An RNAi screen identifies KIF15 as a novel regulator of the endocytic trafficking of integrin

ABSTRACT &agr;2&bgr;1 integrin is one of the most important collagen-binding receptors, and it has been implicated in numerous thrombotic and immune diseases. &agr;2&bgr;1 integrin is a potent tumour suppressor, and its downregulation is associated with increased metastasis and poor prognosis in breast cancer. Currently, very little is known about the mechanism that regulates the cell-surface expression and trafficking of &agr;2&bgr;1 integrin. Here, using a quantitative fluorescence-microscopy-based RNAi assay, we investigated the impact of 386 cytoskeleton-associated or -regulatory genes on &agr;2 integrin endocytosis and found that 122 of these affected the intracellular accumulation of &agr;2 integrin. Of these, 83 were found to be putative regulators of &agr;2 integrin trafficking and/or expression, with no observed effect on the internalization of epidermal growth factor (EGF) or transferrin. Further interrogation and validation of the siRNA screen revealed a role for KIF15, a microtubule-based molecular motor, as a significant inhibitor of the endocytic trafficking of &agr;2 integrin. Our data suggest a novel role for KIF15 in mediating plasma membrane localization of the alternative clathrin adaptor Dab2, thus impinging on pathways that regulate &agr;2 integrin internalization.


INTRODUCTION
Integrins are one of the major cell-surface adhesion receptors, and they are involved in numerous functions, such as cell migration, signal transduction, proliferation, survival and differentiation. Currently, 18 a integrin and 8 b integrin chains are known to be expressed in human cells, forming 24 different heterocomplexes that are specific for diverse ligands (Takada et al., 2007;Humphries et al., 2006;Caswell et al., 2009). The availability and functionality of integrins on the cell surface strongly depends on their biosynthetic trafficking, internalization, recycling and degradation (Margadant et al., 2011).
b1 integrin is frequently used to investigate integrin trafficking; however, it forms heterodimers with 12 different a chains (Hynes, 2002). As different heterodimers exhibit different trafficking kinetics  and might have different molecular requirements (Teckchandani et al., 2009), the analysis of the selected a chain might be more useful to delineate specific molecular mechanisms. a2 integrin interacts only with the b1 integrin subunit (Hynes, 1992), and a2b1 heterodimer is expressed in fibroblasts, keratinocytes, leukocytes, platelets and epithelial and endothelial cells. As one of the most important receptors of collagen (Kramer and Marks, 1989;Hemler et al., 1990;Emsley et al., 2000), a2b1 is crucial for hemostasis, morphogenesis of tubular organs, angiogenesis, platelet aggregation, renewal of the epidermis and wound healing (Chen et al., 2002;Eble, 2005). It is also involved in numerous thrombotic and immune diseases (Eckes et al., 2000;Santoro, 1999;de Fougerolles et al., 2000). Importantly, the expression of a2b1 integrin is strongly reduced in breast cancer, and a2b1 downregulation correlates with increased cell invasion and metastasis and thus with poor prognosis (Zutter et al., 1998;Ramirez et al., 2011). a2b1 integrin also acts as a co-receptor for several pathogens, such as echovirus 1 (EV1), rotavirus and human cytomegalovirus (Bergelson et al., 1992;Londrigan et al., 2003;Feire et al., 2004;Fleming et al., 2011). Binding of EV1 induces clustering of inactive a2b1 and directs virus-bound a2 integrin to a calpain-dependent degradation pathway in multivesicular bodies (Rintanen et al., 2012;Karjalainen et al., 2008;Upla et al., 2004). Binding of a2b1 integrin to collagen might direct the heterodimer to the same degradation pathway (Rintanen et al., 2012). By contrast, little is known about the trafficking route of non-clustered a2b1 integrin, apart from a role for Rab21 binding to the a2 integrin cytoplasmic tail to facilitate endocytosis and a role for binding of the p120 RasGAP (GTPase activating protein; also known as RASA1), which facilitates the recycling of a2b1 integrin to the plasma membrane (Mai et al., 2011).
Here, we used an unbiased quantitative fluorescencemicroscopy-based RNAi screen to identify regulators of the endocytic trafficking of a2 integrin. We targeted the expression of 386 chosen proteins that are functionally related to membrane trafficking and the cellular cytoskeleton. Of these, 122 were demonstrated to affect the intracellular accumulation of a2 integrin, and half of these were novel regulators in the context of endocytosis. We chose to characterize in more detail the role of one of the strongest validated inhibitors of a2 integrin trafficking, the molecular motor KIF15. Our data support the novel suggestion that the internalization of a2 integrin is regulated through motor-dependent plasma membrane localization of the alternative clathrin adaptor Dab2.

Endocytic trafficking of endogenous a2 integrin
To define whether a2 integrin undergoes heterodimer-specific endocytic trafficking, we compared the endocytosis of a2 integrin in HeLa cells (chosen for their ease of transfection and high RNAi efficiency) with the previously characterized trafficking of b1 integrin. Cell-surface a2 integrin was antibody-labeled at 4˚C for 1 h and then internalized at 37˚C for varying lengths of time on plastic dishes. In order to quantify the amount of intracellular a2 integrin, the remaining cell-surface integrin-antibody complexes were removed by treating cells with an acidic buffer [see Methods, (Erfle et al., 2011)]. Internalized a2 integrin accumulated efficiently at the juxtanuclear area and cytoplasmic structures (Fig. 1A,D) after 1 h. a2 integrin localized largely to the scattered cytoplasmic structures after a prolonged incubation time of 6 h (Fig. 1A). There was a progressive reduction in the amount of cell-surface a2 integrin (Fig. 1B,E), similar to that measured for b1 integrin in HeLa cells (Teckchandani et al., 2009). By contrast, antibody-clustered a2 integrin (see Methods) accumulated in cytoplasmic punctate structures with hardly any residual plasma membrane localization (Fig. 1C), as reported previously (Rintanen et al., 2012;Karjalainen et al., 2008;Upla et al., 2004).
