CSN5 inhibition triggers inflammatory signaling and Rho/ROCK-dependent loss of endothelial integrity

RhoGTPases regulate cytoskeletal dynamics, migration and cell-cell adhesion in endothelial cells. Besides regulation at the level of guanine nucleotide binding, they also undergo post-translational modifications, for example ubiquitination. RhoGTPases are ubiquitinated by Cullin RING ligases which are in turn regulated by neddylation. Previously we showed that inhibition of Cullin RING ligase activity by the neddylation inhibitor MLN4924 is detrimental for endothelial barrier function, due to accumulation of RhoB and the consequent induction of contractility. Here we analyzed the effect of pharmacological activation of Cullin RING ligases on endothelial barrier integrity in vitro and in vivo. CSN5i-3 induced endothelial barrier disruption and increased macromolecule leakage in vitro and in vivo. Mechanistically, CSN5i-3 strongly induced the expression and activation of RhoB and to lesser extent of RhoA in endothelial cells, which enhanced cell contraction. Elevated expression of RhoGTPases was a consequence of activation of the NF-κB pathway. In line with this notion, CSN5i-3 treatment decreased IκBα expression and increased NF-κB-mediated ICAM-1 expression and consequent adhesion of neutrophils to endothelial cells. This study shows that sustained neddylation of Cullin RING-ligases leads to activation the NF-κB pathway in endothelial cells, elevated expression of RhoGTPases, Rho/ROCK-dependent activation of MLC and disruption of the endothelial barrier.


Results
Inhibition of the COP9 signalosome disrupts endothelial barrier integrity. CSN5i-3 was designed as an inhibitor of the COP9 signalosome which mediates removal of Nedd8 from the Cullin subunit of Cullin-RING ubiquitin ligases, thus inactivating the complex 24 . To test the effect of CSN5i-3 on endothelial barrier function, we added the compound in different concentrations to confluent primary HUVECs. Electrical cell-substrate impedance sensing (ECIS) was used to quantify changes in endothelial barrier function in real time. Within 1 hour after addition of CSN5i-3 we observed a small increase in barrier function, after which the integrity of the endothelial barrier decreased. This reduction in endothelial barrier was dose-dependent (Fig. 1A). 1 μM CSN5i-3 or higher induced a significant attenuation of endothelial barrier function at 5 hours after addition (Fig. 1B). In addition to reduced resistance of the endothelial barrier, we observed significant barrier disruption by CSN5i-3 in HRP-leakage experiments after prolonged stimulation (>5 hours) (Fig. 1C). Cytotoxicity of CSN5i-3 was reported in cancer cell lines by Schlierf et al. after 72 hours of treatment 24 . To investigate if loss of endothelial barrier integrity was caused by increased apoptosis, we analyzed caspase-3/7 activity in cells treated with CSN5i-3 for 5 and 24 hours (Supplemental Fig. 1). After 5 hours of CSN5i-3 treatment (1 and 4 uM), caspase-3/7 activity was not induced, suggesting that initial disruption of endothelial integrity is not caused by apoptosis. However, we observed increased caspase-3/7 activity after 24 hours of treatment with CSN5i-3 which implicates contribution of apoptosis to the increase of HRP leakage at later timepoints. In subsequent experiments we examined the effect of CSN5i-3 on Cullin-1, Cullin-2 and Cullin-3 neddylation in endothelial cells using immunoblotting (Fig. 1D). We observed a significant mobility shift marking the neddylated isoforms for Cullin-1 and Cullin-3, while this shift was less pronounced for Cullin-2 (Fig. 1D,E). Moreover, CSN5i-3 induced a decrease in total Cullin-2 and Cullin-3 expression, while the levels of Cullin-1 remained unaltered (Fig. 1F). From these data we conclude that CSN5i-3 stabilizes Cullin 1-3 neddylation and differentially affects their expression. Together, this results in a dose-dependent, significant disruption of the endothelial barrier.

