key: cord-0037022-elb93kp6
authors: Li, Shitao
title: Proteomics Defines Protein Interaction Network of Signaling Pathways
date: 2012-12-27
journal: Bioinformatics of Human Proteomics
DOI: 10.1007/978-94-007-5811-7_2
sha: be3bdfa49d9808a8800adec9f9a02108d43defd3
doc_id: 37022
cord_uid: elb93kp6

Protein interactions play fundamental roles in signaling transduction. Analysis of protein–protein interaction (PPI) has contributed numerous insights to the understanding of the regulation of signal pathways. Different approaches have been used to discover PPI and characterize protein complexes. In addition to conventional PPI methods, such as yeast two-hybrid (YTH), affinity purification coupled with mass spectrometry (AP-MS) is emerging as an important and popular tool to unravel protein complex and elucidate protein function through the interaction partners. With the AP-MS method, protein complexes are prepared first by affinity purification directly from cell lysates, followed by characterization of their components by mass spectrometry. In contrast to most PPI methods, AP-MS reflects PPI under near physiological conditions in the relevant organism and cell type. AP-MS is also able to probe dynamic PPI dependent on protein posttranslational modifications, which is common for signal transduction. AP-MS mapping protein interaction network of various signal pathways has dramatically increased in recent years. Here, I’ll present the strategies toward obtaining an interactome map of signal pathway and the methodology, detailed protocols, and perspectives of AP-MS.

Protein interaction plays essential role in cell structure and function. In a simpli fi ed diagram of a signaling pathway, upon interaction of a ligand, the receptor alters its conformation, such as dimerization, phosphorylation, and ubiquitination, leading to recruitment of intracellular molecules and subsequent activation of downstream signal cascades. Each level of the signaling cascades requires protein interaction to work as a well-assembled, multifunctional protein complex essential for signal transduction. The functionality of proteins relies on their ability to interact with one another, whereas pathogenic conditions can re fl ect the perturbations of these protein interactions.

Numerous protein-protein interaction (PPI) methods have been developed, but only a few of them are used for large-scale PPI detection, including yeast twohybrid (YTH), protein fragment complementation assay (PCA), luciferase-mediated interactome (LUMIER), mammalian protein-protein interaction trap (MAPPIT), protein array, and af fi nity puri fi cation coupled with tandem mass spectrometry (AP-MS). The YTH system is the fi rst assay for analysis of large-scale protein-protein interactions and widely accepted method (Fields and Song 1989 ) . In YTH system, interested gene (bait, X) is fused to the DNA-binding (DB) domain of a transcription factor such as Gal4 (DB-X), while the interacting protein (prey, Y) is fused to an activation domain (AD) such as Gal4-AD (AD-Y). Physical interaction between X and Y brings AD and DB together, which reconstitutes the transcription factor and subsequently activates the downstream reporter genes (Fields and Song 1989 ) . Like the YTH, PCA requires that bait and prey are each fused with incomplete fragments of a third protein, which acts as a reporter. Interaction between bait and prey proteins brings the fragments of reporter protein in close enough proximity to allow them to form a functional reporter protein (Rossi et al. 1997 ) . When fl uorescent proteins are reconstituted, the PCA is called bimolecular fl uorescence complementation assay (Kerppola 2009 ) . LUMIER is basically a co-immunoprecipitation assay, in which bait is linked to an epitope for puri fi cation and prey protein is fused to renilla or fi re fl y luciferase for detection (Barrios-Rodiles et al. 2005 ) . In the MAPPIT, bait and prey proteins are linked to signaling de fi cient cytokine receptor chimeras. Interaction of bait and prey restores JAK-STAT cascade after the receptor has been stimulated with ligand, which leads to STAT3-dependent reporter gene activation (Eyckerman et al. 2001 ) . Protein microarray is a microscopic array glass slide on which interested proteins have been af fi xed at separate locations in an ordered manner using a variety of available chemical linkers (MacBeath 2002 ) . Protein microarrays are typically high-density arrays that are used to identify novel proteins or protein-protein interactions. Antibody microarrays are the most common analytical microarray.

AP-MS is biochemical puri fi cation of protein complexes followed by characterization of their components by mass spectrometry. However, unlike the methods discussed above, AP-MS is not designed for one-to-one protein interaction (i.e., binary interaction). Instead, AP-MS detects multi-protein complexes. As with 2 Proteomics De fi nes Protein Interaction Network of Signaling Pathways AP-MS, gene of interests is tagged with desirable epitope for af fi nity puri fi cation. Various tags have been developed, such as FLAG tag, HA tag, glutathione S-transferase (GST) tags, the calmodulin-binding peptide, the streptavidin-binding peptide, or the in vivo biotinylation of the target tagged peptide using coexpression of the BirA ligase (Waugh 2005 ) . With af fi nity tag, protein complexes are enriched fi rst by af fi nity puri fi cation. One early developed AP-MS is to use the tandem af fi nity puri fi cation (TAP) tag (Puig et al. 2001 ) . The original TAP tag is composed of a protein A tag and a calmodulin-binding peptide for two sequential enrichment puri fi cations. In the fi rst puri fi cation step, the protein complex is isolated from the cell lysate using immunoglobulin gamma (IgG) resin with high protein A af fi nity. After protein complex is cleaved from the protein A tag with TEV protease, the eluate undergoes second puri fi cation on an immobilized calmodulin column.

