Variable radius photoproximity labeling of the EGFR interactome in glioblastoma using red-light

Abstract

Photocatalytic proximity labeling has emerged as a valuable technique for studying interactions between biomolecules in a cellular context, providing precise spatiotemporal control over protein labeling. One significant advantage of these methods is their modularity, enabling the exploration of cellular environments at varying labeling radii through the action of different probe-photocatalyst combinations. Despite these advances, fewer methods have been developed using red-light excitation, limiting the use of photoproximity labeling in more complex media such as tissues and animal models. Herein, we develop a variable-radius platform for proximity labeling under red-light excitation, utilizing a single catalyst and two distinct probe types. We first design a short-range labeling system by generating carbene intermediates from sulfonium diazo probes. This system is successfully applied on A549 cells to capture the interactome of epidermal growth factor receptor (EGFR) using a Cetuximab-Chlorin e6 conjugate. Benchmarking against established techniques indicates that this approach performs comparably to leading carbene-based proximity labeling methods. Next, we leverage the strong singlet oxygen generation (SOG) ability of Chlorin e6 to establish a long-range labeling system using aniline and hydrazide probes. EGFR directed chemoproteomics experiments reveal significant overlap with the short-range system, with the short-range approach capturing a subset of interactions identified by the SOG system, reflecting their distinct spatial labeling radii. Finally, we deploy our approach for the characterization of EGFR in resected human glioblastoma (GBM) tissue samples removed from distinct locations in the same tumor, representing the tumor’s infiltrating edge and its viable center, identifying several GBM specific interacting proteins that may serve as a launch point for future therapeutic campaigns.

Graphical Abstract:

INTRODUCTION

Visualizing interactions between biomolecules is critical for our understanding of cellular and molecular biology and has been enabling for the study of human disease.¹ In recent years proximity labeling has emerged as the premier method for realizing this goal, particularly in cases where interactions are scaffolded by lipids or nucleic acids that are disrupted by cellular lysis protocols.^2–5 Generically, these methods require an enzyme or catalyst localized to a protein of interest which can then activate small molecule probes in the immediate vicinity only upon the application of external stimulus, leading to localized tagging of proteins and nucleic acids for downstream analysis.

New chemical technologies have propelled this emerging field forward, with significant advances utilizing visible light catalysis and novel probe designs to deliver innovative new protocols for labeling proteins at the cell surface^6–10, across cellular junctions^6,11, within organelles,^12–14 and at chromatin^15–17. The popularity of these photocatalytic proximity labeling methods has stemmed from the spatiotemporal control and relative flexibility in the catalysts and probes used, including several platforms that can achieve proximity labeling at variable radii only by changing the probe structure.^11,18 Such platforms provide significant value to end users, allowing multiple data sets to be generated from a single biological construct coupled with different commercial probes.

The use of red light (long-wavelength) further enhances the ability to study these interactions in complex biological settings. Red-light excitation offers distinct advantages in biological research, including deep tissue penetration, minimal phototoxicity, and compatibility with intricate biological environments.¹⁹ This makes it particularly valuable for investigating protein interactions in solid tumors, where both short-range interactions (for high-resolution mapping of protein complexes) and long-range interactions (such as cell-cell communication) offer critical insights into the tumor microenvironment.

To date, three reports have described red-light based proximity labeling methods. MacMillan²⁰ and Rovis²¹ independently demonstrated aryl azide activation using organic and osmium catalysts respectively, and Rovis and Shah further extended their system to the activation of aryl diazo probes.²² These innovations have set the stage for proximity labeling in tissue but suffer from several drawbacks. Specifically, aryl azide activation leads to very low labeling intensities, even after long irradiation times of up to 20 minutes. While this is acceptable in cellular models, long labeling times complicate application to in vivo systems. Furthermore, the aryl diazo activation reported by Rovis and Shah is proposed to bear a restricted labeling radius, in line with the µMap technology reported by MacMillan,^6,18 but this has yet to be established via chemoproteomics analysis.

To advance red-light-based photoproximity labeling in more complex biological systems, we identified several key technological requirements: (1) a deep-red light method capable of achieving variable labeling radii using a single photocatalyst, and (2) a method that generates sufficient signal with short irradiation times to minimize the risk of overheating and oxidative damage to biological samples.