CSN5i-3 induces expression and activity of RhoGTPases. Previously, we showed that inhibition of
Cullin-3 activity resulted in increased expression of RhoB due to a reduction of RhoB-ubiquitination with only a modest effect on RhoA 9 . Therefore, we tested the effect of CSN5i-3-mediated Cullin activation on the protein expression of RhoA, RhoB and RhoC. Five hours after addition of 1 and 4 μM CSN5i-3, RhoA and RhoC protein levels were slightly increased ( Fig. 2A,C) with RhoB expression levels 2-and 5-fold increased, respectively (Fig. 2B). Previously, we showed that RhoB ubiquitination controls its subcellular localization in endothelial cells 9 and we therefore analyzed RhoB localization by immunofluorescent staining. Treatment with 1 μM CSN5i-3 induced the upregulated RhoB to localize both to intracellular vesicles and the cytoplasm. This was accompanied by stress fiber formation and a more discontinuous staining for Vascular Endothelial (VE)-cadherin (Fig. 2D). Higher concentration of CSN5i-3 (4 μM) induced similar, but more pronounced effects on RhoB expression and endothelial cell morphology (Fig. 2D). To test if the induction of RhoGTPase expression and change in cell morphology, induced by CSN5i-3, was accompanied by increased activity of RhoGTPases, we performed Rhotekin pulldowns. We found that CSN5i-3 strongly increased the activity of RhoB, with limited effects on RhoA and www.nature.com/scientificreports www.nature.com/scientificreports/ RhoC, directly correlating with the differential effects on their expression (Fig. 2E). Together, these data indicate that treatment of endothelial cells with CSN5i-3 induces the expression of RhoGTPases, primarily RhoB, and the formation of actin stress fibers.
Endothelial barrier disruption by CSN5i-3 is ROCK-mediated. Cell contraction underlies endothelial barrier disruption and is initiated by the activation of Rho GTPases and their downstream effector Rho-kinase (ROCK) 25 . We previously showed that RhoB is important for endothelial cell contraction 4 . Immunofluorescent staining of HUVECs treated with CSN5i-3 showed increased actin stress fiber formation, which is likely mediated by Rho/ROCK activation and Myosin Light Chain (MLC) phosphorylation 26 . Treatment with the ROCK inhibitor Y27632 of otherwise unstimulated cells, reduced stress fiber formation while VE-cadherin staining showed stable, honeycomb-like cell-cell contacts (Fig. 3A). Addition of CSN5i-3 promoted stress fiber formation while the VE-cadherin showed a more jagged distribution, characteristic for remodeling adherens junctions 27 . Pre-treatment of cells with Y27632 clearly abrogated CNS5i-3 -induced stress fiber formation and stabilized VE-cadherin-positive junctions (Fig. 3A). In accordance with this finding, pre-treatment of HUVECs with Y27632 prevented the barrier disruptive effect of CSN5i-3 as shown in ECIS experiments (Fig. 3B,C). Furthermore, CSN5i-3 increased MLC phosphorylation as shown by immunoblot analysis, and this was completely abolished by pre-treating the cells with Y27632 (Fig. 3D,E). Importantly, pre-treatment with Y27632 did not impair CSN5i-3-induced expression of RhoB protein (Fig. 3A,D,F).
In addition to pharmacological inhibition of the Rho/ROCK signaling pathway, we performed a siRNA-mediated knockdown of RhoA, RhoB and RhoC to establish the contribution of these Rho GTPases to the morphological changes caused by CSN5i-3. Individual depletion of RhoA, RhoB or RhoC did not effectively prevent phosphorylation of MLC by CSN5i-3 (Supplemental Fig. 2). This is likely because these GTPases regulate each other's expression such that loss of one drives increased expression of the other 4 , which in turn leads to functional compensation 4 . Therefore, we performed a triple knockdown of all three GTPases simultaneously and analyzed CSN5i-3-induced MLC phosphorylation. We found that only the combined depletion of RhoA, -B and -C effectively abrogated CSN5i-3-induced MLC phosphorylation. In conclusion, our data indicate that the barrier disruption caused by CSN5i-3 is mediated by activation of RhoGTPase/ROCK signaling leading to increased MLC phosphorylation and subsequent cell contraction.