To date, AP-MS has been performed in combination with other techniques, such as biochemical fractionation and chemical cross-linking, for characterization of protein complex. Combining biochemical fractionations, like size fractionation, with AP-MS can provide a more precise characterization of multi-protein complexes according to the factions. For example, a combination of TAP puri fi cation with standard gel fi ltration has allowed for a better characterization of RNA polymerase II complex (Mueller and Jaehning 2002 ) . Crosslinker is used for detecting weak interactions, such as membrane complex, which may be interrupted by detergents in lysis buffer. A combination of TAP with in vivo cross-linking with formaldehyde was used to identify novel proteasome interactors (Tagwerker et al. 2006 ) . AP-MS can also be combined with quantitative proteomics approaches, such as SILAC and ICAT, to better understand the dynamics of protein complex assembly. Stable isotope labeling by amino acids in cell culture (SILAC) is an approach for in vivo incorporation of a label into proteins for mass spectrometry (MS)-based quantitative proteomics (Ong et al. 2002 ) . Isotope-coded af fi nity tags (ICAT) are complementary to SILAC and measure dynamic changes in complexes isolated from tissues or organisms that cannot be metabolically labeled (Gygi et al. 1999 ) . Both entail labeling the samples with isotope labels that allow the mass spectrometer to distinguish between identical proteins in separate samples. Differentially labeled samples are combined and analyzed together, and the differences in the peak intensities of the isotope pairs accurately re fl ect difference in the abundance of the corresponding proteins.

Given the fundamental importance of protein interactions, systematically mapping protein-protein interaction (PPI) in various species has dramatically increased in recent years. Using high-throughput YTH, proteome-wide physical interaction maps have been generated for several organisms: Saccharomyces cerevisiae (Fromont-Racine et al. 1997 ; Uetz et al. 2000 ; Ito et al. 2001 ) , Caenorhabditis elegans (Walhout et al. 2000 ; Reboul et al. 2003 ; Li et al. 2004 ) , Drosophila melanogaster (Giot et al. 2003 ; Guruharsha et al. 2011 ) , and human (Guruharsha et al. 2011 ; Rual et al. 2005 ) . Virus-host protein interactomes were also explored, such as severe acute respiratory syndrome (SARS)-coronavirus (Pfefferle et al. 2011 ) , Kaposi sarcoma herpesvirus (KSHV), and Varicella zoster virus (VZV) (Uetz et al. 2006 ; Rozen et al. 2008 ) . In addition to global mapping, protein interaction networks of several important signal pathways, such as MAPK (Bandyopadhyay et al. 2010 ) , TGF b ( Tewari et al. 2004 , SMAD (Colland et al. 2004 ) , and PI3K-mTOR (Pilot-Storck et al. 2010 ) , have been investigated.

In addition to YTH, AP-MS is another widely used PPI tool to map protein interactomes. Due to many advantages that will be discussed later, AP-MS mapping protein interaction network of various signal pathways has dramatically increased in recent years. Global-wide interactomes have been established in Escherichia coli (Hu et al. 2009 ) and Mycoplasma pneumonia (Kuhner et al. 2009 ) , Saccharomyces cerevisiae (Krogan et al. 2006 ; Gavin et al. 2006 ; Ho et al. 2002 ) , Drosophila melanogaster (Guruharsha et al. 2011 ) , and HIV-host interactome (Jager et al. 2012 ) . In vertebrate, this approach has so far been used to de fi ne proteomic subspaces or speci fi c signal pathways: antiviral innate immunity pathway (Li et al. 2011 ) , autophagy pathway (Behrends et al. 2010 ) , deubiquitinase interactome (Sowa et al. 2009 ) , endoplasmic reticulum-associated protein degradation network (ERAD) (Christianson et al. 2012 ) , TNF pathway (Bouwmeester et al. 2004 ) , proteasome interaction network (Guerrero et al. 2008 ) , and disease-related protein network (Ewing et al. 2007 ) .

Systematic identi fi cation of protein interactions within an organism will facilitate systems-level studies of biological processes. Current binary PPI networks are mainly generated by high-throughput yeast two-hybrid. Due to the small overlap of these maps, it has been assumed that these maps are of low quality containing many false positives (Parrish et al. 2006 ) . Recent efforts to map interactions using AP-MS illustrate the promise to measure speci fi c protein interactions in vivo (instead of in yeast) and provide a more powerful tool to model the in vivo interactome. First, I discuss the advantages of AP-MS versus YTH, and then focus the details of the methodology, applications, and perspectives of AP-MS.

Despite the wide acceptance of YTH system for protein-protein interaction analysis and discovery, high-throughput YTH for protein interaction network bears several major limitations: (1) Reporter analysis method indirectly re fl ects protein-protein interaction which usually leads to high false positives. For example, proteins with transcriptional activity can lead to autoactivation of the reporter genes. (2) Some heterologous protein expressions are incompatible or toxic to yeast, i.e., membrane proteins which are unlikely to be appropriately assayed as a fusion with a reconstituted transcription factor in YTH. (3) YTH cannot re fl ect the endogenous protein interactions in the relevant organism. (4) Lots of signaling pathways in vertebrates do not exist in yeast. Thus, interactions triggered by posttranslational modi fi cations do not occur in yeast, resulting in many intrinsic false negatives. (5) The coverage of prey library usually is not completed. In addition, in high-throughput YTH, the bait expression is not monitored. Heterologous full-length protein expression, especially high-molecular-weight protein, expects to have low expression level in yeast.

