The Repeat Domain of the Melanosomal Matrix Protein PMEL17/GP100 Is Required for the Formation of Organellar Fibers*

Over 125 pigmentation-related genes have been identified to date. Of those, PMEL17/GP100 has been widely studied as a melanoma-specific antigen as well as a protein required for the formation of fibrils in melanosomes. PMEL17 is synthesized, glycosylated, processed, and delivered to melanosomes, allowing them to mature from amorphous round vesicles to elongated fibrillar structures. In contrast to other melanosomal proteins such as TYR and TYRP1, the processing and sorting of PMEL17 is highly complex. Monoclonal antibody HMB45 is commonly used for melanoma detection, but has the added advantage that it specifically reacts with sialylated PMEL17 in the fibrillar matrix in melanosomes. In this study, we generated mutant forms of PMEL17 to clarify the subdomain of PMEL17 required for formation of the fibrillar matrix, a process critical to pigmentation. The internal proline/serine/threonine-rich repeat domain (called the RPT domain) of PMEL17 undergoes variable proteolytic cleavage. Deletion of the RPT domain abolished its recognition by HMB45 and its capacity to form fibrils. Truncation of the C-terminal domain did not significantly affect the processing or trafficking of PMEL17, but, in contrast, deletion of the N-terminal domain abrogated both. We conclude that the RPT domain is essential for its function in generating the fibrillar matrix of melanosomes and that the luminal domain is necessary for its correct processing and trafficking to those organelles.

Melanoma is one of the most notorious tumors because of its poor prognosis (1). Several groups have raised monoclonal antibodies that specifically recognize melanoma cells, with HMB45 being the most popular and frequently used in clinics for the specific detection of melanocytic tumors (2,3). HMB45 is now known to identify the product of the human PMEL17 locus, also known as GP100 or SILV. Interestingly, two other antibodies developed for melanoma detection, HMB50 and NKI/beteb, also react with PMEL17 (4 -7), although the reactivity patterns between HMB45 and HMB50 or NKI/beteb can be quite different (8,9). Melanosomes are one type of lysosome-related organelle that have the unique capacity to produce melanin pigment (10) and that progress through four sequential morphological stages as they mature (11). Stage I melanosomes are round, membrane-bound, and electron-lucent vesicles. Stage II melanosomes result from the elongation of those vesicles and the appearance within of distinct fibrillar structures. The production of internal matrix fibers, which discriminate Stage II from Stage I melanosomes, depends on the maturation and trafficking of PMEL17 (12)(13)(14)(15). In pigmented cells, melanins are synthesized and deposited on those fibers, resulting in a progressively pigmented internal matrix, at which time the organelles are termed Stage III melanosomes. In highly pigmented tissues, melanin synthesis and deposition continue until there is little or no internal structure visible, at which time they are termed Stage IV melanosomes. Several independent groups have used immunoelectron microscopy to show that HMB45 specifically reacts with the fibrillar matrix in Stage II melanosomes (12)(13)(14). We reported recently that PMEL17 produces melanosomal matrix fibers in melanocytic cells in the presence of MART-1 (15). Moreover, even in non-melanocytic cells, when adequately overexpressed, PMEL17 can also produce fibrillar structures within lysosome-like structures (12,16).
Although PMEL17 is studied widely as a melanoma antigen, as noted above, it also plays a crucial role in normal melanocytes as a structural protein in melanosomes (17,18). PMEL17 is a type I membrane protein that consists of several domains predicted by homology modeling (see Fig. 1A). SIG is the signal peptide thought to determine the entry of PMEL17 into the secretory pathway prior to its processing and cleavage (19), as detailed below. PKD is the polycystic kidney disease-like domain, which has an immunoglobulinlike folding structure (20). RPT is a domain that contains 10 imperfect repeats of 13 proline/serine/threonine-rich amino acids (21)(22)(23). Each of those 13 amino acid series has been annotated from a to j in Fig. 1A, and there are 26 potential O-glycosylation sites within the RPT domain. PMEL17 shows high homology to proteins such as lysosome-associ-ated membrane protein (LAMP) 3 and NMB (24), but the RPT domain is unique and specific for PMEL17 (25). KRG is the kringle-like domain, which is a triple disulfide-liked autonomous structural domain found in blood clotting and fibrinolytic proteins (26). It is thought that KRG domains play some role in binding interactions with other proteins necessary for their regulation (20); however, nothing has been reported about the function of the KRG domain in PMEL17. TM is the predicted transmembrane domain. The remaining domains are referred to as the N-terminal domain (NTD), the GAP1, GAP2, GAP3 domains, and the C-terminal domain (CTD). PMEL17 has five potential N-glycosylation sites; however, Asn 321 in the RPT domain is thought not to be utilized (19,25). Three of those N-glycosylation sites are located in the NTD, and another is in the GAP3 domain. Four isoforms of PMEL17 are generated by alternative splicing (6,21), as depicted in Fig. 1A. PMEL17-i is the full-length protein sequence, whereas PMEL17-is and PMEL17-ls have truncated RPT domains. The "l" in PMEL17-l and PMEL17-ls indicates the insertion of 7 additional amino acids in the GAP3 domain immediately following the N-glycosylation site. PMEL17-i is the most abundant form (21). 4 Fig. 1B depicts the complex pattern of maturation and processing of PMEL17 (based in part on Refs. 12, 15, 16, 25, 27, and 28). Briefly, P1/P100 (hereafter termed P1) is the major endoplasmic reticulum (ER; or early Golgi)-modified form of PMEL17, which contains high mannose-type (but not complex-type) N-glycans and which is endoglycosidase H-sensitive. A relatively small amount of P1 then undergoes further glycosylation in the Golgi to generate P2/P120 (hereafter termed P2), the late Golgi-modified form, which still has at least one high mannose-or hybrid-type N-glycan and which is largely endoglycosidase H-resistant (12,27). P2 is then cleaved proteolytically (between Arg 469 and Gln 470 ) within post-Golgi and/or premelanosomal compartments into M␣ and M␤/P26 (hereafter termed M␤) fragments. Those two fragments remain linked by a disulfide bond. M␣ is thought to be further processed to generate the striated fibrils seen in Stage II melanosomes, although the putative cleavage site(s) in the M␣ fragment are unknown. HMB45 is thought to react with the processed form of M␣; however, there is no convincing evidence for this at the molecular level. Although there is no controversy that PMEL17 is found in Stage II melanosomes (12,14), there is some conflict as to which form of PMEL17 (P1 and/or P2) exists in Stage II melanosomes and how it gets there (12,14,27). Several studies have shown that HMB45 reactivity is lost when PMEL17 is digested to remove sialic acids (27,29,30), but there is as yet no report utilizing that characteristic to define the precise epitope or its implications to the processing and trafficking of PMEL17.
