A novel mechanism of keratin cytoskeleton organization through casein kinase Iα and FAM83H in colorectal cancer

Keratins are intermediate filament proteins that form cytoskeletal filament networks in epithelial cells (Moll et al., 2008; Oriolo et al., 2007; Windoffer et al., 2011). To form a filamentous structure, keratin proteins are nucleated, assembled into particles and then elongated (Miller et al., 1991; Miller et al., 1993; Windoffer et al., 2011). Keratin filaments are further bundled to form a mechanically stable cytoskeleton (Moll et al., 2008; Windoffer et al., 2011). Keratin filament assembly and bundling are dynamic and reversible processes (Flitney et al., 2009; Windoffer and Leube, 1999; Windoffer et al., 2004). The keratin cytoskeleton is an important structural stabilizer of epithelial cells (Moll et al., 2008). However, to achieve various dynamic cellular processes, including epithelial cell polarization and migration, keratin filament networks need to be dynamically reorganized (Ameen et al., 2001; Beil et al., 2003; Kölsch et al., 2010; Oriolo et al., 2007; Salas et al., 1997; Windoffer et al., 2011).

Although the molecular mechanism governing the rearrangement of keratin filaments is largely unclear, phosphorylation and protein–protein interactions of keratins have been shown to be involved (Izawa and Inagaki, 2006; Omary et al., 2006; Windoffer et al., 2011). Phosphorylation of keratins induces the disassembly of keratin filaments. Keratin filaments reconstituted using rat keratin 8 and keratin 18 were disassembled by in vitro phosphorylation of the keratins, but reassembled by their dephosphorylation (Yano et al., 1991). Upon treatment of epithelial cells with the protein phosphatase inhibitors okadaic acid and orthovanadate, phosphorylation levels of keratins were enhanced and keratin filaments were concomitantly disassembled (Strnad et al., 2001; Strnad et al., 2002); however, only a few kinases such as p38 MAPK and PKCζ were reported to be involved in keratin filament disassembly in different situations (Sivaramakrishnan et al., 2009; Wöll et al., 2007).

The bundling of keratin filaments seems to be regulated by the association of keratins with specific proteins (Windoffer et al., 2011). The plakin family proteins, which are linker proteins between keratin filaments and various cytoskeletal structures (Wiche, 1998), are major modulators of keratin filament bundling (Boczonadi et al., 2007; Liu et al., 2011; Long et al., 2006; Osmanagic-Myers et al., 2006); however, our knowledge about crucial modulators of keratin filament bundling remains very limited. To further understand the mechanism governing the rearrangement of keratin filaments, identification of novel keratin-associated proteins and kinases responsible for the mechanism is required.

Similar to normal epithelial cells, epithelial cancer cells also express keratin proteins (Moll et al., 2008; Omary et al., 2009). Several studies have reported the involvement of keratins in cancer progression, invasion and metastasis (Omary et al., 2009). In several types of cancer, the expression levels of specific subtypes of keratin correlate with patient survival (Moll et al., 2008). In colorectal cancer, reduced expression of keratin 8 and keratin 20 was associated with shorter patient survival (Knösel et al., 2006). Furthermore, several reports have suggested a direct beneficial impact of keratins in cancer treatment. Forced expression of keratin 18 in MDA-MB-231 breast cancer cells strengthened cell–cell adhesions and markedly regressed tumor growth and metastasis in a xenograft model (Bühler and Schaller, 2005). Pancreatic carcinoma cells also exhibited decreased motility and tumorigenicity upon forced expression of keratin 18 (Pankov et al., 1997); however, additional studies are needed to elucidate the role of keratins in cancer.

FAM83H is an essential protein for amelogenesis because its truncated mutations cause autosomal-dominant hypocalcified amelogenesis imperfecta (Kim et al., 2008; Urzúa et al., 2011). However, very recent reports suggest the involvement of FAM83H in cancer. Microarray analysis of cancer tissues indicate increased expression of FAM83H in various types of cancer, including colorectal cancer (Sasaroli et al., 2011). Furthermore, Hatanaka and colleagues performed a focus formation assay using the cDNA library prepared from biliary tract cancer tissue and showed that fragmented FAM83H cDNA possesses transforming activity (Hatanaka et al., 2010). In a phospho-proteomic study of lung cancer cell lines that are sensitive or resistant to the anticancer drug dasatinib, phosphorylation levels of FAM83H decreased approximately tenfold in dasatinib-resistant cell lines (Klammer et al., 2012), implying the involvement of FAM83H in the resistance to dasatinib; however, there have been no reports on the precise mechanisms.

