim et al., 2016; Matsuda et al., ²⁰¹⁸; Schmitz et al., ²⁰¹³) (Figure 9). 7D12 and GC1118 interact primarily with site 2 and 059‐152 is unique as it is the only crystalized biologic forming widespread interactions with both sites 2 and 3. However, 059‐152 forms the weakest interactions with EGFR residues F357 and D355, residues critical for EGF binding (Table 4). 059‐152 forms a hydrogen bond with D355 (−3 REU) and minimally interacts with F357 (−2.3 REU) (Figure 10). Other observed F357 interactions are stronger, such as the EGF Y13 – F357 pi stacking interaction (−6.3 REU), the creation of hydrophobic pocket into which F357 inserts by GC1118 (−11.6 REU), and the use of the 7D12 R30 aliphatic chain to stack against F357 (−3.6 REU) (Figure 10). Though the 7D12–F357 interaction is the weakest of these, additional R30 utility comes from the salt bridge it forms with EGFR D355 (−3.5 REU) and experimentally the R30A mutation reduces EGFR binding more than 4000 fold (Schmitz et al., 2013). Still, the EGF R41–D355 salt bridge (−4.2 REU) and GC1118 Y59–D355 hydrogen bond (−4.9 REU) interactions (Figure 10) are both more favorable than the D355–7D12 interactions. Overall, these data indicate targeting D355 and F357 with separate amino acids is more favorable than use of a single arginine residue. Finally, biologic binding interface expansion highlights additional residues targeted by biologics.

FIGURE 9.

Domain III of EGFR colored by the per‐residue interaction energy with a variety of binders. Energy units are Rosetta Energy Units (REU). White represents no interaction. Red represents an unfavorable interaction. Blue represents a favorable interaction.

TABLE 4.

The Rosetta energy breakdown of the energy contribution between EGF, GC1118, 7D12, and 059‐152 with the EGFR residues that they interact with.

EGFR residue	EGF		GC1118		7D12		059‐152
	Mean	SD	Mean	SD	Mean	SD	Mean	SD
D323	‐	‐	0.001	7E−4	−1.6	6E−1	−3	2E+0
L325	−1.75	6E−2	−1.353	4E−3	−3.6	1E−1	−6.4	2E−1
L348	−1.8	1E−1	−0.044	9E−4	−3.15	2E−2	−1.3	1E−1
P349	−0.31	8E−3	−0.156	3E−4	−2.91	2E−2	−1	2E−1
V350	−2.27	2E−2	−1.717	4E−3	−4.55	2E−2	0.2	2E−1
R353	‐	‐	−2.3	2E−2	−7.71	6E−2	−2	1E+0
D355	−4.16	2E−2	−4.9	4E−2	−3.5	5E−1	−3	1E+0
S356	0.097	3E−3	−6.82	4E−2	0.043	7E−3	0	1E−3
F357	− 6.29	1E−2	−11.6	3E−1	−3.6	6E−1	−2.25	1E−2

Note: The favorability of each interaction is colored with blue being favorable and white being neutral. A dash denotes no interaction. Mean and standard deviation (SD) values were determined for 20 relaxed structures.

FIGURE 10.

The EGFR–GC1118 (PDB 4UV7), 7D12 (PDB 4KRL), and 059‐152 (PDB 5XWD) interfaces. Residues that make important binding interactions are shown. Dashed lines indicate hydrogen bonding interactions.

2.2.3. Biologics expand the EGFR binding interface

7D12, GC1118, and 059‐152 expand the EGF–EGFR protein–protein interface, highlighting polar, aromatic, and hydrophobic amino acids that EGFR biologic can target. For example, 7D12 residues E110, and D112 form salt bridges with EGFR R353 (−7.7 REU) (Figure 10). Experimentally, the D112A mutation reduces EGFR affinity approximately three fold (Schmitz et al., 2013). GC1118 D33 hydrogen bonds with EGFR S356 (−6.9 REU) and GC1118 Y37 pi stacks with EGFR H359 (−3.8 REU) (Figure 10). 059‐152 hydrogen bonds with D323 of EGFR (−3 REU). Finally, 7D12 and 059‐152 form favorable hydrophobic interactions with EGFR residues including L325, L348, and V350 (−3.2 to −6.4 REU). These data indicate that interactions with EGF critical binding residues and surrounding polar, aromatic, and hydrophobic residues are important for site 2 biologic design.

