If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
* This work was supported, in whole or in part, by the National Institutes of Health through the NHLBI (to R. G.). This work was also supported by the Van Andel Research Institute and the Reynolds Foundation (to R. G.). The on-line version of this article (available at http://www.mcponline.org) contains supplemental Tables 1 and 2, Fig. 1, and other information. 1 The abbreviations used are:
CRPC-reactive proteinAAIMantibody-array interaction mappingKNGkininogenSAPserum amyloid P componentBis-Tris2-[bis(2-hydroxyethyl)amino]-2-(hydroxymethyl)propane-1,3-diolHOCMhypertrophic obstructive cardiomyopathyCAcarbohydrate antigenCEACAMcarcinoembryonic antigen-related cell adhesion moleculePPKplasma prekallikreinADAlzheimer disease. § Present address: Inst. for Hepatitis and Viral Research, Doylestown, PA 18902.
Protein-protein interactions are fundamentally important in biological processes, but the existing analytical tools have limited ability to sensitively and precisely measure the dynamic composition of protein complexes in biological samples. We report here the development of antibody-array interaction mapping (AAIM) to address that need. We used AAIM to probe interactions among a set of 48 proteins in serum and found several known interactions as well potentially novel interactions, including multiprotein clusters of interactions. A novel interaction initially identified between the innate immune system protein C-reactive protein and the inflammatory protein kininogen (KNG) was confirmed in subsequent experiments to involve serum amyloid P instead of its highly related family member, C-reactive protein. AAIM was used in a variety of formats to further study this interaction. In vitro studies confirmed the ability of the purified proteins to interact and revealed a zinc dependence of the interaction. Studies using plasma samples collected longitudinally following a controlled myocardial infarction revealed no consistent changes in the serum amyloid P-KNG interaction levels but consistent changes in KNG activation and interactions with plasma prekallikrein. These results demonstrate a versatile platform for measuring the dynamic composition of protein complexes in biological samples that should have value for studies of normal and disease-related signaling networks, multiprotein clusters, or enzymatic cascades.
Protein-protein interactions form the basis of many types of biological processes, such as signal transduction, immune recognition, and metabolism. Individual proteins may have myriad interaction partners, existing in static conditions in multiprotein units or dynamically in the context of interconnected networks of interactions. Major efforts are underway to identify protein-protein interactions (
). Increased information about protein-protein interactions ultimately will enable a better determination of individual protein functions and a better understanding of biological systems.
Although information regarding interaction partners among proteins has been expanding rapidly, data describing the dynamic variation in the composition of protein complexes in biological samples are still limited. Such information is important for determining the involvement of interactions in particular biological processes, for understanding how interactions are regulated, or for identifying abnormal levels that may be involved in pathologies. The relative lack of knowledge in this area relates to the capabilities of conventional approaches for examining protein-protein interactions. A standard approach is to isolated a tagged or antibody-targeted protein and probe the interacting proteins using Western blots or mass spectrometry (
). Such approaches are valuable for identifying interaction partners directly from biological samples, but these methods do not have the reproducibility required to accurately measure changes in interaction levels between samples. Other methods include yeast two-hybrid screens, which are widely used for identifying or confirming direct, binary interactions between proteins generated from cDNA (
). Although effective for gathering certain types of information, these methods still do not enable an analysis of the composition of protein complexes in biological samples. Therefore, the development of additional methods for identifying and measuring protein-protein interactions is an important goal.
Antibody-based detection allows the sensitive and reproducible detection of specific targets in complex backgrounds. Interactions between individual proteins can be detected in an ELISA-type format using “mismatched” pairs of capture and detection antibodies in which each antibody targets one member of the complex as demonstrated in a study of C-reactive protein (CRP)
). Building on that principle, we developed antibody-array interaction mapping (AAIM) for the detection and measurement of interactions among a larger, defined set of proteins. By combining antibody-based interaction detection with the capabilities of low volume, high throughput microarrays (
), multiple potential interactions, or selected interactions using multiple antibodies, can be examined. AAIM complements the existing repertoire of methods for measuring protein-protein interactions by enabling the sensitive, reproducible, and high throughput detection of multiprotein complexes in their native, biological states.