In this study, we determined that FAM83H and casein kinase Iα (CK-1α) are novel keratin-associated proteins and demonstrate that FAM83H-mediated recruitment of CK-1α to keratin filaments controls the filamentous state of keratins in colorectal cancer cells. Furthermore, we show that overexpression of FAM83H and CK-1α-mediated disassembly of keratin filaments occur in colorectal cancer tissue. FAM83H-overexpressing cancer cells exhibit loss or alteration of epithelial cell polarity and cell–cell adhesions – a hallmark of highly motile cancer cells. Given that FAM83H is involved in reorganization of the keratin cytoskeleton during the migration of colorectal cancer cells, keratin filament disassembly induced by FAM83H overexpression and aberrant localization of CK-1α might be important for the migration and invasion of colorectal cancer.

We first searched for proteins associated with FAM83H using a proteomic approach. HCT116 colorectal cancer cells were transfected with a plasmid vector encoding FLAG-tagged FAM83H or the empty vector, and immunoprecipitation with anti-FLAG antibody was performed using extracts of these cells. Proteins present in the immunoprecipitates were identified using in-gel digestion and LC-MS/MS. By comparing the proteins identified from FAM83H-FLAG immunoprecipitates with those from the control, keratin proteins (subtypes 18 and 19) and CK-1α were identified as FAM83H-associated proteins (supplementary material Table S1).

Western blot analysis of these immunoprecipitates substantiated the association of FAM83H with CK-1α and keratin 8 and 18 (Fig. 1A). Furthermore, endogenous FAM83H and CK-1α were detected in immunoprecipitates using anti-keratin-18 antibody (Fig. 1B). Fig. 1C shows subcellular colocalization of FAM83H with keratin 18 and keratin 19, and Fig. 1D,E show partial colocalization of CK-1α with FAM83H and keratin 18 [Pearson's correlation coefficient: FAM83H vs keratin 18, r = 0.58±0.20 (n = 6 cell images); FAM83H vs keratin 19, r = 0.49±0.06 (n = 6 cell images); FAM83H vs CK-1α, r = 0.35±0.06 (n = 11 cell images); keratin 18 vs CK-1α, r = 0.31±0.07 (n = 11 cell images)]. Given that organization of the keratin cytoskeleton is regulated by protein phosphorylation (Omary et al., 2006), these results suggest that FAM83H and CK-1α is involved in keratin cytoskeleton organization.

To assess whether FAM83H regulates the organization of keratin filaments, HCT116 cells were treated with siRNA to knock down FAM83H and keratin filaments were visualized. Control cells formed an intricate network of thin keratin filaments in the cytoplasm (Fig. 2A). In sharp contrast, FAM83H-knockdown cells constructed a markedly rough network of thick keratin filaments (Fig. 2A). Next, we tested the effect of overexpression of FAM83H–FLAG on keratin filaments in HCT116 cells. Fig. 2B shows that FAM83H–FLAG overexpression completely disassembled keratin filaments (region 1), although adjacent cells without FAM83H–FLAG expression exhibited normal networks of keratin filaments (region 2). FAM83H–FLAG mainly colocalized with the disassembled keratin proteins (Pearson's correlation coefficient, r = 0.78±0.20 (n = 3 cell images); Fig. 2B, region 1). In DLD1 colorectal cancer cells, similar reorganization of keratin filaments by knockdown and overexpression of FAM83H was observed (supplementary material Fig. S1).

In addition, western blot analysis showed that these alterations of the keratin filament networks were not due to changes in the expression levels of keratin proteins because no changes in the expression of keratin 8, keratin 18 or keratin 19 were observed upon knockdown or overexpression of FAM83H (Fig. 2A,B). Given that keratin filaments are dynamic cytoskeletal elements capable of rapid and reversible restructuring (Windoffer et al., 2011), these results indicate that FAM83H controls the equilibrium between the assembly (bundling) and disassembly of keratin filaments.

Next, we examined whether CK-1α is involved in the organization of keratin filaments. HCT116 cells were treated with a CK-1 inhibitor, D4476, or with siRNA to knock down CK-1α (CK-1α siRNA) and analyzed by immunofluorescence for keratin filaments. Inhibition of CK-1α, similar to FAM83H knockdown, promoted bundling of keratin filaments without changes in the protein level of keratins (Fig. 3A,B), implying that FAM83H and CK-1α cooperate to regulate the filamentous state of keratins. On the basis of the amino acid sequence, FAM83H does not have any enzymatic domains; thus, CK-1α kinase activity could mediate the mechanism by which FAM83H rearranges keratin filaments.

