The Lectin LecB Induces Patches with Basolateral Characteristics at the Apical Membrane to Promote Pseudomonas aeruginosa Host Cell Invasion

ABSTRACT The opportunistic bacterium Pseudomonas aeruginosa can infect mucosal tissues of the human body. To persist at the mucosal barrier, this highly adaptable pathogen has evolved many strategies, including invasion of host cells. Here, we show that the P. aeruginosa lectin LecB binds and cross-links fucosylated receptors at the apical plasma membrane of epithelial cells. This triggers a signaling cascade via Src kinases and phosphoinositide 3-kinase (PI3K), leading to the formation of patches enriched with the basolateral marker phosphatidylinositol (3,4,5)-trisphosphate (PIP3) at the apical plasma membrane. This identifies LecB as a causative bacterial factor for activating this well-known host cell response that is elicited upon apical binding of P. aeruginosa. Downstream from PI3K, Rac1 is activated to cause actin rearrangement and the outgrowth of protrusions at the apical plasma membrane. LecB-triggered PI3K activation also results in aberrant recruitment of caveolin-1 to the apical domain. In addition, we reveal a positive feedback loop between PI3K activation and apical caveolin-1 recruitment, which provides a mechanistic explanation for the previously observed implication of caveolin-1 in P. aeruginosa host cell invasion. Interestingly, LecB treatment also reversibly removes primary cilia. To directly prove the role of LecB for bacterial uptake, we coated bacterium-sized beads with LecB, which drastically enhanced their endocytosis. Furthermore, LecB deletion and LecB inhibition with l-fucose diminished the invasion efficiency of P. aeruginosa bacteria. Taken together, the results of our study identify LecB as a missing link that can explain how PI3K signaling and caveolin-1 recruitment are triggered to facilitate invasion of epithelial cells from the apical side by P. aeruginosa.

P. aeruginosa. Apical LecB binding led to Src signaling, followed by local PI3K activation, PIP 3 patch formation at the apical plasma membrane, Rac1 signaling, and actin rearrangement to trigger the formation of protrusions in order to enable host cell invasion of P. aeruginosa. In addition, we show that caveolin-1 is recruited abnormally to apical membranes after LecB stimulation and that PI3K activation requires caveolin-1. These data suggest LecB as a unifying factor that facilitates and modulates many of the invasion mechanisms that have been reported for P. aeruginosa.

RESULTS
Apical LecB treatment triggers Src-PI3K/Akt signaling. To more closely analyze the effects caused by the application of purified LecB to the apical side of polarized MDCK cells, we used MDCK cells stably expressing the green fluorescent protein (GFP)tagged reporter PH-Akt-GFP, which indicates the localization of the lipid PIP 3 (Fig. 1A) (21). In unstimulated cells, PH-Akt-GFP localized mainly to the basolateral plasma membrane, as expected from the role of PIP 3 as a basolateral marker in polarized epithelial cells (21). In cells treated apically with LecB, PIP 3 accumulated at the apical side and protrusions formed that were positive for PH-Akt-GFP (Fig. 1A, white arrows). This replicates the effects that were previously observed after interaction of whole P. aeruginosa bacteria with the apical plasma membrane of MDCK cells (22).
Importantly, we demonstrated already that apical LecB application did not disturb the integrity of tight junctions (19). Thus, LecB-mediated apical PIP 3 accumulation cannot be explained by a loss of the barrier function of tight junctions. We therefore investigated whether activation of PI3K is the cause of apical PIP3 accumulation. Staining cells with antibodies recognizing active PI3K (pP85-Y458 and pP55-Y199) (23, 24) revealed a clearly visible recruitment and activation of PI3K to subapical regions in LecB-treated cells (Fig. 1B). In addition, incubating the cells with the broad-spectrum PI3K inhibitor LY294002 blocked the apical appearance of PH-Akt-GFP after LecB treatment ( Fig. 1C and D). Activation of PI3K was also detectable by Western blotting (WB) and peaked at approximately 15 min after initiation of LecB stimulation (Fig. 1E). Upstream from PI3K, the activation of Src kinases was required, as demonstrated by the ability of the Src kinase inhibitors PP2 and SU6656 to block LecB-induced PI3K activation (Fig. 1F). LecB also activated Akt, for which phosphorylation at S473 was detectable after 30 min of LecB application and peaked at approximately 4 h (Fig. 1G). Akt signaling occurred downstream from PI3K, because the broad-spectrum PI3K inhibitor LY294002 blocked Akt activation (Fig. 1H). Further tests revealed that the PI3K subunit p110a was mainly responsible for LecB-mediated Akt activation, since the p110a-specific inhibitor PIK-75 blocked Akt activation (Fig. S1A in the supplemental material), whereas the p110b-specific inhibitor TGX-221 did not (Fig. S1B). Another fucose-binding lectin, Ulex europaeus agglutinin I (UEA-I), failed to replicate LecB-triggered Akt signaling (Fig. S1C), thus indicating that the observed effects are specific for LecB.