In this study, we focused on characterizing the domain of PMEL17 involved with fibril formation using HMB45 as a fibrilspecific probe. We generated mutants of PMEL17 to determine the domains critical to the formation of fibrils and to assess the correct processing and trafficking of PMEL17 that are essential to its function in melanogenesis. We show that HMB45 reacts preferentially with a detergent-insoluble form of PMEL17 and that its epitope, which contains sialylated O-glycans, is located in the second and third amino acid repeats of the RPT domain. The heterogeneous bands detected by HMB45 staining are due to the multiple cleavage points in the RPT domain. Surprisingly, correct processing and trafficking of PMEL17 to LAMP-2-positive organelles are maintained when the RPT domain or the CTD is deleted, but PMEL17 is missorted when other luminal domain(s) are missing, suggesting that important sorting signals are within that region of PMEL17.

EXPERIMENTAL PROCEDURES
Cell Cultures-Highly pigmented MNT-1 melanoma cells, unpigmented SK-MEL-28 melanoma cells, and HeLa cells were obtained and cultured as described previously (15).
Electron Microscopy-Electron microscopy was performed as described previously (15). Thin sections were examined using a Zeiss EM 912 Omega electron microscope.
Protein Extraction and Immunoblotting-Cell extracts were prepared using the M-PER mammalian protein extraction reagent (Pierce), 1% Triton X-100 (Sigma), or 1% Nonidet P-40 (Calbiochem) in the presence of protease inhibitor mixture (Roche Applied Science). After centrifugation at 20,000 ϫ g for 30 min, the supernatants were harvested as soluble fractions. The protein concentration of each soluble fraction was measured using the BCA protein assay (Pierce). Pellets were washed once with the corresponding lysis buffer and then resuspended in 1ϫ SDS-PAGE buffer as insoluble fractions. Equivalent amounts of soluble and insoluble fractions were used. Immunoblotting was performed as described previously (15). Metabolic Labeling and Immunoprecipitation-HeLa cells were pulsed for 30 min with [ 35 S]Met/Cys (GE Healthcare) and then chased for 4 h. Cells were lysed with lysis buffer containing 1% Triton X-100. Immunopurified samples were electrophoresed and visualized by fluorography as described previously (15).

RESULTS
HMB45 Detects the M␣ Form of PMEL17 by Immunoblotting-Pigmented MNT-1 cells, unpigmented SK-MEL-28 cells, and HeLa cells transfected with empty vector, PMEL17-i, or PMEL17-is were analyzed by immunoblotting using HMB45 and ␣-PEP13h antibody. MNT-1 cells, SK-MEL-28 cells, and HeLa cells overexpressing PMEL17-i or PMEL17-is showed HMB45-positive ϳ30 -50-kDa bands in common ( Fig. 2, left panel), as reported previously (27,29). PMEL17-i showed three distinct bands (or clusters of bands), one at ϳ100 kDa, another at ϳ50 kDa, and another at ϳ30 kDa (with a faint band at ϳ120 kDa). PMEL17-is also showed three distinct bands, but at ϳ80, ϳ45, and ϳ30 kDa (with a faint band ϳ110 kDa). Equivalents of those bands were also found in MNT-1 cells, although the reactivities for some of them were relatively lower. The difference between the PMEL17-i and PMEL17-is constructs is only a small 7-amino acid deletion in the RPT domain, which implies that the differences in HMB45 staining patterns is closely associated with the RPT domain. Only two fast The furin-mediated cleavage site (CS) in the GAP2 domain is indicated. The RPT domain is further subdivided into 10 imperfect series of 13 amino acids. Each series is named from a to j. Potential O-glycosylation sites are indicated by asterisks. Four isoforms of PMEL17 are currently identified. The RPT domains of PMEL17-is and PMEL17-ls lack the underlined 42 amino acids. PMEL17-l and PMEL17-ls contain the underlined 7 amino acids in the GAP3 domain. B, schematics of the processing of PMEL17. P1/P100 is the ER-modified form of PMEL17. P2/P120 is the late Golgimodified form of PMEL17, and mature N-linked glycans are indicated by check marks. M␣ and M␤/P26 are the cleaved products P2/P120, which may remain linked by disulfide bonds. The fourth scheme shows the further processed PMEL17 forms found in Stage II melanosomes. M␣ is further cleaved into M␣N and M␣C. ␣-PEP13h and ␣-mSiN antibodies react with the C and N termini of PMEL17, respectively. HMB50 and NKI/beteb are believed to react with the luminal domain as determined by ultrastructural observations, but the precise epitopes recognized by HMB50 and NKI/beteb are unknown. HMB45 is also believed to react with M␣C (indicated by the question mark). This schematic and nomenclature are based on Refs. 12, 15, 16, 19, 21, and 27. migrating and minor HMB45-reactive bands were detected in SK-MEL-28 cells, the fastest being somewhat larger than that seen in MNT-1 cells or in PMEL17-transfected HeLa cells. This might be due a variation in post-translational modifications, including gly-cosylation and/or sialylation, but is not simply due to the fact that SK-MEL-28 cells are unpigmented (31).