To explore the requirement of CK-1α for FAM83H rearrangement of keratin filaments, HCT116 cells were transfected with FAM83H–FLAG in the presence of CK-1α siRNA or D4476. Immunofluorescence for keratin filaments showed that both CK-1α siRNA and D4476 obviously reversed the effect of FAM83H–FLAG on the disassembly of keratin filaments (Fig. 3C,D). These results suggest that CK-1α mediates FAM83H–dependent rearrangement of keratin filaments.

Keratin proteins have been subjected to phosphorylation to regulate their own structural state (Omary et al., 2006); however, we could not simply explain the mechanism of FAM83H- and CK-1α-mediated rearrangement of keratin filaments by phosphorylation, at least at Ser73 and Ser431 of keratin 8 and Ser33 and Ser52 of keratin 18 (see supplementary material Fig. S2).

What is the precise mechanism by which FAM83H modulates the CK-1α-mediated rearrangement of keratin filaments? It is known that CK-1α is basically constitutive active and its subcellular localization is an important factor in its functional regulation (Knippschild et al., 2005); thus, FAM83H might regulate the subcellular localization of CK-1α. To test this hypothesis, we examined the effect of knockdown or overexpression of FAM83H on the subcellular localization of CK-1α. Immunofluorescence for CK-1α showed that FAM83H knockdown diminished the colocalization of CK-1α with keratin filaments (Fig. 4A). FAM83H knockdown decreased the value of the Pearson's correlation coefficient between CK-1α and keratin filaments (Fig. 4A). Conversely, FAM83H overexpression recruited CK-1α to disassembled keratin speckles (Fig. 4B). In FAM83H-overexpressing cells, CK-1α showed high values of the Pearson's correlation coefficient with FAM83H [r = 0.78±0.20 (n = 3 cell images)] and keratin 18 [r = 0.62±0.10 (n = 6 cell images)]. Coimmunoprecipitation assays using anti-keratin 18 antibody further showed that FAM83H knockdown abolished the physical interaction of CK-1α with keratin 18 (Fig. 4C). Neither knockdown nor inhibition of CK-1α affected the localization of FAM83H to keratins (Fig. 3C,D), excluding the possibility that CK-1α recruits FAM83H to keratins. These results indicate that FAM83H recruits CK-1α to keratin filaments to enable CK-1α to modulate the filamentous state of keratins.

To further elucidate the mechanism governing FAM83H-dependent rearrangement of keratin filaments, we performed proteomic analysis using dominant-negative truncated proteins of FAM83H (Lee et al., 2008; Urzúa et al., 2011; Wright et al., 2009). Similar to FAM83H knockdown, transfection with a plasmid encoding FAM83H-truncated protein with a FLAG-tag [amino acids 1–286, FAM83H-286N–FLAG; 1–296, FAM83H-296N–FLAG] induced the aggregation of severely bundled keratin filaments in HCT116 cells (Fig. 5A and data not shown). LC-MS/MS analysis of coimmunoprecipitated proteins with FAM83H-286N–FLAG and FAM83H–FLAG (full-length) showed that FAM83H-286N–FLAG immunoprecipitates contained higher levels of CK-1α but lower levels of keratins than FAM83H–FLAG immunoprecipitates (supplementary material Table S2). Western blotting of these immunoprecipitates substantiated strong interactions of FAM83H-286N–FLAG with CK-1α but only marginal interaction with keratin 8 (Fig. 5B). These analyses suggest that FAM83H interacts with CK-1α in the N-terminal region and with keratins at the C-terminal region. Additionally, forced expression of FAM83H-286N–FLAG diminished the recruitment of CK-1α to keratin filaments (Fig. 5C). FAM83H-286N–FLAG significantly decreased the value of the Pearson's correlation coefficient between CK-1α and keratin filaments (Fig. 5C). FAM83H-286N–FLAG appeared to exert a dominant-negative effect on the recruitment of CK-1α to keratins. These results suggest that FAM83H tethers CK-1α to keratins and that a FAM83H-mediated interaction between CK-1α and keratins is needed for proper keratin cytoskeleton organization.

