Photoaffinity-engineered protein scaffold for systematically exploring native phosphotyrosine signaling complexes in tumor samples

Significance

Phosphotyrosine (pTyr)-dependent protein complexes are key machinery for regulating cancer signaling. We developed the Photo-pTyr-scaffold approach for unbiasedly capturing and exploring weak and dynamic pTyr protein complexes. By utilizing the Src kinase Src homology 2 superbinder with nanomolar binding affinity, Photo-pTyr-scaffold showed superior sensitivity for profiling native pTyr protein complexes in cancer cells and breast tumor samples. Importantly, we discovered PDGFRB to be a critical signaling node for mediating intercellular cancer signaling, which is highly expressed but independent of ERBB2, the well-established breast cancer therapeutic target. Our results could lead to new targeted therapies for breast cancer and generic approaches for exploring dynamic protein complexes related to other types of protein posttranslational modifications and discovering biomarkers readily from complex clinical samples.

Abstract

Phosphotyrosine (pTyr)-regulated protein complexes play critical roles in cancer signaling. The systematic characterization of these protein complexes in tumor samples remains a challenge due to their limited access and the transient nature of pTyr-mediated interactions. We developed a hybrid chemical proteomics approach, termed Photo-pTyr-scaffold, by engineering Src homology 2 (SH2) domains, which specifically bind pTyr proteins, with both trifunctional chemical probes and genetic mutations to overcome these challenges. Dynamic SH2 domain-scaffolding protein complexes were efficiently cross-linked under mild UV light, captured by biotin tag, and identified by mass spectrometry. This approach was successfully used to profile native pTyr protein complexes from breast cancer tissue samples on a proteome scale with high selectivity, achieving about 100 times higher sensitivity for detecting pTyr signaling proteins than that afforded by traditional immunohistochemical methods. Among more than 1,000 identified pTyr proteins, receptor tyrosine kinase PDGFRB expressed on cancer-associated fibroblasts was validated as an important intercellular signaling regulator with poor expression correlation to ERBB2, and blockade of PDGFRB signaling could efficiently suppress tumor growth. The Photo-pTyr-scaffold approach may become a generic tool for readily profiling dynamic pTyr signaling complexes in clinically relevant samples.

Due to the great success of targeted therapy and biomarker development, phosphotyrosine (pTyr) signaling has become one of the most extensively studied biological systems in cancer research (1, 2). At a molecular level, kinases and phosphatases dynamically regulate pTyr, while Src homology 2 (SH2) domains specifically recognize pTyr sites and assemble these signaling proteins into multiprotein complexes for delivering signaling cues in diverse cancer signaling networks (3, 4). The detection of these signaling proteins in tumors for diagnosis is typically done by immunohistochemistry (IHC) or genomic sequencing, which might not reflect the cancer signaling processes mediated by dynamic pTyr signaling complexes. It is therefore desirable to systematically characterize these protein complexes in tumor samples for a better indication of cancer state and therapeutic response.

The unbiased profiling of pTyr protein complexes in tumor samples is challenging and rarely reported. More than 120 different SH2 domains exist in the human proteome, and taking advantage of their natural affinity for pTyr-motif SH2 domains has been well-characterized and used to probe pTyr proteins from cell or tissue lysates in a high-throughput microarray format (5, 6). These methods do not identify proteins in an unbiased manner, and thus their applications for studies of complex biological samples are limited. The combination of SH2 domain-based affinity purification and mass spectrometry (AP-MS) has been applied for identifying and quantifying dynamic EGFR signaling complexes in cells (7). However, the standard AP-MS approach is largely limited by inefficient purification of weak protein complexes, especially from tumor samples with limited quantities. SH2 domains modified with photoreactive groups by genetic encoding were successfully applied for covalently capturing pTyr protein complexes in cells under mild UV light (8). Although the photocatalyzed reaction is ideal for capturing weak protein complexes, current strategies strictly rely on genetic manipulation and therefore are not straightforward to be applied to tumor samples with clinical relevance.

To overcome these limitations for tumor sample analysis, we designed a photoreactive pTyr protein complex profiling probe (termed Photo-pTyr-scaffold) by engineering SH2 domains with newly designed trifunctional chemical probes. Using this domain–chemical probe hybrid strategy, dynamic pTyr signaling complexes with weak binding affinity were readily recognized by SH2 domains in cell lysates and chemically cross-linked by the photoreactive group, allowing isolation by biotin tag and identification using MS. Dynamic EGFR signaling complexes activated by EGF treatment could be efficiently captured and cross-validated by using PI3K p85α, SHC1, and GRB2 SH2 domain-based Photo-pTyr-scaffolds. We also incorporated the Src SH2 domain superbinder, which has low-nanomolar binding affinity, into a Photo-pTyr-scaffold. Remarkably, this Photo-pTyr-scaffold could efficiently capture native pTyr signaling complexes from both breast cancer cell lines and tumor samples without stimulation. Among more than 1,000 identified pTyr proteins, we validated that the receptor tyrosine kinase PDGFRB has low correlation with ERBB2 expression and functions as a key pTyr signaling node that coordinates intercellular signaling from cancer-associated fibroblasts to breast cancer cells. Blocking PDGFRB signaling with the selective inhibitor crenolanib in the MMTV-PyMT mouse model significantly reduced breast tumor growth, demonstrating the therapeutic potential of targeting PDGFRB in breast cancer treatment. Our results demonstrate the value of the Photo-pTyr-scaffold approach for identifying native pTyr signaling complexes directly from clinical tissue samples.