We also performed immunoblotting using anti-PEP13h antibody, which reacts specifically with the C terminus of PMEL17 (27). P1 or P1s (the short form of P1 derived from PMEL17-is or PMEL17-ls) and M␤ were clearly identified in MNT-1 cells and in PMEL17-i-and PMEL17-is-transfected cells (Fig. 2, right panel). An unknown band, P75 (ϳ75 kDa), was detected in MNT-1 cells (15); however, it was not detected in HeLa cells overexpressing any isoforms of PMEL17 (Fig. 2, right panel).
What are those high molecular mass bands (ϳ100 or 80 kDa) that are recognized by HMB45 except in SK-MEL-28 cells? Similar bands were visible in some previous reports, but were not mentioned, probably because they were considered to be nonspecific (15,27,29,32). Comparison of immunoblots using HMB45 or ␣-PEP13h antibody revealed bands at ϳ120 kDa in PMEL17-i and at ϳ110 kDa in PMEL17-is that represent P2 and P2s (the short form of P2 derived from PMEL17-is or PMEL17-ls), respectively. This raises the question at to why P1 was not detected by HMB45 in that extract. HMB45 requires sialylation to generate its epitope (27,29,30), sialylation being the terminal step of glycosylation. P1 then is the ER-modified form that is not detected by HMB45. Taken together, bands at ϳ120, 110, 100, and 80 kDa correspond to P2, P2s, M␣, and M␣s (the short form of M␣, which are derived from PMEL17-is or -ls), respectively, whereas the smaller ϳ30 -60 kDa bands are derivatives of M␣ or M␣s; thus, we tentatively name them M␣C, M␣C1, M␣C2, and M␣C3 ( Fig. 2, left panel).
HMB45 Recognizes the RPT Domain of PMEL17-To identify the HMB45-reactive domain of PMEL17, we designed a series of 10 deletion mutants of PMEL17 as follows: ⌬SIG, ⌬NTD, ⌬PKD, ⌬GAP1, ⌬RPT, ⌬GAP2, ⌬KRG, ⌬GAP3, ⌬TM, and ⌬CTD. These expression constructs were transfected into HeLa cells, and their immunoreactivity with HMB45 was assessed (Fig. 3A). The ⌬SIG and ⌬RPT constructs did not react at all  The Repeat Domain of PMEL17 Is Required for Fibrillogenesis with HMB45, and reactivity with ⌬TM was greatly reduced. All other constructs reacted well, although HMB45-positive granules in ⌬NTD, ⌬PKD, ⌬GAP1, and ⌬GAP2 were located mainly in the perinuclear region rather than being dispersed in the cytoplasm.
The reactivities of these constructs were also analyzed by immunoblotting using HMB45 and ␣-PEP13h antibody (Fig.  3B). ⌬SIG showed no reactivity with HMB45, and only a single band that migrated faster than P1 was detected by ␣-PEP13h antibody. Signal peptides classically function to target proteins to enter the ER; thus, it is reasonable that ⌬SIG would not undergo glycosylation, would have a relatively smaller P1, and would not react with HMB45. According to a previous report that the P2-to-M␣/M␤ cleavage occurs in post-Golgi and prelysosomal compartments (16), it is also reasonable that M␤ was not detected in ⌬SIG. These results suggest that SIG in PMEL17 is essential for entering the secretory pathway, as expected. ⌬NTD showed two bands recognized by HMB45 at relatively lower intensities, a major one that was smaller than M␣ and the other corresponding to M␣C. However, the major bands (M␣C1 and M␣C3) were not detected by HMB45. P1 and P2 were identified in ⌬NTD by ␣-PEP13h antibody staining, but their molecular masses were smaller than those of PMEL17-i. M␤ was also clearly detected in this mutant by ␣-PEP13h antibody. ⌬PKD and ⌬GAP1 had similar characteristics, showing faint single bands (Fig. 3B, * and **) that were not observed in any other constructs by HMB45. These bands were clearly located between P2 and M␣. M␤ was only faintly detectable in ⌬PKD and ⌬GAP1 by ␣-PEP13h antibody staining; however, P1 was clearly detected in both. This means that cleavage into M␣/M␤ is significantly hampered by deletion of the PKD or GAP1 domain.
