3 Results
3.1 miRNA Screen
3.1.2 miR-149 serves as a positive control for the screen targeting the ErbB3 3’UTR
a control miRNA that directly targets ErbB receptors. Because miR-125a, previously reported to co-target ErbB2 and ErbB3 (Scott et al., 2007), did not affect expression of these recep-tors in MCF7 cells, we validated miR-149 as a potential miRNA control. Using the miRanda prediction algorithm (www.microrna.org) we identified a conserved binding site for hsa-miR-149-5p (miR-149) within the 3’-UTR of the ErbB3 sequence (Figure 8).
Figure 8: miR-149 is predicted to bind ErbB3. Schematic representation of the miR-149 recognition site in the ErbB3 3’UTR spanning nucleotides 527-533, as determined by miRBase (www.microrna.org).
Using a vector containing the 3’UTR of ErbB3 cloned downstream of the luciferase cDNA, we confirmed that miR-149 directly targets ErbB3. Coexpression of a miR-149 mimic reduced luciferase activity in cell lysates compared to the activity measured in lysates from cells co-expressing the control miRNA (Figure 9A). Deletion of the seven nucleotides complementary to the miR-149 seed region from the potential recognition motif in the ErbB3 3’UTR partially restored luciferase activity, indicating that miR-149 blocks luciferase expression by directly binding the ErbB3 3’UTR (Figure 9A). Transient transfection of MCF7 cells with miR-149 fol-lowed by qRT-PCR analysis and immunoblotting revealed potent suppression of ErbB3 tran-script and protein levels, respectively, compared with those in miRNA-control transfected cells (Figure 9B & C). As a positive control, an ErbB3-specific siRNA pool was used, which completely silenced ErbB3 expression. Overexpression of miR-149 further induced the selec-tive loss of ErbB3 from the cell surface as assessed by FACS analysis of cells stained with an ErbB3-specific antibody (Figure 9D), without affecting the surface expression of the other ErbB receptors (data not shown).
52
Figure 9: Expression of miR-149-5p affects Akt activation by directly targeting ErbB3 3’UTR. (A) HEK293T cells were transiently co-transfected with miR-con or miR-149 along with a luciferase re-porter construct containing the wild-type ErbB3-UTR (WT) or a mutated ErbB3-UTR, in which the miR-149 seed region was deleted (mt; Δ527-533). The next day, luciferase activity in the cell lysates was measured and normalized to the activity of the co-expressed Renilla reporter. Data correspond to the mean ± SEM of four independent experiments performed with triplicate samples. Data were analyzed by using two-way Anova, followed by Bonferroni posttests. **p < 0.01, n.s. (non-significant) p > 0.05 (B, C, D) MCF7 cells were transiently transfected with an ErbB3-specific siRNA pool (siErbB3), a con-trol miRNA (miR-con) or miR-149, respectively. Three days post transfection cells were subjected to further analysis. (B) RNA was extracted and ErbB3 transcript levels were determined by RT-qPCR.
Values were normalized to the reference gene GAPDH as an internal control. Data are shown as the mean ± SEM of three independent experiments and were analyzed with one-way Anova followed by Tukey´s multiple comparison test. **p < 0.01. (C) Cells were lysed and ErbB3 expression was ana-lyzed by immunoblotting. The membrane was probed with a tubulin-specific antibody as a loading control. (D) Cells were incubated with a PE-conjugated ErbB3-specific antibody reactive with the ex-tracellular domain and analyzed by flow cytometry. An isotype matched control IgG was used as a negative control (filled in grey).
Having established ErbB3 as a target of 149, we next investigated the impact of miR-149 on HRG-induced phosphorylation kinetics by immunoblotting of the receptors and the downstream kinases Erk1/2 and Akt as readouts for PI3K and MAPK pathways, respectively.
In agreement with the data shown in Figure 9C, miR-149 expression decreased ErbB3 pro-tein levels, thereby impairing HRG-induced phosphorylation and activation of ErbB3 itself and its dimerization partner ErbB2 (Figure 10A). This potent suppression of ErbB2/ErbB3 phosphorylation was accompanied by modestly reduced Akt and Erk1/2 phosphorylation, most likely due to the amplification occurring as the signal is transmitted further downstream.