Previous RNAi experiments demonstrated that the endocytosis of b1 integrin occurs through the clathrin-dependent pathway in various cell types (Nishimura and Kaibuchi, 2007;Chao and Kunz, 2009) and by a caveolin-dependent pathway in myofibroblasts (Shi and Sottile, 2008). The latter pathway is also utilized by a2 integrin in ovarian carcinoma cells when the epidermal growth factor receptor is activated (Ning et al., 2007) or upon clustering of a2b1 integrin by antibodies (Upla et al., 2004), suggesting possible cell-type-specific mechanisms. To investigate whether both of these pathways play a role in the endocytic trafficking of a2 integrin in HeLa cells, we depleted the key components of clathrin-dependent and caveolae-dependent pathways [clathrin heavy chain (CLTC), caveolin-1 (CAV1) and dynamin-2 (DNM2)] (Hansen and Nichols, 2009;Doherty and McMahon, 2009). The efficiency of the respective siRNAs was tested by western blotting or immunofluorescence staining, which demonstrated a 50-80% reduction in the levels of the target proteins after 48 h of transfection (supplementary material Fig.  S1). Even an incomplete reduction in the expression of CLTC, CAV1 or DNM2 was sufficient to inhibit the intracellular accumulation of a2 integrin by 70-80% (Fig. 1F,G).

Potential regulators of a2 integrin intracellular accumulation identified by RNAi assays
To identify regulators of the intracellular accumulation of a2 integrin, we performed a fluorescence-microscopy-based RNAi screen with 1084 siRNAs targeting 386 genes (supplementary material Table S1). The targeted library (provided by R.P.) was designed to include most mammalian motor proteins, including 44 kinesin heavy chains, 13 dynein subunits (the three major heavy chains and ten cargo-binding light chains) and 37 myosins. In addition, proteins regulating membrane trafficking and/or cytoskeletal dynamics were included based on literature data.
This group included 28 small Ras and Rho GTPases, 53 GAPs and 62 GEFs (guanine nucleotide exchange factors) of these GTPases, 41 actin-associated and 14 microtubule-associated proteins, 34 kinases and 9 phosphatases, 5 different b integrin chains, and 23 scaffold and adaptor proteins. For the primary screen, we used Lab-Tek chambers that were pre-spotted with the siRNAs and Lipofectamine [reverse transfection on cell arrays (Erfle et al., 2007)]. These were prepared in one spotting experiment, ensuring a high efficiency and reproducibility of the whole screen (see Methods). We quantified the a2-integrinspecific intracellular fluorescence signal for individual cells after 1 h of internalization, the time-point at which the level of intracellular a2 integrin was the highest in the control sample ( Fig. 1D), thereby achieving the best signal-to-noise ratio during the screen. Inspection of the control samples over large cell populations revealed a large variation in the amount of internalized a2 integrin in HeLa cells. As quantification of intracellular a2 integrin in cells with low accumulation efficiency is unreliable, we focused on cells with efficient a2 integrin accumulation. We set a threshold for cells with high levels of intracellular a2 integrin (defined visually during imaging) and subsequently used the empirical cumulative distribution function to define the respective cell population for every experiment (supplementary material Fig. S2), and the Gaussian Mixture Model to automatically separate cell subpopulations (see Methods). Usually, high levels of intracellular a2 integrin were found in 40-50% of HeLa cells transfected with the negative control, and these were used for further analysis to yield highquality data.
Screen hits were defined according to individual z-scores of each siRNA normalized to that of the negative controls (see Methods). Positive z-scores indicate increased amounts and negative z-scores indicate reduced amounts of intracellular a2 integrin. We considered siRNAs with z-scores of greater than 1 or less than 21 (.|1|) as primary hits (supplementary material Table  S2). In total, 122 genes were assigned as hits in our screen, of which 115 inhibited and 7 accelerated the intracellular accumulation of a2 integrin (Fig. 3A). Next, we separated the primary hits into strong (z-score .|1.5|) and weak effectors (zscore ,|1.5|). Downregulation of CLTC and DNM2 produced zscores of 21.5, corresponding to a strong inhibition of a2 integrin traffic on cell arrays (Fig. 3B). A total of 26 other genes were classified as strong inhibitors, some showing a larger effect that that caused by CLTC and DNM2 downregulation (supplementary material Table S2). For instance, the downregulation of syndecan-4, a protein implicated in the caveolae-dependent internalization of integrins (Bass et al., 2011), led to a strong inhibition (z-score521.78) of the intracellular accumulation of a2 integrin. The majority of the inhibitors were characterized as weak effectors (90 out of 115). The depletion of CAV1 for 48 h showed a reproducible but weak inhibitory effect (z-score521.14). In part, this might reflect the level of protein knockdown achieved during the relatively short incubation period with siRNAs (supplementary material Fig. S1). Seven primary hits of the screen were called 'accelerators' (supplementary material Table S2). One of these is ASAP1, a GAP of ARF6 (Randazzo et al., 2007) that has been previously shown to regulate integrin recycling (Onodera et al., 2012). It was scored as a strong accelerator (z-score52.26); this effect was caused by the enhanced accumulation of intracellular a2 integrin, likely due to inhibited recycling.
With the exception of kinases and phosphatases, which were over-represented (19%) among strong effectors in comparison with their representation in the screened library (11%) (supplementary material Tables S1, S2), no significant representation of protein functional class (assigned based on Gene Ontology terms) was observed. While comparing our primary hits to a genome-wide screen for regulators of the endocytosis of transferrin and epidermal growth factor (EGF) (Collinet et al., 2010), we found that 39 out of 122 of our hits (32%) were reported as regulators of either transferrin or EGF endocytosis, and 83 primary hits from our screen appeared to be specific only to the intracellular accumulation of a2 integrin (Fig. 3C).
