Prominin-1 (CD133) modulates the architecture and dynamics of microvilli

2.1 Distribution and morphology of microvilli at the apical plasma membrane

Prior to the analysis of the impact of prominin-1 on microvilli, we performed a high-resolution characterization of the apical membrane of polarized Madin-Darby canine kidney (MDCK) epithelial cells by scanning electron microscopy (SEM). At 6 days post-confluence (dpc), the distribution of microvilli varied considerably between neighboring cells ranging from densely packed to sparsely scattered (Figure 1A). Microvilli were present as individual units or formed clusters comprising about 10 microvilli (Figure 1A). The mechanism underlying the formation of microvillar clusters is poorly characterized.42, 43 No microvillus was detected in the proximity of the primary cilium, while those at cell borders were arranged in an array (Figure 1A). Microvillar length differed from cell to cell and was categorized into short (Figure 1B, <200 nm), intermediate (200-400 nm) and long (>400 nm). Few cells showed membrane ruffles or branching microvilli (Figure 1B). The distribution and morphology of microvilli were uniform throughout the surface of a given cell, but differed between adjacent cells (Figure 1B). No clusters of cells with a particular pattern were recognized, indicating that the clonal origin does not explain the morphological variability.

A quantitative analysis based on SEM data revealed numerous cells with membrane ruffles at early time points of polarization (Figure 1C, 1-3 dpc; Supporting Information Table S1). From 6 to 12 dpc, short and intermediate microvilli were most common. We rarely observed cells harboring long or branched microvilli or membrane ruffles (<3%). Microvillar heterogeneity was less pronounced in subconfluent cultures, in which many cells did not show any membrane protrusions while others developed membrane ruffles and/or short microvilli (Figure S1A-D). Varying numbers of filopodia were detected at free cell borders and microvilli arrays started to form at cell-cell contacts (Figure S1E-G).

2.2 Expression of human prominin-1 in epithelial cells

To study the influence of prominin-1 on microvilli we overexpressed two splice variants of human prominin-1 (s1 and s2) in MDCK cells. They differ by the short facultative exon 3 in the extracellular N-terminal domain of splice variant s2 that encodes for nine amino acids (PETVILGLK; Figure 2A, red square). Although the function of these residues is unknown, they might interfere with the interaction between prominin-1 and the ganglioside GM₁. As described by Taïeb et al, prominin-1 contains two glycosphingolipid-binding domains.26 One is located in the N-terminal domain and is proposed to bind GM₁ (Figure 2A, orange cylinder) while the other was found in the vicinity of the second transmembrane domain and interacts with GD₃ (Figure 2A, yellow cylinder). The GM₁-binding site contains typical turn-inducing residues (here, Pro₃₇ and Gly₅₀) and an aromatic one (Tyr₄₁). The latter is surrounded by charged amino acids and exposed to the extracellular milieu (Figure 2A and Figure S2A; for details see Reference 26). To determine the impact of this lipid-binding domain on the function of prominin-1 we mutated two of the key residues (Pro₃₇ → Ala and Tyr₄₁ → Ser) in both splice variants (Figure 2A). They were termed 2M mutants. These residues are conserved among other mammalian prominin-1 sequences (Figure S2A). Upon stable transfection, more than 85% of the cells of each cell line expressed the recombinant protein as determined by flow cytometry (Figure S3A) with comparable expression levels as observed by immunoblotting (Figure 2B and Figure S3B,C).

The localization of prominin-1 was then examined by double labeling with mouse monoclonal antibody (mAb) CD133/1 directed against human prominin-1 and fluorescent cholera toxin B subunit conjugates that recognize ganglioside GM₁. Confocal laser scanning microscopy (CLSM) analysis revealed the colocalization of both prominin-1 splice variants with GM₁ at the apical membrane (Figure 2C, top panels and Figure S2B), in agreement with our previous report.25 The punctate pattern is characteristic for prominin-1 associated with microvilli.23, 44 Both 2M mutants also colocalized with GM₁ indicating that the created point mutations did not interfere with the proper microvillar localization of prominin-1 (Figure 2C, bottom panels and Figure S2B).

2.3 Prominin-1 and its mutants increase the number of microvilli

Does prominin-1 influence the architecture of microvilli? To address this question we analyzed the apical plasma membrane using transmission electron microscopy (TEM) and SEM. The analysis suggested that both splice variants of prominin-1, notably their 2M mutants, increase the number of microvilli (Figure 3A and Figure S4A). Besides, we noticed the occurrence of intracellular multivesicular structures in cells expressing 2M mutants (Figure 3A and Figure S4A, green asterisk).

SEM micrographs were used for the quantification of microvilli and revealed an average number of 95 ± 33 microvilli on wild-type cells (MDCK) for a surface area of 12 μm². This number was significantly increased in cells expressing human prominin-1, that is, 127 ± 49 on splice variant s1 and 127 ± 36 on s2 (Figure 3A, SEM top panels and Figure 3B). Interestingly, mutations in the GM₁-binding site further increased their number (see also Figure S4A). The 2M mutants harbored on average 168 ± 56 (s1) and 156 ± 43 (s2) microvilli. On wild-type MDCK cells, about 10 microvillar clusters were observed for a cell surface area of 17.3 μm² (Figure 3A, SEM bottom panels and Figure 3C) and each of them comprised about 10 ± 3 microvilli (Figure 3D). On cells expressing prominin-1.s1 or s2, a significant increase in the number of clusters and microvilli per cluster was detected (Figure 3C,D). The 2M mutants did not exert further effect on microvillar clusters, but the number of microvilli per cluster was increased in prominin-1.s1 2M (Figure 3C,D).