Herein, we describe the development of a variable radii proximity labeling platform (PLP) that operates under red light irradiation. Considering the challenges involved, we began our study by developing a red-light based method for nanoscale proximity labeling via a carbene intermediate.

RESULTS AND DISCUSSION

Development of a short radius red light probe

Carbene intermediates are highly reactive and short-lived, with a half-life of 1 ns and a labeling range of around 4 nm, as shown via super resolution microscopy¹⁸, making them ideal for short-range labeling applications. While diazirines have been shown to be powerful crosslinkers due to the high reactivity of the generated carbenes however, the high triplet energy (60 kCal/mol) of these species makes activation with red light (42 kCal/mol) impractical. We considered diazo species as alternate carbene precursors that may be sensitive to red-light based activation modes.^23,24 Diazo compounds are common precursors to singlet and triplet carbenes and have been shown to be activated through energy transfer, electron transfer and through direct irradiation with low wavelength light.^24,25

One challenge associated with using diazo species as opposed to more stable diazirines, is their intrinsic reactivity with acidic groups via protonation at the α-position, leading to decomposition and high background labeling^26–29. Cognizant of this, we questioned whether the introduction of sulfonium groups at the α-position may simultaneously increase photo reactivity while reducing unwanted protonation via ylide stabilization of the α-anion³⁰ (Figure 1). Furthermore, prior work from Gaunt and coworkers shown these motifs to be biocompatible in the context of protein bioconjugation.³¹

Figure 1:

Photocatalytic activation of sulfonium diazo compounds under red light and their labeling efficiency in cell lysate. (A) Influence of sulfonium diazo compound structure on chemical reactivity. (B) Screening of photocatalysts. (C) Scheme for photocatalytic labeling in cell lysate. (D) Probes used in this study with biotin enrichment handle. (E) In vitro cell lysate labeling efficiency comparison of probe 1 vs probe 2. (F) Labeling dependence on time (G) Labeling dependence on photocatalyst concentration. (H) In vitro cell lysate labeling efficiency comparison of current method vs. previous work.

To test this hypothesis, we synthesized a range of sulfonium-bearing diazo compounds and assessed their reactivity with the red-light photocatalyst Chlorin e6 (Ce6). After 30 minutes of irradiation in DMSO with a commercial 660 nm LED array (see Figure S1 for Reaction Setup), we observed the conversion of both diazo esters and ketones bearing sulfonium groups, while no reactivity was detected with an unsubstituted diazo ester (Figure 1A). More electron-deficient diazos exhibited increased reactivity, although these modifications affected the stability of the probe (see Figure S2 for discussions on the structure-stability relationship of sulfonium diazos). Increasing the steric bulk around the sulfonium group had little effect on the reactivity in vitro. A survey of other red-light dyes and photosensitizers demonstrated that the chlorin scaffold was uniquely effective, with only the related natural product pheophorbide A (Pha) (Figure S3) showing comparable reactivity (Figure 1B, see Figure S4 for discussion of reactivity trends).

To confirm that our system could be applied to photocatalytic protein labeling, we synthesized biotinylated diazo probe 1 and tested their labeling efficiency in HEK293T cell lysates (Figure 1C). Irradiating for 10 minutes with 250 µM of probe 1 and 25 µM of Ce6 resulted in excellent labeling efficiency (Figure S5). Minimal biotinylation signal was detected when using protoporphyrin IX (PPIX) as the photocatalyst, which aligns with our chemical reactivity tests. Both Pha and Ce6 performed similarly in cell lysates at 25 µM, but Ce6 demonstrated better labeling efficiency at lower concentrations (2.5 µM) (Figure S6), likely due to the lower solubility of Pha. We chose Ce6 for this study for its lower cost and higher efficiency at low concentrations.

Control reactions in the absence of a photocatalyst showed minimal background labeling with probe 1, while the more reactive ketone probe exhibited some background labeling (Figure S7). Ethyl substituted diazo 2 (Figure 1D) was marginally more effective than the methyl substituted version, achieving a maximum signal-to-noise ratio of 15:1, so was taken forward as our probe of choice (Figure 1E). Controls experiment shows the labeling intensity of our system is in proportion to irradiation time, photocatalyst concentration and light intensity (Figure 1F&G, Figure S8). Comparison of the labeling efficiency and signal to noise versus the azide-based method, µMap-Red²⁰, showed a three-fold increase in signal under identical conditions (Figure 1H). We also tested other diazo probes with various modifications; however, only diazo probes with sulfonium substitutions led to productive labeling, demonstrating the unique effect of sulfur substitution (Figure S9).