As endocytic trafficking of integrins is fundamental for the regulation of focal adhesion dynamics and cell migration (Morgan et al., 2013;Chao and Kunz, 2009;Caswell and Norman, 2008), we compared the results of our screen with those of two relevant published RNAi screens (Winograd-Katz et al., 2009;Simpson et al., 2008) (Fig. 3C). In addition, 161 genes from our experiments were also screened to identify regulators of focal adhesions and cell migration (Winograd-Katz et al., 2009;Simpson et al., 2008). We identified 24 out of 161 genes (15%) that were primary hits for both a2 integrin intracellular accumulation and focal adhesion formation, including the phosphatidylinositol phosphatase PTEN (Mise-Omata et al., 2005). We found that 9 out of 161 genes (5.5%) were positive both in our assay and in a previously published cell migration assay (Simpson et al., 2008).
We observed a fairly large overlap between our primary hits and the components of integrin-primed complexes. Our screen included 45 out of 467 identified non-redundant components of a5b1 and a4b1 integrins ) and, of these, 19 were assigned as potential regulators of the intracellular accumulation of a2 integrin [e.g. the molecular motors KIF2C and cytoplasmic dynein 1 heavy chain 1 (DYNC1H1)]. A recent search for b1-integrin-interacting proteins (Böttcher et al., 2012) Endogenous a2 integrin (red) was colocalized with the overexpressed GFPtagged Rab5a (A), mCherry-tagged Rab11b (B), GFP-tagged Rab4a (C) and GFP-tagged Rab7a (D) (green) at 1 h after the internalization. The lower row of images shows a higher magnification view of the region indicated by the white box in the upper images. Thick arrows, colocalizing structures; arrowheads, non-colocalizing structures in overlay images; thin arrows, perinuclear localization of the overexpressed Rabs. The figures in the lower right image of each panel show the percentage of a2 integrin that colocalizes with the respective Rab (mean6s.d.). Scale bars: 10 mm (upper row of images), 5 mm (lower row of images).
yielded 96 proteins, of which 11 were tested in our study, and four of these were assigned as primary hits (supplementary material Table S2). These included the integrin-activating protein kindlin-1 (also known as FERMT1) (Moser et al., 2009), LIMS1 [an adaptor protein that connects integrins, small GTPases and EGF signaling pathways (Tu et al., 1999)] and RALBP1 [a GAP of Rac1 and Cdc42 (Matsubara et al., 1997)]. Our screen revealed a number of potential novel regulators of a2 integrin trafficking; for example, KIF18A, a key component of chromosome congression in mitosis (Mayr et al., 2007), strongly inhibited the intracellular accumulation of a2 integrin when downregulated (Fig. 3B).

Validation of potential regulators of a2 integrin intracellular accumulation
To validate the potential regulators of a2 integrin trafficking that were identified in the siRNA screen described above, we selected a fifth of the primary hits (23 proteins). One of the selected groups included the kinesins, as very little is known about the role of these molecular motors in integrin trafficking, and essentially all kinesins were tested in the primary screen (supplementary material Table S1) (Hirokawa et al., 2009). Validation assays were performed in non-coated multi-well plates under direct transfection conditions. Intracellular accumulation was considered as inhibited when the pool of intracellular a2 integrin was reduced by .40% compared with that of the negative control. We scored potential hits as validated when at least two out of four siRNAs showed the same effect as in the primary screening (supplementary material Table S4). From the nine primary hit kinesins that are expressed in HeLa cells (supplementary material Table S3), three, namely KIF15, KIF18A and KIF23 were validated as effectors of a2 integrin intracellular accumulation (Table 1). As kinesins are important trafficking regulators, we extended our investigation beyond the nine kinesins identified as primary hits in our screen. We retested 13 more traffic-related kinesins that were non-hits in the primary screen (supplementary material Tables S3, S4) and demonstrated that only KIF13A, a recycling regulator in melanocytes (Delevoye et al., 2009), affected the intracellular accumulation of a2 integrin in the validation experiments. As this result was obtained only with the additional siRNAs, but not with the ones used for the primary screen, we did not include KIF13A in the list of validated hits.
Next, we tested whether decreased levels of intracellular a2 integrin are a direct consequence of inhibited trafficking or reduced protein expression. To do this, we measured the total amounts of a2 integrin following downregulation of the validated kinesins (Table 1). Expression levels of a2 integrin were not affected when KIF15, KIF18A and KIF23 were downregulated (supplementary material Fig. S3). As these kinesins were shown to play a role in mitosis (Mayr et al., 2007;Tanenbaum et al., 2009;Zhu et al., 2005b), we calculated the number of cells remaining after 48 h of transfection with the respective siRNAs. Downregulation of KIF23 expression reduced the number of cells by nearly twofold, whereas downregulation of KIF15 and KIF18A did not induce a significant change in the number of cells (Table 1).
To get a better overview of the specificity of our screen we retested 14 additional primary hits belonging to diverse functional groups. To do this, we chose the molecules that were scored as effectors with multiple siRNAs in the primary screen (supplementary material Table S2). RNAi-mediated inhibition of seven of these proteins strongly reduced levels of intracellular a2 integrin in the validation assays (Table 1; supplementary  material Table S4). Five of these were also classified as strong inhibitors in the primary screen (ARF1, ABL2, ARHGAP6, Myo1A and PTPN11). Suppressing the expression of two of these proteins (ARF1 and NF2) did not alter the expression levels of a2 integrin, suggesting a role for these proteins in a2 integrin trafficking. ITGB1 and Myo1A reduced a2 integrin expression when downregulated; however, these proteins might play a role in a2 integrin trafficking, as inhibition of the intracellular accumulation of a2 integrin was more pronounced than the effects seen on a2 integrin expression (Table 1). By contrast, the observed effect on intracellular a2 integrin levels upon ABL2 and ARHGAP6 downregulation was largely due to the reduction in total a2 integrin expression. Downregulation of only PTPN1 significantly reduced cell numbers (by nearly 35%).