To demonstrate that LecB-mediated PI3K/Akt activation is not limited to MDCK cells, we carried out experiments in other cell lines. We chose H1975 lung epithelial cells because P. aeruginosa frequently infects lungs. Whereas MDCK cells are Gb3-negative, H1975 cells express Gb3 (Fig. S2). The glycosphingolipid Gb3 has been previously found to be required for LecA-mediated internalization of P. aeruginosa (10). In H1975 cells, LecB also triggered Akt activation, in a dose-and time-dependent manner ( Fig. S3A to D) and dependent on PI3K ( Fig. S3E and F, showing the results of experiments using the pan-PI3K inhibitors wortmannin and LY294002 and the Akt inhibitor triciribine). As a further control, we verified that soluble L-fucose, which prevents LecB from engaging with host cell receptors, is able to inhibit LecB-triggered Akt signaling ( Fig. S3G and H). This demonstrates that LecB binding to fucosylated receptors is necessary to trigger PI3K/Akt signaling and also validates the purity of our LecB preparation.
To identify apical interaction partners of LecB, we applied LecB-biotin apically to polarized MDCK cells, lysed them, and precipitated LecB-receptor complexes with streptavidin beads. Mass spectrometry (MS) analysis revealed 12 profoundly enriched  (Table S1), underscoring the property of LecB of binding to multiple receptors. However, this property also prevented us from singling out a receptor that was responsible for LecB-triggered PI3K signaling, since the list included several proteins for which a capacity to elicit PI3K signaling was known (CEACAM1 [25,26], mucin-1 [27], ICAM1 [28], and podocalyxin [29,30]). Taken together, these findings show that after binding to fucosylated receptors at the plasma membrane of epithelial cells, LecB triggered an Src-PI3K/Akt signaling cascade, which replicated the cellular responses that were observed after binding of live P. aeruginosa cells to apical membranes (13).
Coating beads with LecB and expression of LecB by P. aeruginosa both enhance their apical uptake. To more realistically model the geometry during infection with P. aeruginosa, we utilized bacterium-sized beads that were coated with LecB. In pilot experiments using cell fixation, LecB-coated beads were seen to bind to the apical plasma membrane of polarized MDCK cells and to cause local apical accumulation of PH-Akt-GFP/PIP 3 ( Fig. 2A), and many beads were found to be completely internalized by cells (Fig. 2B). Live-cell microscopy experiments revealed that apical PH-Akt-GFP/ PIP 3 accumulation is a transient event that occurs before apical uptake of beads by MDCK cells (Fig. 2C, Movie S1). Detailed quantification showed that biotin-coated control beads were able to trigger apical PH-Akt-GFP/PIP 3 bursts to some extent, but at a much lower rate than LecB-coated beads (Fig. 2D). Interestingly, the PH-Akt-GFP/PIP 3 bursts caused by control beads were hardly sufficient to mediate cellular uptake, whereas the LecB-coated beads were taken up extensively (Fig. 2E). In addition, LecB treatment stimulated macropinocytotic uptake of dextran in H1975 cells ( Fig. S4A and B), which provides further evidence that LecB activates cellular uptake mechanisms.
