MACSPI enables tissue-selective proteomic and interactomic analyses in multicellular organisms

Significance

Multicellular organisms synthesize a specific set of proteins in each cell type to enable cellular differentiation and functions. These proteins interact with each other to form complexes needed for various cellular processes. Despite recent advances in cell-specific transcriptomics, very few methods can robustly profile cell-specific proteomes and interactomes in intact organisms without the physical isolation of cells. We developed a chemical biology approach to label proteins from specific cells with a bifunctional amino acid probe that carries both a bio-orthogonal tag for affinity purification and a photo-cross-linker for capturing protein–protein interaction. This method only requires cell-specific expression of an engineered aminoacyl-tRNA synthetase that incorporates the probe into proteins and can be applied to a broad range of organisms.

Abstract

Multicellular organisms are composed of many tissue types that have distinct morphologies and functions, which are largely driven by specialized proteomes and interactomes. To define the proteome and interactome of a specific type of tissue in an intact animal, we developed a localized proteomics approach called Methionine Analog-based Cell-Specific Proteomics and Interactomics (MACSPI). This method uses the tissue-specific expression of an engineered methionyl-tRNA synthetase to label proteins with a bifunctional amino acid 2-amino-5-diazirinylnonynoic acid in selected cells. We applied MACSPI in Caenorhabditis elegans, a model multicellular organism, to selectively label, capture, and profile the proteomes of the body wall muscle and the nervous system, which led to the identification of tissue-specific proteins. Using the photo-cross-linker, we successfully profiled HSP90 interactors in muscles and neurons and identified tissue-specific interactors and stress-related interactors. Our study demonstrates that MACSPI can be used to profile tissue-specific proteomes and interactomes in intact multicellular organisms.

Cellular functions and responses are orchestrated by intricate networks of protein–protein interactions (PPIs) that assemble functionally related proteins into complexes, signaling pathways, and macromolecular machineries. The disruption of these molecular interaction networks has been linked to a variety of diseases (1, 2). Although systematic approaches that test the potential interaction between every protein pair provide important information for binary interaction networks (3), such PPI networks do not take cellular context into consideration and thus may lack physiological relevance. In fact, the generation of diverse cell types and tissue-specific structures and functions in multicellular organisms relies on the establishment of tissue-specific PPI subnetworks (4, 5). Therefore, understanding the architecture and dynamics of protein interactomes with tissue specificity can provide critical insights into the molecular basis of many physiological and pathological processes. However, very few methods can robustly map tissue-specific proteomes and interactomes in intact organisms.

Previous interactome mapping approaches that used the yeast two-hybrid (Y2H) screens (6) can yield extensive binary interactomes (7–9) but are limited in only detecting interactions between two proteins and cannot detect collaborative interactions. Affinity purification–mass spectrometry-based approaches can identify protein complexes but often suffer from the low signal-to-noise ratio and the difficulties in identifying transient or weak interactions. Both types of approaches lack sufficient cell specificity. Although tagging tissue-specific proteins with a biotin ligase (e.g., BioID or TurboID) or an engineered ascorbate peroxidase (APEX) could enable the purification and identification of tissue-specific protein complexes (10–13), such proximity-based labeling approaches allow any protein within the labeling radius, including noninteracting proteins, to be captured, making it prone to false positives. In addition, cofactors such as exogenous biotin-phenol and high levels of H₂O₂ used to initiate APEX-based proximity labeling may trigger unwanted cellular responses, leading to experimental artifacts.

Profiling tissue-specific proteomes and interactomes is particularly challenging in small-sized multicellular organisms, such as Caenorhabditis elegans, due to the difficulties in isolating large numbers of cells and maintaining the integrity of cellular structures during the physical isolation. As a result, although C. elegans was one of the first model organisms used to map the proteome-wide interactome through Y2H screens and proximity labeling (9, 10, 12, 14), it still remains difficult to study tissue-specific PPIs in intact worms. A potential solution for the challenge involves tissue-specific expression of a modified aminoacyl-tRNA synthetase that can attach unnatural amino acids with reactive chemical handles [e.g., azidonorleucine, Anl (15, 16) and p-azido-L-phenylalanine, Azf (17)] to corresponding tRNAs, allowing the selective labeling of the proteome from the tissue of interest. This approach was previously used to profile proteins expressed in the pharyngeal muscle of C. elegans using the phenylalanine analog Azf, while attempts to use methionine analogs were unsuccessful (17). More importantly, such bio-orthogonal noncanonical amino acid tagging has not been used to profile protein interactors in specific tissues in C. elegans.

Here, we developed an approach called Methionine Analog-based Cell-Specific Proteomics and Interactomics (MACSPI), in which a bifunctional methionine analog, 2-amino-5-diazirinylnonynoic acid (referred to as photo-ANA) (18) (Fig. 1A), is incorporated into newly synthesized proteins. The alkyne group of photo-ANA enables profiling of the labeled proteins, while its diazirine group mediates covalent linkages of interacting proteins upon photoactivation, allowing the capture of transient or weak PPIs (Fig. 1B). By controlling tissue-specific expression of an engineered methionyl-tRNA synthetase (MetRS) that is capable of attaching photo-ANA to methionyl-tRNA, we can label proteome with cellular selectivity (Fig. 1C). As a proof of principle, we applied MACSPI in C. elegans to selectively label, capture, and profile HSP90 interactors from the body wall muscle and the nervous system under both physiological and stress conditions, demonstrating that this approach can be used to profile tissue-specific proteomes and interactomes in intact animals. Given the ease of engineering the native MetRS and supplementing the photo-ANA probe through diet, we envision that MACSPI can be readily applied to many multicellular organisms to enable the systematic mapping of protein interactions in any selected tissues or cells of interests.