In summary, we could recapitulate the results of the primary screen for 10 out of the 23 hits (43%). For nearly half of the targets this effect could be attributed to changes in the accumulation of intracellular a2 integrin, whereas the other hits primarily affected the expression level of a2 integrin.

Internalization of a2 integrin is KIF15 dependent
To gain more detailed insight regarding what appeared to be novel regulators revealed by our targeted screen, we focused on KIF15, as no trafficking-related function has been described for KIF15 so far. RNAi of KIF15 for 48 h resulted in a 70% reduction in KIF15 protein levels (supplementary material Fig.  S4A,C) and led to a strong inhibition of the intracellular accumulation of a2 integrin without a significant change in a2 integrin expression level and cell numbers (Table 1;  induced an increase in the amount of intracellular a2 integrin by ,40% (Fig. 4A,B), with a2 integrin accumulating at the perinuclear region after 1 h of internalization. Overexpression of GFP alone under these conditions had no apparent influence. GFP-tagged KIF15 mostly localized to the cytoplasm (Fig. 4A), with occasional localization to the plasma membrane, punctate cytoplasmic structures and microtubules. As GFP-tagged murine KIF15 is resistant to human siRNAs, we performed rescue experiments, which showed that the siRNA-mediated inhibitory effect of KIF15 on the intracellular accumulation of a2 integrin was lessened by 30% (Fig. 4B) following overexpression of GFPtagged KIF15. The role of KIF15 in a2 integrin endocytic trafficking was further confirmed by using biotin-capture enzyme-linked immunosorbent assays (ELISAs) (Margadant et al., 2012;Roberts et al., 2001) (Fig. 4C). In addition, endocytosis of b1 integrin was also inhibited, as measured by a ratiometric quenching antibody-based assay (Arjonen et al., 2012) (Fig. 4D). Both assays revealed a lower inhibition rate of integrin trafficking (around 25%) when KIF15 was downregulated as compared with the microscopy-based assay (Table 1). This might be attributed to the difference in quantifying the whole cell population in these assays versus a defined subpopulation of cells for the microscopy-based assay (see Methods). No significant changes in actin and microtubule cytoskeleton were observed following KIF15 downregulation (Fig. 4E,F). Taken together, depletion of KIF15 inhibited the intracellular accumulation of a2 integrin in different types of cells (supplementary material Fig.  S3D).
Next, we analyzed at which stage the intracellular accumulation of a2 integrin was affected when KIF15 was depleted. In cells transfected with the negative control, a2 integrin gradually accumulated within the cells, whereas downregulation of KIF15 strongly inhibited the intracellular accumulation of a2 integrin throughout the assay (5-60 min) (Fig. 5A,D). We then tested the effect of KIF15 downregulation on the trafficking of EGF that is degraded in lysosomes (Carpenter and Cohen, 1976) and transferrin that is recycled to the plasma membrane (Dickson et al., 1983). In agreement with previous results (Collinet et al., 2010), downregulation of KIF15 had no influence on the internalization of either EGF or transferrin cargoes at early time-points (Fig. 5B,C,E,F). However, later steps of transferrin trafficking were significantly affected by KIF15 downregulation (Fig. 5E), with the cargo accumulating in the perinuclear region (Fig. 5B). Presumably, the observed effect is due to inhibited recycling of transferrin to the plasma membrane. By contrast, EGF endocytic trafficking was not influenced by KIF15 downregulation at any time-point (Fig. 5C,F).

KIF15 is required for cell-surface localization of Dab2
We sought to further investigate the integrin-related trafficking function of KIF15. It has been shown that several CLASPs (alternative clathrin adaptors) are required for integrin internalization (Nishimura and Kaibuchi, 2007;Teckchandani et al., 2009). By contrast, internalization of the transferrin receptor is not dependent upon CLASPs such as Dab2 (Teckchandani et al., 2009;Teckchandani et al., 2012;Maurer and Cooper, 2006;Keyel et al., 2006). In our experiments, RNAi of Dab2, ARH and Numb for 48 h caused .80% reduction in the level of the respective transcripts (Fig. 6D), and the same conditions were used to assess the potential role of these CLASPs in a2 integrin trafficking. The strongest inhibition of a2 integrin endocytosis was obtained by depletion of Dab2 ( Fig. 6A-C). RNAi of ARH had a weaker effect, whereas suppression of Numb showed no significant inhibition. Our data agree well with previous studies (Chao and Kunz, 2009;Teckchandani et al., 2012) showing that Dab2 strongly inhibits the internalization of b1 integrin when downregulated. Furthermore, surface levels of integrins b1 and a2 increase in Dab2-deficient HeLa cells (Teckchandani et al., 2009). We next tested whether Dab2 is functionally dependent upon KIF15. In agreement with previous studies (Teckchandani et al., 2009;Chetrit et al., 2011), Dab2 localized to punctate structures that were found both on the plasma membrane and intracellularly in HeLa cells imaged in total internal reflection fluorescence (TIRF) and epifluorescence modes (Fig. 7A). Downregulation of KIF15 induced a .40% loss of the cell-surface-associated Dab2 (Fig. 7C,D), with a large fraction of Dab2-specific structures being redistributed from the plasma membrane to the cytoplasm (Fig. 7B). The total level of Dab2 in cells was not changed significantly under these conditions ( Fig. 7B; supplementary material Fig. S4F). The role of KIF15 in Dab2 distribution was further demonstrated by a 2.7-fold more frequent accumulation of intracellular Dab2 in the perinuclear region following the depletion of KIF15 (Fig. 7D), indicating that KIF15 might play a role in recycling pathways. Surprisingly, no significant colocalization between Dab2 and a2 integrin was obtained at 37˚C, regardless of KIF15 expression levels. It is possible that trapping of integrin at low temperatures is required, as demonstrated for b1 integrin (Teckchandani et al., 2009). Taken together, our data suggest that KIF15 mediates the cellsurface localization of Dab2 that, in turn, might regulate the internalization of a2 integrin.