2.4 Prominin-1 alters the architecture of microvilli

During our EM investigations we recognized alterations in the shape of microvilli (Figure 4A,B and Figure S4A), which prompted us to conduct a quantitative analysis. Using SEM, more than 500 cells per line were randomly picked and assigned to one of the following categories: short, intermediate, long, branched, knob-like/irregularly shaped microvilli or membrane ruffles (Figure 4C and Table S2). About 25% of prominin-1⁺ cells showed branched microvilli irrespective of the splice variant (Figure 4A-C, green). Each branched segment was about the same diameter as the original microvillus. This phenotype was scarcely observed in control cells, that is, MDCK wild type or those expressing an empty vector (MOCK) (Figure 4C). The expression of 2M mutants led to a 2-fold increase in cells with branched microvilli (Figure 4C). Moreover, ≈20% of the cells harbored irregularly shaped microvilli often with a knob-like structure at their tips (Figure 4A-C, purple). These microvilli frequently showed branching at the same time indicating that both phenotypes are linked (Figure 4B). The knob-like phenotype was not observed in control cells and only in ≈2 and ≈10% of cells expressing prominin-1.s1 or s2, respectively (Figure 4C). In line with the observation made in wild-type cells, the altered morphologies of microvilli were distributed uniformly throughout the surface of a given cell, but differed between adjacent cells (Figure S4B). The coincidence of altered phenotypes and an increase in the number of microvilli suggests that prominin-1, notably its mutants, did not induce the formation of microvilli but might affect their stability, dynamics and/or their resorption. Indeed, all altered microvilli are of intermediate size indicating that prominin-1, and notably its 2M mutants, increases the length of microvilli.

Cells expressing the 2M mutants also harbored microvilli with small protruding vesicle-like structures at their tips suggesting that the budding of membrane vesicles might be affected (Figure 4A and Figure S4A, asterisk). This prompted us to examine the release of prominin-1⁺ vesicles. Therefore, cells were incubated for 16 hours with medium, which was then subjected to differential centrifugation followed by immunoblotting for prominin-1 and β-actin (Figure 5A). In cells expressing wild-type protein, the amount of prominin-1 recovered in Vesicle fractions corresponded to ≈10% to 12% of total immunoreactivity (Cell + Vesicle). This number dropped to 5% in both 2M mutants suggesting that the alterations of microvilli affected the proper budding and fission of vesicles (Figure 5B, n = 8).

2.5 PI3K inhibition reduces modified microvilli and restores vesicle release

We hypothesize that the overexpression of prominin-1.s1 or s2, and particularly the 2M mutants, may cause an activation of PI3K, which is an interacting partner of prominin-1. As a result PIP₂ might be converted into PIP₃ in the inner leaflet of the microvillar membrane. Because scaffolding proteins bind to PIP₂ and physically couple the plasmalemma to actin filaments, the phospholipid conversion might favor the detachment of the membrane from the underlying cytoskeleton and explain our phenotypes. To address this issue, cells were incubated for 4 hours with PI3K inhibitor LY294002 or dimethyl sulfoxide (DMSO) as vehicle control. Microvillar phenotypes and the release of prominin-1⁺ vesicles were evaluated. The inhibition of PI3K caused a decrease in the number of cells with altered microvilli and an increase in those harboring short ones (Figure 6A,B). Solely 11% ± 4% of prominin-1.s2 2M-expressing cells displayed knob-like microvilli upon LY294002 treatment while DMSO-treated cells showed 24% ± 6% (Figure 6B, purple). Cells with branched microvilli were also reduced. Similar data were obtained with splice variant 1 (Figure 6B and Table S3).

In 2M mutants-expressing cells treated with LY294002, the reduction of altered microvilli was reflected by a 3-5-fold increase in the release of prominin-1⁺ vesicles in comparison to DMSO alone (Figure 6C). No significant increase was observed in cells expressing wild-type prominin-1 (Figure 6C). Of note, DMSO caused the formation of short microvilli at the expense of intermediate ones (Figure 6B and Table S3). DMSO did not stimulate the release of prominin-1⁺ vesicles, suggesting that the occurrence of short microvilli is not the result of extended release (data not shown, see below).

2.6 Arp2/3 complex inhibition reduces modified microvilli

Using SEM micrographs, we evaluated the angle formed by two branching protrusions within a given microvillus. The observed angle of 64.7° ± 12.5° (n = 21) was highly reminiscent of the activity of Arp2/3 complex that choreographs the formation of branched actin networks.45 To evaluate its role, we incubated 2M mutants-expressing cells for 16 hours with the Arp2/3 complex inhibitor CK-666 or its inactive analogue CK-689.46 No differences were observed between cells treated with CK-689 or DMSO as control (Figure 7A,B). However, cells incubated with CK-666 showed a significant reduction of branched and knob-like microvilli, suggesting that the Arp2/3 complex is involved in the alteration of microvilli (Figure 7A,B and Table S4).

Given that the individual inhibition of PI3K and Arp2/3 complex impaired the formation of both branched and knob-like microvilli we evaluated the combined effect. The incubation of cells with LY294002 and CK-666 created a synergistic effect that almost abolished altered microvilli (Figure 7A,C and Table S5). The number of individual microvilli and clusters of them were also significantly reduced reaching a level comparable to wild-type cells under the same conditions, whereas the number of microvilli per cluster remained similar (Figure 7D-F). Thus, the increased number of microvilli in cells expressing prominin-1 mutants can be explained by their morphological alteration, which might interfere with their turnover. The altered microvilli generated by the overexpression of wild-type prominin-1.s1 and s2 were also lost upon the combined inhibition of PI3K and Arp2/3 complex (Table S5). Likewise, we detected a significant reduction in the number of microvillar clusters in non-transfected cells, which also suggests that their formation relies on the activity of PI3K and Arp2/3 complex (Figure 7E). Collectively, these data suggest that the overexpression of wild-type prominin-1 as well as mutations in its GM₁-binding site induce a crosstalk with the PI3K pathway and the Arp2/3 complex, and hence a remodeling of the actin cytoskeleton.