Fig. 1.

Cell-selective MACSPI analysis in C. elegans. (A) Chemical structure and functions of photo-ANA. (B) A chemical proteomics approach to profile cell type–specific proteome and interactome in C. elegans. (C) A mutant C. elegans MetRS (L43G) is capable of tagging proteins with the reactive noncanonical amino acid, photo-ANA. Spatial selectivity is achieved by controlling the expression of the MetRS L43G mutant using cell-specific promoters in transgenic animals. Proteins synthesized in cells that do not express the MetRS L43G mutant are neither labeled nor detected.

Results

Cell Type–Specific In Situ Labeling of Proteins with Photo-ANA in C. elegans.

To label C. elegans proteins with photo-ANA, we first labeled bacterial proteins in Escherichia coli, which is a common food source for C. elegans. Based on the work on alkyne-functionalized methionine surrogates (15, 16, 19) and our previous results (20), we screened for E. coli MetRS (EcMetRS) mutants that may allow the binding of photo-ANA by substituting leucine 13 (L13), tyrosine 260 (Y260), and histidine 301 (H301) residues located in the methionine-binding pocket (21) (Fig. 2A). Among the tested EcMetRS mutants, a single residue substitution (L13G) had the highest efficiency for incorporating photo-ANA into E. coli proteins (Fig. 2B). The addition of methionine outcompeted photo-ANA for incorporation, confirming that both amino acids followed the same route to be incorporated into synthesized proteins (Fig. 2C). To further increase the efficiency of photo-ANA incorporation, we generated a methionine auxotroph (ΔmetB) by knocking out the metB gene, which is essential for endogenous methionine biosynthesis. As expected, the methionine auxotrophic strain expressing EcMetRS L13G was able to incorporate photo-ANA with high efficiency and specificity (SI Appendix, Fig. S1 A and B). We subsequently used the photo-ANA-labeled ΔmetB E. coli as the food source for C. elegans in the following studies.

Fig. 2.

Engineering EcMetRS and C. elegans MetRS capable of activating photo-ANA. (A) Active site of the wild-type EcMetRS bound with methionine (based on PDB: 1P7P) and the MetRS mutants tested in this study. (B) In-gel fluorescence analysis of cell lysate from 2 mM photo-ANA-labeled E. coli expressing different MetRS mutants. The three-letter notations correspond to the combination of mutations in A. CB, Coomassie blue staining. (C) In-gel fluorescence analysis of cell lysate from 2 mM photo-ANA-labeled, MetRS (L13G)-expressing E. coli in the presence or absence of 0.5 mM methionine. (D) Sequence alignment among EcMetRS, C. elegans MetRS (CeMetRS), and M. musculus MetRS (MmMetRS). BD, binding domain. (E and F) C. elegans express MetRS (L43G) mutant with C-term 2×FLAG-tag under the control of tissue-specific promoters. In situ click chemistry showed tissue-specific labeling of proteins by clicked rhodamine in animals with pan-neuronal (E) or body wall muscle (F) expression of CeMetRS L43G::2×FLAG. Anti-FLAG staining was used to confirm the tissue-specific expression of CeMetRS L43G. (Scale bar, 20 μm.)

Similarly, we screened CeMetRS (C. elegans MetRS) for mutants that allow photo-ANA labeling. Sequence alignment showed that CeMetRS shares similar domain structure as EcMetRS, with Leu 13 in the catalytic domain of EcMetRS homologous to Leu 43 in CeMetRS (Fig. 2D). We hypothesized that an equivalent L-to-G mutation in CeMetRS would allow the attachment of photo-ANA to methionyl-tRNA and subsequent incorporation into proteins. To test this hypothesis, we generated transgenic C. elegans expressing FLAG-tagged CeMetRS L43G mutant in body wall muscles and all neurons using muscle-specific myo-3 promoter and neuron-specific rab-3 promoter, respectively. To test the incorporation of photo-ANA in specific tissues, transgenic C. elegans were fed with photo-ANA-labeled E. coli (ΔmetB), fixed, and then subjected to rhodamine-azide conjugation onto the alkyne group and stained with anti-FLAG antibody. Both rhodamine and FLAG signals were detected only in tissues expressing the CeMetRS L43G mutant (Fig. 2 E and F), confirming that our approach is able to label proteins in selected tissues. We also prepared protein lysate from the animals and performed click chemistry to conjugate proteins with rhodamine, followed by in-gel fluorescent imaging, which confirmed the labeling of proteins with photo-ANA (SI Appendix, Fig. S1C). To validate the feasibility of protein enrichment using the alkyne group as a bio-orthogonal handle, we performed click chemistry to ligate biotin onto photo-ANA-labeled proteins in the lysate of transgenic C. elegans with pan-neuronal expression of CeMetRS L43G and then pulled down a broad range of proteins using streptavidin beads (SI Appendix, Fig. S1D).