We applied this technology to the study of protein-protein interactions in the circulation. For most blood proteins, little is known about whether they circulate alone, with a few other proteins, or as parts of large, multiprotein clusters. Even less is known regarding changes related to various physiological states. Circulating protein-protein interactions are particularly important in the inflammation and coagulation systems in which the maintenance of health depends upon the regulation and accurate processing of multiple signals. Antibody-array interaction mapping offers a unique opportunity for a more comprehensive evaluation of multiple protein-protein interactions in the blood. Furthermore, antibody-based methods can provide high specificity detection even in the presence of high concentration background proteins, which can be problematic when using mass spectrometry detection. Therefore, we used AAIM both as a discovery tool and as a method to study particular blood-based interactions under a variety of conditions. We show that AAIM can be used to measure dynamic changes in interactions using both in vitro experiments and clinical samples. Furthermore, we present the discovery and initial characterization of an interaction that provides new links between innate immunity and inflammation or the contact system of coagulation.
The antibodies and proteins were purchased from various sources (supplemental Tables 1 and 2). The antibodies were purified by dialysis (Slide-A-Lyzer, Pierce) against PBS buffer followed by ultracentrifugation. The concentration of each antibody was adjusted to 500 μg/ml prior to printing. The integrity and purity of each antibody was confirmed by SDS-PAGE under reducing and non-reducing conditions. Antibody biotinylation was performed using EZ-Link sulfo-NHS-LC-biotin (sulfosuccinimidyl-6-(biotinamido) hexanoate; Pierce).
Microarray Fabrication and Preparation
Antibody microarrays were prepared as described previously (
). A piezoelectric non-contact printer (Biochip Arrayer, PerkinElmer Life Sciences) was used to spot ∼350 pl of each antibody solution on the surfaces of ultrathin nitrocellulose-coated glass microscope slides (PATH slides, GenTel Biosciences, Madison, WI). Forty-eight identical arrays were printed on each slide with each array consisting of 16–48 different antibodies as well as control immunoglobulins from several species printed in triplicate. A wax border was imprinted around each of the arrays to define hydrophobic boundaries (SlideImprinter, The Gel Co., San Francisco, CA). The printed slides were stored at 4 °C in a desiccated, vacuum-sealed slide box until use.
The antibody-array assays were performed as described previously with modifications (
). Serum, plasma, or purified protein samples were diluted with PBS buffer containing 0.1% Brij, 0.1% Tween 20, and 50 μg/ml protease inhibitor mixture (Roche Applied Science). A blocking solution consisting of final concentrations of 400 μg/ml goat, mouse, and sheep IgG; 400 μg/ml chicken IgY; and 800 μg/ml rabbit IgG was included in each serum or plasma sample to reduce nonspecific binding to the printed antibodies. Slides were blocked in PBS with 0.5% Tween 20 (PBST0.5) and 1% BSA. Six microliters of each sample solution was incubated on each array for 1 h at room temperature under gentle rocking. The slides were washed in three changes of PBST0.1 for 3 min each with rocking. Captured antigens were detected with biotinylated antibodies at a concentration of 1–10 μg/ml followed by incubation with 1 μg/ml streptavidin-phycoerythrin (Roche Applied Science) using incubation and wash conditions as above. The slides were scanned for fluorescence emission at 575 nm using a microarray scanner (LS Reloaded, Tecan). All arrays assaying the same protein were scanned concurrently at a single laser power and detector gain setting.
Image data were quantified using GenePix Pro 5.1 (Molecular Devices, Sunnyvale, CA). The net fluorescence signal was calculated by subtracting the median local background surrounding each spot from the median intensity of the corresponding spot. The signal intensities from replicate antibody measurements within the same array were averaged (geometric mean).
Immunoprecipitation and Western Blots
Five microliters of 1 mg/ml biotinylated antibody and 10 μl of 10× protease inhibitors in PBS were added to 50 μl of undiluted serum and allowed to incubate overnight with rotation at 4 °C. One milligram of dextran-blocked, streptavidin-coated magnetic beads (Dynabeads M-280 streptavidin, Invitrogen) in 320 μl of PBS was incubated with the mixture for 1 h at room temperature with mixing on a rotator. The beads were washed extensively in PBST0.1 and then PBS alone before being processed for each mass spectrometry analysis or Western blotting.
