There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.
Abstract
Motivation: Reverse-phase protein arrays (RPPAs) allow sensitive quantification of relative
protein abundance in thousands of samples in parallel. Typical challenges involved
in this technology are antibody selection, sample preparation and optimization of
staining conditions. The issue of combining effective sample management and data analysis,
however, has been widely neglected.
Results: This motivated us to develop
MIRACLE, a comprehensive and user-friendly web application bridging the gap between spotting
and array analysis by conveniently keeping track of sample information. Data processing
includes correction of staining bias, estimation of protein concentration from response
curves, normalization for total protein amount per sample and statistical evaluation.
Established analysis methods have been integrated with
MIRACLE, offering experimental scientists an end-to-end solution for sample management and
for carrying out data analysis. In addition, experienced users have the possibility
to export data to R for more complex analyses.
MIRACLE thus has the potential to further spread utilization of RPPAs as an emerging technology
for high-throughput protein analysis.
Protein arrays are described for screening of molecular markers and pathway targets in patient matched human tissue during disease progression. In contrast to previous protein arrays that immobilize the probe, our reverse phase protein array immobilizes the whole repertoire of patient proteins that represent the state of individual tissue cell populations undergoing disease transitions. A high degree of sensitivity, precision and linearity was achieved, making it possible to quantify the phosphorylated status of signal proteins in human tissue cell subpopulations. Using this novel protein microarray we have longitudinally analysed the state of pro-survival checkpoint proteins at the microscopic transition stage from patient matched histologically normal prostate epithelium to prostate intraepithelial neoplasia (PIN) and then to invasive prostate cancer. Cancer progression was associated with increased phosphorylation of Akt (P<0.04), suppression of apoptosis pathways (P<0.03), as well as decreased phosphorylation of ERK (P<0.01). At the transition from histologically normal epithelium to PIN we observed a statistically significant surge in phosphorylated Akt (P<0.03) and a concomitant suppression of downstream apoptosis pathways which proceeds the transition into invasive carcinoma.
Introduction Signaling pathways are prime candidates for regulation by microRNAs (miRNAs) due to their dose-sensitive nature and the fine-tuning role of miRNAs (Inui et al, 2010). Indeed, EGFR signaling and cell-cycle components are known to be regulated by miRNAs in the cancer context (Bueno et al, 2008; Barker et al, 2010). EGFR itself is regulated by miR-7, which can induce cell-cycle arrest and cell death in cancer cell lines (Webster et al, 2009). In addition, let-7 regulates the oncogene Ras, which is the activator of the MAPK pathway upon EGFR activation (Johnson et al, 2005). Similarly, several cell-cycle genes have been identified as direct targets of miRNAs, including the miR-16 family and miR-17/20 that target Cyclin D1 and have important roles in regulating G1 arrest and S-phase entry (Liu et al, 2008; Yu et al, 2008). A systematic analysis, however, of miRNAs regulating the EGFR pathway and downstream cell-cycle proteins has not been performed. The identification of targets of individual miRNAs is commonly achieved by combining transient transfection of miRNA mimics or inhibitors, target prediction algorithms and gene expression profiling (Wellner et al, 2009; Uhlmann et al, 2010). This approach has two main drawbacks: first, it only captures regulatory effects at the transcriptional level, which might deviate from the outputs at the protein level; second, it cannot distinguish direct and indirect effects induced by miRNAs. A more direct, biochemical approach to identify targets of miRNAs is RISC-IP, where miRNA and target mRNAs are co-immunoprecipitated with Ago2 (component of RISC) and then analyzed either by array-based hybridization (Karginov et al, 2007) or by deep sequencing (Hanina et al, 2010). This approach has identified several genes that are regulated both by mRNA degradation and by translational repression (Thomson et al, 2011), and some rare cases of 5′-UTR targeting (Grey et al, 2010). Even with the knowledge of direct targeting, measurements of miRNA effects directly at the protein level are needed to fully comprehend the complexity of their activities. Recent advances in high-throughput proteomics have allowed researchers to begin addressing this issue. For example, two reports used quantitative mass spectrometry to assess changes in protein levels after overexpressing or knocking down selected miRNAs (Baek et al, 2008; Selbach et al, 2008). Both reached a similar conclusion: the miRNAs they studied affected the expression of hundreds of proteins, but in a rather mild manner, fine-tuning protein synthesis. Another study used a combination of a luciferase-based 3′-UTR reporter assay and label-free shotgun proteomics to identify targets of a liver-specific miRNA, miR-122, revealing a network of miR-122-regulated genes that was