FAM83H has FxxxF amino acid sequences (x, any amino acid residue), known to be CK-1-binding motifs of NFAT1, PER1 and PER2 (Okamura et al., 2004), in the N-terminal region (FMWSF, 247–251 aa; FDEEFRILF, 270–278 aa). To assess the requirement of these amino acid sequences for the binding of FAM83H to CK-1α, a coimmunoprecipitation assay using alanine mutants of FAM83H (251A, FMWSA; 274A, FDEEARILF) was performed. The FLAG-tagged FAM83H alanine mutants were transfected into HCT116 cells and immunoprecipitated using anti-FLAG antibody. Compared with wild-type FAM83H and the 251A mutant, immunoprecipitates of the 274A mutant contained very low levels of CK-1α (Fig. 6A), suggesting that the FDEEFRILF sequence is the major motif of FAM83H for CK-1α binding.

We further tested whether the 274A mutation disabled the FAM83H function to control keratin filament organization. HCT116 cells were transfected with FAM83H-274A or FAM83H-251A and analyzed by immunofluorescence of keratin filaments. Overexpression of the 251A mutant, as well as wild-type FAM83H, caused disassembly of keratin filaments, whereas the 274A mutant did not affect keratin filament organization (Fig. 6B). In addition, speckle-like localization of CK-1α was not induced by overexpression of FAM83H-274A, in contrast to wild-type FAM83H or FAM83H-251A (Fig. 6C). Taken together, these results substantiate that interaction of FAM83H with CK-1α is required for keratin filament reorganization.

To examine the expression of FAM83H in colorectal cancer cells in vivo, colorectal cancer tissues were analyzed by quantitative PCR (qPCR) and immunostaining with anti-FAM83H antibody. FAM83H mRNA levels were elevated in 10 of 12 colorectal cancer tissues compared with adjacent non-tumor tissues (Fig. 7A). Increased expression of FAM83H protein was also observed in 75 of 111 specimens, as assessed by the intensity of FAM83H immunostaining in cancer cells and adjacent normal epithelial cells (Fig. 7B). Note that the FAM83H expression pattern was heterogeneous throughout the tumor mass (Fig. 7C) and a subpopulation of cancer cells exhibited markedly high levels of FAM83H expression (Fig. 7C, region 1). Consistent with the in vitro results, FAM83H-overexpressing cancer cells exhibited disassembled keratin filaments and aberrant speckle-like colocalization of FAM83H, CK-1α and disassembled keratins (Fig. 7D,E). Another important feature of FAM83H-overexpressing cancer cells was loss or alteration of epithelial cell polarity (apical-basal polarity) (Fig. 7C, region 1) compared with FAM83H-low cancer cells (Fig. 7C, region 2). In addition, loss of E-cadherin expression, a protein that is essential for cell–cell adhesion (Birchmeier and Behrens, 1994), was also detected in FAM83H-overexpressing cancer cells (Fig. 7F). Given that the loss of epithelial cell polarity and E-cadherin expression is known to be a process during the migration or invasion of cancer cells (Thiery et al., 2009), these in vivo data suggest that FAM83H overexpression and the subsequent disassembly of keratin filaments occur in migrating or invading cancer cells.

Reorganization of the keratin cytoskeleton is required for the migration of epithelial cells (Beil et al., 2003; Kölsch et al., 2010; Windoffer et al., 2011). To examine whether FAM83H was involved in the migration of colorectal cancer cells, we performed a wound-healing assay with FAM83H-knockdown cells. HCT116 cells treated with FAM83H siRNA or control siRNA were scratched, and the wound edge was monitored by microscopy. Treatment with FAM83H siRNA suppressed migration of the wound edge of the cell sheets to less than 50% (Fig. 8A). Visualization of the cells at the wound edge using anti-α-tubulin and anti-keratin-8 antibodies showed that control cells formed a cell-spreading structure at the wound edge and showed a redistribution of keratin filaments that was less abundant in the cell-spreading region compared with the cell body region (Fig. 8B). By contrast, FAM83H-knockdown cells formed severely bundled keratin filaments throughout the cytoplasm and were prevented from spreading outward at the wound edge (Fig. 8B). These results suggest that FAM83H plays a crucial role in cell migration by regulating keratin cytoskeleton organization.

In this study, we identified novel keratin-associated proteins, FAM83H and CK-1α, and demonstrated that FAM83H regulates keratin cytoskeleton organization by recruiting CK-1α to keratin filaments, both in vitro and in vivo. Moreover, loss of epithelial cell polarity and E-cadherin expression, in addition to keratin filament disassembly, was observed in FAM83H-overexpressing cancer cells in vivo. These results strongly suggest that CK-1α-mediated keratin filament disassembly induced by FAM83H overexpression is involved in the invasion and/or metastasis of colorectal cancer.