The expression of CeMetRS L43G mutants and the incorporation of photo-ANA into the proteome did not appear to affect organismal development or cellular functions. We did not observe any developmental delay, morphological defects, or locomotion changes in transgenic animals expressing CeMetRS L43G in body wall muscles or neurons and fed with either regular or photo-ANA-labeled E. coli (SI Appendix, Fig. S2A). We also generated transgenic animals expressing CeMetRS L43G exclusively in the touch receptor neurons using the mec-17 promoter, and the animals displayed normal mechanosensory functions (SI Appendix, Fig. S2B). The above results show that our method is able to selectively label the proteome of specific tissues with photo-ANA, allowing the visualization and enrichment of newly synthesized proteins with spatial selectivity in C. elegans without detectable toxicity.

Tissue-Specific Proteomic Analysis in C. elegans.

To profile newly synthesized proteins in specific tissues using MASCPI, we modified a SILAC (stable isotope labeling by amino acid in cell culture) method for the use in C. elegans (22–25) and applied it to profile proteins from body wall muscles. We first prepared triply labeled ΔmetB E. coli expressing EcMetRS L13G in photo-ANA-containing medium supplemented with either “heavy” (¹³C₆¹⁵N₄-arginine and ¹³C₆¹⁵N₂-lysine) or “light” isotope-labeled (¹²C₆¹⁴N₄-arginine and ¹²C₆¹⁴N₂-lysine) amino acids (see Materials and Methods for details). The heavy isotope labeling efficiency was confirmed to be about 99.9% (SI Appendix, Fig. S2C). Transgenic C. elegans expressing the CeMetRS L43G mutant in body wall muscle were fed with bacteria grown in either heavy medium containing photo-ANA or light medium containing methionine. Tissues were lysed and pooled for conjugation with biotin via click chemistry, followed by enrichment via streptavidin beads and in-gel tryptic digestion (Fig. 3A). The digested peptide mixture was then subjected to liquid chromatography and mass spectrometry. Proteins with a high ratio of heavy to light isotopes (H/L) were considered to be from body wall muscles (see SI Appendix for details on data analysis).

Fig. 3.

Photo-ANA labeling enabled profiling of tissue-specific proteome in C. elegans. (A) Experimental design to profile tissue-specific proteomes of C. elegans. (B) Pie charts show that 94.8% and 81.9% of proteins identified in muscle and neuron-specific proteomes (H/L ratio > 1.5), respectively, had mRNA expression in corresponding tissues based on transcriptomic data. (C) Two-dimensional scatter plot of proteins identified and quantified in two independent proteomic experiments for the body wall muscle. Proteins with average H/L ratio > 1.5 are highlighted in red. Labeled proteins were found to have mRNA expression specifically in muscles. The full list can be found in Dataset S1. (D) KEGG pathway enrichment analysis of identified proteins in body wall muscle by DAVID. (E) The correlation matrix heatmap shows the values of the Pearson correlation coefficient between two experimental groups. The color code for the coefficients ranges from −1 (blue) to 1 (red). (F) A map comparing H/L ratios of proteins in the muscle and neuronal proteomes; the heatmap is constructed by hierarchical cluster analysis. Red and green indicate high and low H/L ratios, respectively. (G) A Venn diagram shows the overlapping of proteins identified in the body wall muscle and neuronal proteomes. Bar graphs show the mRNA expression of selected proteins according to transcriptomic data. (H) Fluorescent images show that GFP signal was mainly found in body muscle cells (labeled by myo-3p::mCherry) in animals carrying the ttr-51p::GFP or inos-1p::GFP transgene. (I) Fluorescent images show that GFP was expressed in many neurons (marked by rab-3p::mCherry) in pab-2p::GFP and R09E10.5p::GFP animals. (Scale bar, 100 μm.)

Among the 498 proteins with H/L values greater than 1.5, 94.8% of them had mRNA transcription in body wall muscles based on transcriptomic data (26, 27) (Fig. 3 B and C and Dataset S1). Among them, 52 proteins were highly enriched in muscles with expression level >100 times higher than the average expression across other tissues (neuron, hypodermis, and intestine), and 16 proteins had mRNA expression exclusively (e.g., UNC-22, SET-18, and OST-1) in the muscle (Fig. 3C and SI Appendix, Fig. S3 A and B). In addition, 85 proteins were defined as muscle-expressed either by fluorescent reporters or immunostaining assay based on Wormbase curation (27). Furthermore, the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis revealed that the identified proteins were mostly enriched in pathways associated with motor proteins, proteasome, pyruvate metabolism, and the citrate cycle [TCA (trichloroacetic acid) cycle] (Fig. 3D and Dataset S2), and the functional enrichment analysis via WormCat (28) and DAVID (29) identified enriched gene ontology (GO) terms related to muscle development and function, as well as actin cytoskeleton which plays an essential role in muscle contraction (SI Appendix, Figs. S3C and S4 and Dataset S2). Many known components of the sarcomere (NMY-2, MYO-3, SAX-7, UNC-22, SET-18, HIP-1, ZIG-12, etc.), striated muscle dense body (ATN-1, UNC-52, UNC-97, UNC-112, GYG-1, GLRX-3, HGO-1, etc.) and cytoskeleton (ACT-1, ACT-2, TBB-1, TWF-2, UNC-116, DLI-1, RHA-1, TLN-1, etc.) were detected (SI Appendix, Fig. S4). These results together confirmed that the captured proteins were indeed muscle proteins.