To prepare proteins for Western blotting, the proteins were eluted from the beads with 60 μl of Laemmli reducing sample buffer while heating to ∼100 °C for 5 min. All samples were separated on 4–12% Bis-Tris 26-well gels (Criterion XT, Bio-Rad), transferred to a PVDF membrane (Sequiblot, Bio-Rad), and blocked overnight.
The primary and horseradish peroxidase-conjugated secondary antibody incubations were carried out using a Miniblotter (Immunetics, Boston, MA). The blots were developed using the SuperSignal West Femto chemiluminescent substrate (Pierce).
Plasma Samples from Patients with Hypertrophic Obstructive Cardiomyopathy (HOCM) Undergoing Septal Ablation
A total of 36 patients undergoing planned myocardial infarction using alcohol septal ablation for the treatment of symptomatic HOCM were included in this study. Inclusion criteria for this cohort were: 1) primary HOCM, 2) septal thickness of 16 mm or greater, 3) resting outflow tract gradient of greater than 30 mm Hg or an inducible outflow tract gradient of at least 50 mm Hg, 4) symptoms refractory to optimal medical therapy, and 5) appropriate coronary anatomy. The most proximal accessible septal branch was instrumented using standard angioplasty guiding catheters, guide wires, and 1.5 or 2.0 × 9-mm MaverickTM balloon catheters. Radiographic and echocardiographic contrast injections confirmed proper selection of the septal branch and balloon catheter position. Ethanol was infused through the balloon catheter at 1 ml/min. Additional injections in the same or other septal branches were administered as needed, causing cessation of blood flow to the isolated myocardium, and to reduce the gradient to <20 mmHg. Blood was drawn at base line (just prior to the onset of the ablation) and at 10 min, 1 h, 2 h, 4 h, and 24 h following the onset of injury. Protocols for obtaining blood from patients in each of these cohorts were approved by the Massachusetts General Hospital Institutional Review Board, and all subjects gave written informed consent.
Profiling Serum Protein Interactions Using AAIM
The method of AAIM is schematically depicted in Fig. 1a. A sample is incubated on multiple, identical antibody arrays to allow the proteins in the sample to bind to the antibodies. Both the target proteins and the interacting partners of the target proteins are captured because native, non-denatured samples are used. Each array is probed with a different detection antibody to detect the level of its target at each capture antibody. The localization of a detection antibody at a particular capture antibody indicates a potential interaction between the respective targets of the two antibodies (Fig. 1b).
The method was applied to the study of a pool of serum samples from 30 patients with adenocarcinoma of the pancreas (Fig. 2a) using arrays, each containing 48 different antibodies (supplemental Table 1). The antibodies were chosen to target a variety of types of proteins known to be involved in inflammation or to be associated with cancer. The specificities of most of the antibodies for their respective targets in a serum background were investigated by Western blot analysis (see supplemental information). Representative images of individual arrays show that a variety of binding patterns were observed with the different detection antibodies (Fig. 2b). Limited binding was observed in the negative control arrays. Certain detection antibodies bound only at their corresponding capture antibody, such as anti-angiogenin and anti-platelet-derived growth factor subunit B (Fig. 2c). In contrast, others bound their corresponding capture antibody but also bound a variety of other capture antibodies. Some detection antibodies shared very similar patterns of reactivity. For example, IL-1β, CEACAM6, and tenascin C all bound essentially the same capture antibodies as did protein S and vascular endothelial growth factor. CRP detection showed a low, but significant, level of unanticipated signal at the anti-kininogen capture antibody (Fig. 2c).
A cluster representation of the entire data set can be used to identify potential interactions and groups of interactions (supplemental Fig. 1). As expected, the AAIM method identified several previously described interactions, including one between haptoglobin and hemoglobin and another between platelet factor 4 and IL-8 (
). The CA 19-9 antibody, which targets a carbohydrate antigen found on multiple proteins, bound at glycoproteins known to carry that antigen, such as MUC1, CEACAM6, and CA 125. Potentially new interactions were found among CA 125, IL-1β, CEACAM1,3,5,6,7 (pan-CEACAM), CEACAM6, and CA 19-9, which could be constituents of one large complex. Another group of interacting proteins included fibronectin, heparin cofactor II, prostate-specific antigen, platelet factor 4, IL-8, and vascular endothelial growth factor.