Having established the feasibility of our photocatalytic PLP we were interested in gaining a deeper mechanistic understanding of this system. Of particular interest was the reactive intermediate responsible for labeling and whether the labeling reaction is dependent on a probe-photocatalyst interaction as proposed in many other PLP.

Performing the reaction in solvents DMSO and MeCN showed adducts reminiscent of carbene reactivity, as previously observed (Figure 2A).^32,33 Stern-Volmer analysis between probe 1 and Ce6 showed non-linear quenching, suggesting that different protonation states of the photosensitizer exhibit varying quenching rates with probe 1 (Figure 2B). To rule out a labeling mechanism that could proceed through a photodegradation product of probe 1, we conducted an in vitro labeling experiment with bovine serum albumin (BSA) (Figure 2C). We measured the biotinylation of BSA by western blotting under three conditions: (A) probe and light only, (B) all components (probe, Ce6, light, and BSA), and (C) a pre-irradiation experiment where the probe and catalyst were irradiated at 660 nm for 10 minutes before BSA was added. Labeling was only observed when all components were irradiated together, consistent with our hypothesized mechanism and previous reports.^20,22

Figure 2:

Mechanistic studies and in vitro labeling site analysis. (A) Carbene product detected during chemical conversions. (B) Stern-Volmer quenching experiment. (C) Control experiment demonstrating that the photodegradation product of Probe 1 is not a labeling-active species. (D) Mass modifications observed in peptides. (E) Peptides modified of BSA during in vitro labeling (F) Labeling sites on BSA. (G-I). Representative MS2 spectra with peptide modifications.

To determine the labeling preferences of our probe, we treated recombinant BSA with probe 1, Ce6, and light and performed MS2 analysis to identify cross-linked residues. An open search common mass adducts found two identifiable masses corresponding to carbene type insertions flowed by either protonation or hydrolysis (Figure 2D). Statistical comparison of these adducts against an uncatalyzed control provided a list of 6 highly enriched sites of labeling that include both protic amino acids (E588, T569, S296) and carbogenic amino acids (A385, V570, L139) (Figure 2E&F) and aromatic amino acids (Y184) (Figure S10) in addition to basic residues that met less stringent cut-offs (K256, H129) (Figure S11). Taken in aggregate, these data strongly support a mechanism where a photoexcited catalyst activates our diazo probe to a carbene intermediate that can subsequently label proteins in a residue agnostic manner.³⁴

We currently cannot discriminate between energy and electron transfer for the diazo activation step, although the electronic requirement of the diazo molecule may favor two subsequent electron transfer steps over direct energy transfer (For further discussion on the reaction mechanism, see Figure S12).

Antibody based interactomics of EGFR

While we have demonstrated that our PLP is highly effective at labeling in vitro and in cell lysate, the most valid test of any PLP is to assess accuracy and selectivity of the tool via chemoproteomics. Across a series of recent studies, the cell surface labeling of epithelial growth factor receptor (EGFR) has become a benchmark for proximity labeling methods because of its well characterized interactome^11,18,20, and reliable and affordable antibodies. We chose to directly conjugate Ce6 to the immunotherapeutic antibody Cetuximab³⁵ (Ctx) using NHS ester ligations which has been shown to be effective in prior studies. We synthesized an NHS ester linked Ce6 molecule with a pendent sulfonic acid to increase the solubility and subsequent stability of the antibody-dye conjugate. Antibody conjugation was performed at 4°C for two hours and the conjugate Ctx-Ce6 was purified by size exclusion chromatography (Figure 3A). We then characterized the conjugate by bicinchoninic acid assay (BCA) to measure protein concentration and fluorescence to measure dye concentration, where comparison of the two values revealed a catalyst to antibody ratio of 4.5 (Figure S13). Successful conjugation of the photocatalyst to the antibody was further confirmed by SDS-PAGE, where luminescence bands corresponding to the molecular weights of both the heavy and light chains of Cetuximab were observed (Figure 3B). To ensure that the Ce6 modification did not affect Cetuximab’s binding affinity, we used flow cytometry to compare the binding of Ctx-Ce6 and unmodified Cetuximab in A549 cells. The results confirmed that Ctx-Ce6 retained its ability to bind EGFR, showing that the conjugation did not impair its function (Figure 3C).