To demonstrate the versatility of our approach in different tissue types, we applied SILAC to profile the neuronal proteome in transgenic C. elegans with pan-neuronal expression of CeMetRS L43G mutants. Like muscles, 81.9% of 426 proteins with H/L ratio above 1.5 showed mRNA expression in neurons (26) (Fig. 3B, SI Appendix, Fig. S5A, and Dataset S3). Many proteins (e.g., RAB-3, TBB-4, MEC-7, IDA-1, and NCS-2) with consistently high H/L ratios between biological replicates (SI Appendix, Fig. S5B) were exclusively expressed in neurons based on the transcriptomic data (26, 27) (SI Appendix, Fig. S5C). Pathway enrichment and GO analyses showed that the identified neuronal proteins were enriched in terms associated with mitochondrial structure and function (e.g., oxidative phosphorylation) and trafficking (SI Appendix, Figs. S5 D and E and S6 and Dataset S4), which is consistent with the high energy demand and active intracellular transport in neurons.

Next, we compared the neuronal proteome with the muscle proteome. Correlation plots showed highly reproducible mass spectrometry data between replicates, but no correlation between the neuronal proteome and muscle proteome (Fig. 3 E and F), supporting that our method enriched proteins from specific tissues. In fact, 92.4% of proteins in the muscle proteome were not found in the neuronal proteome, 91.1% of neuronal proteins were not found in the muscle proteome, and the two proteomes shared only 38 proteins that showed ubiquitous mRNA expression across several tissue types (26, 27) (Fig. 3G). To further validate our results, we selected a few hits in our dataset whose expression patterns had not yet been studied by fluorescent reporters. We generated transcriptional reporters expressing GFP under their respective promoters and confirmed that the candidates from the muscle proteome (e.g., INOS-1 and TTR-16) and neuronal proteome (e.g., PAB-2 and R09E10.5) were specifically or predominantly expressed in the respective tissues (Fig. 3 h and I). These results not only validated our proteomic data but also suggested that our MACSPI approach can effectively identify different tissue-specific proteins.

Tissue-Specific Interactomic Analysis in C. elegans.

We next used MACSPI to profile tissue-specific PPIs using HSP90 as an example. HSP90 is a ubiquitously expressed molecular chaperone whose primary function is to ensure the proper folding and stabilization of client protein substrates. To avoid artifacts associated with HSP90 overexpression, we used CRISPR/Cas9-mediated gene editing to insert two tandem hemagglutinin (HA) tags into the endogenous hsp-90 locus before the stop codon for C-terminal fusion. We then crossed this strain with transgenic animals expressing CeMetRS L43G in either body wall muscles or neurons. After incorporating photo-ANA, we performed photo-cross-linking to covalently link all interacting proteins. Animals were then lysed for the immunoblotting analysis using anti-HA antibodies. Photo-cross-linking led to protein bands with higher molecular weight than the HSP90 monomers (SI Appendix, Fig. S7A), indicating covalent linkage between HSP90 and its interactors. We then performed two rounds of enrichment 1) using the clickable alkyne handle to enrich cell-specific proteins, and 2) using the HA tags to enrich HSP90 and its interacting proteins (Fig. 4A). As expected, after UV-induced cross-linking and double enrichment, we observed several protein bands above 90 kDa, corresponding to enriched HSP90-interactor complexes (SI Appendix, Fig. S7 B and C).

Fig. 4.

Photo-ANA labeling enables the identification of tissue-specific PPIs. (A) Experimental scheme to isolate tissue-specific HSP90-interacting proteins through photo-ANA labeling, click chemistry, and double enrichment. (B) The setup for the SILAC-based mass spectrometry approach to identify tissue-specific HSP90 interactors. (C) A Venn diagram showing the overlapping of the identified HSP90 interactors from body wall muscle and neurons. (D) Two-dimensional scatter plot of proteins identified and quantified (H/L ratio > 1.5) in the HSP90 interactomes from different tissues. The Upper Left Quadrant contains neuron-specific interactors, Lower Right Quadrant for muscle-specific interactors, and Upper Right Quadrant for common ones. Labeled proteins are known HSP90 interactors. The full list can be found in Dataset S5. (E) GO-term enrichment analysis of identified proteins from body wall muscle and neurons. Only significant (P value < 0.05) GO terms are shown. The full list can be found in Dataset S6.

We then performed SILAC experiments to profile tissue-specific HSP90 interactors. C. elegans strains were fed with E. coli grown in either heavy or light medium containing photo-ANA. The heavy isotope-labeled C. elegans were subjected to photoactivation, but not the light C. elegans. Cell lysates extracted from heavy and light C. elegans were then pooled together for two rounds of enrichment followed by mass spectrometry analysis (Fig. 4B). We applied very stringent washing during the enrichment to remove proteins noncovalently bound to HSP90; hence, the identified proteins with high H/L ratios were expected to be HSP90 interactors that were cross-linked with HSP90 during photoactivation. Using this approach, we identified 142 and 125 candidate HSP90 interactors from body wall muscles and neurons, respectively, with 16 proteins shared between the two tissues (Fig. 4C, SI Appendix, Fig. S7C, and Dataset S5). We used STRING to analyze the PPI network and found that the P-value for PPI enrichment was lower than 1 × 10⁻¹⁶, indicating that the identified proteins have significantly more interactions among themselves than what would be expected for a random set of proteins. In fact, based on the information in the STRING database, 81.7% of proteins in our PPI network interact with HSP90 either directly or indirectly through binding partners (SI Appendix, Fig. S7 D and E). In addition, among the identified proteins, 34 were reported to directly interact with HSP90, including UNC-45 (30, 31) (an HSP90 cochaperone for myosin), SIR-2.1 (32) (a client protein directly stabilized by HSP90), and NED-8 (33) (a ubiquitin-like protein that interacts with chaperones to promote proteasomal degradation of target proteins) (Fig. 4D and SI Appendix, Fig. S7D).