Exploration of Novel Interaction
As previously mentioned, a potential novel interaction was observed using antibodies targeting the acute phase reactant CRP and the inflammation/coagulation protein kininogen (KNG). CRP is a member of the pentraxin family of innate immune response proteins, which also includes tissue amyloid P, serum amyloid P component (SAP), which is genetically identical to tissue amyloid P, and pentraxin 3 (
). A physical link between these molecules has not been recognized previously.
Because any given antibody could demonstrate nonspecific binding, we used multiple antibodies against each target as well as purified and recombinant proteins (Fig. 3a) to further investigate this potential interaction. Immunoprecipitations followed by Western blots using various combinations of antibodies supported the existence of the interaction in human plasma (Fig. 3b). An immunoprecipitation from a pool of healthy sera using an anti-CRP antibody showed considerable co-precipitation of KNG. Interestingly, the anti-CRP immunoprecipitation from the cancer sera failed to show KNG co-precipitation. Because the CRP levels were elevated in the cancer patients (data not shown), this finding suggests that the elevated CRP was not bound to KNG. Immunoprecipitations using anti-KNG failed to show associated proteins, which may be due to the much higher concentration of KNG than CRP in healthy individuals (∼60 μg/ml relative to <1 μg/ml), causing most KNG to be free of CRP binding. Measurements in matched serum and plasma samples from the same patients correlated well (not shown), indicating that the measurement of this interaction is not greatly affected by blood preparation protocols or the activation of the intrinsic clotting cascade.
Commercially available, purified kininogen from human plasma, both the low molecular weight (68,000) and the high molecular weight (120,000) varieties, contained a protein that reacted with anti-CRP at the expected size of about 24 kDa (Fig. 3b). The existence of this anti-CRP-reactive molecule in the purified KNG provides strong support for the existence of the interaction in serum because the KNG was stringently purified using a completely distinct method, the capture of KNG using immobilized prekallikrein, which also interacts with KNG.
Further experimentation, however, suggested that the KNG-bound molecule was not CRP because the size of the anti-CRP-reactive band from purified KNG did not exactly match that from purified CRP; the protein in the KNG sample was slightly larger (Fig. 3c). CRP shares close homology with SAP, which is also a pentraxin involved in the innate immune response. SAP monomers (25 kDa) are larger than CRP monomers (23 kDa) by about the same margin observed in the separation. Western blot detection of purified CRP or SAP showed that some anti-CRP antibodies cross-react with denatured SAP (Fig. 3d), similar to previous observations (
). Furthermore, mass spectrometry analysis of the purified KNG showed multiple proteins in the preparation, including SAP but not including CRP (data not shown). These observations suggested that the protein interacting with KNG might actually be SAP and not CRP.
In Vitro Binding Studies Using AAIM
We next used AAIM to examine whether purified versions of these molecules interact with each other under in vitro conditions. Serial dilutions of purified CRP or SAP were added to a constant concentration of KNG in the presence or absence of calcium (the binding to most ligands of CRP and SAP is calcium-dependent) within the buffer, and the solutions were incubated on antibody arrays and probed with various detection antibodies. KNG sandwich detection was not affected by the presence of SAP or CRP (Fig. 4a). The detection of KNG at the anti-CRP capture antibodies showed no KNG bound by CRP (Fig. 4b), but detection at the anti-SAP capture antibodies showed saturable binding by SAP (Fig. 4c). Binding curves that saturate in an asymptotic manner are an indicator of the validity of a biological interaction. The KNG-SAP interaction could be detected both by capturing SAP and detecting KNG and by capturing KNG and detecting SAP (Fig. 4c). The CRP antibody clones used here were previously determined to exhibit very little cross-reactivity with SAP (data not shown). These findings demonstrate that purified versions of KNG and SAP can interact, but purified versions of KNG and CRP cannot.
Next we asked whether the binding of SAP to KNG is affected by the concentration of zinc, which is known to modify KNG conformation and binding properties. SAP-KNG interaction levels were proportionally affected by zinc availability with some binding occurring without zinc but maximal binding occurring by 6 μm (Fig. 4d) (serum zinc concentration is 15–20 μm). The requirement for zinc suggests the specificity of the interaction for particular KNG conformations and suggests that SAP-KNG interactions are physiologically modulated by zinc concentrations.