However, ⌬RPT showed the most striking feature in that HMB45 reactivity was completely lost, but ␣-PEP13h antibody reactivity with (truncated) P1 and M␤ was not affected. The reactivity of ⌬GAP2 with HMB45 also showed only one slower migrating band compared with M␣; thus, the GAP2 domain might affect the processing of M␣. P1 was detected in ⌬GAP2, but M␤ was not detected using ␣-PEP13h antibody because the furin-mediated cleavage site is located in the GAP2 domain, and thus, ⌬GAP2 should not be able to produce M␤ because of its design. ⌬KRG showed a significantly lower level of M␣ detected by HMB45, although reactivity with M␣C1 or M␣C3 was not affected. M␤ was nearly undetectable, although P1 was clearly detected by ␣-PEP13h antibody. ⌬GAP3 showed no difference in staining pattern from PMEL17-i, except for its truncated M␤ band detected by ␣-PEP13h antibody. ⌬TM was nearly undetectable by immunoblotting using HMB45. Immunoblotting of ⌬TM with ␣-PEP13h antibody showed P1 but not M␤, which is similar to the pattern seen for ⌬PKD and ⌬GAP1.
Thus, the disruption of HMB45 reactivity seen for ⌬PKD, ⌬GAP1, and ⌬TM indicates that these domains undergo glycosylation, but are not sufficiently sialylated to be recognized. Interestingly, none of them are significantly cleaved by proprotein convertases. Is that just a coincidence? Sialylation is performed by sialyltransferases, which require acidic pH optima (33), and furin proprotein convertases also require acidic conditions (34). One interpretation that would account for this is the failure to traffic to acidic com-partments such as the trans-Golgi network (TGN). ⌬CTD showed virtually the same immunoblotting pattern as PMEL17-i with HMB45, but was undetectable by ␣-PEP13h antibody, which was quite unexpected and is discussed in further detail below. The ⌬CS mutation inhibited the production of M␣/M␤; however, M␣C1 and M␣C3 remained unchanged.
N-terminal Domains Affect the Trafficking of PMEL17-We then examined whether deletion mutants of PMEL17 (excluding ⌬SIG, ⌬TM, or ⌬RPT) are trafficked to late endosomes or lysosomes. We (15,35) and others (12,21) have reported previously that the trafficking of PMEL17 can be monitored by co-localization with LAMP-2 or LAMP-1, both of which localize in late endosomes, lysosomes, and melanosomes. We characterized the trafficking of mutant PMEL17 in HeLa cells by staining with HMB45 and by cotransfection of LAMP-2-EGFP. In PMEL17-i-transfected HeLa cells, HMB45 and LAMP-2 co-localized well (Fig. 4), and this also occurred in ⌬KRG-, ⌬GAP3-, and ⌬CTD-transfected cells. However, in ⌬NTD-, ⌬PKD-, ⌬GAP1-, or ⌬GAP2-transfected cells, the number of HMB45-positive organelles was significantly less; they were localized only in the perinuclear region, and few co-localized with LAMP-2. Interestingly, in contrast to ⌬PKD and ⌬GAP1, ⌬NTD and ⌬GAP2 showed stronger HMB45 reactivity, and the M␤ fragment was clearly observed in ⌬NTD. Both sialylation (required for HMB45 reactivity) and furin proprotein convertases (required for M␤ cleavage) require acidic pH optima (33,34). This might account for the significant loss of HMB45 reactivity and the M␤ fragment detected by ␣-PEP13h antibody in ⌬PKD and ⌬GAP1 as described above. In contrast, ⌬NTD and ⌬GAP2 could reach the acidic environment, probably the TGN and later compartments. These results suggest that N-terminal domains (NTD, PKD, GAP1, and GAP2) are required for the correct trafficking of PMEL17 in these transfected cells. Moreover, the PKD and GAP1 domains seem to be required for entry into the early secretory pathway. In contrast to those luminal domain deletion mutants, C-terminal domains (KRG, GAP3, and CTD) are not required for the trafficking of PMEL17, at least in transfected HeLa cells. NMB shows high homology to PMEL17, and another study showed that a mutation of the putative dileucine motif in the cytoplasmic domain of QNR-71 (the quail homolog of NMB) clearly abrogated the trafficking of QNR-71 (36). PMEL17 also has di-leucine sorting signal motif at its C terminus. We further introduced di-alanine mutation at di-leucine motif of PMEL17 (LL662AA). As is the case with ⌬CTD, LL662AA did not abrogate the trafficking of PMEL17 (Fig. 4) as it did in QNR-71.
The RPT Domain Is Required for the Proper Maturation of Fibrillar Structure-Previous reports by our group (14) and others (13) have shown that HMB45 specifically recognizes melanosomal matrix fibers. Moreover, HeLa cells can produce melanosomal matrix fiber-like structures when PMEL17 is overexpressed in the absence of other melanosomal proteins or melanosomes (12). When considered with the results of this study, it is clear that the RPT domain comprises the melanosomal matrix fiber itself. We analyzed what is observed in lysosomal organelles of HeLa cells overexpressing ⌬RPT instead of PMEL17-i. We looked first at the intracellular trafficking of ⌬RPT. HeLa cells overexpressing PMEL17-i or ⌬RPT were analyzed by immunocytochemistry using HMB50 (another antibody recognizing PMEL17) and anti-LAMP-2 antibody (Fig.  5A). PMEL17-i and ⌬RPT co-localized similarly with LAMP-2 at high frequency, suggesting that ⌬RPT is trafficked successfully to late endosomes or lysosomes, as is PMEL17-i. We then examined this at the ultrastructural level. Well developed parallel striated fibrillar structures similar to Stage II melanosomes were frequently observed in HeLa cells overexpressing PMEL17-i (Fig. 5B), as reported previously (12). In contrast, only occasional onion-like structures with electron-dense granules (termed lamellar bodies) were observed in HeLa cells overexpressing ⌬RPT. Of all specimens examined, we found only one organelle with a parallel striated fibrillar structure in HeLa cells transfected with ⌬RPT (data not shown). From these results, we conclude that the RPT domain does not affect the trafficking of PMEL17; however, it plays a crucial role in the development of the striated fibrillar structure.