enriched for proliferation, cell cycle and apoptosis, all of which are relevant to liver metabolism, liver diseases and cancer-related processes (Boutz et al, 2011). Researchers have also employed protein lysate arrays, which allow higher throughput compared with mass spectrometry methods, in one case screening >300 miRNAs for regulatory effects on the expression of the estrogen receptor alpha (ERα) protein (Leivonen et al, 2009). Despite these advances, understanding how miRNAs regulate an oncogenic signaling system requires that we simultaneously investigate the interactions between a body of miRNAs and a network of proteins. In this work, we aimed to systematically identify miRNAs that regulate the EGFR pathway on the protein level, and to unravel principles of these regulations. To this end, we screened a genome-wide library of miRNA mimics for effects on the expression of 26 proteins in the EGFR-driven cell-cycle pathway using reverse phase protein arrays (RPPAs). This approach identified a series of miRNAs that downregulate one or more proteins of the EGFR-driven cell-cycle pathway. For most of the proteins with reduced expression levels, sequence-based prediction of targeting miRNAs can explain a significant portion of the regulation. In order to uncover the organizational principles of a mild, yet potentially very dense regulatory network, we developed a novel network-based analysis approach to identify those miRNAs that regulate multiple proteins in the studied system. Network analysis implies that the proteins controlling EGFR-driven G1/S transition are co-regulated by several miRNAs. This principle is supported by the discovery of three novel tumor-suppressor miRNAs that regulate the EGFR-driven cell-cycle pathway. Results miRNome screen to unravel the miRNA–protein interaction network of the EGFR pathway To elucidate how the EGFR-driven cell-cycle protein network is regulated at the genome-wide miRNA level, we performed a gain-of-function screen using a mimic library containing 810 mature miRNAs (miRBase version 10.0). As readout, we measured and quantified the abundances of 26 proteins in this pathway. We employed the MDA-MB-231 breast cancer cell line as our experimental system, where a large body of known human miRNAs is expressed (429 detected with microarray, and 598 by next-generation sequencing data, see Supplementary information, Supplementary Tables S1 and S2 and Supplementary Figure S1 for details). Cells were transfected using an automated robotics system with individual miRNA mimics, and total protein lysates in miRNA overexpressing cells were spotted on RPPAs. Next, we incubated the protein arrays with thoroughly validated antibodies in an automated pipeline (Figure 1A). We chose the proteins to be examined from the EGFR signaling/cell-cycle network (Table I) based on the two criteria. First, the expression of the gene must be detectable in the given cell line. We analyzed published RNA sequencing data (Sun et al, 2011) and chose those genes with at least one transcript detectable in the MDA-MB-231 cell line (Supplementary Table S3). Second, a validated antibody for the RPPAs must be available. To determine the specificity and sensitivity of the antibodies, we validated each antibody using the RNAi-based antibody validation method that we have previously published (Mannsperger et al, 2010). Knockdowns with siRNAs resulted in strong reductions of targeted proteins, confirming the antibody specificity/sensitivity for all proteins analyzed (Figure 1B, raw data provided in Supplementary Table S4). We performed the miRNome screen with two biological replicates, which showed high correlations with an average Pearson's correlation coefficient of 0.78 (Figure 1C; Supplementary Figure S2). We included miR-Control (scramble), known EGFR-targeting microRNA miR-7 (Webster et al, 2009), si-Control (scramble) and si-EGFR in all 34 screening runs (each run involved transfection of 24 different miRNA mimics). These quality control measures assured the robustness of the screen (Supplementary Figure S3). We present the global effect of whole-genome miRNAs on the EGFR/cell-cycle network proteins, summarized as a matrix of 810 × 26=21 060 data points, as a heatmap in Figure 1D (the screen results are provided in Supplementary Table S5). Overall, this experimental platform combining a robust whole-genome miRNA screen with quantitative proteomics allowed us to identify miRNA–protein regulations in the MDA-MB-231 breast cancer cell system. Fine-tuning of EGFR pathway proteins by miRNAs and the resulting miRNA–protein interaction network To build an miRNome–protein interaction network, we first identified miRNA–protein pairs where the regulation was statistically significant. For each protein, we used the z-score method to quantify the effects of all miRNAs on its expression (see Materials and methods). While a positive z-score suggests upregulation of protein expression, a negative value indicated a downregulation upon miRNA expression. For most proteins analyzed, the global effect of miRNAs followed a normal distribution with relatively short tails. This pattern suggests that the effects of miRNAs are rather mild, thereby fine-tuning the protein expression (in Figure 2A, PLCG1 is shown as an example. Histograms for all other proteins are given in Supplementary Figure