Although both YTH and AP-MS detect protein-protein interaction, they have several distinct differences (Table 2 .1 ). AP-MS couples af fi nity puri fi cation with mass spectrometry and requires more labor works and sophisticated equipments. Basically, baits can be expressed in any cell line, which investigator is interested in. After antibiotic selection, bait expression levels are monitored in stable cell lines by western blot, and cell line expressing low bait protein level (close to endogenous level) is usually chosen for following af fi nity puri fi cation. Since the bait expression is close to the counterpart endogenous protein level, we expect the puri fi ed complex re fl ects the endogenous protein interactions under physiological conditions. AP-MS also can be used to detect dynamic protein interactions dependent on protein posttranslational modi fi cation by signal stimulation. Unlike YTH detecting one-to-one interaction (aka binary interaction), AP-MS analyzes the entire bait complex and provides all prey information in one run. However, the puri fi ed complex represents a mix of direct and indirect binding partners since the nature of the interactions identi fi ed in AP-MS data cannot be determined to be either direct or indirect. Last, protein abundance and speci fi city in different cell lines also limits the detection of protein complex. For example, MIB1 and MIB2 have comparable af fi nity with TBK1, but we did not detect MIB2 in TBK1 complex in 293T cells by AP-MS. Using real-time PCR, we found MIB1 predominantly expressed in 293T cell line (Li et al. 2011 ) . Taken all together, AP-MS overcomes the limitations of YTH discussed above except several disadvantages over YTH: high cost, indirect interaction, and cell type speci fi city.

The pipeline of AP-MS from gene construction to interaction network mapping is shown in Fig. 2 .1 (Li et al. 2011 ) . In brief, interested gene is tagged with desirable epitopes such as FLAG, GST, His, and biotin. Depending on the puri fi cation strategy, one or two tags (usually tandem tags) are adopted. These vectors should carry one antibiotic resistance gene for mammalian cell stable line selection. After transfection or infection into the desirable mammalian cell line, cells are selected by designated antibiotics to obtain stably and close to endogenous protein expression. Protein complexes are precipitated from lysates of bulk cells by using various immobilized matrixes, such as resin conjugated with antibody. Protein complexes are then eluated from the matrixes after several washing steps to remove nonspeci fi c interactors. Protein complex is either separated on gel following silver staining or precipitated. Sliced gel bands or solution samples are analyzed by mass spectrometry. After data collection and statistical analysis, protein interaction network is generated and ready for validation and further function analysis. 

To purify protein complex closing to physiological level, cell line stably expressing tagged bait is a prerequisite. Therefore, antibiotic resistance gene should be included in the vector for stable cell line selection. Genes of interest also needs to be tagged in-frame with an epitope (at either the N or C terminus), which is used to af fi nity purify the tagged protein (aka bait) along with its interacting partners (aka prey). Any af fi nity tag can be used for AP-MS in theory, and most successful tags developed to date are FLAG, HA, S-tag, and tandem af fi nity puri fi cation (TAP) tag. Each puri fi cation tag has advantages and disadvantages, and the appropriate technique should be selected depending on the goals of the experiment. For example, a single FLAG or HA epitope only adds 8-11 amino acids (Li et al. 2011 ) , while the TAP tag adds a >20-kDa tag (Krogan et al. 2006 ) which may cause more nonspeci fi c binding. Because tag may interfere with protein expression or interaction, both N-terminal and C-terminal fusion could be tested for optimal AP-MS. For example, membrane protein may need to put the tag on the C-terminal or after signal peptide on the N-terminus. Furthermore, two kinds of puri fi cation methods (single and tandem puri fi cation) are used for AP-MS, which requires bait fused with single or double epitopes, respectively.

Depending on the number of tags on the vector, there are one-step and two-step puri fi cation methods for speci fi c protein complex, cell line, or organism. Originally developed for yeast, the fi rst TAP tag consists of calmodulin-binding peptide (CBP), followed by tobacco etch virus protease (TEV protease) cleavage site and protein A with high af fi nity to immunoglobulin gamma (IgG). Protein complex is fi rst puri fi ed from the cell lysate on an IgG af fi nity resin and cleaved from the protein A tag with TEV protease. The eluate is then enriched in a second af fi nity puri fi cation step on an immobilized calmodulin column. Several variants of TAP with different combinations of tags, such as FLAG-HA double tags, are developed. Usually, one-step puri fi cations on average preserve weaker or more transient protein-protein interactions in the price of a higher number of nonspeci fi c binding proteins. Conversely, the tandem procedure tends to yield cleaner results, but weak interactions can be lost. FLAG and HA double tags are most commonly applied for tandem puri fi cation of protein complexes. We compared the effect of tandem tag versus single tag puri fi cation on the yield of total prey and HCIP by examining four protein complexes puri fi ed by single puri fi cation with FLAG versus a two-step puri fi cation with FLAG followed by HA (Li and Dorf 2013 ) . MS analysis revealed that the number of total interactors was dramatically reduced in all protein complexes (TBK1, NAP1, IRF3, and SINTBAD) isolated by TAP puri fi cation. However, the ratio of HCIP to total prey did not increase. Consistently, more HCIP were detected by single-step af fi nity puri fi cation ( Fig. 2. 2 ). In brief, tandem puri fi cation reduces the NSBP at the price of HCIP loss. Due to on average more than 90% of proteins as nonspeci fi c binding protein in one-step puri fi cation, researchers prefer to tandem af fi nity puri fi cation to get a cleaner background if they only study on a few protein complexes. However, if the study is to map the protein interaction network of a speci fi c signaling pathway, NSBP from one-step puri fi cation can be excluded by statistical analysis of the whole database.

In most proteomics experiments, the puri fi ed proteins are separated by onedimensional SDS-PAGE and stained with a mass spectrometry-compatible dye such as silver, SYPRO ruby, or Coomassie. SDS-PAGE separation removes unwanted contaminants such as buffer components from the protein sample, and the sample complexity is decreased by separating the proteins according to molecular weight. Moreover, it also can be used to compare bands distribution with and without stimulation. In some cases, like IRF3 complexes shown in Fig. 2 .1 , unique bands are only found in the bait complex with stimulation, indicating these interacting proteins are dependent on ligand stimulation.