CSN5i-3 increases RhoB expression partially via transcriptional upregulation. We previously
showed that inhibition of Cullin RING ligases by MLN4924 decreased RhoB ubiquitination, hereby interfering with RhoB degradation 9 . An in vivo ubiquitination assay in HEK293T cells showed that CSN5i-3 treatment, in contrast to MLN4924, did not significantly change the ubiquitination state of RhoB (Fig. 4A). To test a role for de novo protein synthesis, we analyzed the effect of CSN5i-3 on the mRNA expression of RhoA and RhoB. Addition www.nature.com/scientificreports www.nature.com/scientificreports/ of CSN5i-3 (1 and 4 μM) or stimulation with TNF-α, as a positive control, only slightly increased RhoA mRNA expression (Fig. 4B). However, the mRNA expression of RhoB in response to treatment with CSN5i-3 (1 and 4 μM) was induced 2.2-fold, with the TNF-α-mediated induction of RhoB mRNA being 2.8-fold (Fig. 4C). In www.nature.com/scientificreports www.nature.com/scientificreports/ www.nature.com/scientificreports www.nature.com/scientificreports/ accordance with this, inhibition of mRNA translation by cycloheximide significantly impaired induction of RhoB protein by CSN5i-3 ( Fig. 4D,E). These results suggest that the CSN5i-3-mediated increase in RhoB expression is, rather than to altered ubiquitination, mainly due to enhanced de novo synthesis of RhoB protein.

Inhibition of the COP9 signalosome activates NF-κB and enhances ICAM expression and leukocyte adhesion.
Induction of RhoB mRNA by TNF-α was previously described 9,28,29 . Since we found that CSN5i-3 induced RhoB mRNA synthesis similar to TNF-α, and because Cullin-RING ligases have been implicated in TNF-α mediated NF-κB activation 30 , we hypothesized that CSN5i-3 increases RhoB mRNA expression via NF-κB. In resting cells, the cytosolic NFκB p65 subunit is bound to members of the family of inhibitory IκB proteins. Degradation of IκB occurs upon their phosphorylation and subsequent ubiquitination by βTRCp-Cullin-1, followed by proteasomal degradation 30 . Degradation of IκB allows the p65-NFκB complex to translocate to the nucleus and activate transcription of its target genes, including the leukocyte adhesion molecule ICAM-1. Treatment of HUVECs with CSN5i-3 resulted in significantly reduced IκBα expression (Fig. 5A,C). Conversely, phosphorylation of the p65 subunit of NFκB was significantly increased only after prolonged CSN5i-3 (4 μM) treatment (Fig. 5A,D). In addition, we the expression of ICAM-1 was significantly increased (Fig. 5A,B). To confirm the role of the NF-κB pathway in the CSN5i-3-induced upregulation of RhoB, we applied the specific IκB phosphorylation inhibitor BAY11-7085 in combination with CSN5i-3 31 . Treatment of HUVECs with BAY11-7085 significantly reduced both the TNF-α and CSN5i-3-induced increase in RhoB levels (Fig. 5E,G). Also, TNF-α and CSN5i-3-induced ICAM-1 expression was completely blocked by BAY11-7085 (Fig. 5E,F).
Since CSN5i-3 increased ICAM-1 expression, we tested if this increase is physiologically relevant in a leukocyte adhesion assay. We treated HUVEC monolayers with CSN5i-3 or TNF-α and analyzed the adhesion of www.nature.com/scientificreports www.nature.com/scientificreports/ polymorphonuclear neutrophils (PMN) which depend on ICAM-1 for strong adhesion. PMN adhesion was dose dependently increased in endothelial cells treated with CSN5i-3 as compared to control cells (Fig. 5H,I). TNF treatment increased PMN adhesion further, in good agreement with its stronger effect on ICAM-1 upregulation (Fig. 5E,F,H). In conclusion, our data indicate that CSN5i-3 induces degradation of IκBα, most probably via activation of Cullin-1 RING ligase 30 , resulting in induction of the pro-inflammatory NF-κB pathway. This drives the increased expression of RhoB and ICAM-1, resulting in a loss of endothelial barrier function and increased leukocyte adhesion, respectively.