Although it has been undoubtedly accepted that the filamentous state of keratins is regulated by protein phosphorylation (Izawa and Inagaki, 2006; Omary et al., 2006; Windoffer et al., 2011), little is known about the kinases responsible for the rearrangement of keratin filaments; thus, our data showing the role of CK-1α in keratin filament rearrangement are valuable. The filamentous state of keratins is modulated by the equilibrium between assembly (bundling) and disassembly. The assembly of keratin filaments is a self-processing reaction, which does not require any additional proteins, at least in vitro (Steinert et al., 1976), and is highly favored over disassembly (Windoffer et al., 2011). It is therefore conceivable that the mechanism governing the disassembly of keratin filaments plays a crucial role in regulating the equilibrium of filamentous states of keratins. Aberrant speckle-like colocalization of CK-1α to keratin filaments causes disassembly of keratin filaments, whereas dissociation of CK-1α from keratin filaments causes bundling of keratin filaments. These results suggest that correct recruitment of CK-1α to keratin filaments is required to properly maintain the equilibrium of the filamentous state of keratins.

Because CK-1α-mediated disassembly of keratin filaments depends on the kinase activity of CK-1α (Fig. 3B,D), identification of the CK-1α substrates responsible for keratin filament disassembly is an important issue. As keratin proteins are known to be subject to phosphorylation to regulate their own filament status (Omary et al., 2006), keratin proteins themselves are major candidates of the CK-1α substrates; however, we could not explain the CK-1α-mediated reorganization of keratin filaments by phosphorylation, at least at Ser73 and Ser431 of keratin 8 and Ser33 and Ser52 of keratin 18 (supplementary material Fig. S2). Another candidate for CK-1α substrates is plectin, which plays a crucial role in crosslinking keratin filaments to other cellular structures (Wiche, 1998). FAM83H–FLAG immunoprecipitates contained plectin (supplementary material Table S1), and plectin knockdown, as well as the inhibition of FAM83H and CK-1α, caused keratin filament bundling in HCT116 cells (supplementary material Fig. S3) and a human liver cell line (Liu et al., 2011). Our next task will be proteomic analysis to identify the CK-1α substrates responsible for the rearrangement of keratin filaments.

CK-1α is basically constitutive active and thus its subcellular distribution is an important factor for functional regulation (Knippschild et al., 2005); however, the precise mechanism governing the subcellular distribution of CK-1α remains largely unknown. We determined that FAM83H is a novel key regulator of the subcellular distribution of CK-1α. FAM83H was found to be a linker protein between CK-1α and keratins, because FAM83H interacts with CK-1α in the N-terminal region and keratins in the C-terminal region. Consistent with the cases of NFAT1, PER1 and PER2 (Okamura et al., 2004), FAM83H interacts with CK-1α through an FxxxF amino acid sequence in the N-terminal region (FDEEFRILF, 270–278 aa).

Intriguingly, FAM83H overexpression not only induced aberrant recruitment of CK-1α to keratin filaments but also decreased cytoplasmic localization of CK-1α (Fig. 4). Cytoplasmic CK-1α plays an essential role in the phosphorylation and degradation of β-catenin (Knippschild et al., 2005). Loss of CK-1α-mediated phosphorylation of β-catenin causes nuclear accumulation of β-catenin and transcription of Wnt-specific genes responsible for the control of cell fate decisions, resulting in tumorigenesis (Elyada et al., 2011; Polakis, 2000; Valenta et al., 2012). In fact, we observed dephosphorylation and nuclear accumulation of β-catenin upon FAM83H overexpression in vitro and in vivo (unpublished data). These results suggest that the subcellular distribution and function of CK-1α are totally orchestrated by FAM83H expression levels.

Our in vivo data suggested the physiological significance of the mechanism governing the rearrangement of keratin filaments by FAM83H and CK-1α in colorectal cancer. FAM83H overexpression, aberrant localization of CK-1α, and keratin filament disassembly are all detected in colorectal cancer cells exhibiting loss or alteration of epithelial cell polarity. As discussed above, CK-1α is involved in Wnt–β-catenin signaling, which regulates the epithelial cell polarity of cancer cells (Thiery et al., 2009); thus, the aberrant localization of CK-1α might contribute to the loss of epithelial cell polarity. In addition, because the keratin cytoskeleton is an essential element for maintenance of epithelial cell polarity (Ameen et al., 2001; Oriolo et al., 2007; Salas et al., 1997), keratin filament disassembly caused by FAM83H overexpression may also occur during the loss of epithelial cell polarity.