2.7 Impaired prominin-1-PI3K interaction reduces altered microvilli and restores vesicle release

To evaluate the impact of direct interaction between prominin-1 and PI3K on microvillar structure, we generated an additional mutation in the 2M mutants at tyrosine 819/828 (with reference to prominin-1.s1 and s2, respectively; Figure 2A and Figure S2A). This amino acid is found in a tyrosine phosphorylation consensus site [R/K]xxx[D/E]xxY,28 which is conserved in all prominin-1 sequences reported so far (Figure S2A). The mutation impedes the phosphorylation of prominin-1 and thereby its interaction with the 85-kDa regulatory subunit (p85) of PI3K. The changes in tyrosine phosphorylation of cell surface prominin-1 were demonstrated by its immunoisolation using AC133 mAb-conjugated magnetic beads and immunoblotting with anti-phosphotyrosine Ab (Figure 8A). The quantification revealed a significant reduction in the phosphorylation of 2M Y819/828F mutants (Figure 8B). The remaining phosphotyrosine signal can be attributed to the second phosphorylation site in prominin-1 at tyrosine 843/852.28 Interestingly, we detected only a slight but not significant increase in the phosphorylation of tyrosines in 2M mutants compared with their wild-type proteins. Nonetheless, the interaction of prominin-1 with p85 was strongly stimulated in cells expressing 2M mutants, but not in those carrying the additional tyrosine mutation (Figure 8C,D). These data suggest that mutations in the first ganglioside-binding site of prominin-1 influence its interaction with PI3K. This observation together with the impact of the Arp2/3 complex inhibitor on microvillar structure prompted us to probe the collected materials for members of the Arp2/3 complex, that is, Arp2 and p34-Arc/ARPC2. Surprisingly, the two subunits were co-immunoisolated with prominin-1 (Figure 8C,D). Again, the 2M mutations increased these interactions while the additional tyrosine mutation reduced them (Figure 8D).

Morphologically, quantitative SEM analysis revealed that the amount of prominin-1 2M Y819/828F-expressing cells with knob-like/irregular microvilli dropped to 2% to 5% (Figure 9A,B). The number of cells harboring branched microvilli was reduced by half in comparison to 2M mutants (Figure 9B). These data indicate that the lack of tyrosine 819/828 phosphorylation impedes the structural alterations provoked by mutations in the GM₁-binding site. As a result, cells expressing prominin-1 2M Y819/828F show a distribution of microvillar phenotypes similar to their corresponding wild-type proteins (Figures 4C and 9B). The additional Y819/828F mutation also restored the release of prominin-1⁺ vesicles (Figure 9C).

2.8 Ubiquitination of prominin-1 and its interaction with syntenin-1 are stimulated in 2M mutants

As microvilli stability and/or dynamics are changed upon the mutations in prominin-1 we suspected that its recycling and degradation might be altered.14, 47 Therefore, we evaluated whether prominin-1 is subjected to ubiquitination and/or interacts with the adaptor protein syntenin-1.16 Both, the ubiquitination of cell surface prominin-1 and its interaction with syntenin-1, were significantly increased in cells expressing 2M mutants suggesting that mutations in the ganglioside-binding site of prominin-1 can trigger its recycling. The additional Y819/828F mutation reduced both interactions (Figure 8E,F). In contrast, no interaction with ezrin, an adaptor protein of the ezrin/radixin/moesin (ERM) family, was observed (Figure 8E), which is in line with our previous study.16

2.9 Mutating cytoplasmic tyrosine 819/828 in prominin-1 results in short microvilli

Does tyrosine phosphorylation of prominin-1 per se affect the microvillar architecture? To address this question, we analyzed their morphology in cells expressing human prominin-1.s1 or s2 carrying solely the Y819/828F mutation. This mutation impeded the phosphorylation of prominin-1 and its interaction with the p85 subunit of PI3K and Arp2/3 complex (Figure 8A-D). CLSM analysis revealed that both mutants localized to the apical plasma membrane, but the immunoreactive signal of prominin-1 appeared narrower in the mutants in comparison to wild-type protein (Figure 10A, right panels). SEM analysis revealed that ≈75% of cells expressing mutant proteins displayed short microvilli at 6 dpc (Figure 10B,C and Table S6) and 14 dpc (Table S6), which is a significant increase compared with control cells (P < 0.005). This suggests that the phosphorylation of prominin-1 and/or its interaction with PI3K affects the length of microvilli. Moreover, no altered microvilli were detected suggesting that the point mutation neutralizes the effect of prominin-1 overexpression. These data are in line with those obtained upon PI3K inhibition. Finally, the Y819/828F mutation did not stimulate the release of prominin-1⁺ vesicles, indicating that short microvilli are not the result of extensive membrane budding (Figure 10D, see also Figure 5B).

In sum, our data suggest that multiple interactions of phosphorylated prominin-1 with regulatory proteins, such as PI3K and the Arp2/3 complex, are implicated in the architecture and dynamics of microvilli on the apical surface of epithelial cells.