The RPT Domain Undergoes Variable Proteolytic Cleavage-We hypothesized previously that M␣ is processed to M␣C and M␣N, which are nearly the same molecular mass as M␤ (27).
Based on the results shown above (Fig. 2), we now further hypothesize that M␣C is then processed to M␣C1 and M␣C3 (M␣C2 for PMEL17-is or PMEL17-ls). The difference in HMB45 immunoreactivity between PMEL17-i and PMEL17-is suggests that this processing should occur in the RPT domain because PMEL17-is has a truncated RPT domain, and other domains of PMEL17-is are identical to those of PMEL17-i (Fig.  1A). The RPT domain consists of 10 imperfect repeats of 13 amino acids. We tried to identify which repeats correspond to each subdomain as detected by HMB45 and which repeats are identified by HMB45. To examine this, we designed a series of deletion mutants. RPTa-i has 9 of the 10 repeats (from a to i), RPTa-b has two repeats (a and b), RPTb-j has nine repeats (from b to j), and so on (Fig. 1A). Moreover, RPTsa-i has five repeats (a-d and i) and two truncated repeats (e and h) that are specific to PMEL17-is (or PMEL17-ls). We transfected those constructs into HeLa cells and performed immunoblotting using HMB45. The reactivity of HMB45 was lost in RPTa-b, RPTc-j, RPTd-j, and RPTe-j (Fig. 6A). This shows clearly that both repeats b and c are required for the reactivity of HMB45. These two repeats are also predicted to have potential O-glyco- sylation sites (Fig. 1A). Repeats b and c do not have potential N-glycosylation sites. Moreover, digestion with peptide N-glycosidase F did not affect immunoreactivity with HMB45 (data not shown), suggesting that sialic acids recognized by HMB45 are located in repeats b and/or c. M␣, M␣C, M␣C1, and M␣C3 migrated faster in RPTb-j than in PMEL17-i. This suggests that repeat a is involved in M␣, M␣C, M␣C1, and M␣C3. M␣C was not always clearly detected in extracts of PMEL17-i (Figs. 2 and 6A). However, from the results obtained with other transfectants, all repeats (from a to j) are involved in M␣C, which proves that its C-terminal fragment is derived from M␣ (as shown schematically in Fig. 7).
Next, we tried to identify the C-terminal cleavage sites of M␣C1, M␣C2, and M␣C3 (Fig. 6A). Similar to the case in PMEL17-i (data not shown), the ⌬CS mutations (Fig. 1A) did not affect M␣C1, M␣C2, or M␣C3 (Figs. 3B and 6A). This suggests that the furin-mediated cleavage site does not affect M␣C2 or M␣C3 as detected by HMB45. M␣C1 in PMEL17-i was detected in RPTsa-h, but not in RPTa-g (Fig. 6A); thus, the cleavage must occur in repeat h in the RPT domain (Figs. 1A and 7). M␣C2 in PMEL17-is was detected in RPTa-h, but not in RPTa-d (Fig. 6A), meaning that it is cleaved in repeat e or h in the RPT domain (Figs. 1A and 7). M␣C3 was detected in RPTa-c, but not in RPTa-b (Fig. 6A); thus, it must be cleaved in repeat c in the RPT domain (Figs. 1A and 7). In PMEL17-is or PMEL17-ls (Fig. 6A), we could not identify M␣Cs (the short form of M␣C; that derived from PMEL17-is or -ls, which might be due to the fact that its size is too similar compared with M␣C2 and/or that M␣Cs is a transient form. Taken together, these results suggest that the RPT domain undergoes variable proteolytic cleavage. M␣N has been reported to have the same molecular mass as M␤, which is ϳ26 kDa (27). However, in this study, M␣ is ϳ100 kDa and M␣C is ϳ60 kDa, so M␣N is estimated to be ϳ40 kDa (Figs. 1B and  3A). Why should this discrepancy occur? M␣N was detected by pulsechase labeling and immunoprecipitation with an antibody that recognizes the N terminus of PMEL17 (␣-mSiN antibody) (27). We re-examined this question using HeLa cells overexpressing PMEL17-i or ⌬GAP3 (the GAP3 domain is part of M␤) that was radiolabeled and then chased for 4 h. Cells were solubilized and immunoprecipitated with ␣-PEP13h or ␣-mSiN antibody and then electrophoresed and observed by fluorography (Fig. 6B). ␣-PEP13h antibody recognized P1 and M␤ in PMEL17-i, as did ␣-mSiN antibody. ␣-PEP13h antibody also reacted with P1 and with truncated M␤ in ⌬GAP3; and interestingly, ␣-mSiN antibody also recognized that same truncated M␤ band. Thus, what was reported previously as M␣N is the same as M␤. Even after the production of M␤, M␣ and M␤ remain linked, probably via disulfide bonds (12). ␣-mSiN antibody would be expected to immunoreact with P1, P2, and the M␣⅐M␤ complex (Fig. 1B) under the previously reported conditions of immunoprecipitation (27). However, the subsequent electrophoresis was performed under denaturing conditions, in which the M␣⅐M␤ complex would be dissociated (12,27).