CSN5i-3 promotes vascular leakage in zebrafish embryos. CSN5i-3 treatment disrupted endothelial barrier integrity in vitro in primary human endothelial cells. To confirm this finding in vivo, we examined the effect of CSN5i-3 on zebrafish (Danio rerio) vascular integrity. At 24 hours post fertilization (hpf), CSN5i-3 was added to the swimming water of Tg(Fli1:GFP) y1 casper zebrafish embryos for 48-72 hours. In a dose-response experiment, we found that 50 μM CSN5i-3 was required to induce full neddylation of zebrafish Cullin-3 (Fig. 6A). Similar to our observation in in vitro experiments, total Cullin-3 levels were decreased in zebrafish embryos treated with CSN5i-3. As a negative control, zebrafish embryos were treated with 10 μM MLN4924, and as expected, we observed a clear shift to the de-neddylated form of Cullin-3 (Fig. 6A). To assess whether CSN5i-3 induces vascular leakage in the zebrafish embryos, 70 kDa TMR Dextran was injected directly into the www.nature.com/scientificreports www.nature.com/scientificreports/ bloodstream. We performed live fluorescent imaging of the zebrafish embryos 20 minutes after injection of the dextran (Fig. 6B) and quantified the relative dextran extravasation (Fig. 6C). We observed a significant increase in dextran leakage from the intersegmental vessels in 4 days post fertilization (dpf) zebrafish embryos treated with CSN5i-3 (Fig. 6C). In conclusion, our data indicate that CSN5i-3 treatment induces in vivo vascular leakage.

Discussion
Here we show that inhibition of CSN5, the catalytic component of the COP9 signalosome, disrupts endothelial barrier function in vitro and in vivo. The prolonged neddylation of Cullin RING ligases (CRL), consequent to the inhibition of CSN5, resulted in degradation of IκBα and subsequent activation of the NF-κB pathway. This in turn promoted RhoB and, to a lesser extent, RhoA mRNA and protein synthesis, which resulted in increased activity of RhoGTPases, ROCK-mediated actin stress fiber formation, MLC phosphorylation and cell contraction.
Recently, we showed that general inhibition of CRL by MLN4924 enhanced RhoB expression and severly disrupted endothelial barrier function through the induction of RhoB-dependent cell contraction 9 . We identified the Cullin-3-Rbx1-KCTD10 complex as the ligase that mediates the poly-ubiquitination and degradation of RhoB 9 . Based on these findings, we hypothesized that cullin activation by CSN5i-3 would increase ubiquitination and degradation of RhoB, leading to improved endothelial barrier function. In contrast, however, prolonged Cullin-3 neddylation induced by CSN5i-3 resulted in decreased expression of Cullin-3, increased expression of RhoB and reduced endothelial integrity.
Several studies have shown that in endothelial cells RhoB protein levels are upregulated upon TNF-α stimulation of the NF-κB pathway, due to increased mRNA synhtesis 9,28,29 . This TNF-α-induced RhoB localizes to an endosomal compartment, in marked contrast to the pool of RhoB which accumulates following CRL inhibition and localizes to the plasma membrane 9 . Therefore, we analyzed localization of RhoB in HUVEC monolayers treated with CSN5i-3. To our surprise, we found that prolonged CRL neddylation leads to induction of RhoB expression within the endosomal compartment similar to the RhoB localization induced by TNF-α stimulation. Therefore we hypothesized that the NF-κB pathway was upregulated in endothelial cells treated with CSN5i-3. This assumption was further corroborated by the comparable induction of RhoB mRNA expression upon treatment with either CSN5i-3 or TNF-α.
Further analysis confirmed that the NF-κB pathway was indeed activated by CSN5i-3. Originally, the CSN5i-3 compound was designed to inhibit the COP9 metalloprotease in order to prolong Cullin activation 24 . The transition of Cullins between neddylated and non-neddylated states is required in order to exchange the substrate recognition receptors in the CRL complex. Schlierf et al. described that prolonged neddylation of Cullin RING www.nature.com/scientificreports www.nature.com/scientificreports/ ligases by CSN5 inhibition can lead to autodegradation of some but not all substrate recognition receptors. Interestingly, we found that the expression of Cullin proteins in endothelial cells can be affected by CSN5 inhibition as well. The associated deubiquitination activity of the CSN, which is now lost by inhibition of the complex using CSN5i-3, may contribute to the decreased stability and activity of some CRL complexes 19 . CSN5i-3 treatment reduced levels of Cullin-3, but did not significantly affect expression of Cullin-1. This suggests that prolonged CRL activation can have opposing effects on expression levels of substrates of different CRLs.