2.10 Prominin-1 mutants alter membrane protrusions of fibroblast-like cells

Given that the expression of prominin-1 is not limited to epithelial cells we transfected CHO cells, which have fibroblast-like characteristics, with prominin-1.s2 and its corresponding mutants. Interestingly, the expression of prominin-1.s2 increased the length of filopodia emerging from the cell border. Precisely prominin-1.s2-expressing cells showed a length of 3.72 ± 1.41 μm (mean ± SD, more than 1000 filopodia were analyzed) vs 1.71 ± 0.70 (n > 300) and 2.24 ± 0.96 (n > 600) μm of untransfected (CHO) and green fluorescence protein (GFP)-transfected cells, respectively (P < 0.01, two-tailed unpaired Student's t-test; Figure 11A, see cartoon). To visualize the cell surface, cell lines were labeled with fluorochrome-conjugated wheat germ agglutinin (WGA).

Moreover, filopodia of prominin-1.s2-transfected cells were often branched, and we observed the appearance of narrow, sheet-like structures morphologically similar to growth cones (Figure 11A,B, red asterisk).48 Often, long filopodia emerged from the apex of these structures. Growth cone-like structures were barely detected in prominin-1-negative cells. Astonishingly, they were more abundant in prominin-1.s2 2M mutant-expressing cells and the length of filopodia was also further increased (5.52 ± 2.27 μm, n > 600). Cells expressing prominin-1.s2 2M Y828F developed numerous axial and radial filopodial branches emanating from the growth cone-like structures. The latter appeared larger than those of non-mutated or prominin-1.s2 2M mutant cells. The length of filopodia of prominin-1.s2 2M Y828-expressing cells was 5.49 ± 2.04 μm (n > 180). In contrast, cells transfected with prominin-1.s2 carrying only the Y828F mutation showed short filopodia emanating from growth cone-like structures and lamellipodia (Figure 11A,B). Filopodia growing out from the cell border were of similar length as those developed by cells expressing wild-type prominin-1, that is, 4.14 ± 2.03 μm (n > 600), but statistically shorter than those of 2M mutant cells (P < 0.05). In all examined cells, the preferential localization of prominin-1 in highly curved membranes (eg, in filopodia and at the edge of the cell border) is evident.

In contrast to the vertical apical microvilli of MDCK cells, the horizontal formation of filopodia on a planar surface facilitated the investigation of the localization of the p34-Arc/ARPC2 subunit of the Arp2/3 complex. In cells expressing GFP (MOCK) or prominin-1 Y828F, the p34-Arc/ARPC2 staining appeared more distant to the edge of the cell membrane and weaker compared with the staining of those expressing prominin-1 wild-type or 2M mutants (Figure 11C, dashed line). In the latter, a colocalization of prominin-1 and p34-Arc/ARPC2 was observed (Figure 11C, white arrowheads) and p34-Arc/ARPC2 was concentrated in the growth cone-like structures (Figure 11C, red asterisk). However, p34-Arc/ARPC2 is excluded from filopodia in all cell lines except the one expressing prominin-1 2M Y828F, which could explain the presence of highly branched filopodia (Figure 11C, red arrowhead). It is interesting to note that the colocalization of prominin-1 and p34-Arc/ARPC2 was also detected in the “soma” part of the cells, particularly in those expressing prominin-1 2M Y828F (Figure 11C, white arrows).

Using our co-immunoisolation protocol (see above), we observed an increased interaction of prominin-1.s2 2M mutant with Arp2/p34-Arc/ARPC2 subunits of the Arp2/3 complex in comparison to wild-type prominin-1 and Y828F mutant (J.K. and D.C., unpublished data). The same observations were made with PI3K (ie, p85; data not shown). Altogether, prominin-1 influences the structure of filopodia and the localization of the p34-Arc/ARPC2 subunit of the Arp2/3 complex in fibroblast-like cells. Furthermore, the data suggest that this interaction, as well as the one with PI3K, occur in both polarized epithelial and fibroblast-like cells.

2.11 Silencing of prominin-1 alters microvilli of HSCs

All data presented above suggest that mutations in human prominin-1 influence the structure of microvilli and other membrane protrusions. This prompted us to investigate the microvillar morphology of a new murine model (rd19) where a spontaneous mutation in Prom1 gene causes prominin-1 knockdown.34 As demonstrated by immunoblotting, both N-glycosylated prominin-1 found in the kidney as well as its endo H-resistant or PNGase F-sensitive deglycosylated forms were not detected in rd19 mice (Figure 12A). TEM analyses of the kidney and duodenum revealed no major alterations of individual microvilli present in both epithelia (Figure 12B,C, respectively), which is in line with previous observations in another prominin-1 null mice.37 The length, diameter and internal actin cytoskeleton of microvilli are not markedly affected by the lack of prominin-1 (Figure 12C). However, we observed budding-like structures at the basal part of microvilli more often in rd19 mice than in wild-type animals, where they were detected at the tip of microvilli (Figure 12C, red and black arrows, respectively). The latter might generate extracellular membrane vesicles while the former can result in both, vesicles and branched microvilli. Interestingly, the organization of the apical cortex of certain rd19 duodenal cells seemed uneven, which led to the appearance of cluster of microvilli (Figure 12D, dashed line) with fused and/or branched microvilli (Figure 12D, asterisks, red dots). Although fused microvilli can be observed in wild-type cells they were more pronounced in cells lacking prominin-1. The co-expression of prominin-2 in the kidney and small intestine, including the duodenum,19, 33-37 might explain the lack of general phenotype (see Section 3).