Results

Development of the Photo-pTyr-Scaffold Approach.

We were specifically interested in receptor tyrosine kinase (RTK)-mediated signaling, such as EGFR, as RTKs are important therapeutic targets for multiple solid tumors and their activation by growth factors typically forms SH2 domain-scaffolding pTyr protein complexes (2, 9). The key technical challenges for unbiasedly exploring RTK-mediated pTyr signaling complexes from tumor samples are their dynamic nature and low binding affinity, presence in poorly soluble membrane fractions, limited access, and so forth. To this end, we sought to develop a new hybrid affinity reagent which conjugates SH2 domains to multifunctional chemical probes through the convenient and well-established N-hydroxysuccinimide ester chemistry that reacts with the free primary amines on the surface lysine or N terminus of the SH2 domain (Fig. 1). As SH2 domains contain multiple lysines, we carefully optimized the labeling time and protein-to-probe ratio and obtained a major form of engineered SH2 domains which is labeled with one probe 1 and other minor forms labeled with more than one probe (SI Appendix, Fig. S1 A and B). This feature helps to capture multiple proteins in the same pTyr signaling complex. The obtained Photo-pTyr-scaffold therefore comprises the following unique features: (i) engineered SH2 domains with high selectivity and affinity toward pTyr proteins (SI Appendix, Fig. S1C); (ii) photoreactive groups for covalently cross-linking to pTyr protein complexes following exposure to mild UV light (SI Appendix, Fig. S1D); (iii) a biotin group for capturing cross-linked protein complexes, even under harsh washing conditions (SI Appendix, Fig. S1E); and (iv) a flexible spacer arm which ensures efficient cross-linking of multiprotein complexes. We further developed a quantitative chemical proteomics workflow based on label-free quantification (LFQ) for conveniently and accurately distinguishing dynamic pTyr signaling complexes from nonspecific labeling and enrichment.

Fig. 1.

Working principle of the Photo-pTyr-scaffold approach and the chemical structure of the trifunctional probes 0 to 3.

We first evaluated the performance of the trifunctional probes for photoreactive cross-linking of weak protein complexes by using two well-characterized protein complexes: the complex of the N-terminal part of Harmonin (NPDZ1) with the cytoplasmic domain of the membrane protein Cadherin 23 (Cad23-cyto), which has a dissociation constant (K_D) of 5.7 μM (10), and the complex of NPDZ1 with the SAM-PBM region of the scaffolding protein Sans with a K_D of 1.3 nM (11) (Fig. 2A and SI Appendix, Fig. S2A). A standard streptavidin pull-down of probe 1-coupled NPDZ1 or pull-down of GST-tagged NPDZ1 efficiently captured SAM-PBM, but its association with Cad23-cyto was not detected (Fig. 2A and SI Appendix, Fig. S2 B and C), which confirmed that weak protein complexes are missing from standard affinity purification. When UV irradiation was applied before the streptavidin pull-down, probe 1-coupled NPDZ1 efficiently cross-linked both SAM-PBM and Cad23-cyto, with multiple bands appearing in the high molecular mass region. We then identified these cross-linked proteins using in-gel digestion and MS analysis. While most of the identified peptides belonged to NPDZ1 due to intramolecular cross-linking (SI Appendix, Fig. S2D), specific peptides from SAM-PBM and Cad23-cyto were effectively identified with sequence coverage of 21.6 and 35.6%, respectively (SI Appendix, Fig. S2E and Dataset S1). Interestingly, although NPDZ1 and Cad23-cyto exhibit a weak interaction, many more peptides were identified from Cad23-cyto than SAM-PBM (Fig. 2B), indicating that the weak protein complex was efficiently captured by UV cross-linking.

Fig. 2.