Motivated by these results, we investigated whether the expression of LecB influences host cell uptake of live P. aeruginosa bacteria. Indeed, abrogation of LecB expression in P. aeruginosa (dLecB) and blockage of LecB with L-fucose diminished the apical uptake of P. aeruginosa in polarized MDCK cells (Fig. 2F). In accordance with previous studies (13,31), inhibition of Src kinases with PP2 and inhibition of PI3K with LY294002 also decreased P. aeruginosa uptake (Fig. 2F). Due to the easier handling, the experiments whose results are shown in Fig. 2F were carried out with MDCK cells grown in 24-well plates. For verification, we repeated them with transwell filter-grown MDCK cells, which yielded comparable results (Fig. 2G). Of note, the association of wild-type (wt) and dLecB P. aeruginosa with polarized MDCK cells was not significantly different ( Fig. S5), which suggests that the observed decrease of invasion efficiency upon deletion of LecB was due to LecB-mediated signaling and not due to reduced host cell binding. In H1975 cells, the uptake of P. aeruginosa was also lowered by LecB deletion (Fig. S4C) and L-fucose treatment (Fig. S4D).
Taken together, these data demonstrate that LecB promotes the uptake of P. aeruginosa from the apical side in polarized epithelial cells.
LecB-mediated PI3K signaling leads to Rac activation and actin rearrangement. To better understand the cellular response upon apical LecB stimulation, we investigated how PI3K activation is linked to P. aeruginosa uptake. Motivated by the known LecB-Mediated P. aeruginosa Host Cell Invasion mBio correlations between PI3K and Rac activation (32) and the reported implication of Rac in P. aeruginosa internalization (31), we carried out experiments using Rac123-G-LISA assays to test the capability of LecB to activate Rac. We found that apically applied LecB activated Rac in a time-dependent manner in MDCK cells (Fig. 3A) and also in H1975 cells (Fig. 3B). The PI3K inhibitor wortmannin blocked LecB-mediated Rac activation (Fig. 3C), indicating that PI3K activation occurred upstream from Rac activation. To investigate the consequences of LecB-mediated Rac activation on the actin cytoskeleton further, we utilized unpolarized H1975. The reason for this is that this allowed us to use overexpression of dominant-negative (DN) Rac1, which would result in unwanted side effects in polarized MDCK cells, because Rac1 also has roles during the polarization of MDCK cells (33). In sparsely seeded H1975 cells, LecB caused ruffle-like structures (Fig. 3D), and LecB colocalized with transfected Rac1-wt-GFP and actin in the ruffle-like regions (Fig. 3D, white arrows). To verify that LecB induced recruitment of Rac1-wt-GFP toward actin, we determined the Pearson's colocalization coefficient between Rac1-wt-GFP and actin, which increased significantly in LecB-treated cells ( Fig. 3E). This was not the case when DN Rac1-GFP (Rac1-DN-GFP) was overexpressed in H1975 cells ( Fig. 3F and E), showing the requirement of functional Rac for this effect. For verification, we repeated the experiment in untransfected H1975 cells using antibodies recognizing endogenous Rac1 (Fig. S6). Consistently, recruitment of Rac to actin upon LecB stimulation occurred as well in this experiment.
Apical application of LecB also led to substantial rearrangement of actin at the apical cell pole of MDCK cells (Fig. 3G). In untreated cells, dotted structures representing microvilli and the central actin-devoid region of the periciliary membrane and the primary cilium (34)(35)(36) were visible. In cells treated apically for 3 h with LecB, this subapical organization of the actin cytoskeleton was completely lost. Actin was recruited to lateral aspects of the cell membrane, and actin stress fibers constricting around the central position of the outgrowth of the primary cilium (Fig. 3G, white arrows) appeared.
In summary, the results of these experiments show that LecB-triggered PI3K signaling leads to Rac activation and actin rearrangement. All these processes have been previously observed during internalization of P. aeruginosa (22,31,37), thus further underscoring the role of LecB for P. aeruginosa host cell invasion.