However, global interaction metrics demonstrate that targeting site 2 is challenging. For example, 7D12 and GC1118 are site 2 primary biologics and have interaction surface areas of 705 and 650 Ų respectively, but GC1118 has much higher shape complementarity to EGFR (0.80) than 7D12 (0.68) (Table 3). 059‐152 has lower shape complementarity (0.62) and a high interaction surface area (1149 Ų) (Table 3) most of which is in site 3. 059‐159 and GC1118 respectively have low and sub nanomolar affinity for EGFR (Table 3) while 7D12 has 220 nM affinity for EGFR. High EGFR affinity is only achieved by GC1118 via high shape complementarity and 015‐159 via extensive interactions extending into site 3.

The topology of site 2 and the data above allude to its limited clinical success. No site 2 EGFR targeted biologics are FDA approved. 7D12 and 059‐152 have undergone no clinical trials as EGFR targeted biologics. Another antibody thought to bind site 2, nimotuzumab (Tundidor et al., 2014), has undergone phase III clinical trials and improved progression free survival, but did not significantly improve OS (Patil et al., 2019). GC1118 is well tolerated and has good antitumor potential, but has only undergone phase I trials (Oh et al., 2019).

2.3. Site 3 centered binders expand the EGFR binding site beyond the hydrophobic patch bound by L47 of EGF

2.3.1. EGFR site 3 has clinical success

In contrast to site 2, multiple FDA approved EGFR antibodies bind site 3. Cetuximab, panitumumab, and necitumumab have undergone phase III clinical trials (Ciuleanu et al., 2018; Price et al., ²⁰¹⁶; Van Cutsem et al., ²⁰⁰⁹) and are FDA approved (Biller & Schrag, 2021; Garnock‐Jones, ²⁰¹⁶). A phase II clinical trial found that the EGFR, HER2 bispecific antibody duligotuzumab is comparable to cetuximab for HNSCC (Fayette et al., 2016). Further, the MET (N‐methyl‐N′‐nitroso‐guanidine osteosarcoma transforming gene) and EGFR bispecific antibody, amivantamab, binds to site 3 (Neijssen et al., 2021; Yun et al., ²⁰²⁰) and is FDA approved as a second‐line treatment for locally advanced or metastatic NSCLC harboring EGFR exon 20 insertion mutations (Syed, 2021), but has no EGFR bound, determined structure. Additional biologics with EGFR bound, determined structures, 059‐152 and a repebody, have undergone no clinical trials. As demonstrated by the numerous biologics that bind site 3, the relatively flat surface of site 3, compared to site 2, makes it a favorable location for biologics. Key site 3–EGF interactions include hydrogen bonds formed between EGFR Q384 and the main chain of EGF residues Q43 and R45 as well as the interaction of EGF L47 with an EGFR hydrophobic pocket (Figure 4).

2.3.2. Some antibodies bind to Gln384, an EGFR residue critical for EGF binding

EGFR Q384 is a residue that makes key interactions with both EGF and antibodies. Similarly to the backbone–sidechain hydrogen bonds EGF makes with Q384 (−6.1 REU) (Figure 11), the backbone carbonyl of 059‐152 M102 hydrogen bonds with Q384 (−3.4 REU) in the Rosetta relaxed structure (Figure 11). This hydrogen bonds requires flipping the Q384 amine crystal structure orientation (Matsuda et al., 2018). In contrast, panitumumab (−4.0 REU) and necitumumab (−2.0 REU) utilize a tyrosine sidechain to hydrogen bond with the amine of Q384 (Bagchi et al., 2018; Sickmier et al., ²⁰¹⁶) (Figure 11). For necitumumab, the Q384A mutation reduced binding by approximately 13 fold (Li et al., 2008). Though, no biologic utilizes Q384 as both a hydrogen bond donor and acceptor like EGF does. These data show that designed biologics could target Q384 and its proximity to the site 3 hydrophobic pocket suggests possibility of designing a biologic to interact with both features.