So the challenge is how can we detect M␣N? To do this, we artificially inserted an HA tag into the NTD of PMEL17-i between the potential third N-glycosylation site and the PKD domain (PMEL17-i-HA). HeLa cells were transfected with PMEL17-i-HA and then digested with or without peptide N-glycosidase F because M␣N contains N-glycosylation sites (Fig. 1B). We then analyzed the digests by immunoblotting with anti-HA antibody. We observed only P1 and P2 at shorter exposure times (Fig. 6C, right panel). However, with longer exposure times, several additional bands were detected at ϳ40 kDa, which is the estimated size of M␣N. Those bands were clearly sensitive to digestion by peptide N-glycosidase F (Fig. 6C, left  panel), and we conclude that M␣N is ϳ40 kDa. Because there are multiple bands, there might be multiple cleavage sites rather than a single one. The saturated bands at ϳ100 kDa should contain both M␣ and P1 (Fig. 1A), but we cannot make any conclusions about M␣. Similarly, we were unable to visualize M␣ clearly by pulse-chase labeling (Fig. 6B), so why is M␣ so difficult to detect? One interesting report showed that M␣ is buried in the Triton X-100-insoluble fraction (16). We used M-PER cell lysis buffer because HMB45 reactivity is stronger with this buffer than following extraction with Triton X-100 or Nonidet P-40 (27). We prepared soluble and insoluble fractions of MNT-1 cells solubilized with M-PER-, 1% Triton X-100-, or 1% Nonidet P-40-based lysis buffers and then immunoblotted these extracts using HMB45 or ␣-PEP13h antibody (Fig. 6D). In the soluble fractions, the reactivity of HMB45 was strongest with M-PER solubilization, as reported previously (27). Interestingly, in the insoluble fractions, which were not examined in the previous study (27), HMB45 reactivity was stronger following extraction with Triton X-100 or Nonidet P-40 and was weaker following extraction with M-PER reagent. In contrast, ␣-PEP13h antibody-detectable forms of PMEL17 (P2, P1, P1s, and M␤) were found mainly in the soluble fractions of all lysis buffers (Fig. 6D). From these results, it can be concluded that M␣ and M␣s were highly enriched in the Triton X-100-and Nonidet P-40-insoluble fractions. Moreover, their derivatives (M␣C, M␣C3, M␣C2, and M␣C1) were also enriched in the Triton X-100-and Nonidet P-40-insoluble fractions. The exact composition of M-PER reagent has not been released, but HMB45-reactive materials are more efficiently solubilized by M-PER reagent. Finally, we tried to clarify the N-terminal cleavage site(s) of M␣C and M␣Cs, including M␣C1, M␣C2, and M␣C3. According to the immunoblotting of ⌬NTD using HMB45, M␣C was detected (Fig.  3B, left panel), showing that the NTD is not involved. Immunoblotting of RPTb-j (deletion of repeat a in the RPT domain) using HMB45 showed faster migration of M␣C1 and M␣C2 (Fig. 6A), so the RPT domain is also not involved. HMB45 did not detect M␣C in ⌬PKD or ⌬GAP1 (Fig. 3B, left panel), although the levels of M␤ were very low in these extracts, and perhaps it did not undergo cleavage. Taken together, the N-terminal cleavage site(s) of PMEL17 should be in ⌬PKD and/or in ⌬GAP1, but cannot be further identified from these results. Thus, it is clear that M␣ is processed into M␣N (ϳ40 kDa) and M␣C (ϳ60 kDa), after which M␣C is further processed into M␣C1 or M␣C3 (M␣C2 in the case of PMEL17-is or PMEL17-ls).

DISCUSSION
In this study, we have demonstrated that HMB45 reacts with the RPT domain of PMEL17. HMB45 was initially developed for melanoma detection, but was later shown to react with PMEL17 and melanosomal matrix fibers (12)(13)(14). PMEL17 can produce fibrillar structures even following expression by transfection in non-melanocytic cells (12). Combining previously published results with those of this study, independent lines of evidence can now be linked at the subdomain FIGURE 6. HMB45 recognizes the second and third repeats in the RPT domain. A, HeLa cells overexpressing full-length PMEL17-i or PMEL17-i containing nine (RPTa-i or RPTb-j), eight (RPTa-h or RPTc-j), seven (RPTa-g or RPTd-j), six (RPTa-f or RPTe-j), five (RPTa-e), four (RPTa-d), three (RPTa-c), or two (RPTa-b) repeats in the RPT domain were analyzed by immunoblotting with HMB45. HeLa cells overexpressing PMEL17-is, RPTsa-i (deletion of repeat j in the RPT domain of PMEL17-is), RPTsa-h (deletion of repeats i and j), PMEL17-is-⌬CS, or PMEL17-ls-⌬CS were also analyzed. B, HeLa cells overexpressing PMEL17-i or ⌬GAP3 were metabolically pulsed with 35 S for 30 min and then chased for 4 h. Extracts were immunoprecipitated (IP) with ␣-mSiN or ␣-PEP13h antibody, separated by electrophoresis, and visualized by autoradiography. C, HeLa cells overexpressing PMEL17-i containing an HA tag in the NTD were digested with or without peptide N-glycosidase F (PNGaseF). Cell lysates were analyzed by immunoblotting. Films were developed after relatively long (L) or short (S) exposure times. The numbers to the right indicate sizes in kilodaltons, and the specific bands as discussed under "Results" are indicated to the right. D, MNT-1 cells were solubilized with M-PER-, 1% Nonidet P-40-, or 1% Triton X-100-containing lysis buffer. Soluble (S) and insoluble (I ) fractions were analyzed by immunoblotting using HMB45 (upper panel ) and ␣-PEP13h antibody (lower panel ).