Measurement of Changes in Interaction Levels in Acute Phase Responses
We further investigated whether AAIM could be used to detect SAP-KNG interactions in serum and changes in the interaction levels in acute phase responses. Changes in association with acute phase responses were thought to be possible due to the involvement of both molecules in inflammatory and coagulation processes. A valuable resource for the study of changes associated with acute phase responses was the plasma samples collected immediately prior to, 10 min after, and 24 h after a controlled myocardial infarction, a therapeutic treatment for the excessive cardiac muscle mass of subjects with hypertrophic cardiomyopathy. This human injury model faithfully recapitulates spontaneous myocardial infarction and allows each person to serve as his or her own biological control. We used dilution curves of pooled plasma samples to determine the optimal dilutions to measure the respective proteins and their interactions in the linear ranges of the assays (data not shown).
Measurements of the individual proteins showed that only CRP changed consistently with the acute phase response. CRP significantly increased after 24 h (Fig. 5a), which is consistent with the rate of transcriptional regulation of acute phase production of CRP. The CRP base-line levels varied between the patients; patients 1 and 2 had undetectable base-line levels, patient 3 had a high base line and only a minimal increase after ischemic induction, and patients 4 and 5 had slightly elevated base-line levels with significant acute phase inductions. The antibody microarray results agreed well with the Western blot measurements, confirming their accuracy. Neither the KNG nor the SAP levels consistently changed after ischemic induction.
An interaction between SAP and KNG was detected using two different anti-SAP antibodies that do not cross-react with CRP (determined in separate AAIM experiments; not shown). The levels of this interaction did not change greatly with the presence of an acute phase response (Fig. 5b). Similar results were obtained using anti-SAP detection of anti-KNG capture antibodies (not shown).
A valuable feature of antibody arrays is that multiple measurements can be obtained in parallel. The arrays used in these experiments also contained antibodies targeting plasma prekallikrein (PPK) and activated KNG. Nearly all PPK circulates bound to high molecular weight kininogen. After PPK is processed to its active form (plasma kallikrein) via cleavage by activated Factor XII (
), both KNG and PPK may become activated in controlled myocardial infarction samples. The detection of KNG (using an antibody recognizing both the active and inactive forms) at the anti-activated KNG (anti-KNG (
)) antibody provides measurements of KNG activation, and the detection of KNG at the anti-PPK antibody gives measurements of non-activated kallikrein bound to KNG.
As measured by AAIM, activated KNG increased in every patient, some dramatically, 10 min after the induction of myocardial infarction and dropped to levels below base line after 24 h (Fig. 5b). In contrast, the PPK·KNG complex showed decreases after 10 min that remained steady after 24 h (Fig. 5b). The approximate negative correlation between these measurements was expected because a decrease in non-activated kallikrein should lead to an increase in activated KNG. These findings indicate a rapid peripheral activation of the kallikrein-kinin system after injury and a readjustment of the relative levels of the molecular complexes after 24 h. Furthermore, these results demonstrate the ability to measure dynamic changes in multiple protein isoforms and complexes in biological samples, and in particular, they demonstrate a platform for the sensitive probing of circulating complexes involved in inflammation and coagulation.
Practical methods for examining variation in multiprotein complexes in biological samples have been missing from the bioanalytical tool kit. AAIM addresses this need by combining the advantages of antibody-based detection with the capabilities of low volume, high throughput microarrays. The combined features of AAIM include sensitive measurements in low volumes of biological samples, sufficient precision to allow comparisons of interaction levels between samples, and the ability to detect multiprotein clusters. The existing methods such as immunoprecipitation followed by mass spectrometry, yeast two-hybrid screens, and protein arrays do not provide these features but rather offer a broader de novo discovery opportunity. AAIM can be used for interaction discovery within a set of antibodies defined by the size of the array, or it can be used to study particular interactions, possibly using multiple antibodies against each member of an interacting set of proteins. The use of multiple antibodies against each protein may be particularly valuable if each antibody targets a different epitope or conformation, which would allow the monitoring of changes in the exposure of certain domains or conformations. Such an approach may be useful for monitoring cascades of enzymatic cleavage or changes in the composition of multiprotein complexes. We demonstrated the use of AAIM to probe for novel interactions and to examine particular interactions both among proteins in their native biological samples and among purified proteins in vitro.