FIGURE 11.

The EGFR–cetuximab (PDB 1YY9), pantimumab (PDB 5SX4), necitumumab (PDB 3B2U), duligotuzumab (PDB 3P0Y), relaxed 059‐152 (PDB 5XWD), and repebody (PDB 4UIP) interfaces separated into overview, site 3 central and site 3 edge interactions. Residues that make important binding interactions are shown. Dashed lines indicate hydrogen bonding interactions.

2.3.3. Site 3 biologics make favorable interactions with the site 3 hydrophobic patch

The site 3 hydrophobic pocket that L47 of EGF binds into is a key EGFR feature for ligand and biologic binding. A415 of EGFR and surrounding hydrophobic amino acids, L382; F412; V417; and I438, form a pocket complementary in shape to EGF L47 (Figure 4). However, no biologic binds as deeply into this pocket as EGF does, as the higher interaction energies that biologics have with A415 demonstrate (Table 5). Though, biologics do reveal alternative targeting mechanisms to target this site with a variety of aromatic and aliphatic amino acids (Figure 11). Some of the most intriguing interactions with this hydrophobic pocket include targeting by cetuximab Y102, necitumumab I102 and F103, 059‐152 V103, and duligotuzumab Y56 (Figure 11). For duligotuzumab, the F412A mutation reduces binding by approximately 80 fold (Schaefer et al., 2011). In contrast, panitumumab V102 forms weaker interactions with this pocket (Figure 11) and the repebody makes minimal interactions with key features of the EGF–EGFR interaction (Table 5). However, all of these biologics make complex stabilizing interactions with a site 3 expanded protein–protein interaction interface.

TABLE 5.

The Rosetta energy breakdown of the energy contribution between EGF, cetuximab (CET), duligotuzumab (DUL), necitumumab (NEC), panitumumab (PAN), 059‐152, and the repebody (REP) with the EGFR residues that they interact with.

EGFR residue	EGF		CET		DUL		NEC		PAN		059–152		REP
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD
L382	−6.08	6E−2	−2	3E−1	0.05	2E−2	−2	1E+0	−4	6E−1	−3.4	3E−1	0	5E−4
Q384	−2.56	8E−2	−3.05	7E−2	−1.62	8E−2	−4.28	4E−2	−2.75	1E−2	−3	1E+0	−0	6E−3
H409	−3.82	3E−2	−2	3E−1	−3.3	1E−1	−2.61	2E−2	−0.43	2E−2	−1.99	2E−2	−0.27	2E−2
F412	−1.32	2E−2	−0.6	1E−1	−0.08	3E−2	−0.78	1E−2	0.001	2E−4	−1.1	2E−1	‐	‐
A415	−2.26	3E−2	−5	1E−1	−0.87	8E−2	−5	3E−2	−4.3	1E−1	−1.81	5E−2	−1.44	9E−3
V417	−0.05	5E−2	−0.43	9E−2	‐	‐	−5	3E−1	−2.4	3E−1	−0.6	1E−1	−0.26	3E−2
S418	‐	‐	‐	‐	−5.4	7E−1	‐	‐	‐	‐	‐	‐	0	4E−4
I438	−3.62	6E−2	−2.09	8E−2	−3.7	3E−1	−3.12	1E−2	−2.38	3E−2	−1.8	2E−1	−0.47	5E−3
S440	−1.1	2E−1	−2.42	6E−2	−0.61	5E−2	−2.05	6E−2	−0.7	6E−1	−0.9	3E−1	−1.97	3E−2
K443	‐	‐	−7.96	4E−2	‐	‐	−3.29	4E−2	−0.8	3E−1	−5.15	2E−2	0.02	8E−4
K454	‐	‐	‐	‐	−6.5	1E−1	‐	‐	‐	‐	‐	‐	−2	2E+0
K463	‐	‐	−0.05	9E−2	−5.4	5E−1	0.001	7E−4	0.003	2E−3	‐	‐	−0.8	5E−1
K465	−1.57	4E−2	−2.9	1E−1	−7.5	1E−1	−5.6	6E−2	−5	1E+0	−3.9	2E−1	−3	2E−1
I467	−3	1E+0	−7	1E−1	−4.3	2E−1	−6.5	1E−1	−7.9	4E−1	−2.5	1E−1	−5.01	6E−2
S468	−0.56	6E−2	−5.3	4E−1	−0.12	1E−2	−4.1	1E−1	−4.3	2E−1	−0.37	4E−2	−1.89	3E−2
N469	‐	‐	−4.9	3E−1	‐	‐	−3.5	1E−1	−4.2	1E−1	‐	‐	−1.39	4E−2
E472	‐	‐	0.3	2E−1	−0.01	1E−2	−0.03	2E−2	0.14	2E−2	‐	‐	−4.1	4E−1
N473	‐	‐	−0.9	2E−1	‐	‐	−0.62	5E−2	−0.01	1E−2	‐	‐	−5	2E+0