S4). To better characterize the properties of the underlying miRNA–protein regulation network, we tested different significance thresholds and chose two commonly used ones, P 1.96 and ∣z∣>3.29, see Materials and methods for details). We observed that networks of different significance thresholds show distinct patterns with respect to the number of remaining edges (Figure 2B). The resulting miRNA–protein regulation network is very dense at the lower threshold (∣z∣>1.96), as can be seen in Figure 2C. This high density is not entirely unexpected, since (1) it has been computationally predicted that one miRNA can regulate several genes, (2) one gene can be regulated by dozens of miRNAs and (3) indirect regulations are possible. Our data are well in line with these hypotheses, and provide the first large-scale experimental evidence, to the best of our knowledge, at the miRNome level. The number of links decreased sharply with increasing stringency (Figure 2B) and we observed a relatively sparse graph with a low number of regulations when applying a high significance threshold (∣z∣>3.29) (Figure 2D). These observations reveal potential challenges in understanding the complex network: if the significance threshold is set too low, regulatory relationships are complicated and are noisy; on the other hand, if it is set too high, too few edges will remain to reflect structural features of the network. To overcome this dilemma, one possibility is to combine the low-stringency network with prior knowledge of miRNA-direct target predictions. In order to construct such a network, one can use pre-compiled miRNA-mRNA sequence mappings provided by the microRNA.org database (http://www.microrna.org/microrna/getDownloads.do, Release August 2010) by choosing different stringencies of the sequence-matching requirements (seed sequence mapping). We set the stringency filter so that all target predictions having a 6-mer or more seed sequence pairing are included. To construct the subnetwork, we filtered the whole network by keeping miRNA–protein edges that (1) has an absolute z-score larger than 1.96 and (2) the miRNA is mapped to gene's 3′-UTR with at least one 6-mer match or more. This approach led to 241 miRNAs and 25 genes (PIK3CB was not a predicted miRNA target), with 355 edges connecting them (Supplementary Figure S5; Supplementary Table S6). We could further reduce the complexity of the resulting network by using an evolutionary conserved sequence-matching filter. To this end, we compared the low-stringency network obtained from experimental data with miRNA-direct target predictions made by TargetScan (Friedman et al, 2009). We only kept those miRNA–protein links where (1) the miRNA significantly downregulates the protein with ∣z∣>1.96 and (2) the miRNA is predicted to directly target the 3′-UTR of the gene in an evolutionarily conserved manner. The resulting network, supported by both computational predictions and experimental evidence, ended up with 120 potential miRNA/protein target interactions (Figure 3; Supplementary Table S7). In order to find out how many interactions in this network were validated by previously published independent studies, we manually searched the miRWalk database (http://www.ma.uni-heidelberg.de/apps/zmf/mirwalk/) for predicted and validated miRNA targets. Indeed, some of the miRNA direct targeting interactions had been validated in other studies (16 validated interactions), e.g., the miR-143/KRAS (Chen et al, 2009), miR-200b/c/429/PLCG1 (Uhlmann et al, 2010) and miR-7/EGFR (Webster et al, 2009). However, our results identified many more new potential direct regulations. These results suggest that our experimental platform, aided by computational predictions to filter results, can indeed identify miRNA–protein interactions, even if their effects are merely moderate. Seed/3′-UTR matches dominate the negative regulation of protein expression by miRNAs To test whether the regulation by miRNAs is dominated by the nucleotide matching mechanism between the miRNA seed sequences and 3′-UTRs of target genes, we assessed whether protein expression differs in the presence of predicted targeting miRNAs compared with non-targeting ones for each protein in the network. For this purpose, we used two distinct target prediction algorithms (TargetScan, Friedman et al, 2009; and MicroCosm, Griffiths-Jones et al, 2008; see Bartel, 2009 for detailed discussions on the prediction algorithms) to compare predicted versus non-predicted miRNAs for 26 proteins in our network. We applied the Kolmogorov–Smirnov (K–S) test, a non-parametric, threshold-free statistical method (Smirnov, 1948), to assess whether distributions of z-scores of the two groups differ significantly (or three groups, in case of TargetScan with conserved and non-conserved predicted targets). For most of the proteins with predicted targeting miRNAs, we observed a significant negative aberration of expression distributions (18 proteins out of 25, with two-sided P 0.01. Note that A and B can, in principle, be connected by all three kinds of edges but, biologically, we expect that A and B are (a) in only one relationship, or (b) co-upregulated by some miRNAs (pattern I) and co-downregulated by others (pattern II) or (c) that they are always antagonistically co-regulated (pattern IIIa/b). Inconsistent mixtures of these