Apical LecB treatment reversibly removes primary cilia. Motivated by our observation that LecB treatment led to the formation of actin stress fibers that appeared to constrict around the basis of the primary cilium, we investigated the effects of LecB on the primary cilium. Interestingly, apical application of LecB removed primary cilia from polarized MDCK cells (Fig. 4A and B) within 12 h. This effect was reversible after LecB-Mediated P. aeruginosa Host Cell Invasion mBio washout of LecB ( Fig. 4C and D). Although the potential physiological consequences of loss of primary cilia during P. aeruginosa infection remain to be investigated, this finding underscores the massive extent of LecB-mediated actin rearrangement. LecB triggers a feedback loop between caveolin-1 recruitment and PI3K activation. Interestingly, we also found caveolin-1 in the MS screen of LecB interactors (Table S1). Since caveolin-1 is a cytosolic protein, it presumably coprecipitated with LecB-interacting receptors. Motivated by this finding, we further investigated the behavior of caveolin-1 after LecB treatment. In undisturbed MDCK cells, caveolin-1 preferentially localized to the basolateral plasma membrane, as observed before (Fig. 5A) (38). However, apical LecB treatment resulted in abnormal recruitment of caveolin-1 toward the apical cell pole (Fig. 5A). In addition, the recruitment of caveolin-1 to LecB-receptor complexes was verifiable by WB and increased in a time-dependent manner (Fig. 5B). Interestingly, blocking Src kinases with SU6656 or PP2 and blocking PI3K with LY294002 diminished the coprecipitation of caveolin-1 in complexes with LecB-biotin (Fig. 5C). To directly investigate the requirement of caveolin-1 for LecB-mediated PI3K activation, we knocked down caveolin-1 in MDCK cells using small hairpin RNA (shRNA) (Fig. S7). Caveolin-1 knockdown almost completely suppressed PI3K activation upon LecB treatment (Fig. 5D).
Taken together, these data demonstrate that caveolin-1 is apically recruited by LecB stimulation and that this recruitment requires activation of Src kinases and PI3K, whereas caveolin-1 is also required for LecB-triggered PI3K activation. This constitutes a positive feedback loop between caveolin-1 recruitment and PI3K activation.

DISCUSSION
Here, we demonstrate that LecB is able to trigger an Src-PI3K-Rac signaling cascade, which is modulated by caveolin-1 and leads to actin rearrangement and protrusion formation in order to promote cellular uptake of P. aeruginosa bacteria. This adds LecB- LecB-Mediated P. aeruginosa Host Cell Invasion mBio triggered signaling to the growing list of P. aeruginosa host cell invasion mechanisms, which provokes the question of why this bacterium has evolved so many invasion mechanisms and how LecB fits in. The multitude of invasion mechanisms might be rooted in the adaptability of this opportunistic pathogen. P. aeruginosa can infect the respiratory tract, urinary tract, eye, and skin (39), and it was demonstrated that this bacterium can invade epithelial cells from the lung (9), cornea (2), and kidneys (22,40). Considering this diversity, it makes sense that P. aeruginosa possesses many invasion mechanisms, which might be used by the bacterium depending on the type of host cell encountered. One example is lipid zipper-type invasion, which requires interaction between LecA from P. aeruginosa and the glycosphingolipid Gb3 as the host cell factor (10). However, this lipid is not expressed in all epithelial cell types. For example, the MDCK cells used in this study do not express Gb3 (Fig. S2) (41). Nevertheless, P. aeruginosa successfully invaded MDCK cells, and thus, it uses alternative pathways like LecB-mediated signaling, as we demonstrated here. In addition, we show that LecB deletion in P. aeruginosa also decreased the invasion efficiency in H1975 cells, which we identified as Gb3 positive (Fig. S2), and it has been demonstrated previously that Gb3 expression in MDCK cells increased the invasion efficiency (10). These examples suggest that invasion mechanisms, such as LecA-and LecB-dependent invasion, are not exclusive but rather function in an additive manner. Our data provide an additional line of evidence for a cooperative function of invasion mechanisms. Coating of bacterium-sized beads with LecB markedly stimulated their uptake into cells, thus demonstrating that LecB alone is sufficient for stimulating cellular uptake. But LecB deletion or blocking LecB with L-fucose did not decrease the internalization of P. aeruginosa bacteria to the same extent as inhibition of Src kinases and PI3K did. This hints at other bacterial factors that are also able to cause PI3K-dependent uptake into host cells. A potential candidate is type IV pili, since deletion of pili led to a small but significant reduction of PI3K/Akt activation upon apical application of P. aeruginosa to polarized Calu-3 cells (12).