Functional enrichment analysis revealed that the HSP90 interactors were enriched in pathways related to protein translation, unfolded protein binding, and protein transport, which is in line with the roles of HSP90 in the folding, intracellular transport, maintenance, and degradation of proteins (Fig. 4E and Dataset S6). Interestingly, HSP90 interactors in muscles and neurons were enriched in different GO terms, suggesting that HSP90 may regulate distinct biological processes in different tissues. For example, UNC-45 is required for muscle myosin assembly and was found only in the muscle interactome, while neuronal HSP90 interactors showed enrichment in the dendrite terminus, a structure that does not exist in muscles. The above results show that the MACSPI can profile PPIs with tissue selectivity in intact animals.

Identification of Tissue-Specific, Stress-Induced HSP90 Interactors.

One promising application of the MACSPI method is to monitor the dynamics of PPIs under different cellular states. Since HSP90 plays important roles in stress response and maintenance of cellular homeostasis, we used MACSPI to profile tissue-specific HSP90 interactors in response to thermal stress. Transgenic C. elegans strains expressing CeMetRS L43G in muscle and neurons were labeled with heavy or light isotope; both were also labeled with photo-ANA, but only the heavy isotope-labeled animals were subjected to heat shock. After photoactivation, cell lysates were pooled together for double enrichment and mass spectrometry (Fig. 5A) to detect proteins with high H/L ratio, which were considered as thermal stress–induced HSP90 interactors. We identified 149 and 250 HSP90 interactors in body wall muscles and neurons, respectively, with 43 proteins shared between the two tissues (Fig. 5 B and C and Dataset S7). The P-value of the PPI enrichment among the identified proteins was below 1 × 10⁻¹⁶, and 89.8% of the proteins were biologically related (SI Appendix, Fig. S8A).

Fig. 5.

Identification of thermal stress–induced HSP90 interactors in selective tissue. (A) A SILAC-based mass spectrometry approach using photo-ANA to identify tissue-specific HSP90-interacting proteins induced by thermal stress. (B) A heatmap comparing the H/L ratio of identified stress-induced HSP90 interactors between muscle and neurons. The heatmap was constructed by hierarchical cluster analysis; red and green indicate high and low H/L ratios, respectively. (C) Venn diagram comparison of identified HSP90 interactors upon thermal stress from body wall muscle and neurons. The full list can be found in Dataset S7. (D) Known HSP90 interactors were identified from muscle and neuronal HSP90 interactomes. Nodes with purple borders represent orthologs of known HSP90 interactors in humans. (E and F) Coimmunoprecipitation results validating the HSP-90/HSP-4 interaction in body wall muscle (E) and HSP-90/VDAC-1 interaction in neurons (F) upon thermal stress. Transgenic animals with endogenous HA-tagged HSP-90 and muscle- or neuron-specific expression of myc-tagged candidates were used for the co-IP. (G) Functional enrichment analysis of HSP90 interactors from body wall muscle (green) and neurons (red) by REVIGO. Only pathways with FDR < 0.05 are shown. The cluster was created using Markov’s clustering algorithm. Annotation was created based on the most frequent and adjacent words of the names of pathways within a cluster and rephrased for better readability.

Among the stress-induced HSP90 interactors, 54 proteins or their homologs in other species were reported to interact directly with HSP90 and have roles in stress response (Fig. 5D). Among the HSP90 interactors commonly found in both muscles and neurons were HSP-60 and CYN-7. The mitochondrial chaperone HSP-60 is essential for the folding and assembly of newly imported proteins in the mitochondria and plays an important role in unfolded protein response and innate immune response (34), whereas CYN-7 is a peptidyl-prolyl cis/trans isomerase chaperone that binds to unfolded proteins and regulates folding at proline residues (35). In the muscle-specific HSP90 interactome, we found PKN-1, a homolog of mammalian protein kinase N that regulates muscle contraction and force transmission (36), and GDI-1, a homolog of GDP dissociation inhibitor 1 that modulates muscle degeneration (37). In the neuron-specific HSP90 interactome, we found KLC-2, a kinesin light chain that regulates synaptic vesicle transport (38), SPC-1, a homolog of human HSP90 client SPTAN1 that regulates neuronal migration and dendrite formation (39), and CHA-1, a choline acetyltransferase responsible for synthesizing acetylcholine (40). These results further supported that our method can selectively detect interactors in muscle or neurons.

To further validate the HSP90 interactors, we selected a few hits in our dataset and performed coimmunoprecipitation experiments in C. elegans expressing HA-tagged HSP90 endogenously and myc-tagged candidate proteins in specific tissues from a transgene. Using this approach, we confirmed the HSP90/HSP-4 interaction in muscle (Fig. 5E) and HSP90/VDAC-1 interaction in neurons (Fig. 5F) upon thermal stress, which showed that our MACSPI method can uncover PPIs with tissue specificity.