Apart from its effect on ErbB3, miR-149 expression also reduced Erk1 protein levels. Be-cause miRNAs often co-regulate several targets within a specific signaling pathway, it is pos-sible that miR-149 also regulates Erk1 post-transcriptionally; alternatively, miR-149 may af-fect Erk1 expression indirectly.
53
Figure 10: Expression of miR-149-5p affects ErbB2/3 downstream signaling. (A) MCF7 cells were transiently transfected with an ErbB3-specific siRNA pool (siErbB3), a control miRNA (miR-con) or miR-149, respectively. Three days post transfection; cells were either left unstimulated (0 min) or stim-ulated with 10 ng/ml heregulin for the indicated times prior to lysis. Equal amounts of cell lysate were analyzed by immunoblotting using phosphospecific antibodies specific for ErbB3(pY1289), ErbB2(pY1221/1222), Akt(pT308) and Erk1/2(pT202/pY204). Membranes were further probed with antibodies that detect the total level of these proteins. Tubulin was detected to confirm equal loading.
(B) MCF7 cells were transiently transfected with a control miRNA (miR-con) or miR-149, respectively.
Three days post transfection, cells were stimulated with 10 ng/ml heregulin for 1 h, fixed with PFA and subjected to In-Cell Western analysis using an phosphospecific antibodies for Akt(pT308) and an anti-body specific for total Akt protein level. Data are presented as the relative Akt activation after heregulin stimulation (ΔAkt) calculated as described in materials and methods and section 3.1.1. One repre-sentative experiment performed with triplicate samples is shown.
For screening purposes, we transferred the analysis of HRG signaling to a 96-Well format using the In-Cell Western protocol. MCF7 cells transfected with control miRNA, miR-149, a control siRNA (siLacZ), and siRNAs against all members of the ErbB family were therefore stimulated with 10 ng/ml HRG for 1 h followed by staining with pAkt and Akt antibodies, re-spectively. In parallel, untreated cells were stained to determine basal Akt activity. pAkt/Akt ratios were determined for each sample and the basal values were then subtracted from the HRG-stimulated ones, yielding ΔpAkt (see Material and Methods or section 3.1.1 for details).
Compared with the controls, ErbB2 and ErbB3 knockdowns almost completely abolished HRG-induced Akt activation, whereas ErbB1 and ErbB4 had minimal effects (Figure 10B), confirming that ErbB2/3 is the relevant signal heterodimer in this setting. Ectopic expression of miR-149 reduced ΔpAkt by ~40%, demonstrating that miRNA-mediated modulation of ErbB receptor signaling can be quantified using this method. As Akt phosphorylation is a dy-namic process, the optimal incubation time with HRG was experimentally assessed. There-fore, cells were exemplarily transfected with one plate of the whole miRNA library screen according to the screening protocol and three days later cells were stimulated with 10 ng/ml HRG for 20 min, 60 min or 180 min or left untreated. Values of ΔpAkt after 20 min and 60 min of HRG stimulation were comparable for the identical set of miRNAs or siRNAs. Howev-er, after 180 min, values for ΔpAkt generally increased for positive as well as negative con-trols compared to the previous time-points (data not shown). This is probably due to
second-54
ary effects, like signal amplification. Because of practical reasons, we choose an incubation time of 60 min for the screening procedure.
3.1.3 Genome-wide miRNA screening for regulators of HRG-induced Akt activation In-Cell Western screening of a miRNA mimic library comprising 879 miRNAs was performed as detailed in Figure 11. The screen was repeated to obtain a second independent data set for both the basal and HRG-stimulated condition.
Figure 11: Workflow of the screening procedure. MCF7 cells were transiently transfected in 96-Well format with a human mimic microRNA library comprising 879 human microRNAs. Three days post transfection, cells were either left unstimulated (basal) or stimulated with heregulin (HRG) for one hour followed by the detection of total Akt and phosphorylated Akt(pT308) protein level using the In-Cell Western technique.
Screen data were then normalized by determining Akt-activity (pAkt/Akt ratios) for all repli-cate samples. For quality control, the replirepli-cates were correlated yielding an average Pear-son’s coefficient of 0.74 (basal) and 0.68 (HRG), respectively (Figure 12A). Additionally, crys-tal violet data were correlated with the corresponding values of Akt activity under basal and HRG-stimulated condition, resulting in a Spearman coefficient of 0.39 (basal) and 0.03 (HRG), respectively. This indicates a minimal correlation of cell number and Akt activity only under basal conditions (Figure 12B).