Note: The favorability of each interaction is colored with blue being favorable and white being neutral. A dash denotes no interaction. Mean and standard deviation (SD) values were determined for 20 relaxed structures.

2.3.4. Biologics expand the site 3 binding interface

Expansion of the site 3–EGF binding interface is observed for all site 3 biologics. Many site 3 biologics interact with residues in the segment K463 to N469 (Figure 11). For example, cetuximab, panitumumab, and necitumumab form similar backbone‐backbone hydrogen bonds with S468 and N469 (−3.5 to −5.3 REU) (Figure 11). Moreover, duligotuzumab, necitumumab, panitumumab, and 059‐152 utilize glutamate and aspartate residues to form salt bridges and hydrogen bonds with nearby K465 (−3.9 to −7.5 REU) (Figure 11). For duligotuzumab, the K465A mutation reduces binding by approximately 80 fold (Schaefer et al., 2011). Interestingly, the EGFR binding repebody utilizes Y186 to form a cation–pi interaction with K465. Finally, in this segment, except for the repebody, all site 3 biologics form extensive hydrophobic interactions with I467 (−4.3 to −7.9 REU) (Figure 11). For duligotuzumab, the I467A mutation reduces binding by approximately 3000 fold. Additional interactions of interest are polar interactions, such as cetuximab–EGFR K443 (−8.0 REU); necitumumumab–EGFR S418 (−8.0 REU); and duligotuzumab–EGFR D436 (−5.4 REU), K454 (−6.5 REU), and K463 (−5.4 REU) (Figure 11). Though, the duligotuzumab–EGFR K454 interaction is 4.6 Å in the crystal structure, so this is likely a weaker charged interaction (Schaefer et al., 2011). Additionally, the K443A mutation had little effect on cetuximab binding (Li et al., 2008) and the D436A and K463A mutations respectively reduced duligotuzumab binding by approximately 800 and 1000 fold (Schaefer et al., 2011).

Interestingly, the repebody–EGFR interactions differ from all other biologic–EGFR interactions due differences in interaction location on site 3 (Figure 9). Repebody–EGFR hotspot interactions are limited to the aforementioned Y188–K465 interaction and two hydrogen bonding interactions: repebody E119–EGFR E472 (−4.1 REU) and repebody Y95–EGFR E473 (−5 REU) (Figure 11). The interaction between the repebody and EGFR is intriguing because it seems to interact with fewer hotspots than other biologics.

All site 3 biologics have low or sub nanomolar EGFR affinity; though, their EGFR interaction surface area and shape complementarities vary (Table 3). For example, the repebody and 059‐152 have EGFR shape complementarities of 0.63 and 0.62 respectively and interaction surface areas greater than 1000 Ų. Meanwhile the other antibodies have EGFR shape complementarities of 0.66–0.72 and interaction surface areas less than 1000 Ų. These data indicate that at site 3, with lower interaction surface area as high shape complementarity is needed, but high affinity with a low shape complementarity can be achieved with a larger interaction surface area even if interactions are weaker.

Overall, site 3 biologic data demonstrate that for EGF competition, interactions with Q384 and the site 3 hydrophobic patch are interesting and can be supplemented with interactions to surrounding residues. The biologics described here display some mechanisms of targeting those residues. However, novel EGFR targeted biologics should additionally consider emerging EGFR ectodomain resistance mutations that induce resistance to current EGFR targeted biologics to avoid mutation‐induced resistance or to specifically address resistance mutations.