E-cadherin expression is also suppressed in FAM83H-overexpressing cancer cells in vivo. Loss of E-cadherin expression, as well as that of epithelial cell polarity, is a hallmark of epithelial–mesenchymal transition (EMT), which is a process by which cancer cells escape from the primary tumor mass for invasion and metastasis (Thiery et al., 2009). Moreover, E-cadherin expression is controlled by Wnt/β-catenin signaling (Thiery et al., 2009). These results imply that the aberrant localization of CK-1α caused by FAM83H overexpression might be involved in the suppression of E-cadherin and EMT of colorectal cancer cells.

Several reports have suggested that expression levels of keratin proteins are correlated to the cancer grade and patient survival (Knösel et al., 2006; Moll et al., 2008) and that keratin 18 contributes to suppression of the invasiveness of breast and pancreatic cancer cells (Bühler and Schaller, 2005; Pankov et al., 1997). In addition, Beil and colleagues proposed that reorganization of the keratin cytoskeleton by sphingosylphosphorylcholine induced cellular elasticity and enhanced cell migration of pancreatic cancer cells (Beil et al., 2003). Given our in vitro data suggesting the role of FAM83H-mediated rearrangement of keratin filaments in the migration of colorectal cancer cells, these results suggest that the disassembly of keratin filaments by overexpression of FAM83H contributes to the invasion and metastasis of colorectal cancer.

In conclusion, this work elucidated a novel mechanism governing the rearrangement of keratin filaments in colorectal cancer cells. Our data also suggested that aberrant localization of CK-1α and keratin filament disassembly induced by FAM83H overexpression are involved in the migration of colorectal cancer cells. FAM83H overexpression was suggested in various types of cancer (Sasaroli et al., 2011), and CK-1α is known to play a tumor suppressive role (Elyada et al., 2011; Valenta et al., 2012). Thus, tumor progression of various types of cancer might be commonly promoted by FAM83H overexpression, malfunction of CK-1α and subsequent disassembly of keratin filaments.

HCT116 and DLD1 colorectal cancer cell lines were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA). Cells were grown at 37°C in 5% CO₂ in Iscove's Modified Dulbecco's Medium (IMDM; Invitrogen, Carlsbad, CA) supplemented with 10% fetal bovine serum (FBS; Invitrogen). Plasmids and siRNA were transfected using Lipofectamine 2000 (Invitrogen) and Lipofectamine RNAiMAX (Invitrogen), respectively. D4476, a CK-1 inhibitor, was used at a concentration of 100 µM (Abcam, Cambridge, UK).

To generate FAM83H–FLAG-expressing plasmid, cDNA encoding human FAM83H (NM_198488.3) was amplified from HCT116 cells by PCR using the forward primer 5′-ATAGAATTCAACATGGCCCGTCGCTCTCAGAG-3′ and the reverse primer 5′-ACGGGATCCTCCCTTCTTGCTTTTGAACG-3′ and cloned into the p3XFLAG-CMV-14 vector (Sigma-Aldrich, St Louis, MO). FAM83H cDNA from HCT116 cells has three silent mutations (1896C/T, 2001C/A and 2953C/T). To generate plasmids expressing FLAG-tagged N-terminal fragments of FAM83H (amino acids 1–286, FAM83H-286N–FLAG; 1–296, FAM83H-296N–FLAG), the corresponding regions of FAM83H cDNA were amplified by PCR using the common forward primer 5′-ATAGAATTCAACATGGCCCGTCGCTCTCAGAG-3′ and the reverse primer 5′-ATAGGATCCGGGCACAAGCGGCTCGGACTG-3′ for FAM83H-286N–FLAG and 5′-ATAGGATCCGGCGTCCATGCGGGCCAG-3′ for FAM83H-296N–FLAG and cloned into the p3XFLAG-CMV-14 vector. FAM83H-251A–FLAG and FAM83H-274A–FLAG were generated using a PrimeSTAR Mutagenesis Basal Kit (Takara Bio, Shiga, Japan) and FAM83H–FLAG vector. For the 251A mutant, 5′-TGGTCCGCTGAGAAGATCCACCGCAGC-3′ and 5′-CTTCTCAGCGGACCACATGAAGCTGTA-3′ were used as PCR primers. For the 274A mutant, 5′-GAGGAGGCCCGCATCCTCTTCGCGCAG-3′ and 5′-GATGCGGGCCTCCTCGTCGAAGCTGGA-3′ were used as PCR primers. FAM83H siRNAs (FAM83H siRNA S1, FAM83H-HSS138852; FAM83H siRNA S2, FAM83H-HSS138851) were purchased from Invitrogen. CK-1α siRNAs targeting the sequence 5′-CAGAATTTGCGATGTACTT-3′ or 5′-GAATTTGCGATGTACTTAA-3′ and plectin siRNA targeting the sequence 5′-CCAAGAACTTGCAGAAGTT-3′ were purchased from Sigma-Aldrich. The control siRNA duplexes were purchased from Invitrogen (Medium GC Duplex #2) and Sigma-Aldrich (Mission negative control SIC-001).