55
Figure 12: Correlation analysis. (A) For quality control the Akt activities, representing the ratios of the detected phosphorylated Akt and total Akt signal intensities (pAkt/Akt ratio) of each replicate were correlated for the unstimulated (basal, left) and heregulin stimulated (HRG, right) condition. (B) The mean values of crystal violet signal and Akt activity (pAkt/Akt ratio) of both replicates were correlated for the unstimulated (basal, left) and heregulin stimulated (HRG, right) condition. Dotted lines repre-sent the linear regression of the screen data points
HRG-induced Akt activation (ΔpAkt) was determined and plotted for all miRNAs (Figure 13A). The majority of miRNAs had no effect on ΔpAkt values and clustered in the range of the negative controls miR-con and siLacZ (Figure 13B). By contrast, ΔpAkt values for the positive controls miR-149, siErbB2 and siErbB3 were ≤1 (Figure 13B) and therefore used to define a cut-off: Only those miRNAs were considered as screen hits for which ΔpAkt chang-es were significant and either below 1 or greater than 2 (Figure 13A).
56
Figure 13: MicroRNA screen data. (A) Screen data visualization. Akt activation upon heregulin stimulation (ΔAkt) for each miRNA is plotted. Data are shown as the mean of the two replicates and miRNAs are presented in ascending number. Screen hits are depicted in light grey. (B) Depicted in the scatter blot is the Akt activation upon heregulin stimulation (ΔpAkt) for all controls and plates. The miRNA library was presented in 11 96-Well plates and transfected in two biological replicates. Each plate contained a duplicate set of five controls; two negative (miR-con, siLacZ) and three positive con-trols (miR-149, siErbB2, siErbB3); to monitor transfection efficiency and the dynamic range of Akt acti-vation. As marked by the dotted lines ΔAkt values values <1 indicate a reduction and ∆pAkt values >2 indicate an enhancement of Akt activation upon heregulin stimulation. Data were analyzed by one way Anova followed by Bonferroni’s multiple comparison test. ***p < 0.001, **p < 0.01, n.s. (non-significant) p > 0.05.
42 miRNAs plus the positive control miR-149 met these criteria, 19 of which significantly re-duced (ΔpAkt < 1) and 24 of which significantly increased HRG-inre-duced Akt activation (ΔpAkt > 2) (Figure 14).
Figure 14: Bioinformatical analysis of the screen data. For each replicate the ratio of pAkt/Akt signal intensity was calculated and the difference between unstimulated and stimulated condition was computed. miRNAs which significantly (p < 0.01) alter Akt-activity and had a value < 1 or ˃ 2 were considered as screen hits.
In Figure 15 the ΔpAkt values as well as the basal and HRG-stimulated pAkt/Akt ratios are depicted for the miRNA screen hits. Red coloring indicates an increase, green coloring a de-crease of the pAkt/Akt level compared to that in miR-con expressing cells. This reveals that those miRNAs that enhance HRG-stimulated pAkt/Akt levels generally suppressed basal pAkt/Akt levels. This is most pronounced for miR-886-3p, for which the greatest difference in
57
HRG-induced Akt activation was observed (ΔpAkt = 2.8). In the case of 1304 and miR-654-3p, basal pAkt/Akt levels were unaffected, thus, the increase in ΔpAkt was specifically due to enhanced Akt activity upon HRG stimulation. For those miRNAs that negatively af-fected ΔpAkt, there was no uniform trend regarding basal pAkt/Akt levels, as some miRNAs such as miR-204 enhanced, whereas others, e.g. miR-520a-3p, reduced basal pAkt/Akt lev-els. Regardless of the effect on basal Akt activity, all miRNA hits in this category attenuated Akt activation in response to HRG. This is best reflected by miR-148b, for which the lowest ΔpAkt value was obtained, indicating almost complete suppression of HRG-induced Akt acti-vation. Furthermore, a general reduction of pAkt/Akt levels for the basal and HRG-stimulated situation is seen for miR-520a-3p, miR-519c-3p, miR-485-3p, miR-302c, and miR-520d-3p.