Figure 3:

Cetuximab-based cell surface EGFR labeling using diazo probes. (A) Preparation of cetuximab-Ce6 conjugate. (B) Confirmation of photocatalyst conjugation to the antibody via SDS-PAGE. (C) Assessment of cetuximab-Ce6 binding to A549 cells using flow cytometry. (D) Western blot analysis of EGFR cell surface labeling and protein enrichment. (E) Quantitative proteomics: volcano plot of enriched proteins, with red data points representing known EGFR interactors from the BioGrid database and blue data points representing membrane protein. (F) Subcellular localizations of enriched proteins. (F) Gene ontology (GO) analysis of the enriched proteins. (G) Comparison of the diazo dataset with previously published carbene-based EGFR interactomes. (I) DepMap co-dependence of EGFR with observed interactors. (J) Proximity ligation assay confirming interactions between NEO1 and EGFR: representative images and quantification.

As photosensitization is a common method of promoting apoptosis, we wanted to assess whether the conditions employed in our method were toxic to the cells under study. We exposed A549 cells to our labeling conditions, antibody only, or DMSO, and measured proliferation after 12 and 24 hours. We found no change in cellular growth at these time points under any condition, showing that our protocol displays minimal toxicity (Figure S14).

We then evaluated the labeling ability of Ctx-Ce6 using diazo probes on A549 cells via flow cytometry, which resulted in a strong biotinylation signal. Control conditions, including the use of an undirected IGG-Ce6 conjugate, showed weaker labeling intensity (Figure S15). In addition, performing the reaction in the presence of excess epidermal growth factor (EGF), which competitively blocks the binding of the antibody-catalyst conjugate to EGFR, also reduced the biotinylation signal (Figure S11). Subsequently, we assessed the selective labeling of EGFR in A549 cells via western blot (Figure 3D). Labeling reactions showed substantial enrichment of EGFR after streptavidin pull-down. EGFR labeling was abrogated when using an undirected IGG-Ce6 conjugate or when the reaction was performed in the presence of excess EGF, consistent the flow cytometry data. These results demonstrate the applicability of our red-light-activated protein labeling system in live cell systems.

Following this western blot validation, we measured the EGFR interactome using our red light PLP. We treated A549 cells for 30 min with Ctx-Ce6 or IGG-Ce6 before irradiating for 1 minute in the presence of probe 2 (100 µM). All experiments were performed with three biological replicates with two technical replicates each and analyzed on a TIMS-Tof Pro2 mass spectrometer in DIA-PASEF mode.

Our protocol provided 59 enriched proteins (Log2FC>0.5, -LogP>3) (Figure 3E), of which 46 (78%) were membrane proteins (Figure 3F), and 20 were known EGFR interactors (from BioGRID database³⁶). Gene ontology (GO) analysis of the enriched proteins revealed significant enrichment in biological processes related to EGFR function and the cellular membrane, such as cell-cell adhesion and cell morphogenesis (Figure 3G). Comparing our dataset with previously published carbene-based EGFR interactomes was informative. µMap captured 20 proteins, with 50% being established EGFR interactors¹⁸, MultiMap captured 72 proteins with 21% being established interactors¹¹, and our method captured 59 with 33% being previously annotated (Figure 3H). AlphaFold Multimer prediction of all novel hits showed 27 out of 37 interactions were classified as high confidence based on prediction scores (Figure S16).

Interactions between EGFR and surface proteins correlated with EGFR expression may provide options for treatment of EGFR-overexpressing cancers. To assess this, we searched our list of enriched genes for co-dependencies in the pendent in cancer. Four of these were previously annotated as EGFR interactors, and one hit, NEO1³⁷ (Figure 3I), is not known to interact with EGFR (per BIOGRID) and was not identified in the µMap and multimap datasets. We validated this novel interaction using proximity ligation (Figure 3J), AlphaFold-multimer predication and co-IP (Figure S16), demonstrating a physical relationship between NEO1 and EGFR in addition to the existing functional relationship.