patterns have not occurred with any combinations of t 1 and t 2 that we have tested so far. The edges can be assigned with the miRNAs that induce the relationship between the two proteins The consensus graph As we reasoned in the Results section, the choice of t 1 influences the resulting bipartite graph. It will also influence the identification of those co-regulations that are statistically significant as edges are added to the system. To keep only those co-regulations that are assigned statistical significance under different t 1 thresholds, we built the consensus graph. For this purpose, we chose three relatively relaxed thresholds for t 1, namely 1.28 (P<0.2), 1.64 (P<0.1) and 1.96 (P<0.05). For those, we computed the P-values of all relationships and the resulting protein co-regulation graphs for a series of five t 2 values. The t 2 thresholds were chosen based on the change of the average clustering in function of the P-values. These graphs vary in their size and the number of edges. We concentrated on those edges that all of the three graphs (corresponding to one strictness level t 2) find statistically significant. This results in a consensus graph, in which each edge only exists if it is contained in all three graphs at the same t 2 value. Supplementary Material Supplementary Material Supplementary Methods, Supplementary Figures and Supplementary Tables with less than 50 rows. Table S1 List of miRNAs expressed in MDA-MB-231 cell line determined by an NGS platform. Table S2 List of miRNAs expressed in the MDA-MB-231 cell line determined by array-based hybridization with an Illumina platform. Table S4 Results of RNAi-based antibody validation with RPPA output. Table S5 Results (z-scores) of the miRNome screen with RPPA output. Table S6 List of miRNAs/targets in the "intermediate" network shown in Supplementary Figure S5. Table S7 List of miRNAs/targets identified in the regulatory network shown in Figure 3. Table S8 Rank of the miRNAs based on their frequency in the consensus graphs. Table S11 Sequences of siRNAs to validate sensitivity/specificity of the primary antibodies used in RPPAs. Review Process File
The lack of large panels of validated antibodies, tissue handling variability, and intratumoral heterogeneity potentially hamper comprehensive study of the functional proteome in non-microdissected solid tumors. The purpose of this study was to address these concerns and to demonstrate clinical utility for the functional analysis of proteins in non-microdissected breast tumors using reverse phase protein arrays (RPPA). Herein, 82 antibodies that recognize kinase and steroid signaling proteins and effectors were validated for RPPA. Intraslide and interslide coefficients of variability were <15%. Multiple sites in non-microdissected breast tumors were analyzed using RPPA after intervals of up to 24 h on the benchtop at room temperature following surgical resection. Twenty-one of 82 total and phosphoproteins demonstrated time-dependent instability at room temperature with most variability occurring at later time points between 6 and 24 h. However, the 82-protein functional proteomic "fingerprint" was robust in most tumors even when maintained at room temperature for 24 h before freezing. In repeat samples from each tumor, intratumoral protein levels were markedly less variable than intertumoral levels. Indeed, an independent analysis of prognostic biomarkers in tissue from multiple tumor sites accurately and reproducibly predicted patient outcomes. Significant correlations were observed between RPPA and immunohistochemistry. However, RPPA demonstrated a superior dynamic range. Classification of 128 breast cancers using RPPA identified six subgroups with markedly different patient outcomes that demonstrated a significant correlation with breast cancer subtypes identified by transcriptional profiling. Thus, the robustness of RPPA and stability of the functional proteomic "fingerprint" facilitate the study of the functional proteome in non-microdissected breast tumors.
1Lundbeckfonden Center of Excellence in Nanomedicine NanoCAN,
2Molecular Oncology, Institute of Molecular Medicine,
3Human Genetics, Institute of Clinical Research,
4Epidemiology, Biostatistics and Biodemography, Institute of Public Health and
5Department of Mathematics and Computer Science, University of Southern Denmark, 5000
Odense, Denmark
Author notes
*To whom correspondence should be addressed.
†The authors wish it to be known that, in their opinion, the first two authors should
be regarded as Joint First Authors.
This is an Open Access article distributed under the terms of the Creative Commons
Attribution Non-Commercial License (
http://creativecommons.org/licenses/by-nc/3.0/), which permits non-commercial re-use, distribution, and reproduction in any medium,
provided the original work is properly cited. For commercial re-use, please contact
journals.permissions@oup.com
History
Page count
Pages: 8
Categories
Subject:
Eccb 2014 Proceedings Papers Committee
Subject:
Original Papers
Subject:
Methods and Technologies for Computational Biology
scite shows how a scientific paper has been cited by providing the context of the citation, a classification describing whether it supports, mentions, or contrasts the cited claim, and a label indicating in which section the citation was made.
Smart Citations