The following antibodies were purchased: anti-FAM83H (HPA024604; Sigma-Aldrich), anti-keratin-8 (TS1, Thermo Scientific, Fremont, CA; EP1628Y, Epitomics, Burlingame, CA), anti-keratin-18 (DC10; Thermo Scientific), anti-keratin 19 (RCK108; Thermo Scientific), anti-α-tubulin (DM1A, Sigma-Aldrich; YOL1/34, Santa Cruz Biotechnology, Santa Cruz, CA), anti-FLAG (M2; Sigma-Aldrich), anti-E-cadherin (36B5; Thermo Scientific), anti-CK-1α (C-19; Santa Cruz Biotechnology), anti-phospho-keratin-8-Ser431 (EP1630; Epitomics), anti-phospho-keratin-8-Ser73 (E431-2; Abcam), anti-phospho-keratin-18-Ser33 (sc-101727; Santa Cruz Biotechnology), anti-phospho-keratin-18-Ser52 (sc-17032; Santa Cruz Biotechnology), anti-plectin (sc-7572; Santa Cruz Biotechnology) and anti-actin (C-11; Santa Cruz Biotechnology). Mouse IgG1 control (MOPC-21; Exbio, Vestec, Czech Republic) was used for immunoprecipitation. Alexa Fluor 488 and Alexa Fluor 594 donkey anti-mouse IgG and Alexa Fluor 488 and Alexa Fluor 594 donkey anti-rabbit IgG, Alexa Fluor 488 donkey anti-goat IgG, and Alexa Fluor 488 donkey anti-rat IgG antibodies were used for immunofluorescence (Invitrogen). HRP-conjugated horse anti-mouse IgG (Cell Signaling Technology, Beverly, MA), donkey anti-rabbit IgG (GE Healthcare, Little Chalfont, UK), and donkey anti-goat IgG (Santa Cruz Biotechnology) antibodies were used for western blotting.

Tumor and adjacent non-tumor tissue samples were collected from patients with colorectal cancer in the Department of Frontier Surgery, Chiba University Hospital. For qPCR, tissue samples were rapidly frozen and stored at −80°C. For immunofluorescence, tissue samples were embedded in OCT compound (Sakura Finetek, Tokyo, Japan), snap frozen in liquid nitrogen, and cut into 5 µm sections using a cryostat (Hyrax C50; Carl Zeiss, Jena, Germany). Paraffin-embedded blocks were cut into 2.5 µm sections using a microtome, REM-710 (Yamato, Saitama, Japan). The protocol for the collection and use of the tissue samples was approved by the ethics committees of the Graduate School of Medicine, Chiba University and the Proteome Research Center, National Institute of Biomedical Innovation. Written informed consent was obtained from each patient before surgery.

Extraction of total RNA and cDNA synthesis were performed as described previously (Tomonaga et al., 2004). qPCR reaction was performed using Power CYBR Green reagents (Applied Biosystems, Foster City, CA). The levels of mRNA encoding FAM83H and β-actin were examined, and the value of FAM83H was normalized by that of β-actin in each tissue sample. The following primer pairs were used: FAM83H, forward primer 5′-cgacaagtgccgtgtcaacc-3′ and reverse primer 5′-acttcccagtgcggcagtag-3′; β-actin, forward primer 5′-agaaaatctggcaccacacc-3′ and reverse primer 5′-ggggtgttgaaggtctcaaa-3′.