DISCUSSION
Defective endocytosis of integrins leads to numerous alterations in signaling, cell cycle progression, cell-extracellular-matrix interactions and the cytoskeleton, crucial processes of cancer initiation, progression and metastasis (Mosesson et al., 2008;Shin et al., 2012;Desgrosellier and Cheresh, 2010). Integrin a2b1 is known as a potent suppressor of breast cancer metastasis (Ramirez et al., 2011); however, little is known about the endocytic trafficking of this heterodimer. We show here that endogenous a2 integrin in HeLa cells passes through a Rab5specific compartment and is potentially recycled to the plasma membrane by Rab4-and Rab11-specific structures (Fig. 2). Interestingly, depletion of not only clathrin but also caveolin-1 induced a comparably strong inhibition of the intracellular accumulation of a2 integrin (Fig. 1F,G). Both pathways might act in parallel; however, RNAi-based perturbation of individual pathways could invoke broad cellular responses (Doherty and McMahon, 2009). Alternatively, our data opens an intriguing possibility that clathrin-and caveolin-dependent mechanisms might act in a sequential manner. Nevertheless, the endocytic route of unbound a2 integrin differs strongly from the calpaindependent degradation pathway taken by the antibody-clustered a2 integrin (Rintanen et al., 2012). With the goal of defining a substantial proportion of the proteins affecting the cell-surface expression of a2 integrin, we performed a fluorescent microscopy-based RNAi screen (Fig. 3).
In total, 386 genes associated with membrane traffic and cytoskeletal organization, and encompassing most mammalian motor proteins, were tested (supplementary material Table S1), with 122 molecules scored as potential regulators of the intracellular accumulation of a2 integrin ( Fig. 3A; supplementary material Table S2). Predominantly, a reduction in a2 integrin intracellular accumulation was recorded (for 115 potential regulators), and only seven molecules increased the intracellular accumulation of a2 integrin when downregulated. In the subsequent validation experiments, some of the inhibitors were shown to reduce the overall expression of a2 integrin in cells; however, specific inhibitors of trafficking, such as KIF15, were also identified. We found that 83 primary hits of our screen might be specific to the intracellular accumulation of a2 integrin, as they were not identified as regulators of transferrin or EGF endocytosis [ Fig. 3C; (Collinet et al., 2010)]. Possibly, the different outputs of the screens reflect differences between data acquisition (wide field versus confocal microscopy), read-out (monoparametric versus multiparametric), data normalization procedures and testing different siRNA libraries. By contrast, both screens were performed in HeLa cells and, unsurprisingly, there was a fairly large overlap of our hits (39 out of 122) with the regulators of transferrin and EGF endocytosis [ Fig. 3C; (Collinet et al., 2010)]. Interestingly, nearly half of the interacting partners of b1 integrin (Böttcher et al., 2012) present in our library were also effectors of a2 integrin trafficking (supplementary material Table S2), but little overlap could be found with the proteomics analysis of interacting partners of the overexpressed and tagged a2 integrin in HT1080 cells (Uematsu et al., 2012). Potentially, this is due to a different cellular context and/or expression levels of integrin. Surprisingly, 24 out of 122 potential regulators of a2 integrin trafficking are known as potential regulators of focal adhesion formation [ Fig. 3C; (Winograd-Katz et al., 2009)], indicating a functional crosstalk between the trafficking of unbound and ligand-bound integrins.
For data validation, we chose several subsets of the primary hits. One subset contained 14 functionally heterogeneous molecules that were recorded as hits by multiple siRNAs in the primary screen (supplementary material Table S2) and seven of them were validated (Table 1). Curiously, only two of them -NF2 (also known as Merlin) and a small GTPase, ARF1predominantly influenced the intracellular accumulation of a2 integrin, whereas others appeared to influence primarily a2 integrin expression levels. ARF1 is a key component of the COPI coatomer complex and one of the most important trafficking regulators (Donaldson and Klausner, 1994;D'Souza-Schorey and Chavrier, 2006;Volpicelli-Daley et al., 2005). Potentially, ARF1 inhibits a2 integrin trafficking by regulating its recycling to the plasma membrane as was demonstrated for a5 integrin (Krndija et al., 2012); however, other functions of ARF1 cannot be excluded. NF2/Merlin might act indirectly by providing a linkage between membrane-associated proteins and the actin cytoskeleton, thus regulating membrane organization (Li et al., 2012;McClatchey and Giovannini, 2005). By contrast, the Drosophila ortholog of NF2/Merlin localizes to the endocytic compartment in S2 cells (McCartney and Fehon, 1996), and it was reported that NF2/Merlin binds to b1 integrin in Schwann cells (Obremski et al., 1998), suggesting a more direct role for these proteins in integrin traffic.
Another subset of potential regulators chosen for validation analysis were kinesins, microtubule-based molecular motors. Systematic analysis of kinesins has already been performed in the context of cell cycle regulation (Tanenbaum et al., 2009;Goshima and Vale, 2003;Neumann et al., 2010;Zhu et al., 2005a) and membrane trafficking (Collinet et al., 2010;Paul et al., 2011;Simpson et al., 2012), but information about their roles specifically in integrin trafficking is missing. We observed no effect on a2 integrin trafficking when downregulating 13 kinesins that had no effect in the primary screen (supplementary material Tables S2, S4), indicating high fidelity of our assays. Of the nine kinesins recorded as primary hits, three (KIF15, KIF18A and KIF23) strongly inhibited the endocytic trafficking of a2 integrin (Table 1). As expected, the validation rate of kinesins was lower than that of the regulators targeted with the multiple siRNAs. However, the validated ones are most likely actual trafficking regulators as none of them changed the expression level of a2 integrin. Indeed, KIF15 and KIF23 have been identified as potential regulators of the biosynthetic trafficking of ts-O45-G (a temperature-sensitive mutant of vesicular stomatitis virus) (Simpson et al., 2012).