Individual protein bands of interest are excised, or the entire lane is cut into approximately 1-mm 3 pieces. Gel pieces were then subjected to an in-gel trypsin digestion procedure to produce peptides for mass spectrometry analysis. But the extraction ef fi ciency of peptides from a gel is low and dependent on the primary structure of the peptide. As an alternative approach to in-gel digestion, protein mixtures can be digested in solution without prior separation (Behrends et al. 2010 ) . Because buffer components, such as detergents, interfere with the mass spectrometry ionization process, protein samples need to be precipitated with trichloroacetic acid (TCA), washed, and redissolved in a digestion buffer. The main advantages of solution digestion are the reduction of the time and a higher recovery of peptides compared to in-gel digestion. However, bear in mind that some proteins like membrane proteins are resistant to be redissolved. The peptide mixture can be directly introduced into the mass spectrometer or separated by HPLC before mass spectrometric analysis (LC-MS). The two primary mass spectrometry methods developed for identi fi cation of proteins are electrospray ionization (ESI) (Fenn et al. 1989 ) and matrix-assisted laser desorption/ionization (MALDI) (Hillenkamp et al. 1991 ) . Electrospray ionization mass spectrometry is a desorption ionization method. A sample solution is sprayed from a small tube into a strong electric fi eld in the presence of a fl ow of warm nitrogen to assist desolvation. The droplets formed evaporate in a region maintained at a vacuum of several torr causing the charge to increase on the droplets. The multiply charged ions then enter the analyzer. The most obvious feature of an ESI spectrum is that the ions carry multiple charges, which reduces their mass-to-charge ratio compared to a singly charged species. This advantage allows mass spectra to be obtained for large molecules. A major disadvantage is that this technique cannot analyze mixtures very well. The other most used technique, MALDI, is a two-step process. First, desorption is triggered by a UV laser beam. Matrix material heavily absorbs UV laser light, leading to the ablation of upper layer (~micron) of the matrix material. A hot plume produced during the ablation contains many species: neutral and ionized matrix molecules, protonated and deprotonated matrix molecules, matrix clusters, and nanodroplets. The second step is ionization (more accurately protonation or deprotonation). In the most common instrumental designs, ESI and MALDI are performed with mass spectrometers capable of tandem mass spectrometry (MS/ MS) experiments. Ion traps, quadrupole time-of-fl ight instruments (Q-TOF), Fourier transform ion cyclotron resonance (FT-ICR) mass spectrometers (FTMS), and the Orbitrap are the most common types of instrumentation now used in high-end protein analysis.

Most protein interactomes only represent as static entities, which however only poorly captures the dynamics of complex composition. There has been increasing efforts to detect dynamic views of interactomes using various modi fi ed AP-MS. Systematic methods to map dynamic changes include semi-quanti fi cation based on total spectral counts or ion intensities of precursor peptide (MS1) or fragment ions (MS2) and use of isotopic labeling approaches to obtain more accurate relative quanti fi cation. Relative quanti fi cation methods such as the stable isotope labeling by amino acids in cell culture (SILAC) detect differences in protein abundance among samples using nonradioactive isotopic labeling. Although relative quantitation is more costly and time-consuming, and less sensitive to experimental bias than label-free quantitation, it entails labeling the samples with stable isotope labels that allow the mass spectrometer to distinguish between identical proteins in separate samples. Differentially labeled samples are combined and analyzed together, and the differences in the peak intensities of the isotope pairs accurately re fl ect difference in the abundance of the corresponding proteins. Thus, relative quantitation may discover the dynamic interactions by comparing the change of identical protein abundances from same bait cells with and without extracellular stimulation. Absolute quantitation of proteins is also developed by using isotopic peptides entails spiking known concentrations of synthetic, heavy isotopologues of target peptides into an experimental sample (Mirgorodskaya et al. 2012 ) . However, the cost of absolute quantitation is too high and not realistic for large-scale interactome mapping.

As quantitative methods become more robust, there will be increasing demand for detection of dynamic protein interaction upon extracellular stimulation. For example, we revealed that ~20% protein interactions are dependent on ligand stimulation, such as viral dsRNA mimics poly(dI:dC), in the h uman i nnate i mmunity i nteractome for type I i nterferon (HI5) (Li et al. 2011 ) . Another example in insulin pathway, Glatter et al. de fi ned the interaction network of insulin receptor/target of rapamycin pathway in Drosophila (Glatter et al. 2011 ) . They found that 22% of the detected interactions were regulated by insulin. In addition to the quantitative power of mass spectrometry, it is also crucial to establish a stable cell line sensitive to stimulations. When overexpressed in cells, bait protein may not respond to stimuli as sensitive as the corres ponding endogenous protein.

In most cases, the raw data fi les are fi rst processed by the software controlling the respective mass spectrometry instrument. The generated data sets are then searched against a protein database using search engines such as MASCOT (Hirosawa et al. 1993 ) or SEQUEST (MacCoss et al. 2002 ) . A valid approach for validation of the chosen parameters is to search the obtained data sets against a decoy protein database. The data also need to be further fi ltered by setting speci fi c thresholds such as a minimum peptide length or a speci fi c number of peptides to consider a protein identi fi cation. Mass spectrometry has some intrinsic problems, such as the common problem of carryovers between mass spectrometry runs. To circumvent the carryover problem in mass spectrometry, we usually analyze the repeated sample in different batch. The carryovers in two independent AP-MS of the same bait will not be possible to show up twice. The record of each batch of MS runs will also help to discriminate the carryovers.