These analyses indicate that our PLP can provide proximity labeling data comparable to established carbene based methods, generate novel leads for biological evaluation, and highlights the variability between even very similar interactomics techniques.

Long radius probes based on singlet oxygen generation

With a short radius labeling method in place, we next turned our attention to the development of a PLP that exhibits a longer labeling radius. Many recently reported PLP function via the generation of singlet oxygen which has a half-life of approximately 4 microseconds and a labeling radius of 30 nm³⁸. Singlet oxygen labels proteins by oxidizing histidine residues into 2-oxo-histidine, which can subsequently undergo nucleophilic attack by anilines and hydrazides.^39–42 These protocols have been shown to function through both blue and green light irradiation but have yet to be achieved with red light. As blue light produces oxidative damage and background labeling by itself⁴³ (Figure 4B), a red-light approach would be particularly valuable. We reasoned that Ce6 may produce sufficient singlet oxygen to facilitate labeling through this mechanistic paradigm.⁴⁴ In principle, this mechanism should provide significantly enhanced signal over carbene based methods, enabling general PL in tissue samples where light penetration is restricted.

Figure 4:

Singlet oxygen based photocatalytic proximity labeling. (A) Probes used in this study. (B) Evaluation of blue light-induced oxidative damage and background labeling. (C) Ce6 cell lysate labeling using common PL probes. (D) In vitro comparison of labeling efficiency: singlet oxygen probes vs. previous methods using recombinant carbonic anhydrase. (E) In vitro comparison of labeling efficiency: singlet oxygen probes vs. previous methods in cell lysate. (F) Western blot analysis of EGFR cell surface labeling and protein enrichment using the aniline probe. (G) Western blot analysis of EGFR cell surface labeling and protein enrichment using the hydrazide probe. (H) Quantitative proteomics: volcano plots of enriched proteins in labeling experiments with hydrazide and aniline probes. Red data points represent known EGFR interactors from the BioGrid database and blue data points represent membrane proteins. (I) Overlap of EGFR interactomes identified through singlet oxygen labeling with BioGrid database interactors. (J) Comparison of long and short radii EGFR interactomes.

To test this, we examined the reactivity of singlet oxygen probes—biotin-aniline 6 and biotin-hydrazide 7 in cell lysate and compared their labeling efficiency to other commonly used PL probes (azide 3, diazirine 4, phenol 5,). As expected, no reactivity was observed with the diazirine probe 4, minimal labeling occurred with the azide probe 3, and strong protein biotinylation was detected with both singlet oxygen probes (Figure 4C). Moderate labeling was also observed with phenol probe 5, which is consistent with previous reports⁴⁵. This comparative experiment strongly supports a singlet oxygen-based labeling mechanism. Crucially, the use of both aniline and hydrazide probes provided an 8-fold increase in labeling intensity over the same period compared to azide-based approaches, as demonstrated in the labeling of recombinant proteins (Figure 4D) and cell lysate (Figure 4E).

To validate the applicability of this labeling system under cellular conditions, we performed same EGFR labeling experiment on A549 cells. Flow cytometry revealed a strong biotinylation signal when Ctx-Ce6 was used, which was ablated in IGG-Ce6 and EGF control samples (Figure S17 and Figure S18). Ctx-Ce6 directed labeling of EGFR was also successful for both probes, providing 9-fold enrichment by western blot (Figure 4F and Figure 4G). In contrast, no biotinylation signal was detected by flow cytometry nor enrichment of EGFR in pulldown assays when azide probe 3 were used (Figure S19 and S21), which is consistent with the in vitro observation that azide 3 is not activated by Ce6. Interestingly, although phenol-biotin 5 exhibited moderate labeling in cell lysate, no biotinylation signal was detected in the flow experiment of Ctx-Ce6 labeling in A549 (Figure S20). Similarly, EGFR enrichment was not observed when phenol-biotin probe 5 were used (Figure S21), indicating a significant difference between in vitro and in vivo systems.