In this study, we have focused on KIF15, one of the strongest validated effectors of intracellular accumulation of a2 integrin. Downregulation of KIF15 inhibited intracellular accumulation of a2 and b1 integrins, as shown by several complementing approaches (Table 1; Fig. 4C,D; Fig. 5A,D) without having any effect on cell numbers. The latter observation is in agreement with the notion that KIF15 can be replaced by Eg5 (also known as KIF11) (Tanenbaum et al., 2009;Florian and Mayer, 2011) in the organization of bipolar mitotic spindles (Sueishi et al., 2000;Vanneste et al., 2009;Sturgill and Ohi, 2013). In addition, KIF15 was not identified as a regulator of cell cycle progression in a genome-wide RNAi screen (Neumann et al., 2010). KIF15 is expressed in interphase and shows a remarkably different localization in various cell types -it is localized at centrosomes in HeLa cells (Sueishi et al., 2000), actin bundles in fibroblasts and microtubules in terminally post-mitotic neurons (Buster et al., 2003). GFP-tagged murine KIF15 localized largely to the cytoplasm in HeLa cells, with occasional localization to the plasma membrane, microtubules and punctate structures. A similar localization pattern was observed in lung A549 epithelial carcinoma cells and the normal human fibroblasts BJ-5ta. Furthermore, downregulation of KIF15 induced inhibition of a2 integrin intracellular accumulation in these cells (supplementary material Fig. S3D), indicating that KIF15 plays a role in integrin trafficking in cancer and normal cells. Altered expression levels of KIF15 did not induce apparent alterations in actin or microtubule cytoskeleton and centrosomes, ruling out the possibility that the intracellular accumulation of a2 integrin is altered owing to general cytoskeletal changes (Fig. 4E,F).
Overexpression of GFP-tagged murine KIF15, which is resistant to human siRNAs, partially rescued inhibition of a2 integrin intracellular accumulation caused by KIF15 depletion (Fig. 4A,B). In addition, increased cell-surface a2 integrin expression was measured following KIF15 downregulation (supplementary material Fig. S3C). These data somewhat contradict the directionality of this motor (Boleti et al., 1996). For instance, downregulation of the microtubule plus-enddirected motors KIF1C and KIF16B induced a reduction in the cell-surface localization of a5b1 integrin (Theisen et al., 2012) and transferrin recycling (Hoepfner et al., 2005), respectively. Downregulation of KIF15 induced accumulation of transferrin at the perinuclear region (Fig. 5B), suggesting that it might function in trafficking from the perinuclear recycling compartment. On the contrary, internalization of EGF and transferrin was not influenced by the depletion of KIF15 [ Fig. 5B,C,E,F; (Collinet et al., 2010)].
Taken together, our data indicate the possibility that KIF15 might be involved in the surface delivery of an entry factor required for a2 integrin internalization. Indeed, CLASPs are known to specifically regulate b1 integrin internalization (Nishimura and Kaibuchi, 2007;Chao and Kunz, 2009;Ezratty et al., 2005). In agreement with this, we showed that endocytic trafficking of non-clustered a2 integrin might require not only the clathrin-independent but also clathrin-dependent pathways in HeLa cells (Fig. 1F,G). It is known that the FXNPXY-signalsorting CLASPs (Keyel et al., 2006) Dab2, ARH, and Numb are required for b1 integrin internalization in a number of migrating cells, as well as for the disassembly of focal adhesions (Nishimura and Kaibuchi, 2007;Chao and Kunz, 2009). Our RNAi data using a microscopy-based a2 integrin endocytosis assay largely confirms these observations, with Dab2 being the strongest effector (Fig. 6). Similar to the endocytosis of LDLR (Keyel et al., 2006, Maurer and, downregulation of ARH also inhibited the intracellular accumulation of a2 integrin by 40% (Fig. 6B). By contrast, no effect of ARH depletion on the surface expression of b1 integrin was reported (Teckchandani et al., 2009), which might reflect the use of different methodologies. Depletion of Numb, the adaptor with a preference for a5 integrin (Teckchandani et al., 2009), resulted in no significant change in the intracellular accumulation of a2 integrin. As Dab2 is known to inhibit the internalization of unbound b1 integrin (Teckchandani et al., 2009;Teckchandani et al., 2012) and appeared to be the strongest effector of the intracellular accumulation of non-clustered a2 integrin, we focused on Dab2 for further analysis. We could show that the downregulation of KIF15 induced redistribution of punctate Dab2 structures from the plasma membrane (Fig. 7A,C). These are most likely endocytic structures, as they were shown to colocalize with clathrin and AP-2 by .70% in HeLa, NIH3T3 and ES-2 cells (Chetrit et al., 2011;Teckchandani et al., 2009;Morris and Cooper, 2001). In addition, Dab-2-specific structures were shown to colocalize to unbound b1 integrin by .25% (Teckchandani et al., 2009). It is plausible, therefore, that their displacement from the plasma membrane inhibits the internalization of nonclustered a2 integrin. Dab2 is redistributed from the periphery and accumulates at the perinuclear compartment following downregulation of KIF15 (Fig. 7D), but does not colocalize with transferrin under these conditions. In agreement, ce-Dab-1, the Caenorhabditis elegans ortholog of Dab2, does not colocalize with Rab11-positive structures (Kamikura and Cooper, 2006). By contrast, the transferrin distribution might be affected by the mislocalized Dab2, which was shown to regulate the recycling of numerous proteins, including transferrin (Fu et al., 2012). As Dab2-mediated internalization of integrins occurs through the clathrin-dependent pathway (Teckchandani et al., 2012;Ezratty et al., 2009;Chao and Kunz, 2009), KIF15 might contribute to the regulation of the endocytic fate of a2 integrin. Whether KIF15 influences Dab2 distribution through direct binding, as demonstrated for the actin-based molecular motor myosin 6 (Spudich et al., 2007;Morris et al., 2002), or indirectly remains to be answered, particularly within the context of cell cycle progression, as Dab2 cell-surface localization and function in endocytosis change throughout the cell cycle (Chetrit et al., 2011). The involvement of KIF15 in clathrin-independent trafficking cannot be ruled out presently, as it was recently demonstrated that kinesins, like KIF13B, enhance cargo recruitment to caveolae (Kanai et al., 2014). The emerging role of KIF15 in the trafficking of a2 integrin, and possibly other cargoes, will be interesting to investigate in a disease context, as KIF15 is overexpressed in various tumours (Bidkhori et al., 2013).