In addition to mass spectrometry, af fi nity puri fi cation also has its own inherent false positives and false negatives, which is critical general limitation encountered in the interpretation of the AP-MS due to lack of binary interaction information. False positives are nonspeci fi c binding proteins and contaminants found in puri fi ed bait complex. Several types of false positives are present in typical af fi nity puri fi ed protein samples. The most common ones are from researchers' hands when they perform puri fi cation and handle samples. These contaminants usually are keratin proteins and easy to remove from the dataset. There are also other various kinds of nonspeci fi c binding proteins: (1) proteins binding to af fi nity matrices, like STK38 and PRMT5; (2) proteins bind to af fi nity tag, like KIF11 binding to FLAG tag; (3) abundant proteins (e.g., actin, tubulin); (4) proteins prefer binding to speci fi c domain, like ribosomal proteins binding to baits with nucleic acid-binding domain; (5) and heat-shock proteins for protein folding. Therefore, it is important to use cell line stably expressing baits at near physiological levels to avoid NSBPs, as transient overexpression may probably result in protein aggregation and improper intracellular localization. To discriminate NSBP from the protein complex, repetition of AP-MS is mandatory. In our experiences, NSBPs are dramatically different in two independent AP-MS of the same bait. Proper controls including cells expressing GFP with the same epitope will be also useful to exclude NSBPs. Last, large database with the same af fi nity tag and the same cell line background from high-throughput study will be a good resource for identi fi cation of NSBPs and HCIPs. If a protein is often isolated with many unrelated bait proteins, it is easily recognized through analysis of the high-throughput data. However, systematic large-scale experiment does not allow for the subjective and individual evaluation of their results, which means the removal of potential contaminating proteins cannot be based on judging individual puri fi cations. Therefore, statistic tools for analysis of database are required to fi lter out nonspeci fi c proteins and yield high-con fi dence interacting proteins.

For statistical analysis of AP-MS data, three main parameters are protein abundance, uniqueness (the frequency of observed protein in database), and reproduci bility. Total spectral counts (TSC) have gained acceptance as a practical, label-free, semiquantitative measure of protein abundance in proteomics study. Several computational tools have been developed for the processing of AP-MS data, like CompPASS (Sowa et al. 2009 ) , SAINT (Breitkreutz et al. 2010 ) , and MiST (Jager et al. 2012 ) .

We designed a simpli fi ed method for analysis of AP-MS data, combining three main parameters: protein abundance, uniqueness (the frequency of observed protein in the database), and reproducibility. Total spectral counts (TSC) have gained acceptance as a practical, label-free, semiquantitative measure of protein abundance for proteomics studies. We adopted the z -score statistic to compare protein abundance because z -score calculates the probability of TSC occurring within a normal distribution. However, z -score does not re fl ect reproducibility. In our protocol, each protein complex is tested in 4 MS runs, so reproducibility can be readily factored into the analysis. z -Score also does not analyze information about prey occurrence (i.e., prey uniqueness). To explore the likelihood that an interaction is speci fi c, we set a value of prey occurrence at <5%. We now propose a simple 3-stage scoring system to identify HCIP. This algorithm combines z -score plus prey occurrence and reproducibility (ZSPORE) (Li and Dorf 2013 ) . In the ZSPORE scoring system, each interaction must pass all three criteria to merit classi fi cation as HCIP. The fl owchart of ZSPORE is shown as in Fig. 2.3 , and a detailed description is provided in Sect. 2.4.6 . Taken together, the ZSPORE method combines three parameters ( z -score based on TSC, prey occurrence, and reproducibility) and is a simple, ef fi cient, and robust way to analyze AP-MS data.

As with any large screening database, AP-MS also has false negatives, like lacking many known protein-protein interactions documented previously. There are several reasons why a known interaction fail to be found in AP-MS. First, statistical analysis tool may fi lter out the known interaction as a nonspeci fi c binding. Second, the nature and location of the tag might interfere bait protein function and disrupt its interactions. Third, to parallel comparison, all AP-MS experiments are performed in a same single condition. The generic conditions of af fi nity puri fi cation may be too harsh to preserve some protein interactions, such as the buffer for membrane proteins should be different from other ones. Fourth, the known protein interaction depends on different stimulation. Some proteins may be involved in several pathways and have different interactors in response to the relevant stimulation. Last, the absence of detection is often due to the protein expression level in the speci fi c cell type, especially when the cells have relative low abundances of the protein.

To visualize the protein interaction network formed by HCIPs and baits, graphic representation of two protein interactions basically consists of drawing two circles (nodes) linked by a line (edge). All interactions are combined to generate a map of (Smoot et al. 2011 ) . For comprehensive and dynamic visualization of the network, various kinds of attributes can be applied to the node and the edge by representation of different color and line thickness. In addition, the functional classi fi cations of HCIPs can be analyzed by a few online programs. For example, HCIP list can be uploaded to PANTHER (Thomas et al. 2003 ) or DAVID (da Huang et al. 2009 ) via a web interface. These programs group these proteins by protein domains, molecular functions, biological processes, and signal pathways. The functional classi fi cations may help discover common threads underlying the proteins of interest. Another approach is to obtain clues from known protein interactions to discover regulation mechanisms. Several protein-protein interaction databases are available for online search, repository, and free download, such as BioGRID, STRING, IntAct, and MINT. The BioGRID database is an online protein interaction repository with data compiled through comprehensive curation efforts. The latest version searches 31,739 publications for 510,188 raw protein and genetic interactions from major model organism species . The STRING is a database of known and predicted protein interactions. The interactions include direct (physical) and indirect (functional) associations. STRING quantitatively integrates interaction data from these sources for a large number of organisms and transfers information between these organisms where applicable (Szklarczyk et al. 2011 ) . The IntAct database provides a freely available, open-source database system and analysis tools for molecular interaction data (Kerrien et al. 2012 ) . All interactions are derived from literature curation or direct user submissions and are freely available. The MINT database focuses on experimentally veri fi ed protein-protein interactions mined from the scienti fi c literature by expert curators (Licata et al. 2012 ) .