2.3.5. Antibodies are susceptible to ectodomain resistance mutations

Antibody treatment influences development of domain III resistance mutations. For example, the S468R mutation, observed in 16% of cetuximab treated patients and only 1% of panitumumab treated patients (Price et al., 2020), induces cetuximab but not panitumumab or necitumumab resistance (Bagchi et al., 2018). All three antibodies interact favorably with S468 (Table 5) and the larger light–heavy chain gap for panitumumab and necitumumab, compared to cetuximab, is hypothesized to enhance their position 468 mutation tolerance (Bagchi et al., 2018; Sickmier et al., ²⁰¹⁶). Additional resistance mutations, that emerge post EGFR antibody treatment include R427C, S440L, G441R, G441E, K443T, and I467M (Bagchi et al., 2018; Sickmier et al., ²⁰¹⁶). Except for R427, these resistance mutations occur on antibody epitopes and alter the binding site's chemical environment. As these resistance mutations are a pressing clinical issue, future EGFR targeted biologic design should aim to specifically counteract these mutations.

2.4. EGF non‐competitive inhibitors sterically impede EGFR dimerization

2.4.1. Matuzumab has a different mechanism of inhibition from EgA1 and 9G8

EgA1, 9G8, and matuzumab are unique in that they noncompetitively inhibit EGFR activation (Figure 12). Matuzumab prevents the conformational changes required for dimerization (Schmiedel et al., 2008), and nanobodies EgA1 and 9G8 both stabilize the inactive tethered conformation by binding between domains II and III of EGFR at a concave site inaccessible to flatter antibody paratopes (Schmitz et al., 2013). EgA1 and 9G8 have highly similar epitopes and matuzumab's epitope is distinct from other EGFR targeted biologics, but all three epitopes are more polar than that of EGF competitive biologics.

FIGURE 12.

Domain III of EGFR colored by the per‐residue interaction energy with a variety of binders. Energy units are Rosetta Energy Units (REU). White represents no interaction. Red represents an unfavorable interaction. Blue represents a favorable interaction.

In contrast to previous biologics, matuzumab interacts most favorably with a polar epitope including residues such as K454, T459, S460 G461, and K463 (−4.6 to −6.5 REU) (Table 6). EGFR residues K454 and K463 form hydrogen bonds with matuzumab residues D100 and D49 respectively. Additionally, matuzumab residue E50 forms hydrogen bonds with EGFR residues T459 and S460. Finally, matuzumab residue T31 forms a hydrogen bond with the backbone of EGFR residue G461 (Figure 13). Experimental data shows that the K463A and combined T459A/S460A mutations respectively reduce binding by approximately 140 and 250 fold. However, the K454A mutation has little effect on binding (Schmiedel et al., 2008). While these interactions are interesting, it is difficult to make any conclusions useful for rational design from these data beyond that a large biologic can target the matuzumab epitope.

TABLE 6.

The Rosetta energy breakdown of the energy contribution between matuzumab (MAT), EgA1, and 9G8 with the EGFR residues that they interact with.

EGFR residue	MAT		EgA1		9G8
	Mean	SD	Mean	SD	Mean	SD
M294	‐	‐	−3.5	3E−1	−0.7	3E−1
E306	‐	‐	−3.3	7E−1	0.1	3E−1
R310	‐	‐	−4.2	1E−1	−4.14	6E−2
K375	‐	‐	−0.63	6E−2	−2.5	5E−1
E400	‐	‐	−6.01	3E−2	−0.66	9E−2
I401	0.015	8E−4	−2.78	4E−2	−2.71	4E−2
R403	0.054	3E−3	−4.03	2E−2	−3.7	8E−1
R405	‐	‐	−3.97	2E−2	−4	1E−1
E431	−1.29	3E−2	−1.49	4E−2	−3.8	6E−1
K454	−6.2	1E−1	−0.17	4E−2	−1.4	5E−1
G458	−2.28	3E−2	−5.43	4E−2	−6.84	7E−2
T459	−5	1E+0	0.57	2E−2	0.38	1E−2
S460	−6.5	3E−1	−0.097	9E−3	−0.164	4E−3
G461	−4.1	8E−1	0.052	2E−3	‐	‐
K463	−5.4	1E−1	‐	‐	‐	‐

Note: The favorability of each interaction is colored with blue being favorable and red being unfavorable. A dash denotes no interaction. Mean and standard deviation (SD) values were determined for 20 relaxed structures.