For the extraction of whole cellular proteins of cell lines, cells were directly lysed in SDS-PAGE sample buffer. For preparation of cell lysates used for immunoprecipitation (IP lysates), cells were suspended in PBS containing 1% NP40, Complete protease inhibitor cocktail (Roche, Basel, Switzerland), and PhosSTOP phosphatase inhibitor cocktail (Roche), and then homogenized by sonication. After centrifugation at 100,000 g for 30 minutes, the supernatant was collected. Immunoprecipitation was performed using antibodies crosslinked to Protein G Dynabeads (Invitrogen) using dimethyl pimelimidate dehydrochloride (MP Biochemicals, Santa Ana, CA). IP lysates were reacted with antibody-coated Dynabeads for 1 hour at 4°C, and the absorbed proteins were eluted with 100 mM glycine-HCl (pH 3.0) or SDS-PAGE sample buffer. Western blotting was performed using the chemiluminescence detection system ECL or ECL Prime (GE Healthcare). The images were obtained with LAS4000 (Fuji Film, Tokyo, Japan) and processed with Multi Gauge V3.2 (Fuji Film) and Photoshop CS5 (Adobe, San Jose, CA).

Immunoprecipitates with anti-FLAG antibody were resolved by SDS-PAGE and the gel lane was divided into nine pieces corresponding to different molecular masses. In-gel tryptic digestion of proteins was performed as described previously (Adachi et al., 2007). The digested peptides were analyzed using an LTQ-Orbitrap Velos mass spectrometer (Thermo Scientific) equipped with a nanoHPLC system (Paradigm MS2; Michrom, Auburn, CA) and an HTC-PAL autosampler (Zwingen, Kanton Bern, Switzerland). A 0.3×5 mm trap column [L-column ODS; Chemicals Evaluation and Research Institute (CERI), Tokyo, Japan] and an analytical column made in-house by packing L-column2 C18 (CERI) into a self-pulled needle (0.1×200 mm) were used (Muraoka et al., 2012; Narumi et al., 2012). The mobile phases consisted of buffer A (0.1% formic acid and 2% acetonitrile) and B (0.1% formic acid and 90% acetonitrile). The nanoLC gradient was delivered at 500 nl/minute and consisted of a linear gradient of buffer B developed from 5 to 35% B in 45 minutes. The dynamic exclusion function of LTQ-Orbitrap was turned off. For protein identification, peptide mass data were matched by searching the UniProtKB/Swiss-Prot database (2011_12) using the MASCOT search engine v2.3. Database search parameters were: the charge of the precursor ion, 2+ and 3+; peptide mass tolerance, 3 ppm; fragment tolerance, 0.6 Da; allowing up to one missed cleavage; fixed modification, carbamidomethylation of cysteine; variable modification, oxidation of methionine. Proteins were identified based on at least two unique peptides. The number of assigned spectra was calculated using Scaffold 3 software (Proteome Software, Portland, OR) for semi-quantitation.

Cells or tissue sections from snap-frozen tissues were fixed with methanol at −20°C for 2 minutes or 4% paraformaldehyde in PBS at 30°C for 20 minutes, permeabilized with 0.5% Triton X-100 in PBS for 5 minutes on ice, blocked in PBS containing 0.01% Tween 20 and 3% BSA on ice, and sequentially incubated with primary and secondary antibodies at room temperature. For immunofluorescence, DNA was stained with 100 ng/ml of 4′-6-diamidino-2-phenylindole (DAPI; Sigma-Aldrich) and stained samples were viewed under an LSM710 confocal microscope with Zen software (Carl Zeiss, Jena, Germany). The objective lenses were EC Plan-NEO FLUAR 10× /0.3 or 40× /1.3 and Plan APOCHROMAT 63× /1.4. Pearson's correlation coefficient was calculated using Zen software, according to the manufacturer's instructions. The value can range from 1 to −1, with 1 indicating a complete positive correlation, −1 a negative correlation and with 0 indicating no correlation. For immunohistochemistry, visualization was performed with diaminobenzidine (DAB) chromogen (EnVision+ Kit/HRP; Dako, Glostrup, Denmark) and nuclei were stained with Mayer's hematoxylin (Muto Pure Chemicals, Tokyo, Japan) (Seimiya et al., 2008). Stained sections were scanned on a NanoZoomer RS digital slide imaging system (Hamamatsu Photonics, Hamamatsu, Japan). Two pathologists evaluated the immunostaining of colorectal cancer tissues. Composite figures were prepared using Photoshop CS5.

We would like to thank Hiromi Saito, Masayoshi Kuwano, Shio Watanabe and Yuuki Hashimoto for technical assistance.