Photo-pTyr-scaffold efficiently captures weak and dynamic protein complexes. (A) UV irradiation-induced NPDZ1-SAM-PBM and NPDZ1-Cad23-cyto cross-linking by incubating probe 1-coupled NPDZ1 with either SAM-PBM or Cad23-cyto. The green asterisk indicates the bait protein, and the red asterisk indicates specifically stained proteins. (B) The identified peptides of SAM-PBM and Cad23-cyto from the regions in the dotted green boxes in A. (C) WB validation of tyrosine phosphorylated EGFR captured by the N-PI3K Photo-pTyr-scaffold from the HeLa cell lysates. (D) Schematic outline of the quantitative chemical proteomics workflow. (E) Volcano plots of the enriched proteins upon EGF stimulation of HeLa cells. Red dots indicate significant enrichment after EGF treatment [red, EGF-treated; false discovery rate (FDR) = 0.05, s0 = 2, two-sample t test, n = 3] (43). (F) An interaction map of identified and cross-validated EGFR signaling complexes in HeLa cells by using probe 1.

We then went on to develop a Photo-pTyr-scaffold based on chemical probe 1 and the N-terminal SH2 domain of PI3K p85α (termed N-PI3K), which has direct pTyr-dependent association with EGFR signaling complexes. After HeLa cells were treated with EGF for 5 min to activate EGFR signaling (SI Appendix, Fig. S3A), the cells were lysed with mild lysis buffer for preserving potential weak protein complexes. As shown in Fig. 2C, N-PI3K Photo-pTyr-scaffold could efficiently capture EGFR upon EGF stimulation from the lysate. Importantly, a greater amount of EGFR was recovered after UV cross-linking compared with standard streptavidin pull-down. Surprisingly, the N-PI3K Photo-pTyr-scaffold also captured a small fraction of EGFR in the absence of EGF stimulation, suggesting a high sensitivity for capturing weak pTyr signals in poorly soluble membrane fractions. We next applied the chemical proteomics workflow to capture other SH2 domain-scaffolding EGFR signaling complexes (Fig. 2D). Nine well-known EGFR signaling proteins were selectively captured by the N-PI3K Photo-pTyr-scaffold approach and identified by MS (Fig. 2E and Dataset S2). This high enrichment selectivity is largely due to the covalent cross-linking of enriched proteins which allows harsh washing of noncovalently interacting proteins and nonspecifically adsorbed proteins. Encouragingly, since the N-PI3K SH2 domain is known to bind GAB1 in a pTyr-dependent manner while EGFR has known interaction with GAB1, this finding indicates that N-PI3K might bind EGFR indirectly through GAB1 (12). To further cross-validate that these signaling proteins were scaffolded onto the EGFR-associated multiprotein complex by SH2 domains on various pTyr signaling proteins, we further developed SHC1 SH2 and GRB2 SH2 domain-based Photo-pTyr-scaffolds. We also adopted the well-established stable isotope labeling of amino acids in cell culture-based quantification to validate the quantification performance of the LFQ strategy (SI Appendix, Fig. S3B). EGFR and other associated signaling proteins, such as GRB2 and SHC1, were reversibly pulled down and validated (Fig. 2F, SI Appendix, Fig. S3 C and D, and Dataset S3). The obtained results well-demonstrated the performance of Photo-pTyr-scaffold based on N-PI3K, GRB2, and SHC1 SH2 domains for capturing pTyr signaling protein complexes, especially in poorly soluble membrane fractions.

Since SH2 domains often form multiprotein complexes together with large membrane-associated RTKs, a flexible spacer arm with sufficient length is critical for capturing more components in these macromolecular protein complexes. To this end, we further engineered N-PI3K with both probe 0 and probe 1 with spacer arm lengths of 26 and 56 Å, respectively (Fig. 1). Compared with probe 0, probe 1 with the longer spacer arm showed slightly better performance for capturing one more EGFR signaling protein, PTPN11 (Figs. 2E and 3A and Dataset S2). Furthermore, compared with small cytoplasmic proteins such as SHC1, a longer spacer arm also facilitated the more efficient capture of the large transmembrane receptor EGFR with a molecular mass of ∼130 kDa (Fig. 3B and SI Appendix, Fig. S4A). These data confirmed that Photo-pTyr-scaffold with a long spacer arm is preferred for identifying dynamic EGFR-associated multiprotein complexes.

Fig. 3.

Probe 1 with a 56-Å spacer arm and benzophenone group is ideal for Photo-pTyr-scaffold. (A) Volcano plots of the enriched proteins upon EGF stimulation of HeLa cells by using N-PI3K engineered with probe 0 (red, EGF-treated; FDR = 0.05, s0 = 2, two-sample t test, n = 3). (B) Comparison of the identified peptides from EGFR and SHC1 by using N-PI3K engineered with probe 0 and probe 1. Error bars represent means ± SE from three independent experiments. (C) Volcano plots of the enriched pTyr proteins upon EGF stimulation of HeLa cells by using N-PI3K engineered with probes 1 to 3 (red, EGF-treated; FDR = 0.05, s0 = 2, two-sample t test, n = 3). (D) Pairwise comparison of the nonspecific labeling proteins in unstimulated HeLa cells with and without UV irradiation as indicated in C.