FIGURE 13.

The relaxed EGFR (blue) and matuzumab (pink) interface (PDB 3C09). Residues that make important binding interactions are shown. Dashed lines indicate hydrogen bonding interactions.

9G8 and EgA1 similarly have a largely polar epitope compared to most EGFR biologics. The intersection of EGFR epitope residues that both nanobodies form highly favorable interactions with includes polar interactions with R310, R403, R405, and G458 (−4.0 to −6.8 REU) (Table 6). The similar nanobody epitopes derives from their highly similar sequences. Both complexes contain a backbone–backbone hydrogen bond between nanobody V2 and EGFR G458 and a nanobody T100–EGFR R403 hydrogen bond. Both D115 of 9G8 and D118 of EgA1 form salt bridges with EGFR R405 (Figure 14) and experimental mutagenesis of these residues reduces binding by more than 12 fold (Schmitz et al., 2013). Further, Y107 of 9G8 forms a pi–cation interaction with R310 of EGFR and D104 of EgA1 makes a hydrogen bond with R310. E115 of 9G8 makes an additional sidechain salt bridge with R403 (Figure 14). Differences in the nanobody epitopes include EgA1 making more favorable interactions with M294, E306, and E400, and 9G8 making more favorable interactions with K375 and E431 (Figure 14). Overall, both nanobodies primarily interact with domain III of EGFR and highlight an interface with EGFR targeting potential. This interface requires a convex paratope to target and polar interactions dominate its interface.

FIGURE 14.

The EGFR–9G8 (PDB 4KRP) and relaxed EgA1(PDB 4KRO) interfaces. Residues that make important binding interactions are shown. Dashed lines indicate hydrogen bonding interactions.

2.4.2. Noncompetitive EGFR inhibitors have had limited clinical success

Unfortunately, current noncompetitive inhibitors of EGFR have been clinically unsuccessful. Matuzumab underwent phase II clinical trials for ovarian and primary peritoneal malignancies and use as a second‐line treatment for NSCLC (Schiller et al., 2010; Seiden et al., ²⁰⁰⁷) and was unsuccessful. Additionally, Depatuxizumab an antibody–drug conjugate that binds to domain II of EGFR (Johns et al., 2004; Sivasubramanian et al., ²⁰⁰⁶) failed phase II clinical trials for glioblastomas (Clement et al., 2021). A limitation of 9G8, EgA1, and matuzumab has been their greater than 100 nM binding affinities (Table 3). For matuzumab, the combination of 758 Ų binding interface and low shape complementarity of 0.62 lead to lower EGFR affinity. 9G8 and EgA1 have similar issues with higher, but still moderate shape complementarities of 0.67 and 0.65 and smaller binding interfaces of 636 and 701 Ų respectively. While these biologics present interesting EGFR inhibitory mechanisms, their lower EGFR affinities show the challenges they face.

3. CONCLUSIONS

Current FDA‐approved, EGFR therapeutics that target the EGFR ectodomain are monoclonal antibodies. However, monoclonal antibodies are not effective for all EGFR‐driven cancers. When monoclonal antibodies are effective, they may only extend median overall survival by a few months and are susceptible to resistance mutations. These limitations suggest alternate EGFR targeted ectodomain therapeutics need to be explored. To this end, we analyzed the binding interfaces of all crystalized EGFR inhibitory biologics to identify the structural determinants of EGFR binding. In this analysis, we highlight binding motifs observed across multiple biologics and suggest they could be used as the basis of rational design for new EGFR targeted biologics. This rational design could explicitly include these motifs in computational or experimental library screening for molecules that stabilize them and expand upon their interactions. In addition to antibodies, exploring peptides, nanobodies, and other scaffolding molecules will expand the diversity of chemical properties and conformations of molecules for targeting EGFR.