The Repeat Domain of PMEL17 Is Required for Fibrillogenesis
level. The fibrillar structures found in the Stage II melanosomal matrix correspond to the RPT domain of PMEL17, and the RPT domain is O-glycosylated with sialic acid termination. Moreover, the RPT domain undergoes variable proteolytic cleavage. The LAMP-1, LAMP-2, CD68, and NMB proteins show high homology to PMEL17, especially NMB, which shows the closest homology to PMEL17; but the other proteins lack the RPT domain (25). NMB is expressed in melanocytic cells (37), but it is also expressed in gliomas and in cancers of the breast, stomach, and pancreas, in which no melanosomal fibrillar structures have been reported (38). Truncation of the RPT domain (PMEL17-is or PMEL17-ls) does not affect the trafficking of PMEL17 or the morphology of the fibrillar structures (21). These results suggest that the RPT domain is important for proper fibrillar structure, but has some redundancy.
Recombinant M␣ reconstituted from Escherichia coli inclusion bodies has been shown very recently to have the characteristics of amyloid (39). Typical amyloid fibers were observed that could bind melanogenic precursors and hasten their polymerization, even in a prokaryotic system. PMEL17-i, ⌬KRG, ⌬GAP3, and ⌬CTD (all of which have an intact M␣ fragment) showed similar trafficking and similar processing that could be detected by HMB45 (Figs. 3B and 6). The results of this study also suggest that the M␣ fragment is important for amyloidogenesis. M␣ seems to be a minimum requirement for amyloidogenesis of PMEL17 and binding of melanogenic precursors. An important question remaining is what is the role of O-glycosylation in the RPT domain? O-Glycosylation plays important roles in the attachment and invasion of cancers, in the maintenance of protein conformations, and in the immune system (40). However, O-glycosylation does not seem to be required for the formation of fibrils (39). From this point of view, O-glycosylation in the RPT domain might play some role in its ability to bind melanogenic precursors, and we are currently trying to clarify this issue.
The trafficking routes PMEL17 uses to get to early melanosomes are diverse (12,14,16,25,27). One proposed pathway is that PMEL17 is trafficked from the ER or cis-Golgi to Stage I melanosomes and then is further processed concurrent with the formation of the fibrillar matrix seen in Stage II melanosomes (14,27). This route is supported by the fact that subcellular fractions of human melanoma cells enriched in Stage II melanosomes contain P1, but not P2 (14). Proteomic analysis of those fractions revealed ER-resident proteins (35). The endoglycosidase H-sensitive form of PMEL17 (P1) was detected, but the largely endoglycosidase H-resistant form (P2) could not be detected by immunoblotting of melanoma cell lysates (27). Another proposed sorting route is that PMEL17 is trafficked from the TGN to early endosomes and Stage I melanosomes and then finally to Stage II melanosomes (12). This is supported by the fact that P2 and M␣ are largely endoglycosidase H-resistant, which means that N-acetylglucosaminyltransferase in the medial-Golgi is involved (12,25). In a recent study (28), we showed that both trafficking pathways are in fact involved in the sorting of PMEL17. Both the major (P1) and minor (P2) forms of PMEL17 are delivered to early melanosomes, but via distinct sorting pathways. In this present study, we have demonstrated that it is only the minor late Golgi-modified form (P2) that is critical to the formation of fibrils in melanosomes. What function(s) the major ER-modified form of PMEL17 (P1) may play remains open.
This study has also shown that the HMB45-reactive RPT domain of PMEL17 is O-glycosylated with sialic acid modifications. Indeed, sialic acid modifications for N-glycans occur in the trans-Golgi or TGN (31). However, sialyltransferases responsible for the sialylation of O-glycans are located throughout the entire Golgi stack, so one cannot conclude from those results alone whether proteins have gone through the late Golgi (41,42). Interestingly, M␣ detected by HMB45 was largely endoglycosidase H-resistant and peptide N-glycosidase F-sensitive (data not shown), suggesting that the HMB45-reactive form of PMEL17 has been modified in the late Golgi. P2 and M␣ are transient forms and are usually difficult to detect, especially PMEL17 expressed endogenously in melanocytic cells (Figs. 3A and 4A). Moreover, M␣ is buried in the detergent-insoluble fraction, and its molecular mass is quite similar to that of P1 (Figs. 3A and 4A) (16). Thus, failure to detect P2 or M␣ does not directly suggest that only P1 (but not P2) will be trafficked to melanosomes. M␣ and its derivatives can be clearly identified in M-PER reagent-soluble fractions using the sialylation-dependent antibody HMB45. The specificity of HMB45 allows it to recognize only M␣ and P2, whereas the specificity of ␣-PEP13h antibody allows it to recognize only P1 and P2; thus, M␣ and P1 can be readily distinguished with those two reagents. Derivatives of M␣ (such as M␣C1 and M␣C3) that are detected by HMB45 indicate the existence of M␣ and P2. It is also clear that HMB45 recognizes the fibrillar matrix in Stage II melanosomes at the ultrastructural level (12)(13)(14). Taken together, the fragment of PMEL17 found in Stage II melanosomes recognized by HMB45 is late Golgi-modified and must be trafficked from the TGN.