How is LecB able to trigger the Src-PI3K-Rac-actin signaling cascade? By MS analysis, we showed that LecB binds multiple apical receptors capable of triggering PI3K-signaling: CEACAM1 (25,26), Mucin-1 (27), ICAM1 (28), and podocalyxin (29,30). This makes it on one hand more robust for the bacterium to trigger the desired response, but it also makes it difficult for us to isolate a detailed mechanistic picture of LecB action at the apical cell membrane. We hypothesize that LecB has, due to being a tetramer that offers four binding sites (42), the capacity to cross-link and cluster different receptors (19,43), which is a general mechanism to activate receptor-mediated signaling cascades at the cell membrane. The data we present here provide two independent lines of evidence for this hypothesis. The first line derives from our control experiments with the lectin UEA-I. UEA-I is also able to bind fucose, but it has only two binding sites (44). This makes UEA-I a less ideal cross-linker than the tetrameric LecB, which was shown to be capable of crosslinking fucosylated lipids and integrins (19,43). Consequently, we found that UEA-I was not capable of eliciting PI3K signaling. This confirms that binding to fucosylated receptors is not enough and additional cross-linking, as in the case of LecB, is required for triggering PI3K signaling. The second line of evidence can be deduced from our experiments regarding caveolin-1. It has been shown that receptor cross-linking is sufficient to aberrantly induce caveolin-1-containing caveolae at the apical plasma membrane of epithelial cells (38,45). LecB application at the apical plasma membrane also caused the abnormal recruitment of caveolin-1 to the apical plasma membrane, which can be explained by assuming that LecB cross-linked receptors. In addition, the fact that caveolin-1 knockdown abrogated LecB-mediated PI3K activation, together with our finding that caveolin-1 recruitment could be blocked by PI3K inhibitors, suggests that there exists a positive feedback loop between PI3K activation and caveolin-1 recruitment. This is strongly supported by our observation that caveolin-1 coprecipitation with apical LecB receptors increased in a time-dependent manner. This also offers an explanation for the previously reported role of caveolin-1 for P. aeruginosa host cell invasion (9).
There has been speculation in the literature about the initial events that trigger the basolateral patch formation at the apical membrane by P. aeruginosa, and two possible hypotheses were offered (11): Either membrane damage could be responsible, or a still unknown bacterial factor causes the required PI3K activation. Our results favor the second hypothesis. Binding and cross-linking of apical receptors by LecB offer a direct explanation for PI3K activation and, thus, identify LecB as the unknown bacterial factor. In addition, we previously reported that application of purified LecB to the apical plasma membrane of MDCK cells does not induce membrane damage, as measured by trypan blue assays that use the fluorescence of trypan blue as a sensitive readout (19). Likewise, tight junction integrity was not affected by apical application of LecB (19). This is in agreement with the finding by others that the formation of PIP 3 -rich protrusions during infection with P. aeruginosa did not compromise tight junctions (11). This finding also excludes the possibility that LecB-triggered apical PIP 3 accumulation occurred by diffusive spreading of PIP 3 from the basolateral plasma membrane and additionally proves that apical PIP 3 accumulation was due to LecB-mediated local PI3K activity at the apical plasma membrane.
The involvement of Rac1 for P. aeruginosa internalization through the LecB-triggered cascade we describe here will need further clarification. Specifically, the P. aeruginosa exotoxin S and exotoxin T are known to contain N-terminal RhoGTPase activating protein (RhoGAP) domains, which can hydrolyze GTP to GDP in Rho, Rac, and Cdc42, leading to cytoskeletal depolymerization and countering host cell invasion (46,47). It will be interesting to investigate whether varying expression levels of LecB, exotoxin S, and exotoxin T cause more or less invasive behavior of P. aeruginosa.
In conclusion, our results identify LecB as a novel bacterial factor that promotes uptake of P. aeruginosa bacteria from the apical side of epithelial cells. Our data suggest that LecB represents a missing link that provides a unifying explanation for many observations that have been made during host cell invasion by P. aeruginosa. We revealed that LecB is sufficient to trigger the well-known Src-PI3K-Rac signaling cascade (11), which is required for basolateral patch formation at the apical plasma membrane and host cell invasion. LecB-mediated signaling also provides additional rationales for the previously found implication of caveolin-1 in P. aeruginosa invasion (9), since we identified here a LecB-triggered positive feedback loop between PI3K activation and caveolin-1 recruitment to the apical plasma membrane.