Initially, we tried to adopt the commercially available sulfo-SBED probe, which has a trifunctional chemical structure similar to probe 0 but has not been further improved specifically for use in chemical proteomics experiments since its development about 20 y ago (13, 14) (SI Appendix, Fig. S4B). However, we found significant nonspecific photoreactive labeling by the aryl azide group, which was also confirmed in multiple other reports (15, 16) (SI Appendix, Fig. S4C). Discouraged by the serious background labeling with the sulfo-SBED probe with the aryl azide group, we systematically synthesized three types of trifunctional probes carrying the three most popular photoreactive groups (i.e., aryl azide, benzophenone, and diazirine), engineered N-PI3K with these probes, and investigated their nonspecific labeling and efficiency for identifying pTyr protein complexes (Fig. 3 C and D and SI Appendix, Quality Control Data S3–S5). After HeLa cells were treated with EGF to activate EGFR signaling, probe 1 with the benzophenone group efficiently captured more EGFR signaling pTyr proteins than did probe 2 with the aryl azide group and probe 3 with the diazirine group (Fig. 3C and Dataset S2). Furthermore, probe 2 and probe 3 showed 558 and 156 significantly enriched proteins only after UV irradiation, respectively (Fig. 3D and Dataset S4). Since SH2 domains almost exclusively bind pTyr-modified proteins and the basal level of pTyr is minimal, the 558 and 156 proteins should be mostly counted as nonspecifically labeled proteins. Unexpectedly, probe 1 with the benzophenone group showed the lowest level of background labeling compared with probe 2 and probe 3. These results suggest that benzophenone as a photoreactive group showed a lower level of nonspecific labeling and a higher efficiency for capturing pTyr protein complexes.

Src Superbinder Photo-pTyr-Scaffold Is Superior for Profiling Native pTyr Protein Complexes.

As the capture of native pTyr protein complexes without artificial stimulation is extremely challenging but very useful for translational applications, we aimed to further improve the sensitivity of the Photo-pTyr-scaffold approach by adopting an engineered Src kinase SH2 domain with markedly increased pTyr peptide binding affinity (termed Src superbinder) (17). As the Src superbinder contains three mutations in the pTyr-binding pocket (T138V, C188A, and K206L), it shows increased protein binding affinity to RTK signaling proteins in vivo (18) and a broad specificity for tryptic-digested pTyr peptides (19). Compared with the N-PI3K Photo-pTyr-scaffold, the Src superbinder Photo-pTyr-scaffold showed greatly improved performance for identifying 15 EGF stimulation-dependent signaling proteins (Fig. 4A and Dataset S5). More importantly, the Src superbinder significantly outperformed N-PI3K for capturing native EGFR (Fig. 4B) and most of the EGFR-associated pTyr protein complexes in unstimulated cells (SI Appendix, Fig. S5A).

Fig. 4.

Src superbinder Photo-pTyr-scaffold approach is superior for profiling native pTyr protein complexes. (A) Volcano plots of the enriched proteins upon EGF stimulation of HeLa cells (red, EGF-treated; FDR = 0.05, s0 = 2, two-sample t test, n = 3). (B) WB validation of native EGFR capture by N-PI3K and Src superbinder Photo-pTyr-scaffold in unstimulated HeLa cells. (C) Volcano plots of the enriched proteins from unstimulated human breast cancer cells (MDA-MB-468 vs. BT-474). Red and green dots indicate significantly enriched proteins (FDR = 0.01, s0 = 2, two-sample t test, n = 3). (D) Volcano plots of the enriched proteins from unstimulated human breast cancer cells (MDA-MB-231 vs. MDA-MB-468). Red and green dots indicate significantly enriched proteins (FDR = 0.01, s0 = 2, two-sample t test, n = 3). (E) Pathway map of EGFR and ERBB2 signaling-related proteins. The pathway map was adapted from a previous report (21). LFQ intensities of the identified proteins are indicated. (F) Protein–protein interaction map of the identified proteins. The thickness of the edge indicates the STRING confidence for protein interaction prediction.