AP-MS raw data also can be deposited in the Tranche repository (Smith et al. 2011 ) , which is a distributed fi le system into which any sort of proteomics data may be uploaded. The data then are distributed on the internet and downloaded by anyone who has access to the hash key identi fi ers for the data, which may be kept private or publicly released. In summary, all these free online programs are useful and convenient research tools for mapping, analysis, and repository of AP-MS data.

AP-MS has applied for mapping of protein interactome of various cellular signaling pathways in mammalian cells. Our lab has established an ef fi cient AP-MS pipeline for de fi ning protein interaction network and successfully applied in several pathways including h uman i nnate i mmunity i nteractome for type I i nterferon (HI5) (Li et al. 2011 ) , mi RNA pathway i nteractome (Mii), and in fl uenza-host (iHost) protein interaction network (Li and Dorf, unpublished data) . Detailed pipeline of our AP-MS is provided in this section, and how this applies on different pathways in mammalian cells will be discussed.

Genes known to regulate the studied signaling pathway are usually selected as primary baits. Baits cover from extracellular signals like ligands to cognate receptors on cell membrane and to signaling intermediates, kinases, and transcription factors involved in these signaling pathways and their family members. After analysis of primary bait AP-MS, some new and important HCIPs with primary baits are also chosen to be as secondary baits. Secondary baits will validate the association with primary baits but also expand the protein interaction network, provide new insights into this signaling pathway, and cross talk with other pathways.

Bait cDNAs can be tagged with various epitopes, such as FLAG or HA epitope. As we discussed earlier, commercially available anti-FLAG beads have much higher af fi nity than anti-HA beads. We use two mammalian expression vectors, pCMV-3Tag8 (Stratagene) and viral expression vector, pLPCX (Clontech), for transfection and infection, respectively. Vector pCMV-3Tag8 harbors a hygromycin resistance gene, while pLPCX confers cells' resistance to puromycin.

Transfection and transduction are two common DNA delivery methods into mammalian cells. For cell lines easy to be transfected like HEK293 cells, bait constructs are directly transfected into cells. For cell lines with low transfection ef fi ciency, such as THP-1 cell line, bait gene needs to be fi rst packaged into retroviral virion. The following infection will allow bait gene to integrate into cell genome DNA and subsequent expression in cells. Two days after transfection and infection, cells are treated with puromycin or hygromycin for 14 days. Single colonies are picked and expanded in 6-well plates. Protein expression levels in each colony are determined by immunoblotting. Colony with protein expression close to endogenous level is picked up for AP-MS.

Most protein interactomes are descriptions of homeostasis of a speci fi c signaling pathway, such as DUB network (Sowa et al. 2009 ) , autophagy interaction network (Behrends et al. 2010 ) , and ERAD interactome (Christianson et al. 2012 ) . However, many protein interactions depend on protein posttranslational modi fi cations induced by different stimuli. For example, we found that about 20% interactions were ligand dependent in HI5 protein interaction network (Li et al. 2011 ) . We also noticed many new interactions between in fl uenza virus protein and human host after viral infection (Li and Dorf, unpublished data) . Therefore, in our pipeline for AP-MS, each stable cell line is divided into two groups, and cells are treated with ligand speci fi c for the signaling pathway or infected with virus for studying virus-host interactome.

Each group of cells is cultured in four or fi ve 15-cm 2 culture dishes (about 5 × 10 7 cells) to scale up for af fi nity puri fi cation. Cells are lysed in 10 ml TAP buffer (50 mM Tris HCl [pH 7.5], 10 mM MgCl 2 , 100 mM NaCl, 0.5% Nonidet P40, 10% glycerol, phosphatase inhibitors, and protease inhibitors). After shaking on ice for 30 min, cell lysates were centrifuged for 30 min at 15,000 rpm. Supernatants are collected and precleared with 50 m l of protein A/G resin. After shaking for 1 h at 4°C, resin is removed by centrifugation. Cell lysates are added to 20 m l anti-FLAG M2 resin (Sigma) and incubated on a shaker for 12 h. Then the anti-FLAG resin is 3× washed (15 min/time) with 10 ml TAP buffer. After removing the wash buffer, the resin is transferred to a spin column (Sigma) and incubated with 40 m l 3× FLAG peptide (Sigma) for 1 h at 4°C in a shaker. Eluates are collected by centrifugation and stored at −80°C.

Puri fi ed complexes are loaded on 4-15% NuPAGE gels (Invitrogen) and run about 1 cm 2 distance for 8 min at 200 V. Gels were stained using the SilverQuest Staining Kit (Invitrogen). Each entire stained lane was excised and rinsed twice with 50% acetonitrile.

The Taplin Biological Mass Spectrometry Facility (Harvard Medical School) performs MS analysis for our samples. Excised gel bands were cut into approximately 1-mm 3 pieces. Gel pieces are then subjected to a modi fi ed in-gel trypsin digestion procedure. Gel pieces were washed and dehydrated with acetonitrile for 10 min followed by removal of acetonitrile. Pieces were then completely dried in a speed-vac. Gel pieces were rehydrated with 50 mM ammonium bicarbonate solution containing 12.5 ng/ m l modi fi ed sequencing grade trypsin (Promega, Madison, WI) at 4°C. After 45 min, the excess trypsin solution was removed and replaced with 50 mM ammonium bicarbonate solution to just cover the gel pieces. Peptides were later extracted by removing the ammonium bicarbonate solution, followed by one wash with a solution containing 50% acetonitrile and 1% formic acid. The extracts were then dried in a speed-vac (~1 h) and stored at 4°C until analysis.