Elucidating the molecular mechanisms influencing the endocytic trafficking of a2b1 integrin, one of the most important collagen receptors, is emerging as an exciting research area. This knowledge is essential for the prevention and treatment of diseases associated with integrin-mediated adhesion. Here, we present a microscopy-based assay that is suitable for high-throughput applications and is helpful in the systematic identification of novel regulators of the intracellular accumulation and/or expression of a2 integrin.

Transfection of siRNAs and cDNAs
For the reverse transfection on cell arrays, 5 ml of 30 mM siRNA was mixed with 3.5 ml of Lipofectamine 2000 (Invitrogen) and 3 ml of Opti-MEM (Invitrogen) containing 0.4 M sucrose and incubated for 20 min at room temperature. Thereafter, it was mixed with 7.25 ml of 0.2% gelatine (w/v) in 0.01% fibronectin (v/v), and used for contact printing with the automated liquid handling robot 'Microlab Star' (Hamilton) on one-chamber Lab-Tek slides (Nunc) using eight solid pins (PTS 600), giving a spot size of ,400 mm diameter. The spot-to-spot distance was set to 1125 mm; thus, a single Lab-Tek chamber fitted 384 spots organized in 12 columns and 32 rows. The whole library of 1084 siRNAs was spotted onto four Lab-Tek chambers, with 6-12 negative control siRNAs distributed randomly across each layout. 5610 5 cells were seeded per Lab-Tek chamber. Liquid-phase direct transfection with siRNAs or cDNAs in eight-well m-slides (Ibidi) was performed using Lipofectamine 2000 according to the manufacturer's protocol with 2610 4 cells seeded per well. Transfection of cells with siRNA and cDNAs was performed 48 h and 24 h before the assay, respectively.
Western blotting 1.5610 5 HeLa cells were plated in a 12-well plate, transfected with the respective siRNAs and cDNAs, lysed in 80 ml of hot Laemmli buffer, supplemented with 100 mM DTT and separated by 8% SDS-PAGE. The proteins were transferred to polyvinylidene difluoride (PVDF) membrane (Immobilion-P, Merck-Millipore), unspecific proteins were blocked with 5% milk in PBS-Tween, and incubation with the anti-KIF15 and anti-Dab2 antibodies was performed for 1 h at 4˚C or RT, respectively. The proteins of interest were detected by using the ECL system (GE Healthcare). The luminescence was recorded by using the chemiluminescence detection system (Intas) and quantified by using the ImageJ (NIH) 'Analyze Gel' option. The intensity of the relevant band of the blot was expressed as the area of the peak after subtraction of the average background level of the whole blot.

Quantitative RT-PCR and RT-PCR
Preparation of total cellular RNA was performed using TRIzol reagent (Invitrogen) according to the manufacturer's protocol. cDNA was prepared with MMLV reverse transcriptase (Ambion) and oligo-dT primer. Power SYBR Green PCR Mastermix was used for the reaction, which was performed using a StepOnePlus Real-Time PCR system (Applied Biosystems). Relative Dab2, ARH and Numb expression after RNAi was calculated by the 2 DDCT method using the expression level of GAPDH as a reference for the quantification. Expression of genes of interest in HeLa cells was tested by RT-PCR with up to three sets of primers that were determined by nucleotide BLAST analysis against the human genome and transcriptome. Primers were either designed using Primer BLAST or used as published previously (Jaulin et al., 2007).
Fluorescence-microscopy-based a2 integrin measurements Cells were serum starved for 14 h, overlaid with 10 mg/ml anti-a2integrin antibody in MEM-BSA for 1 h on ice, then briefly washed twice with ice-cold MEM-BSA to remove excess antibody and incubated with pre-warmed MEM-BSA at 37˚C to internalize non-clustered a2 integrin. After the internalization, cells were rinsed with PBS, surface-bound antibody was removed with acetic buffer [0.5% (v/v) acetic acid, 0.5 M NaCl, pH 2.6] for 30-40 s, and fixed with 2% paraformaldehyde (PFA) for 20 min at room temperature. Cells were permeabilized with 0.2% (w/ v) saponin in 10% (v/v) FCS in PBS and incubated with the secondary antibodies for 1 h at room temperature. When measuring total a2 integrin levels, treatment with the acetic buffer was omitted, cells were permeabilized with 0.2% (w/v) saponin in 10% (v/v) FCS in PBS and stained with the primary and secondary antibodies. a2 integrin on the cell surface was measured by fixing cells with 2% PFA for 20 min at room temperature, quenching with 30 mM glycine for 5 min at room temperature and staining with the primary and secondary antibodies. Clustering by antibody was induced as described previously (Upla et al., 2004) by sequential incubation of cells with the primary anti-a2-integrin and then with secondary antibodies on ice for 1 h each. Then cells were briefly washed twice with ice-cold MEM-BSA and incubated with the pre-warmed MEM-BSA at 37˚C to internalize antibody-clustered a2 integrin. Nuclei were stained with 0.1 mg/ml Hoechst 33342 dye. In order to estimate the efficiency of acid stripping, we compared the average vales of total a2 integrin specific fluorescence, surface-localized a2 integrin and intracellular non-clustered a2 integrin following 60 min of internalization using the same parameters for imaging and image analysis. The highest average fluorescence intensity was obtained for total a2 integrin (100%). The level of non-clustered intracellular a2 integrin obtained after acid stripping was 19.2% and a2 integrin surfacespecific fluorescence was 69.5%, together making up to 88.7% of the total a2 integrin fluorescence.