Next, we applied the Src superbinder Photo-pTyr-scaffold to analyze native pTyr protein complexes in lysates of three unstimulated breast cancer cell lines that endogenously express different RTKs: MDA-MB-468 for EGFR, BT-474 for ERBB2, and MDA-MB-231, which does not express either receptor (20). Intriguingly, native EGFR and ERBB2 pTyr signaling complexes with distinct protein profiles were clearly identified in the three cell lines (Fig. 4 C and D, SI Appendix, Fig. S5B, and Dataset S6). The performance of the Src superbinder Photo-pTyr-scaffold in capturing EGFR and ERBB2 was also validated by Western blot (WB) analysis (SI Appendix, Fig. S5C). Pathway analysis and gene ontology annotations confirmed that many canonical signaling proteins associated with EGFR and ERBB2 signaling pathways were enriched in BT-474 and MDA-MB-468 cells, respectively, compared with MDA-MB-231 cells (21) (Fig. 4E, SI Appendix, Fig. S6, and Datasets S6 and S7). Consistent with SH2 domain-mediated pTyr signaling complexes, the identified proteins that have tighter association with BT-474 and MDA-MB-468 cells are those located in the cytosol and well-known components of the EGFR and ERBB2 signaling pathways. To obtain a comprehensive view of their interaction networks, we queried these proteins in the STRING protein interaction database (22), and the analysis showed that these proteins are highly connected in the same signaling network (Fig. 4F). These results confirmed that the Src superbinder Photo-pTyr-scaffold is superior for capturing native pTyr protein complexes.

Exploration of pTyr Signaling Complexes in Breast Cancer Patient Samples.

Finally, we applied the Src superbinder Photo-pTyr-scaffold approach to explore pTyr protein complexes from human breast tumor samples. We reasoned that dynamic pTyr protein complexes may represent unique activation states and function as potential biomarkers for breast cancer diagnosis and treatment. Due to the great success of ERBB2-based targeted therapies for breast cancer, we selected 12 breast cancer tumor samples with distinct ERBB2 expression levels as quantified by immunohistochemical staining (Fig. 5A). The chemical proteomics analysis of these samples identified more than 2,000 proteins in each sample with good LFQ accuracy (Fig. 5B and SI Appendix, Quality Control Data S8). Compared with the largest pTyr proteome dataset (19), more than 1,000 known pTyr proteins were identified among them, while 716 of them were significantly enriched compared with the set of identified proteins directly from the total lysate without enrichment (SI Appendix, Fig. S7A and Dataset S8). The identified proteins exhibited different expression patterns and clustering for kinase-related molecular function activities (Dataset S9). Because protein kinases are the driving machinery for activating pTyr protein complexes and downstream signaling pathways, we next focused our analysis on 64 identified protein kinases (23) (Fig. 5C). A big portion of these protein kinases are highly connected as predicted by the STRING protein interaction database, indicating their association in the same signaling network (SI Appendix, Fig. S7B). For example, serine/threonine-protein kinase RAF1 should be captured by the Src superbinder Photo-pTyr-scaffold indirectly through PDGFRB (24). Furthermore, we found that some of them have high abundance and association with RTK signaling (Fig. 5C, highlighted in red). Excitingly, we identified ERBB2 with similar expression patterns as characterized by the immunohistochemical method. But, in addition, the Src superbinder Photo-pTyr-scaffold confidently identified ERBB2 from all of the ERBB2-negative samples that have about 100 times lower expression level than the ERBB2-positive samples (Fig. 5D). Interestingly, a minimum level of EGFR was identified in some patient samples. Because ERBB2 and EGFR belong to the same RTK family with highly similar protein structures, the Src superbinder Photo-pTyr-scaffold showed excellent selectivity for the activated signaling molecules in breast cancer (25). Unexpectedly, we detected another RTK, PDGFRB, in all patient samples, which has comparable abundance to ERBB2 in the positive samples but no obvious correlation with ERBB2 expression (Fig. 5D). This result might guide a new PDGFR signaling-based therapy for breast cancer, as a number of PDGFR inhibitors have been approved by the Food and Drug Administration for other cancer indications, such as hepatocellular carcinoma (26). In summary, these results demonstrated the powerful performance of the Src superbinder Photo-pTyr-scaffold approach for profiling native pTyr signaling complexes directly in cancer patient samples.

Fig. 5.

Exploration of native pTyr protein complexes from breast tumor samples by the Src superbinder Photo-pTyr-scaffold. (A) IHC staining of ERBB2 in 12 breast tumor samples. (Scale bar, 20 μm.) (B) Summary of identified and quantified total proteins, pTyr proteins, and protein kinases. (C) Hierarchical clustering of protein kinases identified by the Src superbinder Photo-pTyr-scaffold approach or directly from the indicated cell lysates. RTK signaling-related protein kinases are in red. (D) The protein abundance distribution of ERBB2, EGFR, and PDGFRB as indicated by LFQ intensity.

PDGFRB Is a Potential Therapy Target for Breast Cancer.