Comparable to our findings, Schweitzer et al. showed that knockdown of CSN2, also a component of the COP9 complex, leads to decreased expression of IκBα in HeLa cells, which eventually resulted in increased phosphorylated p65 in the nucleus after TNF-α stimulation 32 . The essential role of ubiquitination of IκB in the regulation of the NF-κB pathway was established previously 30 . Furthermore, the knockdown of CSN5 in endothelial cells increased NF-ĸB activity, ICAM-1 expression, and PMN adherence to the EC monolayer 18 . In the present study, in accordance with the previous work, we found that CSN5i-3 treatment decreased expression of IκBα and increased the phosphorylation of p65, resulting in increased ICAM-1 expression which promotes adhesion of PMNs.
The CSN5i-3 compound was found to be a promising candidate for potential treatment against cancer in vitro and in animal studies 24 . However, in our current study we found that prolonged neddylation of CRLs, induced by CSN5i-3, is not beneficial in endothelial cells, as it reduced barrier integrity and induced an inflammatory phenotype. The same effect on endothelial integrity was found upon general inhibition of CRLs by MLN4924, a compound which is already being tested in clinical trials [21][22][23] . In conclusion, our findings demonstrate that both prolonged activation, as well as -inhibition of CRLs can induce unwanted side effects, in the case of CSN5i-3 leading to endothelial inflammation and loss of endothelial barrier function. More specific inhibitors, targeting a smaller subset of E3 enzymes will be required to specifically modulate substrate degradation in cancer-and other cells and at the same time preserve the endothelial integrity and prevent adverse effects on the cardiovascular system.
Cell culture. Freshly isolated HUVECs. Primary Human Umbilical Vein Endothelial Cells (HUVECs) were isolated from umbilical cords of healthy donors. Umbilical cords were provided by the Amstelland Ziekenhuis, Amstelveen. Informed consents were obtained from all donors in accordance with the institutional guidelines and the Declaration of Helsinki. The cells were isolated and characterized as described by Jaffe et al. 33 . The primary HUVECs were cultured in M199 medium supplemented with: penicillin 100 U/mL and streptomycin 100 μg/ mL, L-glutamine 2 mMol/L (all from Bio Whittaker/Lonza), heat-inactivated human serum 10% (Sanquin, Amsterdam, The Netherlands), heat-inactivated new-born calf serum 10% (Gibco), crude endothelial cell growth factor 150 μg/mL (locally prepared from bovine brains) and heparin 5 U/mL (Leo pharmaceutical products, Weesp, The Netherlands). Cells were cultured at 37 °C and 5% CO 2 , and medium was refreshed every second day. For all experiments, pools of HUVECs of 3 donors in passages 1-2 were used.
Lonza HUVECS. Primary HUVECs were purchased from Lonza (#CC-2519) and cultured on fibronectin-coated plates in Endothelial Cell medium (ECM), supplemented with singlequots (Sciencell Research Laboratories). Cells were cultured at 37 °C and 5% CO 2 and the medium was refreshed every second day. Experiments were performed with cells until passage 7. . For ECIS measurements, cells were seeded on gelatin-coated ECIS plates containing gold intercalated electrodes (Applied biophysics). When primary isolated HUVECs were confluent, they were serum-starved in M199 medium supplemented with 1% human serum albumin (HSA, Sanquin) for approximately 90 minutes. Subsequently, compounds were added to the cell medium. For Lonza HUVECs, ECIS plates were coated with 5 μg/ml Fibronectin and the compounds were directly added to the ECM medium.
Macromolecular permeability of the endothelial barrier was measured by passage of horse radish peroxide (HRP) through human endothelial barriers. Lonza HUVECs were seeded on top of gelatin-fibronectin coated Thin-Certs TM (Greiner Bio-One) and cultured in ECM with a medium change every second day. When a stable barrier was formed the medium in the upper compartment was replaced by complete medium containing HRP 5 μg/mL and CSN5i-3 or a vehicle control. At several time-points a sample was taken from the lower compartment. The HRP concentration was calculated by measuring absorbance after adding tetramethylbenzidine (TMB) (Upstate/Millipore) and sulfuric acid to stop the reaction.