Given that MDCK cells express both canine prominin-1 and prominin-249 (K.T. and D.C., data not shown), we decided to study the consequences of the lack of human prominin-1 in HSCs. These cells express solely prominin-1.s27, 8 and, importantly, are devoid of prominin-2.19, 33-37 To that end, primary human CD34⁺ HSCs were immunoisolated from patients and prominin-1 was knocked down using short interfering (si) RNA oligonucleotides as described previously.50, 51 Cells were then cultured on primary mesenchymal stromal cells (MSCs), which were used as feeder cell layer. MSCs secreted various soluble factors, including stromal cell-derived factor-1, and thus mimic to a certain degree the physiological conditions associated with the bone marrow niche.51 We have previously demonstrated that the silencing of prominin-1 did not affect the migration and general polarization of CD34⁺ HSCs, that is, the formation of the uropod at the rear and lamellipodia at the leading edge.50 At that time, we did not determine the impact on microvillus-like structures, particularly those associated with the uropod where prominin-1 is concentrated.51 ^,52 Strikingly, SEM analysis revealed that the number of microvillus-like per uropod was significantly reduced in cells treated with siRNA against prominin-1 in comparison to cells transfected with control-siRNA (Figure 13A,B). In contrast, the knockdown of CD82, a tetraspanin protein associated with the HSC-uropod,52 did not affect microvilli found thereon (Figure 13A,C).

4.1 Animals and ethics statement

The prominin-1 knockout mouse (Prom1 ^rd19) carrying a spontaneous point mutation in Prom1 gene99 was obtained from The Jackson Laboratory (Sacramento, CA). The C57BL/6J mouse strain was used as control. Animal experiments were performed in strict accordance with German Animal Welfare legislation, and all procedures were performed in compliance with the UK Animals (Scientific Procedures) Act 1986, which were approved by the Institutional Animal Welfare Officer. All efforts were made to minimize the suffering of animals, and the ones for organ removal were sacrificed according to the applicable European and National regulations (Tierschutzgesetz). The Max Planck Institute of Molecular Cell Biology and Genetics holds necessary licenses (74-9165.40-9-2000-1 and 74-9168.25-9-2001-1) for keeping and breeding laboratory animals and collecting organs and tissues, both issued by the Regierungspräsidium Dresden, Saxony.

4.2 Plasmid construction

Expression plasmids (hprom-1.s1 and hprom-1.s2) were constructed by introducing the full-length coding sequence of human prominin-1 splice variant 1 (GenBank AF507034.1, cDNA clone MGC: 20041 IMAGE: 4644690) and 2 (AF027208.1; Reference 100) into the enhanced yellow fluorescent protein (EYFP)-N1 vector (BD Bioscience Clontech). XhoI and NotI restriction sites were used for directional cloning.100 Note that XhoI/NotI digestion of pEYFP-N1 vector causes the excision of the fluorescent protein.

4.3 Site-directed mutagenesis

Two amino acids (proline 37 and tyrosine 41) in the GM₁-binding site of prominin-126 and tyrosine 819/828 (referring to prominin-1.s1 and s2, respectively) in its cytoplasmic C-terminal end were mutated using QuikChange II Site-Directed Mutagenesis Kit (catalog number #200523; Agilent Technologies) according to the manufacturer's instructions. The hprom-1.s1 and hprom-1.s2 plasmids were used as templates. The applied forward and reverse oligonucleotides were 5′-GGAATTATGAATTGGCTGCAACAAATTCTGAGACCCAAGACTCC-3′ [kt-5′^2M] and 5′-GGAGTCTTGGGTCTCAGAATTTGTTGCAGCCAATTCATAATTCC-3′ [kt-3′^2M] and 5′-GGATTCGGAGGACGTGTTCGATGATGTTG-3′ [kt-5′^Y828F] and 5′-CAACATCATCGAACACGTCCTCCGAATCC-3′[kt-3′^Y828F], respectively. In the 2M mutant, two single nucleotide changes (underlined) led to the encoding of alanine (Ala37) and serine (Ser41) instead of proline and tyrosine, respectively. In the Y819/828F mutants, the single nucleotide change (underlined) resulted in the encoding of phenylalanine instead of tyrosine. Expression plasmids were sequenced to confirm the absence of any reading mistakes introduced by the DNA polymerase and/or subcloning steps. Sequencing reactions were performed using Applied Biosystems 3730 Genetic Analyzer at the Sequencing Facility of the Max Planck Institute of Molecular Cell Biology and Genetics (MPI-CBG, Dresden, Germany).

4.4 Cell culture

MDCK cells (strain II, American Type Culture Collection, #ATCC CCL-34) were cultured in Dulbecco's modified Eagle medium (DMEM, 1×, 1 g/L d-glucose, l-glutamine, pyruvate, GIBCO, Thermo Fisher Scientific) supplemented with 10% fetal calf serum (FCS; PAA, GE Healthcare), 100 U/mL penicillin and 100 μg/mL streptomycin (GIBCO), 10 mM HEPES and MEM non-essential amino acids (1×, PAA). CHO cells7 were cultured in Ham's F-12 nutrient mixture (+l-glutamine, GIBCO) containing 10% FCS and 100 U/mL penicillin and 100 μg/mL streptomycin. Both cell lines were maintained in a humidified incubator at 37°C under 5% CO₂ atmosphere as described previously.49