To further investigate the possibility of PDGFR signaling-based tumor therapy, we first characterized the expression patterns of PDGFRB in breast tumors. In agreement with previous reports that PDGFRs are critical regulators for mesenchymal cells (27), our data showed that PDGFRB abundantly expressed in a breast fibroblast cell line (human mammary fibroblasts; HMFs) but not in malignant breast cells or normal breast cells (MCF 10A) (Fig. 6A). In vivo analysis also validated that PDGFRB was prominently expressed in tumor-associated stromal cells, including spindle-shaped fibroblasts, but not in tumor tissues by using the polyoma middle T (PyMT) oncogene-driven breast cancer mouse model, which mimics all identifiable stages of human breast cancer progression (28) (Fig. 6B). More importantly, our data indicated that high stromal PDGFRB expression is positively associated with breast cancer progression (Fig. 6 B and C), implying a strong scientific rationale for targeting the PDGFRB-associated fibroblasts in anti-breast cancer therapy. To further analyze the paracrine effects of breast cancer cells on fibroblast expansion, we assessed the ligand secretion by breast cancer cells (Fig. 6D and SI Appendix, Quality Control Data S9). Results showed that high concentrations of the PDGFRB ligands PDGFB and/or PDGFC were secreted by breast cancer cells into the conditioned media (29–31) (Dataset S10), which in turn acted on fibroblasts to activate known PDGFR downstream targets such as AKT and ERK but not STAT3 (Fig. 6 E and F).

Fig. 6.

PDGFRB is a potential therapy target for breast cancer. (A) WB analysis of PDGFRB expression in breast fibroblasts and normal and tumor cells. (B) IHC staining for PDGFRB in mammary gland sections and tumors from wild-type and MMTV-PyMT mice at different ages. (Scale bar, 20 µm.) (C) Quantification of IHC score for PDGFRB staining in breast sections from wild-type and MMTV-PyMT mice at different ages (n = 4 or 7). Means ± SE, unpaired Student’s t test. (D) The LFQ intensities of PDGFB and PDGFC from breast cancer cell line conditioned medium (CM). Error bars represent means ± SE; n = 3. (E) WB validation of PDGFRB activation in HMFs by recombinant PDGFB and MDA-MB-468 cell CM. Starved HMFs were pretreated with crenolanib (0.5 μM) for 12 h and then treated with CM from MDA-MB-468 or 100 ng/mL PDGFB for 5 min. (F) A schematic explanation of PDGFRB-mediated intercellular signaling. CAFs, cancer-associated fibroblasts. (G) Gross examination of tumor loads in vehicle- or crenolanib- (20 mg⋅kg⁻¹⋅d⁻¹) treated MMTV-PyMT mice at 12 wk of age. (H) Antitumor effects of the crenolanib in vivo. Eight-week-old female MMTV-PyMT mice were treated with vehicle or crenolanib (20 mg/kg) intraperitoneally every day for 35 d. Tumor volumes were measured every 5 d. Error bars represent means ± SE, unpaired Student’s t test. (I) Tumor weights were determined from MMTV-PyMT mice treated with vehicle or crenolanib (n = 6). Means ± SE, unpaired Student’s t test.

To test the potential of PDGFR as a tumor therapeutic target in vivo, we administered crenolanib, a highly selective inhibitor of PDGFRB (32) (Dataset S11), to 8-wk-old MMTV-PyMT mice to block stromal PDGFRB signaling. Tumor volumes in crenolanib-treated and untreated mice were measured every 5 d after dosing initiation. Beginning at day 10 after treatment, tumors derived from crenolanib-treated mice grew significantly more slowly than those derived from vehicle-treated mice, as evaluated by standard external calipers (Fig. 6 G and H). In addition, tumor weight per mouse at sacrifice (12 wk) was 55% lower in crenolanib-treated mice than in vehicle-treated mice (Fig. 6I). Taken together, our findings might lead to a novel therapeutic approach targeting stromal PDGFRB signaling to benefit patients with different breast cancer subtypes.

Discussion

More than 100,000 protein interactions occur in a human cell at any one time (33). These interactions assemble proteins into multiprotein complexes for controlling various molecular machineries (34). These protein complexes are often mediated by various posttranslational modifications (PTMs) and related binding domains for forming dynamic signaling networks (35). As a typical example, the SH2 domain specifically recognizes the pTyr motif on various signaling proteins and helps assemble dynamic signaling complexes (4). Exploration of dynamic protein complexes is therefore important for understanding diverse cell signaling processes and their physiological and pathological outputs.

AP-MS has been the most widely used approach for analyzing protein complexes on a global scale (36, 37). Although this approach has been successfully applied to explore thousands of human protein complexes and their dynamic changes (38, 39), AP-MS is limited by the need to use transfected cell lines and by the loss of weak protein complexes during the affinity purification step. In this study, the Photo-pTyr-scaffold approach was successfully developed and used to explore dynamic and weak pTyr protein complexes in clinical tumor samples. The trifunctional chemical probe 1 functionalizes SH2 domains with photoreactive cross-linking and enrichment groups in a simple but efficient manner. Taking advantage of the natural binding affinity of SH2 domains for the pTyr motif, Photo-pTyr-scaffold provides a powerful approach for capturing, cross-linking, and identifying dynamic pTyr protein complexes directly from complex cell or tumor lysates (Fig. 2F). Compared with standard affinity purification by genetically introduced affinity tags, Photo-pTyr-scaffold significantly increased pull-down efficiency, especially for the large transmembrane receptor EGFR, which is located in poorly soluble membrane fractions (Fig. 2C).