MATERIALS AND METHODS
Antibodies, plasmids, and reagents. The antibodies used are listed in Table S2. The plasmid pPH-Akt-GFP encoding PH-Akt-GFP was a gift from Tamas Balla (Addgene plasmid no. 51465). The plasmids encoding wild-type Rac1 tagged with GFP (Rac1-wt-GFP) and a mutant protein bearing a change of T to N at position 17 (Rac1-T17N) and tagged with GFP (Rac1-DN-GFP) were kindly provided by Stefan Linder (University Hospital Hamburg-Eppendorf, Germany).
Mammalian cell culture and creation of stable cell lines. MDCK strain II cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 5% fetal calf serum (FCS) at 37°C and 5% CO 2 . H1975 cells were maintained in Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 10% FCS at 37°C and 5% CO 2 . For generating polarized MDCK monolayers, 3 Â 10 5 MDCK cells were seeded on transwell filters (12-well format, 0.4-mm pore size, polycarbonate membrane, product number 3401; Corning) and cultured for 4 days before experiments. For experiments with H1975 cells, 3 Â 10 4 cells were seeded per 12-mm glass cover slip placed in a 24-well plate and cultured for 1 day. For the creation of the MDCK cell line stably expressing PH-Akt-GFP, cells were transfected with the plasmid pPH-Akt-GFP using lipofectamine 2000 (Thermo Fisher). After allowing the cells to express the proteins overnight, they were trypsinized and plated sparsely in medium containing 1 mg/mL G418. After single colonies had formed, GFP-positive colonies were extracted with cloning rings. At least 6 colonies were extracted for each cell line, grown on transwell filters for 4 days, fixed, and stained against the basolateral marker protein b-catenin and the tight junction marker protein ZO-1 to assay their polarized morphology. Based on these results, we chose one colony for each cell line for further experiments. the instructions of the manufacturers and purified using PD-10 desalting columns (GE Healthcare). StxB was labeled with NHS-ester conjugated with Alexa Fluor 488 (Thermo Fisher).
Preparation of LecB-coated beads. Biotinylated LecB (LecB-biotin) was incubated with a solution containing streptavidin-coated polystyrene beads containing the dye flash red with 1-mm diameter (Bangs Laboratories). To ensure homogenous coverage with LecB-biotin, a 10-fold molar excess of LecBbiotin compared to the available streptavidin binding sites on the beads was used, and then beads were washed three times with PBS. In control beads, the streptavidin binding sites were saturated with biotin.
Mass spectrometry-based identification of LecB interaction partners. MDCK cells were cultured in medium for stable-isotope labeling by amino acids in cell culture (SILAC medium) for 9 passages and then seeded on transwell filters and allowed to polarize for 4 days. For the first sample, biotinylated LecB was applied to the apical side of light-SILAC-labeled cells and on the basolateral side of medium-SILAC-labeled cells, whereas heavy-SILAC-labeled cells received no stimulation and served as a control. For the second sample, the treatment conditions were permuted. After lysis with immunoprecipitation (IP) lysis buffer, the different SILAC lysates were combined and LecB-biotin-receptor complexes were precipitated using streptavidin agarose beads as described before. Eluted LecB-biotin-receptor complexes were then prepared for MS analysis using SDS-PAGE gel electrophoresis. Gels were cut into pieces, the proteins therein digested with trypsin, and the resulting peptides were purified by stop-and-go-extraction (STAGE) tips. MS analysis was carried out as described previously (19) using a 1200 HPLC (Agilent Technologies, Waldbronn, Germany) connected online to a linear trap quadrupole (LTQ) Orbitrap XL mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). From the list of MS-identified proteins generated, we defined those proteins as LecB interaction partners that showed more than 2-fold enrichment on a log 2 scale over controls in both SILAC samples (Table S1).
Statistics. If not stated otherwise, data obtained from n = 3 independent experiments were used to calculate arithmetic means, and error bars represent standard errors of the means (SEM). Statistical significance analysis was carried out using GraphPad Prism 5. For determining the significance in experiments with multiple conditions, one-way analysis of variance (ANOVA) with Bonferroni's post hoc testing was applied. For determining the significance in experiments in which values were measured for one condition relative to the control condition, one-sample t testing was applied. n.s. denotes not significant, * denotes P , 0.05, ** denotes P , 0.01, *** denotes P , 0.001, and **** denotes P , 0.0001. All primary data are available from the authors upon request.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.