Primary human HSCs and MSCs were collected from healthy donors after informed consent and approval of the local ethics committee (ethic board nos. EK201092004 and EK263122004). Isolation and cultivation of MSCs were performed as described.77 CD34⁺ HSCs were isolated from mobilized peripheral blood directly after leukapheresis by immunomagnetic separation using the magnetic-activated cell sorting system with CD34 MicroBeads (CD34 MicroBead Kit, #130-046-703; Miltenyi Biotec, Bergisch Gladbach) and LS columns following the manufacturer's instructions, as reported previously.77 Mobilization was achieved by subcutaneous injection of 7.5 μg/kg granulocyte colony-stimulating factor per day for 5 days (Granocyte; Chugai Pharma). Freshly isolated CD34⁺ HSCs were transfected (see below) and cultured on MSCs in HSC medium (serum-free medium [Cell Gro SCGM; CellGenix]) supplemented with early-acting cytokines (50 ng/mL stem cell factor, 50 ng/mL fms-related tyrosine kinase-3 ligand [CellGenix] and 15 ng/mL interleukin-3 [R&D Systems]) at a density of 15 x 10⁴/cm² for 2 days in a humidified 5% CO₂ atmosphere at 37°C.

4.5 Plasmid and siRNA transfections

MDCK and CHO cells were stably transfected with expression plasmids described above by electroporation using Amaxa Nucleofector Kit L (MDCK cells) or T (CHO cells) according to the manufacturer's instructions (#VCA-1005, VCA-1002; Lonza). As control, cells were either transfected with the expression plasmid only or with one that contained a GFP reporter gene (both referred to as MOCK).

Cells expressing the neomycin resistance gene were selected by introducing 500 μg/mL of Geneticin (G418 Sulfate; Thermo Fisher Scientific) into the culture medium. After selection, cells were further expanded in a medium containing 250 μg/mL of the selective agent. Note that the microvillar morphology did not change when cells were cultured in medium with 0.5 or 1 mM geneticin or without (data not shown). These data rule out a direct effect of the drug on polyphosphoinositides.101, 102

To enrich MDCK and CHO cell populations expressing human prominin-1, cells were sorted using paramagnetic labeling with CD133 MicroBeads (CD133 MicroBead Kit, #130-097-049; Miltenyi Biotec) and LS columns. To enhance the expression of the transgene 10 mM sodium butyrate was added to the culture medium for 16 hours except for SEM and TEM analysis.44 If not stated otherwise, cells were used at 6 dpc. To investigate the influence of the PI3K, MDCK cells were incubated with 5 μM LY294002 (Sigma-Aldrich) or DMSO (1:2000) as negative control for 4 hours prior to their analysis. To investigate the influence of the Arp2/3 complex, cells were incubated with 200 μM CK-666 or its inactive analogue CK-689 (Merck) for 16 hours. Again, DMSO (1:500) was used as control. For combined inhibition of PI3K and Arp2/3 complex, cells were first incubated with CK-666 or CK-689 (200 μM) for 16 hours and, thereafter, with CK-666/CK-689 and LY294002 (5 μM) for additional 4 hours.

Freshly isolated CD34⁺ HSCs were transfected with siRNA (1 μg) by electroporation using the human CD34 Cell Nucleofector Kit (VPA-1003; Amaxa). Prominin-1 (CD133, ID.HSS113055), CD82 (ID.HSS105654) and negative control (lo GC Duplex; ID.12935-200) Stealth siRNA oligonucleotides were purchased from Invitrogen. After the transfection, cells were washed in pre-warmed medium, spun down, and the cell pellets were incubated for 15 minutes at 37°C in a 5% CO₂ humidified atmosphere.51 Transfected HSCs were then distributed onto the MSC monolayer (see above). Under these conditions, the knockdown of targeted proteins was more than 75% as observed by immunoblotting and/or immunofluorescence50 (data not shown).

4.6 Flow cytometry

MDCK cells were harvested with 0.05% trypsin/0.5 mM EDTA solution (Life Technologies) for 15 minutes at 37°C, centrifuged, and the pellet was resuspended in PBS containing 0.2% gelatin and 5 mM EDTA (VWR International). Cell suspensions (2 x 10⁵ cells in 100 μL) were incubated with or without phycoerythrin (PE)-conjugated mAb against human prominin-1 (1:10, clone AC133; Miltenyi Biotec) or human CD34 (1:10, clone 581; BD Bioscience) for 30 minutes at 4°C. Cells were washed with PBS containing 5 mM EDTA. After centrifugation, cells were resuspended and incubated for 10 minutes in 4% paraformaldehyde (PFA). After washing with PBS containing 5 mM EDTA, 20 000 events were acquired on a FACSCanto II System (BD Bioscience). Instrument settings and gating strategies were established using unstained or CD34-labeled cells.

4.7 Differential centrifugation of prominin-1-containing membrane vesicles

Cells were grown in six-well dishes for 6 dpc. After washing with PBS they were supplied with fresh culture medium (2 mL) and incubated for 4 or 16 hours. Afterward, conditioned media were collected, supplemented with Complete protease inhibitor cocktail (Roche Molecular Biochemicals) and subjected to differential centrifugation as follows: 5 minutes at 300g (pellet 1 [P1]); 20 minutes at 1200g (P2); 30 minutes at 10 000g (P3) and 1 hour at 200 000g (P4). Pellets were resuspended in Laemmli buffer. In parallel, adherent cells were harvested and centrifuged for 5 minutes at 300g. The resulting pellets were lysed for 30 minutes at 4°C in solubilization buffer A (1% NP-40, 0.5% deoxycholate, 0.1% sodium dodecylsulfate, 150 mM NaCl, 50 mM Tris-HCl, pH 8.0) supplemented with complete protease inhibitor. Detergent extracts were centrifuged at 4°C (10 minutes, 10 000g) and resulting supernatants were mixed with Laemmli buffer. All samples were analyzed by immunoblotting for prominin-1 and β-actin (see below). For quantification, we combined prominin-1 immunoreactivity associated with cell extract, P1 and P2 fractions and designated them as Cell fraction while materials found in P3 and P4 were referred to as Vesicle fraction.