Surface-biotinylation-based a2 integrin internalization assay
The internalization of biotinylated a2 integrin was measured as described previously (Roberts et al., 2001) and the amount of biotinylated a2 integrin was determined by using capture-ELISA.

EGF and transferrin internalization assays
Cells were serum starved for 14 h before the assay, then EGF-Alexa-Fluor-555 was added to MEM-BSA medium to a final concentration of 100 ng/ml and cells were incubated at 37˚C for different lengths of time. Surface attached EGF-Alexa-Fluor-555 was removed with acidic buffer (50 mM glycine, 150 mM NaCl, pH 3.0). For the transferrin endocytic trafficking assay, cells were serum starved for 1 h, then transferrin-Alexa-Fluor-568 was added to MEM-BSA medium to a final concentration of 25 mg/ml. Cells were fixed with 3% PFA for 20 min at room temperature.
Image acquisition by wide-field, confocal and TIRF microscopy Image acquisition in wide-field mode was performed on an Olympus IX81 ScanR automated inverted microscope (Olympus Biosystems) controlled by ScanR acquisition software. A 106 0.4 NA air objective (UPlanSApo; Olympus Biosystems) was used for the primary screening on cell arrays and a 206 0.75 NA air objective (UPlanSApo; Olympus Biosystems) was used for all other experiments. An excitation wavelength of 450-490 nm and emission wavelength of 500-550 nm was used to image a2 integrin, an excitation wavelength of 426-446 nm and emission wavelength of 460-500 nm was used to image EGF-Alexa-Fluor-555, and an excitation wavelength of 545-580 nm and emission wavelength of 610-700 nm was used to image transferrin-Alexa-Fluor-568. An excitation wavelength of 325-375 nm and emission wavelength of 435-475 nm was used to image nuclei in all assays. Confocal imaging was performed on a confocal laser scanning microscope TCS SP5 (Leica Microsystems) using a 636oil-immersion objective (HCX PL APO 636/ 1.4-0.6 Oil CS) and 94-mm diameter pinhole. Laser lines of 488, 561 and 633 nm were used for the excitation of GFP or Alexa Fluor 488, mCherry or Alexa Fluor 568, and Alexa Fluor 647, respectively. TIRF was performed on a Nikon Eclipse Ti TIRF microscope with Nikon Apo TIFR 606 NA 1.49 objective. Laser lines of 488 and 640 nm were used for the excitation of Alexa Fluor 488 and Alexa Fluor 647, respectively.

Colocalization analysis
Colocalization of internalized a2 integrin with overexpressed Rabs in the multi-channelled single confocal plane was performed in ImageJ. The image background was subtracted using a rolling ball algorithm, noise was removed by Gaussian Blur filter and individual structures of .10 pixels in size were binarized and overlapped. In total, peripheral regions of more than five cells/Rab were used for the analysis, with .120 structures analyzed per cell.

Statistical data analysis
During the primary screening, one image completely encompassed an individual spot. Prior to quantification of the internalized integrin-specific fluorescence intensity, background signal was subtracted by applying a rolling ball algorithm [ScanR Analysis module (Olympus)]. All images were subjected to visual control. The nuclei of individual cells were identified by intensity module, and the regions of nuclei were expanded so as to encompass a maximum area of the cell without touching the neighboring ones. Cell densities were optimized to reach 70-90% confluence to secure clear cell-to-cell separation. As the majority of intracellular a2-integrin-specific fluorescence was clustered in the region of interest, the method allowed unbiased data quantification. For the statistical analysis of the primary screening data, R package (http://cran.r-project.org/) and cellHTS from Bioconductor (http://www. bioconductor.org/) were used. At first, the 2% of cells with the highest integrin-specific fluorescence intensities were excluded for each imaged spot, remaining cell intensities were averaged and Bscore normalization (Brideau et al., 2003) was applied to calculate a correction factor for each spot, which would account for spatial effects and between-plate artifacts.
To reliably quantify the intracellular accumulation of a2 integrin, we removed cells with weak endocytic trafficking using a threshold that was set for every experiment separately, according the negative controls, and was kept the same for all samples within the experiment. A Gaussian Mixture Model (Bowman and Azzalini, 1997) was used to automatically cluster the cells into two subpopulations (with high and low efficiencies of a2 integrin intracellular accumulation), assuming that in each subpopulation cells are normally distributed. The median intracellular signal intensities of a2 integrin and cell numbers of each subpopulation was calculated for every spot. Then, the median signal intensity of cells with efficient accumulation was multiplied by the number of such cells and divided by the total cell number in every spot. This ratio was normalized against the median of the negative controls on each cell array. Eventually, the median z-score of 5-8 individual replicates of every siRNA was calculated and a one-sided, one-sample Welch's t-test was used to compute significance values. A gene was considered as a primary hit when either one out of two tested siRNAs, at least two siRNAs out of six tested or at least three siRNAs out of more than eight siRNAs tested had a consistent inhibitory or acceleratory effect on the intracellular accumulation of a2 integrin. More than 3000 cells were analyzed for every siRNA in the primary screen.
For the validation experiments in eight-chamber m-slides, 30-42 images/well were taken and analyzed. All other steps of the analysis were the same as in the primary screening, except that a threshold, separating cells with high and low endocytosis, was set manually. For transferrin and EGF trafficking assays all cells in population were taken for the analysis. Statistical significance of difference between experiments throughout this study was tested by Student's two-tailed t-test for samples with uneven variance.