Notably, three of these miRNAs possess the same seed region, which is critical for target recognition. The similar ΔpAkt values obtained for the different seed family members under-scores the reliability of the screen and suggests that the effects of these miRNAs are medi-ated by common targets (Figure 15B).
Figure 15: Analysis of the screen data. (A) Screen hits including the positive control miR-149 de-picted in list format. Akt activity (pAkt/Akt ratio) under unstimulated (basal, column 2) and heregulin-stimulated (HRG, column 3) condition is color-coded with green indicating a reduction and red indicat-ing an enhancement of Akt activity compared to the mean of miR-con (black). Akt activity is calculated based on the ratio of the log2 transformed signal intensities of pAkt and Akt (pAkt/Akt). For each
con-58
dition a separate color key is depicted as the range of Akt activity differs between unstimulated and heregulin stimulated cells. Specific Akt activation upon heregulin stimulation (ΔpAkt) for each miRNA is shown as a bar chart in the last column of the table. ΔpAkt value represents the difference between the heregulin-stimulated and unstimulated Akt activity (pAkt/Akt). Because miRNAs are selected based on the ΔpAkt value the cut off for screen hits is shown as dotted lines, with values < 1 indicating a reduction and ∆pAkt values > 2 indicating an enhancement of heregulin-dependent Akt activation.
pAkt/Akt ratios and ∆pAkt values correspond to the mean of the replicates. (B) miRNAs were clustered according to their seed region and for each miRNA the ∆pAkt is visualized as a color-coded value. All seed families with more than three members are listed. Values correspond to the mean of the two replicates.
3.1.4 Identification of a miRNA-ErbB interaction network.
For the construction of ErbB/Akt signaling network we used data from the Reactome data-base and identified 24 pathways connected to ErbB/Akt signaling (Croft et al., 2011). These pathways were parsed into R using r-package rBiopaxParser as directed graphs (Kramer et al., 2013). Next, 24 graphs were merged into a single signaling network comprising 312 nodes representing genes and 3582 edges representing activation or inhibition effects. Be-cause some of the genes are components of protein complexes and to simplify the network we merged these gene-nodes into protein-complex-nodes. This was done for PIK3 (PIK3CA, PIK3R1), mTORC1 (AKT1S1, MLST8, MTOR) and mTORC2 (RICTOR, MLST8, MTOR).
This results in 309 nodes and 3524 edges.
To acquire miRNA-target information for the 43 screen hits miRNAs including miR-149, we searched 3 databases with computationally predicted miRNA targets: MicroCosm Targets release v5 (Griffiths-Jones et al., 2008), miRDB v4.0 (Wang, 2008) and microRNA.org Au-gust 2010 release (Betel et al., 2008). For three miRNAs (miR-1304, miR-1259, miR-1915) no targets were found in any database. The remaining 40 miRNAs were divided into two groups: a first group with negative effects on ΔpAkt comprising 19 miRNAs and a second group with positive effects on ΔpAkt comprising 21 miRNAs. We pooled miRNA – target in-formation from all three databases and identified 298 target genes in the ErbB/Akt signaling network. Next, we separately connected both groups of miRNAs with the network by 959 and 750 miRNA – target gene edges for the first and second miRNA group, respectively. This resulted in the construction of two miRNA-ErbB/Akt signaling networks. To visualize the net-works, the target genes in each network were ranked by the number of targeting miRNAs and the top target genes were then selected. For the network with miRNAs that negatively affected ΔpAkt, a sub-network of the top 14 genes, which were targeted by 9 or more miRNAs, was extracted with corresponding miRNAs. In the case of the network with positive-ly acting miRNAs, a sub-network of the top 17 genes, targeted by 6 or more miRNAs, was extracted with corresponding miRNAs. A list of the miRNA-target gene interactions is given in Figure 16.
59 .
Figure 16: miRNAs-target gene interaction network. The most frequently targeted genes for those miRNAs that significantly reduced ΔpAkt activity (left) and significantly enhanced ΔpAkt (right) are shown. The frequency accounts for the number of miRNA targeting this particular genes and predicted miRNA-protein interactions are marked by crosses.