We then examined the EGFR interactome using aniline and hydrazide probes. After 30 seconds of red-light irradiation, the proteomic analysis showed similar EGFR enrichment (aniline: 2.17 Log2FC, hydrazide: 2.69 Log2FC), with each experiment labeling 433 and 538 proteins, respectively (Figure 4H). These results support the notion that these probes offer a longer labeling radius compared to carbene-based approaches. Both experiments demonstrated strong enrichment of membrane proteins, with 86% and 79% of the enriched proteins being membrane-localized for aniline and hydrazide, respectively (Figure 4I). Comparison with annotated EGFR interactors revealed good overlap, with 28% of aniline and 30% of hydrazide-enriched proteins being previously annotated EGFR interactors (Figure 4J).

Performing the labeling experiment with 5 minutes of irradiation resulted in a significant loss of selectivity for EGFR, while the fold change for EGFR remained unchanged (Figure S22), emphasizing the importance of careful optimization in chemoproteomic studies of new PLP systems. Comparison of the three datasets (diazo, aniline, hydrazide) revealed that the proteins enriched by the diazo probe represented a subset of those identified by the aniline and hydrazide probes, consistent with our proposed labeling mechanism.

Interestingly, analysis of proteins uniquely captured by our short radius probes identified BASP1, an established interactor of EGFR in metastatic lung cancer⁴⁶, only using the diazo probe. Sequence analysis of BASP1 shows no histidine residues in the protein structure (Figure S23), rendering it unreactive to singlet oxygen-based approaches, but susceptible to carbene reactivity, highlighting the differences and strengths of both approaches.

Proximity labeling of EGFR in primary glioblastoma tissue

GBM is the most aggressive brain tumor, with a median survival time of 15 months. The GBM tumor microenvironment is notoriously heterogenous with each tumor displaying many cellular phenotypes, genetic mutations, and cellular drivers of growth, making treatments challenging. EGFR is broadly implicated in driving GBM proliferation with 30–60% of tumors overexpressing this gene. Anti-EGFR therapies have been trialed extensively but have been ultimately unsuccessful due to poor penetration of the blood brain barrier, toxicity, and drug resistance.⁴⁷ One strategy to overcome these challenges in the context of immunotherapies and antibody-drug conjugates (ADCs) is to employ bispecific antibodies that target two neighboring proteins on the cell surface that only interact in the cancer context.⁴⁸ In principle, this approach can increase cancer specificity through avidity and combat resistance by targeting two separate genes. Critical to this approach is the identification of targetable antigens adjacent to EGFR that are unique to GBM and bind EGFR throughout the whole of the tumor.

To answer this question, we applied our red-light based approach to study the interactome of EGFR in resected human glioblastoma tissue (Figure 5A). Fresh GBM patient samples taken from three regions of a single patient’s tumor were dissociated and deployed in our workflow. These dissociated samples are opaque and resistant to low wavelength photoirradiation, so provide a significant challenge for current photoproximity labeling approaches (Figure S24).

Figure 5:

Profiling the EGFR interactome in primary human GBM tissue. (A) Graphic depicting antibody-based strategy. (B&C) Quantitative chemo proteomics. Volcano plots of enriched proteins in labeling experiments with diazo and aniline probes of labeling experiments performed on GBM tumor samples (superficial). Red data points represent known EGFR interactors from the BioGrid database and blue data points represent membrane proteins. (D) Overlap of EGFR interactomes identified through proximity labeling with BioGrid database interactors. (E)Venn diagram showing overlap between enriched EGFR interactors in GBM sample (superficial)(F) Graphic illustrating three GBM tissue sections used in this study. Tissue staining of superficial, lateral, and middle sections of tissue. (G) GO analysis of enriched hits from diazo based labeling experiment across all three samples. (H) Differential enrichment of EGFR adjacent synaptic genes across three samples. (I) Differential enrichment of EGFR adjacent genes from A549 cells also found in GBM tissue samples.