4. METHODS

4.1. Protein structure preparation

The following PDB ID were used as starting structures: 1IVO (EGF), 1MOX (TGF‐α), 5WB7 (Epiregulin), 5WB8 (Epigen), 1YY9 (Cetuximab), 5SX4 (Panitumumab), 3B2U (Necitumumab), 3P0Y (Duligotuzumab), 3C09 (Matuzumab), 4UV7 (GC1118), 4KRO (EgA1), 4KRP (9G8), 4KRL (7D12), 5XWD (059‐152), 4UIP (Repebody) 3QWQ (Adnectin 1). For patitumumab, the EGFR N328D and N420D mutations were reverted to asparagine using PyMOL. For cetuximab, the S474K mutation was reverted to serine using PyMOL. All HETATM were removed from each structure. The Rosetta relax application (Conway et al., 2014; Khatib et al., ²⁰¹¹; Maguire et al., ²⁰²¹; Nivon et al., ²⁰¹³; Tyka et al., ²⁰¹¹) generated relaxed structures with the following representative command:

   main/source/bin/relax.default.linuxgccrelease ‐in:file:s EGF_bound.pdb ‐nstruct 20 ‐in:file:fullatom ‐ignore_zero_occupancy false ‐linmem_ig 10 ‐in:detect_disulf ‐relax:constrain_relax_to_start_coords true ‐coord_constrain_sidechains true ‐ramp_constraints false ‐out:prefix EGF_bound_1_ ‐out:file:scorefile scorefile.sc

Then, the round one minimum energy structure was relaxed with the following representative command:

   main/source/bin/relax.default.linuxgccrelease ‐in:file:s EGF_bound_1_EGF_bound_0001.pdb ‐nstruct 20 ‐in:file:fullatom ‐ignore_zero_occupancy false ‐linmem_ig 10 ‐in:detect_disulf ‐relax:constrain_relax_to_start_coords true ‐coord_constrain_sidechains false ‐ramp_constraints false ‐out:prefix EGF_bound_1_EGF_bound_0001_ ‐out:file:scorefile scorefile.sc

For relax round two, atom pair restraints were added to preserve interaction of GC1118 residue Glu56 with Arg353 and the interaction between 7D12 residues Arg30 with Asp355. Ref2015_cst weights were used for atom pair constraints.

4.2. Energy decomposition

The Rosetta energy breakdown application (Alford et al., 2017) was used on all 20 relaxed complexes from the second relax round to determine a per‐residue interface interaction energy decomposition (Vu et al., 2022) with the following representative command:

   main/source/bin/residue_energy_breakdown.bcl.linuxgccrelease ‐in:file:s EGF_bound_1_EGF_bound_0001_EGF_bound_1_EGF_bound_0001.pdb ‐out:file:silent EGF_bound_1_EGF_bound_0001_EGF_bound_1_EGF_bound_0001_eng_brk.out

For each complex, the mean and standard deviation of their EGFR per‐residue interface energy were calculated.

4.3. Surface area and surface complementarity calculations

Interaction surface area and shape complementarity were calculated with the Rosetta 3.13 InterfaceAnalyzerMover and ShapeComplementarity Filter respectively.

AUTHOR CONTRIBUTIONS

Claiborne W. Tydings: Conceptualization; writing – original draft; writing – review and editing; visualization; formal analysis; methodology. Bhuminder Singh: Writing – review and editing. Adam W. Smith: Writing – review and editing. Kaitlyn V. Ledwitch: Writing – review and editing. Benjamin P. Brown: Writing – review and editing; conceptualization; supervision. Christine M. Lovly: Writing – review and editing. Allison S. Walker: Supervision; writing – review and editing; conceptualization; funding acquisition. Jens Meiler: Conceptualization; writing – review and editing; supervision; funding acquisition.

Tydings CW, Singh B, Smith AW, Ledwitch KV, Brown BP, Lovly CM, et al. Analysis of EGFR binding hotspots for design of new EGFR inhibitory biologics. Protein Science. 2024;33(10):e5141. 10.1002/pro.5141