Next, we compared the stress-specific HSP90 interactome with unstressed interactome obtained under normal physiological conditions. As expected, the SILAC analysis showed that most interactors did not overlap between the two datasets (SI Appendix, Fig. S8B and Dataset S8). Pathway and functional enrichment analyses showed that the stress-specific interactors were enriched in pathways related to translation and cellular component organization, supporting the strong involvement of HSP90 in protein expression and cellular homeostasis (Fig. 5G, SI Appendix, Fig. S8C, and Datasets S9 and S10). Neuronal interactors were specifically enriched in pathways related to extracellular matrix-receptor interaction, mitochondrial organization, cell adhesion, cytoskeleton, etc., which is consistent with the involvement of HSP90 chaperones in the development and maintenance of the nervous system, especially at the synaptic connections between neurons (41) (SI Appendix, Fig. S9A). Interestingly, interactions between HSP90 and several stress-related proteins (e.g., NSUN-1, EPI-1, PRMT-1, HSP-16.41, TAX-6, HSP-16.1, and PAR-5) were found only in the neuronal interactome under thermal stress, supporting the dynamics of HSP90 interaction networks under different cellular states. In fact, the heat shock–induced neuronal HSP90 interactors were enriched in GO terms for unfolded protein binding and calcium/calmodulin binding, which are presumably related to the stress response (SI Appendix, Fig. S9B).

Among the stress-induced muscle HSP90 interactors, we found seven chaperone proteins (HSP-60, HSP-4, CCT-7, UNC-45, CNX-1, ENPL-1, and CRT-1), 20 stress response proteins (HSP-3, HSP-4, HSP-60, HSP-16.48, IRG-7, PEPT-1, CRT-1, SEL-11, EIF-2BDELTA, ENPL-1, PRMT-5, VAB-10, C14B9.2, T24H7.2, VSRA-1, DSC-4, DPY-18, CATP-3, GLY-5, and CNX-1), and 25 Adenosine triphosphate (ATP)-binding proteins (e.g., HAF-3, DDX-46, DDX-19, CATP-3, SUCA-1, PAT-4, MCCC-2, IARS-1, ABTM-1, UNC-43, KARS-1, PKN-1, PPK-2, etc.). The KEGG pathway enrichment analysis showed that the muscle interactors were related to phagosome and citrate cycle, which were also pathways significantly enriched in the muscle proteome (Fig. 3D and SI Appendix, Fig. S10A). The heat shock–induced HSP90 interactors were also enriched in stress-related pathways, such as IRE1-mediated endoplasmic reticulum unfolded protein response and ubiquitin-dependent ERAD pathway (SI Appendix, Fig. S10B). Overall, our findings highlighted key pathways regulated by HSP90 in response to thermal stress in different tissues of C. elegans. These results demonstrated that the MACSPI method can robustly profile tissue-specific PPIs in response to stress in intact animals.

Discussion

In this study, we applied the MACSPI approach to analyze tissue-specific proteomes and interactomes in the model multicellular organism C. elegans. This approach leverages the incorporation of a bifunctional methionine analog, photo-ANA, into proteins of selected tissues. The alkyne group of photo-ANA enables profiling of the labeled proteins, while its diazirine group mediates covalent linkages of interacting proteins upon photoactivation, allowing the capture of transient or weak PPIs. The incorporation of photo-ANA in tissues can be easily controlled by the expression of tissue-specific MetRS mutants capable of attaching photo-ANA to methionyl-tRNA. For example, by expressing tissue-specific MetRS mutants in body wall muscles and all neurons, we showed that MACSPI can profile muscle- and neuron-specific proteomes and interactomes. Future efforts will focus on profiling proteomes and interactomes in specific neuron types in intact animals, such as the six touch receptor neurons. In fact, our method was able to label these cells out of ~1,000 somatic cells. We are currently working on improving the efficiency of protein isolation and the sensitivity of mass spectrometry detection to achieve higher cellular resolution.

We used the ubiquitously expressed molecular chaperone HSP90 as an example to assess the tissue-specific interactomics approach and identified tissue-specific interactors involved in the development and function of muscles and neurons. Many of these interactors were predominantly expressed in only one tissue, explaining the tissue-specific PPIs. However, some tissue-specific interactors were ubiquitously expressed, suggesting that some PPIs are truly context dependent. We reason that other tissue-specific factors may mediate the tissue-specific PPIs between two ubiquitously expressed proteins. Alternatively, tissue-specific posttranslational modifications of the proteins may be important for such interactions (42). Besides finding many known HSP90 interactors, we also identified numerous previously unknown interactors; further investigation of these proteins will likely advance our understanding of the tissue-specific functions of HSP90.

Moreover, by comparing HSP90 interactomes under normal and thermal stress conditions using SILAC, we identified stress-induced HSP90 interactors in muscles and neurons. Some of these interactors were common or function in the same pathways in both muscles and neurons, whereas others were enriched in specific pathways depending on the tissue type. These results suggest that different tissues may have distinct HSP90-dependent cellular responses toward the same stress. Besides HSP90 chaperones, we expect our method to be widely applicable in identifying tissue-specific PPIs of many proteins, especially more broadly expressed proteins with distinct functions across tissue types.