To gain better insight into the regulatory sub-networks of negatively and positively acting miRNAs we created two protein-protein interaction networks using STRING database, which integrates predicted protein interaction based on physical and functional associations (Figure 17) (Franceschini et al., 2013).
miRNAs negatively affecting ΔpAkt miRNAs positively affecting ΔpAkt
target frequency miR-302c miR-520a-3p miR-520d-3p miR-148b miR-204 miR-539 miR-155 miR-30d miR-19a miR-149 miR-326 miR-520f miR-649 miR-519c-3p miR-1301 miR-1202 miR-486-3p
target frequency miR-181d miR-488 miR-34c-5p miR-548c-3p miR-122 miR-382 miR-579 miR-18b miR-193b miR-146a miR-509-3p miR-574-5p miR-632 miR-654-3p miR-558 miR-640 miR-363* miR-708*
SOS1 12 x x x x x x x x x x x x INPP5B 8 x x x x x x x x
RPS6KA5 11 x x x x x x x x x x x PDE4D 7 x x x x x x x PDE4D 10 x x x x x x x x x x EPS15L1 7 x x x x x x x
RAP1A 10 x x x x x x x x x x MYH11 7 x x x x x x x
mTORC2 10 x x x x x x x x x x EGFR 7 x x x x x x x
FRS2 9 x x x x x x x x x PRLR 7 x x x x x x x
CALD1 9 x x x x x x x x x MYLK 7 x x x x x x x
ADAM17 9 x x x x x x x x x mTORC2 6 x x x x x x
ERBB3 9 x x x x x x x x x TAB2 6 x x x x x x
APBB1IP 9 x x x x x x x x x STAM 6 x x x x x x
EGFR 9 x x x x x x x x x PHLPP2 6 x x x x x x
PIK3 9 x x x x x x x x x GAB1 6 x x x x x x
CPA3 9 x x x x x x x x x TPM3 6 x x x x x x
CDKN1B 9 x x x x x x x x x RAP1A 6 x x x x x x
TBL1XR1 6 x x x x x x
MAP3K8 6 x x x x x x
HPGD 6 x x x x x x
60
Figure 17: Regulatory protein/protein interaction network predicted by STRING database. (A) Sub-network of miRNAs negatively affecting ΔpAkt. The network includes the 14 most frequently tar-geted genes tartar-geted by at least 9 miRNAs. (B) Sub-network of miRNAs positively affecting ΔpAkt.
The network includes the 17 most frequently targeted genes targeted by at least 18 miRNAs. Type of interaction is indicated as colored lines. The single proteins Akt1 and Akt2 were included manually and are merged into Akt. This was also done for the protein complexes PIK3 (PIK3CA, PIK3R1), mTORC1 (AKT1S1, MLST8, MTOR) and mTORC2 (RICTOR, MLST8, MTOR).
Next, we focused our attention on those miRNAs that reduced Akt activation because their effect can be explained by the direct targeting of key players within the ErbB-Akt pathway.
Interestingly, ErbB3 and PIK3 were among the most frequently predicted target genes. To investigate the potential co-regulation of these genes, we created a simplified miRNA-protein interaction graph comprising only directly connected proteins and miRNAs predicted to target ErbB3, namely 520a-3p, 520d-3p, 302c, 19a, 148b, 204, miR-155, miR-149, miR-326 and additionally integrated information of protein-protein interactions from STRING and Reactome databases into this sub-network (Figure 18).
61
Figure 18: miRNA-protein interaction network for negative regulators of HRG signaling. The 12 most frequently targeted genes plus Akt are shown. The edges between them are based on signaling data of Reactome and protein-protein interactions of STRING databases. Only those miRNAs that co-target ErbB3 are depicted.
Intriguingly, all nine miRNAs including miR-520a-3p, miR-520d-3p and miR-302c, which are members of the same seed family, had largely overlapping target spectra. Apart from ErbB3 and PIK3, genes such as RAP1A, a member of the Ras GTPase family, mTOR complex 2 and the Ras GDP exchange factor SOS1 were predicted as targets. This indicates that this subset of miRNAs negatively regulates HRG-induced Akt activation by targeting the PI3K pathway at multiple levels, thereby multiplying the effects caused by the post-transcriptional suppression of single pathway components.
3.1.5 Expression of miR-148b, miR-149, miR-326 and miR-520a-3p reduces