On the day of analysis, the samples were reconstituted in 5-10 m l of HPLC solvent A (2.5% acetonitrile, 0.1% formic acid). A nanoscale reverse-phase HPLC capillary column was created by packing 5 m m C18 spherical silica beads into a fused silica capillary (100-m m inner diameter x ~12-cm length) with a fl ame-drawn tip. After equilibrating the column, each sample was loaded via a FAMOS auto sampler (LC Packings, San Francisco, CA) onto the column. A gradient was formed and peptides were eluted with increasing concentrations of solvent B (97.5% acetonitrile, 0.1% formic acid).

As peptides eluted, they were subjected to electrospray ionization and then entered into an LTQ Velos ion trap mass spectrometer (Thermo Fisher, San Jose, CA). Peptides were detected, isolated, and fragmented to produce a tandem mass spectrum of speci fi c fragment ions for each peptide. Dynamic exclusion was enabled such that ions were excluded from reanalysis for 30 s. Peptide sequences (and hence protein identity) were determined by matching protein databases with the acquired fragmentation pattern by the software program SEQUEST (Thermo Fisher, San Jose, CA). The human IPI database (ver. 3.6) was used for searching. Precursor mass tolerance was set to ±2.0 Da, and MS/MS tolerance was set to 1.0 Da. A reversed-sequence database was used to set the false discovery rate at 1%. Filtering was performed using the SEQUEST primary score, Xcorr, and delta-Corr. Spectral matches were further manually examined, and multiple identi fi ed peptides (>1) per protein were required.

As with many screening methods, un fi ltered AP-MS data contain many nonspeci fi c binding proteins due to some intrinsic characteristics, such as nonspeci fi c binding to bead or tag, protein aggregation, and carryover during MS runs. We now describe a simple ef fi cient statistic method, z -score plus prey occurrence and reproducibility (ZSPORE) scoring system, for identi fi cation of HCIP. Using this pipeline, we achieve a higher ef fi ciency of AP-MS and better identi fi cation of high-con fi dence interacting proteins. The methods and criteria used to remove nonspeci fi c binding proteins and identify high-con fi dence interacting proteins include:

(a) GFP and controls. AP-MS of GFP-FLAG and various controls, such as non-FLAG IgG conjugated resin for AP-MS, were used to identify nonspeci fi c binding proteins in the database. (b) z -Score. A z -score (aka a standard score) indicates how many standard deviations an element is from the mean. To calculate z -score, mass spectrometry data were transformed into a "stats table," where the columns are total spectral counts (TSC) from 4 MS runs, the rows are bait-associated proteins (Table 2. 2 ). Then we calculated z -score of each X i,j (i prey interacts with j bait) based on the maximum total spectral counts (TSC) of 4 MS runs. For HI5 database analysis, we set the cutoff of z -score as 2.

z is the z-score, X is the value of the element, m is the population mean, and s is the standard deviation. (c) Prey occurrence. We considered any prey associated with a single bait as an HCIP while preys associated with all baits as NSBP. Generally, we set the bar of prey occurrence as <5%, which means one speci fi c prey interacts less than 5% of total baits in the entire database. In HI5, we showed that preys that interact with less than 5 baits represented statistically signi fi cant interactions in HI5 dataset. So the threshold for prey occurrence in HI5 is set as 4. Due to known high interconnectivity among selected baits, bait-to-bait interactions were considered as HCIP. (d) Reproducibility. Each prey must appear in at least 2 out of 4 MS runs. (e) Batch reproducibility. To account for possible variations in the list of background contaminants observed in our dataset that were not identi fi ed by other statistical approaches, we intentionally sequenced each duplicate puri fi ed complex in different experiments. Any protein that did not appear in different puri fi cations was considered an NSBP and manually removed from HCIP list.

After statistical analysis of dataset, all pairwise interactions are collected and analyzed by Cytoscape. Several important attributes, such as z -score and TSC, can be integrated into the interaction map. Except generating interaction map, the functional classi fi cations of HCIPs also need to be analyzed. Interactors can be grouped by protein domains, molecular functions, biological processes, and signal pathways, which may help discover common mechanism underlying the proteins of interest. To fi gure out the new interactions in database, several protein-protein interaction databases such as BioGRID, STRING, IntAct, and MINT can be used to identify the known interaction. However, protein interactions in new publication will not be included in these databases. The interaction information is also not completed, and many known interactions may not be found in these database. Therefore, it is important to dig out protein interaction information in curated literature. Take together, all AP-MS data must be interpreted with care and validated with additional experiments. As with any screening approach, the database does not represent a fi nal or complete interaction network.

Understanding how proteins interact in complex and dynamic networks is the key to dissect the complexity of many genotype-to-phenotype relationships. The systematic mapping of physical interactions is therefore critical for post-genomic research. Comprehensive analysis of protein-protein interactions is still a challenging endeavor of functional proteomics. Since intrinsic negatives are inherent to every technique, the physical interaction data generated by AP-MS may carry many false positives and negatives. Thus, AP-MS is unlikely to grasp the entire interactome. It is also still a challenge to develop optimal computational tools to visually and computationally represent the multiple layers of data and integrate existing biological knowledge and functional data in literature with the interactome data. Since most AP-MS data represent static graph of PPI map, advanced methods have to be developed and focused on dynamic and spatial changes in PPI.