In summary, we showed that MACSPI is effective in profiling the proteome and interactome of different tissues in C. elegans under different conditions. As C. elegans is a model organism for various aspects of biology, we expect our method to be useful in many scenarios for characterizing the proteomes and interactomes of specific tissues or cells throughout development and in various physiological and pathological conditions. Coupling with triple SILAC (43) or tandem mass tag-based quantitative proteomics (44), MACSPI could enable high spatiotemporal resolution of proteomic and interactomic studies during development, aging, and stress responses. As C. elegans is also a popular organism for studying microbe-host interactions, our method may also be applied to interrogate interactions between C. elegans and infectious microbes or gut-colonizing bacteria by identifying the interactions between host proteins and microbial proteins during infection or colonization.

In terms of technical limitation of MACSPI, a key concern is the false positives generated by endogenously biotinylated proteins that are not labeled by photo-ANA but can still bind to the streptavidin beads. This problem exists for essentially all biotin-based affinity purification methods but can be partially mitigated by the SILAC approach we adopted in this study. Since endogenously biotinylated proteins are enriched from both heavy and light samples, they are not likely to show significant differences in the signal intensities between the heavy and light forms. In contrast, photo-ANA-labeled proteins only exist in the heavy sample and can easily stand out in the proteomic analysis. Another issue concerns the false negatives. Given that the cross-linking radius of photo-ANA is around 3.4~4.5 Å, the covalent linkage between the protein of interest and its interactors depends on the presence of photo-ANA at or near the PPI interface. Thus, the position of methionine residues (which can be replaced by photo-ANA) in the proteins involved in the interaction would likely affect the efficiency of the cross-linking. Inevitably, some interactors may not be captured through MACSPI due to the lack of properly positioned methionine residues. In that case, combining the use of photo-ANA with other photoreactive amino acid [e.g., photo-lysine (45)] may help improve the coverage.

At last, although we only presented data for the use of MACSPI in C. elegans, this method can be easily applied to many multicellular organisms based on the following two reasons. First, the MetRS gene is highly conserved across metazoan. A mutation homologous to the CeMetRS L43G can be easily introduced into the native MetRS gene, and the mutant MetRS proteins can then be expressed from a cell-specific promoter to allow the labeling of proteins in specific tissues of the organism. Second, photo-ANA is a methionine analog that follows a similar metabolic route as methionine. After simple dietary supplementation, the probe can enter any tissue of the organism and be incorporated into the proteome of any cell that expresses the engineered MetRS. These features of MACSPI suggest its universal applicability to profile cell-specific proteomes and interactomes in multicellular organisms.

Materials and Methods

Transgenic C. elegans Strains.

C. elegans strains were mostly maintained at 20 °C, unless otherwise indicated, using previously described methods (46). CeMetRS expression constructs were generated by first cloning the mars-1 genomic coding region and ligating it to the downstream of either a 1.2 kb rab-3 promoter (for pan-neuronal expression) or a 2.5 kb myo-3 promoter (for body wall muscle expression) in a pDEST (Invitrogen) backbone containing unc-54 3′-untranslated region using Gibson Assembly (New England Biolabs or NEB). L43G mutation was then introduced into mars-1 coding sequence through Q5 site-directed mutagenesis (NEB), and DNA fragments encoding two tandem FLAG tags were inserted before the stop codon of mars-1. The resulted tissue-specific CeMetRS L43G expression vectors were injected into unc-119(ed3) mutants using Cb-unc-119(+) as a coinjection marker. Stable lines were then established and maintained. In the case of the pan-neuronal CeMetRS L43G expression strain, we integrated the extrachromosomal array through γ-irradiation and then outcrossed the integrated strain five times to establish the CGZ379 strain. The body wall muscle-specific CeMetRS L43G expression strain CGZ116 was kept as a stable line. To create the endogenous 2×HA knock-in at the hsp-90 locus before the stop codon, we adopted a CRISPR/Cas9-mediated gene editing protocol (47). We chose 5′-TTTAGTCGACCTCCTCCATG-3′ as the Cas9 target and used single-stranded DNA oligos as the repair template. Recombinant spCas9 (NEB) and sgRNA synthesized using the NEB sgRNA synthesis kit, along with the repair template, were injected into CGZ379, and the successfully edited animals were identified and confirmed by genotyping. The resultant strain was named CGZ498 for neuronal HSP90 interactomics. The same editing was done to generate CGZ743 for body wall muscle studies. All strains used in this study are listed in SI Appendix, Table S1. Additional strain information can be found in SI Appendix.

Stable Isotope Labeling by Amino Acids in C. elegans for Tissue-Specific Proteomics.

For stable isotope labeling by amino acids, we modified a previously reported C. elegans SILAC protocol (25). We deleted metB in the LSE1 E. coli strain and transformed it with a plasmid expressing EcMetRS L13G. The resulted CGZ1760 strain is a ΔargA ΔlysA ΔmetB triple knockout expressing both EcMetRS L13G and dsRNAs against the C. elegans orn-1 gene from T7 promoter. A single colony of the above bacteria was cultured in heavy medium (M9 minimal medium with 30 mg/L L-methionine, 30 mg/L arginine-¹³C₆¹⁵N₄¹H₁₄¹⁶O₂, 30 mg/L lysine-¹³C₆¹⁵N₂¹H₁₄¹⁶O₂) or light medium (M9 minimal medium with 30 mg/L L-methionine, 30 mg/L arginine-¹²C₆¹⁴N₄¹H₁₄¹⁶O₂, 30 mg/L lysine-¹²C₆¹⁴N₂¹H₁₄¹⁶O₂) at 37 °C. Overnight bacterial culture was diluted 1:100 into freshly prepared corresponding medium supplemented with 1 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG), grown to OD₆₀₀ = 1.0, and then were pelleted, washed twice with phosphate-buffered saline (PBS), and resuspended in corresponding medium with 1 mM IPTG and 2 mM photo-ANA or methionine. After overnight incubation at 37 °C, triply labeled bacteria were concentrated and plated on NGM plates.