4.8 Immunoisolation of prominin-1

Cells were grown in T75-flasks for 6 dpc and then solubilized for 30 minutes on ice in solubilization buffer B (0.2% Triton X-100, 150 mM NaCl, 20 mM Tris/HCl, pH 7.5) containing EDTA-free Complete protease inhibitor cocktail (Roche Diagnostics GmbH), 2 mM sodium orthovanadate, 10 mM sodium fluoride (both from New England Biolabs Inc.) and 10 mM β-glycerophosphate (Sigma-Aldrich). After centrifugation (10 minutes, 16 000g, 4°C) for removal of insoluble materials, detergent lysates were divided into two fractions that were incubated for 2 hours at 4°C with 140 μL of either CD133 mAb-conjugated to magnetic beads or anti-fluorescein isothiocyanate (FITC) antibody-coupled magnetic beads as negative control, prior to magnetic separation using MS columns (Miltenyi Biotec). Flow-through materials were collected. Materials retained in columns were washed six times with 1 mL of ice-cold PBS containing 2 mM EDTA and 0.1% Triton X-100 and then flushed with the same buffer. Flow-through and bound fractions were subjected to methanol/chloroform precipitation. Proteins were resuspended in Laemmli buffer and analyzed by immunoblotting for various antigens (see below).

4.9 Endoglycosidase digestion

Kidney lysates were prepared from adult wild-type or rd19 mice using solubilization buffer C (1% Triton X-100, 150 mM NaCl, 20 mM Tris/HCl, pH 7.5) containing Complete protease inhibitor cocktail (Roche Diagnostics GmbH). After 30 minutes on ice, samples were centrifuged (10 minutes, 16 000g, 4°C) to remove insoluble materials. Solubilized proteins (50 μg) were then incubated for 6 hours at 37°C in the absence or presence of either 10 mU endo H (endo-β-N-acetylglucosaminidase H) or 1 U PNGase F (peptide-N-glycosidase F) according to the manufacturer's instructions (Roche Diagnostics GmbH).

4.10 Immunoblotting

Proteins were separated by SDS-polyacrylamide gel (7.5% or 8%) electrophoresis and transferred to polyvinylidene fluoride membranes (pore size 0.45 μm; Millipore Corp.) using a semi-dry transfer cell system (Cti). After transfer, membranes were incubated overnight at 4°C in blocking buffer I (PBS containing 0.3% Tween 20 and 5% milk powder) prior to 1 hour incubation at room temperature (RT) with primary antibody against prominin-1 (1:2000; mouse mAb anti-CD133, clone 80B2584) diluted in blocking buffer I. As loading controls, mouse mAbs AC-74 against β-actin (1:10 000; Sigma-Aldrich) or 6-11B-1 against α-tubulin (1:1000; Sigma-Aldrich) were used. For the analysis of immunoisolated prominin-1, membranes were incubated in blocking buffer II (TBS containing 0.1% Tween 20 and 3% bovine serum albumin) either overnight at 4°C or 1 hour at RT prior to 1 hour incubation at RT with the primary antibodies against prominin-1 (see above), ubiquitin (1:000; rabbit polyclonal [p] Ab; Dako), phosphotyrosine (1:330; mouse mAb, clone PY99; Santa Cruz Biotechnology), p85 subunit of PI3K (1:1000; rabbit pAb; Millipore Corp.), Arp2 (1:250; rabbit pAb [H-84]; Santa Cruz Biotechnology), p34-Arc/ARPC2 subunit of Arp2/3 complex (1:1000; rabbit pAb; Millipore Corp.), syntenin-1 (1:5000; rabbit pAb; Abcam) and ezrin (1:5000; rabbit mAb, clone EP886Y; Abcam). Antibodies were diluted in blocking buffer II. In all cases, antigen-antibody complexes were detected using appropriate horseradish peroxidase-conjugated secondary antibodies (1:5000; Jackson ImmunoResearch Inc.) and visualized with enhanced chemiluminescence reagents (ECL system; GE Healthcare Life Sciences). Membranes were exposed to Amersham Hyperfilms ECL (GE Healthcare). Fiji was used for the quantification of immunoblotted materials.103 Note that none of the current available antibodies directed against human or mouse prominin-1 cross-reacted with canine and/or hamster orthologs.1, 7, 49