Furthermore, the combination of probe 1 with the Src superbinder provides a powerful chemical proteomics approach for global profiling of native signaling complexes with high sensitivity (Fig. 4B). Since pTyr is present at minimal basal levels in most human cells, this high sensitivity is unexpected and allows us to well-differentiate EGFR and ERBB2 signaling complexes in different breast cancer cell lines without external stimulation (Fig. 4 C–F). Taking advantage of its superior sensitivity, we successfully used the Src superbinder Photo-pTyr-scaffold approach to explore native pTyr signaling complexes directly from clinical tumor samples. More than 1,000 pTyr proteins were identified across 12 breast tumor samples with diverse ERBB2 expression levels. Application of Photo-pTyr-scaffold to other tumor samples would enable the unbiased discovery of potential diagnostic markers and drug targets. Because the protein-binding domains for the major PTMs have been well-characterized but have relatively low binding affinity, particularly for ubiquitination and methylation, chemical probe 1 should become a generic approach for functionalizing protein-binding domains and profiling other PTM-dependent protein complexes (40).

This systematic proteome profiling provided a unique resource for studying pTyr signaling complexes with direct clinical relevance. Prominent from this resource is the discovery of 64 protein kinases. As a preliminary effort for validation, we focused on the receptor tyrosine kinase PDGFRB because it has a consistently high expression level across all of the 12 tumor samples, especially compared with the variable expression level of ERBB2 (Fig. 5D). Previous studies of tumor microenvironments have indicated the expression of PDGFRs in cancer-associated fibroblasts and its function mainly through intercellular signaling (41). In this study, we found that breast cancer cells can secrete PDGFs (Fig. 6D) but barely express PDGFRs (Fig. 6A), indicating that PDGFs elicit their action primarily through paracrine mechanisms that involve other cell types, such as fibroblasts. Indeed, the cancer cell-secreted PDGF can activate fibroblast PDGFR (Fig. 6E), and targeting this receptor by using the kinase inhibitor crenolanib can significantly reduce mammary tumor growth in MMTV-PyMT mice. Crenolanib is a highly selective inhibitor for PDGFR and FLT3 but not for other RTKs such as VEGFR, and is under development in a phase II clinical study for gastrointestinal tumors with PDGFRA mutation (32) (Dataset S11). PDGFRA and FLT3 were not detected in our human breast tumor sample data, suggesting that crenolanib may directly block the PDGFRB signaling pathway in activated stromal cells.

Materials and Methods

Preparation of Photo-pTyr-Scaffold.

Fifty micrograms of SH2 domain (N-PI3K, GRB2 SH2 domain, SHC1 SH2 domain; provided by Y. Zheng, National Center for Protein Sciences, Beijing, China) (39), Src SH2 domain superbinder (17), and NPDZ1 domain (10, 11) was purified and conjugated to probes (protein-to-probe ratio of 1:10) at room temperature for 1 min. Glycine (probe-to-glycine ratio of 1:10) was then added to stop the reaction, and the hydrolyzed and nonreacted probe was removed by ultrafiltration with a 3-kDa cutoff Amicon Ultra centrifugal filter device (Merck Millipore).

Cell Culture and Cell Lysate Preparation.

The human cancer cell lines HeLa, BT-474, MDA-MB-468, MDA-MB-231, SK-BR-3, and MCF-7 and human breast cell line MCF 10A were purchased from American Type Culture Collection. Human mammary fibroblasts were purchased from ScienCell. All cell lines were cultured according to the supplier’s instructions. When the cells reached 80% confluence, they were starved with FBS-free medium for 4 h at 37 °C and then treated with 100 ng/mL EGF or PDGFB for 5 min or 1 mM pervanadate for 10 min. The cells were washed twice with ice-cold 1× PBS. Total cell lysates were prepared by scraping cells from the Petri dishes in ice-cold Nonidet P-40 buffer containing freshly added protease and phosphatase inhibitor mixtures (Roche). The lysate was then clarified by centrifugation at 14,000 × g for 10 min at 4 °C. The protein concentration was measured using the bicinchoninic acid (BCA) assay.

Tissue Lysate Preparation.