Transgenic worms were subsequently transferred to the plates with triply labeled bacteria. After labeling, worms were washed off the plates, incubated in M9 medium for 30 min to empty the intestine, harvested, and lysed in sodium dodecyl sulfate (SDS) lysis buffer [1% SDS, 150 mM NaCl, 50 mM 4-(2-Hydroxyethyl)piperazine-1-ethane-sulfonic acid (HEPES), 2 mM MgCl₂, 0.1% TWEEN 20, 20% glycerol, pH 7.5 with 2 mM phenylmethylsulfonyl fluoride (PMSF), and 1× cOmplete protease inhibitor cocktail] with sonication. Two milligrams of lysate extracted from the heavy and light groups, respectively, were pooled together at 1:1 ratio and reacted with 100 μM cleavable biotin azide by “click” chemistry with 1 mM tris-(2-carboxyethyl)phosphine, 100 μM tris(benzyltriazolylmethyl)amine, and 1 mM CuSO₄ at room temperature for 1 h. Ice-cold acetone was added to precipitate proteins, which were washed twice with methanol. Protein pellets were dried and dissolved in PBS with 20 mM ethylenediaminetetraacetic acid, 4% SDS, and 10% glycerol, heated at 75 °C for 8 min, and then diluted with PBS to an SDS concentration of 0.5%. Streptavidin agarose beads were added to samples and incubated at room temperature for 2 h. The beads were then washed with PBS containing 0.2% SDS, 6 M urea in PBS with 0.1% SDS, and 250 mM NH₄HCO₃ with 0.05% SDS. Enriched proteins were eluted by 25 mM Na₂S₂O₄ and 0.05% SDS in 250 mM NH₄HCO₃ for 1 h. The eluate was collected, dried, and resuspended with 1× lithium dodecyl sulfate (LDS) sample loading buffer containing 50 mM dithiothreitol (DTT), heated at 85 °C for 8 min, and then reacted with 130 mM iodoacetamide for 30 min. Samples were then separated by a 4 to 12% Bis-Tris gel and subjected to in-gel digestion and desalting before mass-spectrometry analysis. Two biological replicates were performed for each experiment. Details about each step in the experimental procedure can be found in SI Appendix. Key antibodies and reagents were listed in SI Appendix, Table S1.

SILAC Profiling of Tissue-Specific HSP90 Interactors in C. elegans.

C. elegans strains expressing HA-tagged HSP-90 and CeMetRS L43G mutant were fed with E. coli grown in either heavy or light medium, both supplemented with photo-ANA. The heavy isotope-labeled C. elegans were subjected to photo-cross-linking by 365-nm UV light for 30 min at room temperature in PBS, while the light C. elegans were not. Animals were then lysed in the SDS lysis buffer. Four milligrams of the cell lysate (diluted to 1 mg/mL) from the heavy and light worms, respectively, were pooled together at 1:1 ratio and then subjected to two rounds of enrichment: first using the clickable alkyne handle to enrich cell-specific proteins (as mentioned above) and then using the HA tag to enrich HSP90 and its interacting proteins. The photo-ANA-labeled proteins eluted from the streptavidin beads were precipitated out by 20% TCA and resuspended in 1 mL of immunoprecipitation buffer (150 mM NaCl, 50 mM HEPES, 2 mM MgCl₂, 0.1% TWEEN 20, 20% glycerol, pH = 7.4, 2 mM PMSF, and 1× cOmplete protease inhibitor cocktail) and then incubated with 50 μL of anti-HA magnetic beads overnight at 4 °C. The beads were washed with PBS containing 0.1% Triton X-100 and eluted in 50 μL of SDS lysis buffer containing 50 mM DTT. The isolated proteins were heated at 75 °C for 8 min, alkylated in 130 mM iodoacetamide for 30 min, and then separated by a 4 to 12% Bis-Tris gel. After staining, each lane of the gel above 90 kDa was sliced into 5 pieces and further diced into 1 mm cubes, which were destained, digested by trypsin, and prepared for mass-spectrometry analysis. To profile heat stress–induced HSP90 interactors, the heavy isotope-labeled worms were incubated at 30 °C for 3 h before photo-cross-linking, while the light C. elegans were kept at 20 °C. Two biological replicates were performed for each experiment. Details about each step can be found in SI Appendix.

Other Methods.

Detailed information of the above methods and other methods, including bacterial engineering, plasmid construction, metabolic labeling of E. coli and C. elegans, click chemistry in solution and in situ, western blotting, immunoprecipitation, mass spectrometry, and proteomic data analysis, can be found in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (XLSX)

Dataset S07 (XLSX)

Dataset S08 (XLSX)

Dataset S09 (XLSX)

Dataset S10 (XLSX)

Data, Materials, and Software Availability

Materials generated in this study and additional information required to reanalyze the data are available upon reasonable request. The raw proteomic data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with accession no. PXD048632 (48). All processed data are included in the main text or SI Appendix.