The probing for interactions among a set of 48 proteins revealed several new potential interactions, including some interactions that seemed to take place among multiple proteins, forming multiprotein clusters. Scant details are known about the composition of multiprotein complexes in the circulation because of limitations in the conventional technologies for probing that information, so these results provide some novel insights. The preliminary indication from our data is that many proteins might form associations with each other (supplemental Fig. 1) perhaps through some common scaffold. Further information about this finding and the validation of newly observed protein-protein interactions await further study. The need for validation of antibody-based observations was highlighted by the discovered cross-reactivity of some anti-CRP antibodies with its close homolog SAP. After recognizing potential cross-reactivities between SAP and CRP antibodies, we chose antibodies displaying no cross-reactivity in AAIM experiments for the further study of the interactions.
The validity of the KNG-SAP interaction was supported by the facts that multiple anti-SAP and anti-KNG antibodies, as well as anti-CRP antibodies that cross-react with SAP, detected the complex; that SAP was associated with highly purified, commercially available KNG; that purified KNG and SAP, but not purified KNG and CRP, interacted in vitro in a saturable manner; and that the interaction level was modulated by zinc, which affects KNG conformations. Previous reports found that SAP does not have a major interaction partner in the circulation (
) may have disrupted interactions with KNG by non-physiological pH and buffer conditions or by the induction of KNG activation. Alternatively, the possibility exists that in our work the SAP interactions were dynamically formed as proteins localized to the capture antibodies because SAP can have higher affinity for clustered ligands and can itself have altered binding properties upon self-aggregation (
). In any case, AAIM may be particularly useful for studying complex interaction situations such as this because multiple potentially interacting partners may be probed in pure biological samples with minimal disruption. Even if dynamically formed instead of existing in the normal circulation, the interaction would have biological significance at sites of KNG or SAP binding and clustering.
The observation of the activation of the kininogen system after induction of myocardial infarction (Fig. 5c) shows the capability of this method to measure normal physiological changes to protein complexes in response to invasive insults. The detection of this change in the peripheral blood only 10 min after injury indicates that the response is rapid and intense with systemic consequences. The higher concentrations of activated KNG in the circulation may have effects at distant sites because activated KNG has biological activity distinct from the non-activated version (
). The ability to monitor the SAP-KNG interaction and KNG activation states may be relevant for studying diseases in which both SAP and KNG are involved, such as Alzheimer disease (AD). SAP is found ubiquitously at amyloid plaques, and the KNG contact system is activated in the cerebrospinal fluid of AD patients (
). It is possible that the binding of SAP leads to the localized concentration of and activation of KNG at amyloid plaques. The resultant bradykinin generated by activation of KNG could play a role in neuronal cell signaling leading to neurofibrillary tangles (
) as well as disruptions to the blood-brain barrier. Our observation that the SAP-KNG interaction is influenced by zinc levels could have pathological interest in AD as elevated zinc levels in the brain have long been believed to be a risk factor for the development of AD (
). The results and methods reported here provide a framework for further studies into these possibilities.
In summary, we have demonstrated a method that can provide new insights into protein-protein interactions in biological samples, and we have used the method to discover and study a novel serum interaction. AAIM was used to screen for novel interactions, perform in vitro studies on a particular interaction, and monitor changes among interacting proteins in serially collected serum samples. The interaction between KNG and SAP may have functional implications for inflammatory responses and diseases such as Alzheimer disease, particularly by linking the ligand recognition properties of SAP with the vascular regulation and signaling functions of KNG. The AAIM platform should be useful for the study of dynamic changes in signaling networks, multiprotein clusters, or enzymatic cascades in various biological contexts.
We thank Andrew Porter and Steven Kluck for expert assistance preparing the antibody microarrays, Kevin Maupin and Thomas LaRoche for assistance with the experiments, and Greg Cavey of the Van Andel Research Institute Proteomics core and Curtis Wilkerson of the Michigan State University Proteomics core for the mass spectrometry analyses.