Human breast cancer tissue samples were obtained from Shenzhen People’s Hospital and then deidentified to protect patient confidentiality. All patients provided written informed consent, and all human studies were approved by the Medical Ethics Committee of Shenzhen People’s Hospital institutional review board. The samples were stored in liquid nitrogen or fixed with 10% (wt/vol) buffered formalin for 12 h. Fifty milligrams of freshly frozen human breast cancer tissue samples was washed with ice-cold 1× PBS and then lysed in Nonidet P-40 buffer as described above assisted by the PowerLyzer 24 Bead-Based Homogenizer (MO BIO, catalog no. 13155) for four rounds of 10 s at a speed of 3,500 rpm, with a 60-s incubation on ice between rounds. The obtained cell lysate was sonicated four times for 5 s each at a power of 200 W, with a 10-s incubation on ice between each sonication. The lysate was then clarified by centrifugation at 14,000 × g for 10 min at 4 °C. The protein concentration was measured using the BCA assay.

Photoreactive Cross-Linking and Pull-Down.

Fifty micrograms of Photo-pTyr-scaffold was incubated with freshly isolated total lysate in mild Nonidet P-40 lysis buffer containing freshly added protease and phosphatase inhibitor mixtures for 2 h at 4 °C with end-over-end rotation. Then, the mixtures were placed in a quartz colorimetric cuvette and subjected to UV irradiation in the UVP CL-1000L UV Crosslinker (365 nm) at a distance of ∼3 cm on ice. After UV irradiation for 30 min, the samples were incubated with 30 μL streptavidin beads for 2 h at 4 °C with end-over-end rotation. The beads were washed three times with 1 mL of harsh modified RIPA buffer, and subsequently the samples were subjected to WB analysis.

Animal Studies.

All animal procedures were approved by the Institutional Animal Care and Use Committee at Southern University of Science and Technology. Twelve 8-wk-old female MMTV-PyMT mice were divided into six groups. The two mice in each group were littermates. Each group of mice was treated with crenolanib (20 mg/kg, dissolved in vehicle; Selleck) or vehicle [5% (vol/vol) glycerol formal; Sigma] intraperitoneally once daily for 35 d. Once palpable, tumor sizes were measured with a digital caliper every 5 d, and the volumes were calculated using the formula (length × width²)/2. Data were reported as means ± SE. At the end of crenolanib treatment, mice were killed, and tumors were harvested and weighted.

SDS/PAGE and Western Blot Analysis.

The labeling performance of probe 1 to various bait proteins was characterized by high-resolution SDS/PAGE using the PROTEAN II xi Cell electrophoresis instrument (gel size, 16 × 20 cm; Bio-Rad) and Coomassie brilliant blue staining.

Equal amounts of protein were separated by SDS/PAGE and transferred to membranes. Membranes were blocked in 5% (wt/vol) BSA and then incubated with a primary antibody overnight at 4 °C, followed by an incubation with an HRP-conjugated anti-rabbit or anti-mouse secondary antibody (Cell Signaling Technology). Proteins were visualized using a Western ECL Substrate Kit (Bio-Rad). The primary antibodies used in this study were streptavidin-HRP (Thermo Fisher), anti-4G10 (Merck Millipore), anti-His (Cell Signaling Technology), anti-EGFR (Cell Signaling Technology), anti-ERBB2 (Cell Signaling Technology), anti-PDGFRB (Cell Signaling Technology), anti–p-STAT3 (Cell Signaling Technology), anti–p-AKT (Cell Signaling Technology), anti-AKT (Cell Signaling Technology), anti–p-ERK 1/2 (Cell Signaling Technology), anti-ERK 1/2 (Cell Signaling Technology), and anti–β-actin (Beyotime).

Immunohistochemical Analysis.

Paraffin-embedded mouse and patient tissue slides (4 μm thick) were immunostained with the ERBB2 or PDGFRB primary antibody (Cell Signaling Technology) overnight at 4 °C, followed by incubation with HRP-conjugated secondary antibody (DAKO). The signal was detected using a 3,3′-diaminobenzidine detection kit (DAKO). Slides were counterstained with hematoxylin, dehydrated, mounted, and analyzed by light microscopy. Quantification of PDGFRB was scored using the H score according to staining intensities and percentage of positive cells within the whole tissue section (200× magnification). For each mouse, three random fields were counted.

MS Analysis.

After the Photo-pTyr-scaffold labeling and streptavidin pull-down, the beads were washed three times with 1 mL of harsh modified RIPA buffer and then washed with 1 mL of 50 mM ammonium bicarbonate (ABC). After being reduced and alkylated, the beads were washed with 1 mL of 50 mM ABC and incubated with trypsin (Promega) overnight at 37 °C. The digested peptides were collected by washing the beads with 1% (vol/vol) formic acid, and the peptides were subjected to StageTip C18 (42) desalting and MS analysis.

Details about trifunctional probe design and chemical synthesis, preparation of Photo-pTyr-scaffold, cell culture, tissue and cell lysate preparation, Photo-pTyr-scaffold photoreactive cross-linking and pull-down, SDS/PAGE and Western blot analysis, immunohistochemical analysis, affinity purification, MS sample preparation and analysis, data analysis, and animal studies are described in SI Appendix, Materials and Methods.