4.11 Immunofluorescence and confocal laser scanning microscopy

Cells were grown on fibronectin (BD Biosciences)-coated glass coverslips or in eight-well chambers (μ-Slide 8 Well, #80826; Ibidi) for 6 to 7 dpc. They were incubated with 10 mM sodium butyrate 16 hours before their analysis. For the cell surface labeling, cells were rinsed with Ca/Mg buffer (phosphate-buffered saline containing 1 mM CaCl₂ and 0.5 mM MgCl₂) and incubated in blocking buffer III (Ca/Mg buffer containing 0.2% gelatin) for 30 minutes at 4°C. Cells were incubated with anti-human prominin-1 mouse mAb CD133/1 (1:50; clone AC133, #130-090-422; Miltenyi Biotec) alone or in combination with Alexa Fluor 488-conjugated cholera toxin subunit B (1:1000; Invitrogen) diluted in blocking buffer III for 1 hour at 4°C. Alternatively, cells were incubated after prominin-1 immunolabeling with FITC- or rhodamine-conjugated WGA (10 μg/mL; Vector Laboratories) for 30 minutes at 4°C. Cells were washed with PBS and fixed in 4% PFA in PBS for 30 minutes at RT. They were then rinsed and quenched for 10 minutes in PBS containing 50 mM NH₄Cl prior to the incubation with the appropriate Alexa Fluor 488/546/555-conjugated goat anti-mouse IgG₁ (1:500; Molecular Probes, Life Technologies). For the intracellular labeling, PFA-fixed cells were permeabilized with 0.1% saponin diluted in PBS containing 0.2% gelatin (blocking buffer IV)104 for 20 minutes at RT prior to immunolabeling for prominin-1 (see above) and p34-Arc/ARPC2 subunit of Arp2/3 complex using rabbit antiserum (5 μg/mL; Merck). Primary antibodies were observed using Alexa Fluor 488-conjugated goat anti-mouse IgG₁ and Alexa Fluor 555-conjugated donkey anti-rabbit IgG. All antibodies were diluted in blocking buffer IV. In some experiments, nuclei were counterstained with 4′-6-diamidino-2-phenylindole (DAPI, 1 μg/mL; Sigma-Aldrich). All samples were then washed three times in PBS containing 0.2% gelatin, two times in PBS and H₂O and mounted in Mowiol 4.88 (Merck). Cells seeded in eight-well chambers were not mounted, but directly observed at RT with a confocal Zeiss LSM 700 using ZEN software and a 100× 1.4 oil DIC Plan-Apochromat objective, 4× zoom (Figure 2C and Figure S2B) or a 63× 1.4 oil DIC Plan-Apochromat objective (Figure 10A). All figures were processed with Fiji.103

For the length measurement of filopodia emerging from CHO cells, CLSM micrographs of prominin-1 and/or WGA staining were used. Only cells with spread morphology, and not those of a rounded shape, were taken into consideration. Note that we did not measure the absolute number of their length since they were often tilted. Instead we measured their length from the tip to the planar cell surface with an angle of 90° (see cartoon in Figure 11A).

4.12 Transmission electron microscopy

Culture cells growing on 30-mm dishes were pre-fixed by the addition of 0.1% glutaraldehyde into the cell culture medium and subsequently fixed in 1% PFA, 1% glutaraldehyde in 0.1 M PIPES, pH 7.4 at RT. After washing steps, cells were post-fixed in 1% aqueous osmium tetroxide for 1 hour at RT. After washing with water, cells were incubated with 0.1% tannic acid in water for 10 minutes and contrasted with 1% aqueous uranyl acetate for 30 minutes at RT. The samples were processed through a graded series of ethanol for standard plastic embedding in LX 112 resin (LADD Research Industries). Sheets of resin embedded cells were removed from the dishes by liquid nitrogen. Transverse sections were cut at 70 nm on an UCT ultramicrotome (Leica) and post-stained with uranyl acetate and lead citrate. The samples were viewed in a Morgagni electron microscope (Thermo Fisher Scientific) and images acquired on a Morada CCD camera (EMSIS GmbH).

Adult mice, 4 to 9 months of age, were fixed by perfusion with 2% PFA, 2% glutaraldehyde in 0.12 M phosphate buffer at RT. Tissues were dissected and fixed in the same buffer overnight. After washing steps, the tissue was further dissected into small pieces, which were post-fixed in 1% aqueous osmium tetroxide, 1.5% potassium ferrocyanide for 1 hour at RT. Subsequent steps were performed as described above. Note that Epon replacement (Carl Roth) was used as of resin and embedding was performed in silicon molds.

4.13 Scanning electron microscopy

Samples for SEM analysis were prepared as described.19 Briefly, MDCK cells were grown on fibronectin-coated coverslips for the indicated periods and then fixed in 2% glutaraldehyde overnight at 4°C. The HSC-MSC co-cultures were fixed in the same way. After dehydration in an acetone gradient (25%-100%) all samples were critical point dried in a CO₂ system (Critical Point Dryer CPD 300; Leica). Samples were mounted on aluminum stubs, coated with gold (≈20 nm thickness) in a sputter coater (SCD 050; BAL-TEC GmbH) and examined with a SEM (JSM-7500F; JEOL). SEM micrographs were used for the quantification of microvilli of MDCK cells and HSCs.

4.14 Statistical analysis

Statistical analyses were performed using Graph Pad Prism version 6.00 for Mac (Graph Pad Software; www.graphpad.com). Two-tailed unpaired Student's t-test was used for the analysis of the immunoreactivity ratio of prominin-1/β-actin or other antigens, the number of microvilli or microvillar clusters and the prominin-1 immunoreactivity associated with membrane vesicles. The length of filopodia was also compared with this test. Mann-Whitney test was used for the analysis of the number of microvilli per cluster. Two-tailed Fisher's exact test was applied for the evaluation of microvillar phenotypes. The number of events is indicated in the figures and their legends. Observed differences were regarded as significant if the calculated P-values were ≤0.05. *P < 0.05; **P < 0.01; ***P < 0.005.

Author contributions

K.T., D.S., J.K., W.B.H., M.W.-B. and D.C. conceptualized and designed the project. K.T., D.S., J.K., V.B., D.R., A.M. and M.W.-B. conducted the experiments and analyzed the data. D.C. and W.B.H. analyzed the data. K.T. and D.C. prepared the manuscript. K.T., J.K., M.B., W.B.H., M.W.-B. and D.C. finalized the manuscript. All authors read and approved the final manuscript.