Immunofluorescence imaging of cultured endothelial cells. Lonza HUVECs were seeded on 2 cm 2 coverslips (Thermo Scientific, Menzel-gläser) (#10319303) which were pre-coated with 5 μg/ml fibronectin. Cells were grown until confluency with a medium change every second day and upon reaching confluency, experiments were performed. Cells were fixed with warm (37 °C) 4% paraformaldehyde (Sigma Aldrich) (#158127) in phosphate buffered saline (PBS) (B Braun) (#3623140) and incubated at room temperature for 15 minutes. The PFA was washed away with PBS, cells were permeabilized with 0,2% triton X-100 in PBS for 3 minutes and blocked for 30 minutes with 1% HSA in PBS. Hereafter, coverslips were stained with primary antibodies in 1% HSA/PBS for 1 hour at room temperature or overnight at 4 °C. After washing with PBS, coverslips were incubated with a FITC-labeled secondary antibody (anti-rabbit or anti-mouse 1:100 in 1% HSA/PBS), Acti-stain 670 phalloidin (Cytoskeleton) (#PHDN1-A) and DAPI (Thermo Fisher Scientific) for 1 hour at room temperature. Coverslips were mounted with Mowiol4-88/DABCO solution (Calbiochem, Sigma Aldrich). Confocal scanning laser microscopy was performed on a Nikon A1R confocal microscope (Nikon). Images were analyzed and equally adjusted with ImageJ software.
Zebrafish husbandry, embryo care and compound treatment. Adult Tg(Fli1:GFP) y1 casper zebrafish were maintained at 26 °C in aerated 5-L tanks with a 10/14 hour dark/light cycle 34,35 . The Tg(fli1:GFP) y1 zebrafish line expresses GFP in the endothelial cells of the entire vasculature under the control of the fli1 promoter. Zebrafish were raised, staged and maintained according to standard procedures (zfin.org). Zebrafish embryos www.nature.com/scientificreports www.nature.com/scientificreports/ were collected within the first hours post fertilization (hpf) and kept at 28 °C in E3-medium (5.0 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl·2H 2 O, 0.33 mM MgCl 2 ·6H 2 O) supplemented with 0.3 mg/L methylene blue. For compound treatment, zebrafish embryos were manually dechorionated at 24 hpf and transferred to separate wells. After experiments were performed zebrafish embryos were anesthetized in 0.02% (w/v) buffered 3-aminobenzoic acid methyl ester (pH 7.0) (Tricaine) (Sigma-Aldrich) (#A5040) and euthanized by hypothermic shock. All experiments involving zebrafish embryos were according to local animal welfare regulations. VU University medical center animal welfare committee approved the breeding procedure of the adult zebrafish. Experimental procedures were performed in zebrafish larvae from 1-4 days post-fertilization prior to the stage of free living, which is in accordance with the EU Animal Protection Directive 86/609 EEC.

Preparation of zebrafish embryo lysates for western blot.
For western blot analysis of Cullin-3 expression, whole lysate of the zebrafish embryos was prepared between 72 hpf and 96 hpf, the same time-frame in which the dextran was injected for analysis of leakage. Prior to lysis, zebrafish embryos were anesthetized in 0.02% (w/v) buffered 3-aminobenzoic acid methyl ester (pH 7.0) (Tricaine) (Sigma-Aldrich) (#A5040), collected in an Eppendorf tube and euthanized by hypothermic shock. The water was removed and 10 ul 2x SDS sample buffer per fish was added. The lysate was boiled for 3 minutes at 95 °C and homogenized by sonication.
Caspase-3/7 assay. HUVECs were grown to confluency in a Black Falcon 96-well plate with clear bottom.
For the analysis of caspase-3/7 activity, medium was replaced by fresh medium containing 0, 1 or 4 μM CSN5i-3. Cells were treated for 5 hours with 200 nM Staurosporin as a positive control. After 5 and 24 hours of treatment with CNS5i-3, caspase-3/7 activity was analyzed using the Apo-ONE ® Homogeneous Caspase-3/7 Assay kit and following the manufacturer's protocol.
RhoB ubiquitination assay. HEK293T cells were co-transfected with mCherry-RhoB and HA-Ubiquitin using TransIT-LT1 (Mirus) (#MIR 2300) and following the manufacturer's protocol. The next day, cells were treated for five hours with 1 or 4 μM CSN5i-3 or 500 nM MLN4924, with addition of 2.5 μM MG132 for the last four hours. Next, denaturing HA-immunoprecipitation was performed as described previously 36 . Statistical analysis. Data is represented as mean ± SD. Statistical significance was tested by one-way ANOVA or repeated measures ANOVA with Dunnet's post-hoc test, unless indicated differently. P-values were considered statistically significant if p < 0.05. Analysis was performed using GraphPad Prism 7 software.