Is the proposal that P1 is trafficked from the ER or cis-Golgi to early melanosomes correct? For a long time, it has been believed that melanosomes mature from Stages I to IV (11). When considered with other published studies, our results may raise the question, "Do Stage I melanosomes really mature to Stage II melanosomes?" The difference between Stage I and II melanosomes depends completely on the presence of the fibrillar structure (11). We reported recently that MART-1-negative but PMEL17-positive melanoma cells produce Stage I (but not Stage II) melanosomes (15). P1 was detected in subcellular fractions enriched in Stage I melanosomes, but HMB45 was negative in all subcellular fractions as well as in whole cell lysates. This suggests that P1, which is the major form of PMEL17, can be trafficked to Stage I melanosomes, but is not required for the generation of fibrils involved in the maturation to Stage II melanosomes. In contrast, the minor late Golgi-modified form of PMEL17 (P2) is trafficked from the TGN to Stage II melanosomes. From these points of view, both hypotheses for the trafficking of PMEL17 are compatible and convincing. PMEL17 is also secreted, a phenomenon not seen for TYR (tyrosinase), TYRP1 (tyrosinase-related protein 1), or dopachrome tautomerase in normal melanocytes (21). Thus, it is reasonable that multiple forms of PMEL17 might take different trafficking routes, although careful investigation is required to identify them (28).
By clarifying the characteristics of HMB45 recognition of PMEL17 and its fragments, our study has shed light on the trafficking as well as the processing of PMEL17 linked to early melanogenesis. Is late Golgi-modified PMEL17 sufficient to produce Stage II melanosomes? In this study, we focused only on PMEL17 and HMB45 using transfection of non-melanocytic cells. In recent studies, mice harboring mutations in pigmentrelated genes other than PMEL17 produced fibrillar structures; however, the numbers were few, and/or the shapes were not ellipsoidal (43,44). Such factors should be considered for a better understanding of early melanogenesis.
In this study, we have shown that M␣ is processed to M␣N and M␣C and that M␣C is then further processed because of the variable proteolytic cleavage of the RPT domain. Where does such processing occur? Processing of M␣ was observed for ⌬NTD, which was not trafficked to LAMP-2-positive organelles (Figs. 3B and 6). The cleavage of P2 to M␣/M␤ has been reported to occur in post-Golgi and prelysosomal compartments (16). This suggests that the processing of M␣ to M␣C/M␣N might occur in post-Golgi and/or prelysosomal compartments. However, such processing might occur independently because ⌬PKD and ⌬GAP1 showed some cleavage of P2 to M␣/M␤, but did not show any cleavage of M␣ to M␣C/ M␣N (Fig. 3B). M␣C1 and M␣C3 were observed only in PMEL17-i-, ⌬KRG-, ⌬GAP3-, and ⌬CTD-transfected cells, and all of them were efficiently trafficked to late endosomes or lysosomes (Figs. 3B and 6). ⌬NTD was localized mainly in the perinuclear area, but could produce M␤ and M␣C (Figs. 3A and 6). Taken together, the variable proteolytic cleavage of the RPT domain probably occurs in late endosomes or lysosomes, in other words, in melanosomes. Which enzymes are involved is another interesting issue, and future work will attempt to identify them.
The function of the KRG domain in PMEL17 is currently unknown. M␤ was undetectable in ⌬KRG-transfected cells by ␣-PEP13h antibody (Fig. 3B, right panel). Accordingly, M␣ detected in ⌬KRG-transfected cells by HMB45 was significantly weaker than that detected in ⌬GAP3-, ⌬CTD-, or PMEL17-i-transfected cells (Fig. 3B, left panel). Interestingly, ⌬KRG trafficked correctly to LAMP-2-positive organelles (Fig.  4). Recently, the kringle-2 domain of tissue plasminogen activator was shown to be necessary for the cleavage of plateletderived growth factor CC at its tribasic processing site, 231 RKSR (45). Moreover, the kringle-2 domain of tissue plasminogen activator directly binds platelet-derived growth factor CC. The enzyme that cleaves P2 into M␣/M␤ has not yet been specifically identified (16). However, it is possible that the KRG domain is necessary for efficient processing by that enzyme.
An unexpected and interesting result was found for ⌬CTD. ⌬CTD showed no detectable disruption in the processing or trafficking of PMEL17 in transfected cells (Figs. 3B and 6). Many melanosomal proteins have putative sorting signals in their cytoplasmic domains (46). PMEL17 has a putative dileucine-based signal (ENSPLL) in its CTD. Equivalent dileucine-based sorting signals are found in TYRP1, TYR, and QNR-71 (the quail homolog of NMB), and mutations of those signals cause misrouting of those proteins (36,(47)(48)(49). Surprisingly, this may not hold true for PMEL17 because deletion of the CTD or mutation of the dileucine motif in PMEL17 did not disrupt sorting to late endosomes or lysosomes in transfected non-melanocytic cells. Whether the CTD plays a role in the sorting of PMEL17 in melanocytic cells is an important question to resolve, as is the identification of the precise signal(s) involved in the luminal region. The sorting determinants found in luminal domains of PMEL17 (such as NTD, PKD, GAP1, and GAP2) seem to be critical to the regulation of PMEL17 trafficking. This might account for the distinct nature of PMEL17 trafficking in melanocytes, which is different from the sorting of other melanosomal proteins such as TYR, TYRP1, and dopachrome tautomerase (14,28). The PKD and GAP1 domains seem to be required for earlier events in the secretory pathway, whereas the NTD and GAP2 domain seem to be required for later events. Probably, they are involved in distinct sorting mechanisms. Among those domains, the PKD domain is the only conserved domain, and it might be involved in proteinprotein and protein-carbohydrate interactions (50), although the function of the PKD domain is still unresolved at this time. This is a novel finding and a quite intriguing issue and will require further study. Future investigation will be directed to further identify sorting components that might utilize the luminal domains of PMEL17.