The three tumor samples studied here represent increasing depth into the tumor tissue, based on the MRI-guided neuronavigation deployed during surgery, so we can assess the EGFR interactome in distinct environments of the tumor surface, where cancer cells can readily interact with cells of the normal brain tissue, through the dense actively dividing tumor center and avoid the necrotic core. Each of the three tumor sampleswas divided into two prior to addition of the probe so datasets with both short and long radius probes could be collected. We initiated our study with a superficial sample of GBM where the tumor infiltrates normal brain parenchyma. We identified 229 significant proteins using diazo probe 2 (Figure 5B) and 426 protein using aniline probe 6 (Figure 5C). Of these, ~30% were known EGFR interactors, and 90 proteins were shared between the two datasets (Figure 5D and 5E). Cellular component analysis showed significant enrichment of the plasma membrane (FDR= 1.21×10⁻⁵⁷), providing confidence in our protocol (Figure S25). Functional analysis of the hits showed enrichment of proteins related to cell-cell adhesion including the transmembrane glycoprotein L1CAM, which is known to regulate tyrosine kinase signaling and is associated with metastasis and poor prognosis⁴⁹ (Figure S25). L1CAM has been reported to bind and activate EGFR through a cellular interaction in trans in a model.⁵⁰ Other cell adhesion proteins such as NCAM1 and a set of integrins are also interactors in our data set. As cell-cell adhesion is a major driver of EGFR signaling, these specific interactors may be viable targets for blockade or dual therapeutic strategies.

Furthermore, we also identified a set of proteins related to synaptic signaling, including GABAergic genes, glutamate receptor 2B⁵¹, and NTRK2⁵² (Figure S25). Of particular interest, NTRK2 is a receptor kinase that has synergistic activity with EGFR, where NTRK2 activation promotes resistance to anti-EGFE therapies in esophageal and colorectal cancers.^53–55 Furthermore, NTRK2 blockade has been shown to be a viable strategy for treating GBM in cell lines.⁵⁶ These data, suggesting that EGFR and NTRK2 are not only functionally related but positioned proximal to each other supports their synergistic role. AlphaFold modeling of this interaction supports a physical interaction between the two proteins (Figure S26).

Investigation of tumor resections from the lateral and middle sections using our short radius probe demonstrated significant heterogeneity within the same tumor (Figure 5F). Of the 229 interactors in the superficial resection 167 were found in the lateral resection (out of 990) and 13 were found in the middle resection (out of 51) and only 11 proteins were enriched in all samples (Figure S25). These data suggest that the peripheral regions of the tumor which have greater interactions with the normal brain are more similar in their interactome, and that the tumor core is distinct. This is supported by GO analysis where synaptic terms are progressively enriched towards the periphery of the tumor and GO terms for cellular growth and adhesion are enriched at the core of the tumor (Figure 5G and 5H). Further, comparison of each interactome with a list of established EGFR interactors found in cancer (BioGRID), shows that the middle resection more strongly resembles datasets obtained in immortalized cancer cell lines (Figure 5I).

CONCLUSION

Together, we have demonstrated a variable radii PLP driven by red-light photocatalysis. Our platform functions through a commercial red-light organic photosensitizer that can be paired with commercial long radius probes or a custom diazo probe for nanoscale labeling. Both reactions are activated by a cheap, commercially available red light LED array and require minimal irradiation times for effective protein enrichment (<5 min). We benchmark our system using antibody-based labeling of EGFR and find our nanoscale probe to be comparable with best-in-class carbene based PL methods, and that the same catalyst can be used for long radius singlet oxygen-based labeling. Furthermore, the significant overlap between short and long radius interactomes provides confidence in any novel interactors identified. We then apply our platform in human tumor tissue to characterize the heterogenous EGFR interactome in GBM, demonstrating the dramatic differences in interactome at the tumor periphery when compared to the tumor core. Additionally, we identify several proximal proteins that may serve as candidates for therapeutic intervention in this deadly disease. Investigation of these relationships and how they can be targeted therapeutically are currently underway.

By providing an increased labeling signal and options for alteration of the labeling radius, this multi-scale PLP solves two problems that remain for translation of photocatalytic PL to complex tissue samples. Despite this, the challenge of photolabeling cell surface proteins in live animals remains significant, with further advances required to better understand biologically important interactomes in a native context. Furthermore, future studies on a cohort of GBM patients will be needed to address the interpatient heterogeneity in EGFR interactome, so the data here is unlikely to well represent the GBM patient population.

Supplementary Material

The Supporting Information contains supporting figures, experimental methods, data, compound characterization (PDF) and mass spectrometry compiled data (XLXS)

The Supporting Information is available free of charge on the ACS Publications website.