We have presented the general principles of the AP-MS approach and highlighted some recent developed technologies and successful applications on various signaling pathways. Despite of the increasing AP-MS data and analysis tools, there are still many major challenges. It includes (1) the speci fi city of protein complex in different cells and tissues, (2) the dynamics of protein complex with different stimulations or posttranslational modi fi cations, (3) the absolute and relative quantitation of proteins, (4) mapping of transient or weak PPI and endogenous PPI from native cells and tissues, (5) the integration of PPI data sets with the other functional data sets, (6) the standardization and benchmarking for interactome mapping, and (7) the challenges for primary cells like neuronal cells and the detection of weak endogenous interaction. Given the different types of mass spectrometric instrumentation, ionization processes, and software platforms, the assessment of published data becomes increasingly dif fi cult. To facilitate sharing experimental data, common standards in data acquisition, data interpretation, and data storage are required. Many processes in a cell depend on PPI, and perturbations of these interactions can lead to diseases. Comprehensive knowledge of PPI network of signaling pathways will not only give us insights on how the cells respond to stimulation but will also provide new drug targets for therapeutic application. Moreover, many viral and bacterial pathogens rely on host PPIs to survive in host cells and tissues and exert their damaging effects. Ultimately, such high-quality PPI networks will become invaluable resources for better understanding the mechanisms underlying major human diseases and will enable the better de fi nition of drug targets. Shitao Li , Ph.D., USA Shitao Li is a research fellow in Department of Microbiology and Immunobiology, Harvard Medical School. He obtained his Ph.D. from Wuhan University, China. He is a recipient of Kaneb Fellowship and AAI Abstract Award (2010). Dr. Li studies on protein interaction network using proteomics approach. He has mapped a dynamic antiviral innate immunity protein interaction network and currently is working on virus-host protein interaction network. By examining the protein network, he is investigating the signaling mechanisms controlling innate antiviral immunity and new drug targets for host defense to viral infection. He published his research on several prestigious journals such as Nature, Immunity, and Molecular Cell.

A human MAP kinase interactome

High-throughput mapping of a dynamic signaling network in mammalian cells

Network organization of the human autophagy system

A physical and functional map of the human TNFalpha/NF-kappa B signal transduction pathway

A global protein kinase and phosphatase interaction network in yeast

De fi ning human ERAD networks through an integrative mapping strategy

Functional proteomics mapping of a human signaling pathway

Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources

Large-scale mapping of human protein-protein interactions by mass spectrometry

Design and application of a cytokine-receptor-based interaction trap

Electrospray ionization for mass spectrometry of large biomolecules

A novel genetic system to detect protein-protein interactions

Toward a functional analysis of the yeast genome through exhaustive two-hybrid screens

Proteome survey reveals modularity of the yeast cell machinery

A protein interaction map of Drosophila melanogaster

Modularity and hormone sensitivity of the Drosophila melanogaster insulin receptor/target of rapamycin interaction proteome

Characterization of the proteasome interaction network using a QTAX-based tag-team strategy and protein interaction network analysis

A protein complex network of Drosophila melanogaster

Quantitative analysis of complex protein mixtures using isotope-coded af fi nity tags

Matrix-assisted laser desorption/ionization mass spectrometry of biopolymers

MASCOT: multiple alignment system for protein sequences based on three-way dynamic programming

Systematic identi fi cation of protein complexes in Saccharomyces cerevisiae by mass spectrometry

Global functional atlas of Escherichia coli encompassing previously uncharacterized proteins

A comprehensive two-hybrid analysis to explore the yeast protein interactome

Global landscape of HIV-human protein complexes

Visualization of molecular interactions using bimolecular fl uorescence complementation analysis: characteristics of protein fragment complementation

The IntAct molecular interaction database in 2012

Global landscape of protein complexes in the yeast Saccharomyces cerevisiae

Proteome organization in a genome-reduced bacterium

A map of the interactome network of the metazoan C. elegans

Optimization and ZSPORE analysis of af fi nity auri fi cation aoupled with tandem mass spectrometry in mammalian cells

Mapping a dynamic innate immunity protein interaction network regulating type I interferon production

MINT, the molecular interaction database: 2012 update

Protein microarrays and proteomics

Yates 3rd JR. Probability-based validation of protein identi fi cations using a modi fi ed SEQUEST algorithm

Absolute quantitation of proteins by acid hydrolysis combined with amino acid detection by mass spectrometry

Ctr9, Rtf1, and Leo1 are components of the Paf1/RNA polymerase II complex

Stable isotope labeling by amino acids in cell culture, SILAC, as a simple and accurate approach to expression proteomics

Yeast two-hybrid contributions to interactome mapping

The SARS-coronavirus-host interactome: identi fi cation of cyclophilins as target for pan-coronavirus inhibitors

Interactome mapping of the phosphatidylinositol 3-kinasemammalian target of rapamycin pathway identi fi es deformed epidermal autoregulatory factor-1 as a new glycogen synthase kinase-3 interactor

The tandem af fi nity puri fi cation (TAP) method: a general procedure of protein complex puri fi cation

elegans ORFeome version 1.1: experimental veri fi cation of the genome annotation and resource for proteome-scale protein expression

Monitoring protein-protein interactions in intact eukaryotic cells by beta-galactosidase complementation

Virion-wide protein interactions of Kaposi's sarcoma-associated herpesvirus

Towards a proteome-scale map of the human proteinprotein interaction network

Tranche distributed repository and ProteomeCommons. org

Cytoscape 2.8: new features for data integration and network visualization

De fi ning the human deubiquitinating enzyme interaction landscape

The BioGRID Interaction Database: 2011 update

The STRING database in 2011: functional interaction networks of proteins, globally integrated and scored

A tandem af fi nity tag for two-step puri fi cation under fully denaturing conditions: application in ubiquitin pro fi ling and protein complex identi fi cation combined with in vivo cross-linking

Systematic interactome mapping and genetic perturbation analysis of a C. elegans TGF-β signaling network

PANTHER: a browsable database of gene products organized by biological function, using curated protein family and subfamily classi fi cation

A comprehensive analysis of protein-protein interactions in Saccharomyces cerevisiae

Herpesviral protein networks and their interaction with the human proteome

Protein interaction mapping in C. elegans using proteins involved in vulval development

Making the most of af fi nity tags