The Kringle-like Domain Facilitates Post-endoplasmic Reticulum Changes to Premelanosome Protein (PMEL) Oligomerization and Disulfide Bond Configuration and Promotes Amyloid Formation*

The formation of functional amyloid must be carefully regulated to prevent the accumulation of potentially toxic products. Premelanosome protein (PMEL) forms non-toxic functional amyloid fibrils that assemble into sheets upon which melanins ultimately are deposited within the melanosomes of pigment cells. PMEL is synthesized in the endoplasmic reticulum but forms amyloid only within post-Golgi melanosome precursors; thus, PMEL must traverse the secretory pathway in a non-amyloid form. Here, we identified two pre-amyloid PMEL intermediates that likely regulate the timing of fibril formation. Analyses by non-reducing SDS-PAGE, size exclusion chromatography, and sedimentation velocity revealed two native high Mr disulfide-bonded species that contain Golgi-modified forms of PMEL. These species correspond to disulfide bond-containing dimeric and monomeric PMEL isoforms that contain no other proteins as judged by two-dimensional PAGE of metabolically labeled/immunoprecipitated PMEL and by mass spectrometry of affinity-purified complexes. Metabolic pulse-chase analyses, small molecule inhibitor treatments, and evaluation of site-directed mutants suggest that the PMEL dimer forms around the time of endoplasmic reticulum exit and is resolved by disulfide bond rearrangement into a monomeric form within the late Golgi or a post-Golgi compartment. Mutagenesis of individual cysteine residues within the non-amyloid cysteine-rich Kringle-like domain stabilizes the disulfide-bonded dimer and impairs fibril formation as determined by electron microscopy. Our data show that the Kringle-like domain facilitates the resolution of disulfide-bonded PMEL dimers and promotes PMEL functional amyloid formation, thereby suggesting that PMEL dimers must be resolved to monomers to generate functional amyloid fibrils.

Amyloid is a "cross-␤-sheet" polymeric protein conformation in which ␤-strands are stacked perpendicular to the long axis of the amyloid fibril (1). The amyloid fold is traditionally associated with pathological protein misfolding in neurodegenerative diseases such as Alzheimer and Parkinson diseases (2). However, the amyloid fold has also been exploited for diverse physiological processes in a wide range of organisms (3) including biofilm formation in bacteria (4), learning and memory in Drosophila (5), and programmed necrosis in mammals (6). The molecular mechanisms that limit the toxicity of such "functional amyloids" or their folding intermediates during synthesis are incompletely understood but likely regulate when and where these proteins assemble into amyloid fibrils. For example, curli biofilm formation in Escherichia coli is regulated by the accessory proteins CsgB, CsgC, and CsgF. CsgB is a minor curli subunit required to nucleate amyloid fibril formation in vivo (7). CsgC is a chaperone protein that prevents premature amyloid formation in the bacterial periplasm (8), and CsgF is an extracellular protein important for the attachment of curli fibers to the cell surface (9). However, much less is known about functional amyloid regulation in mammals, where the formation of large, self-assembling, insoluble protein structures likely carries increased risk because of the multicellular nature and longer life span of higher organisms.
The best characterized mammalian functional amyloid protein is premelanosome protein (PMEL 2 ; also referred to as Pmel17, Silver, ME20, or gp100). Within immature melanosomes of pigment cells in the skin and eye, PMEL polymerizes into amyloid fibrils (10) that associate laterally into sheets upon which melanins are deposited as they are synthesized during melanosome maturation (reviewed in Ref. 11). The amyloid sheets are critical determinants of the ellipsoid melanosome shape (12,13), which is required for proper melanosome motility into the apical processes of retinal pigment epithelial cells (14). The amyloid sheets have also been proposed to accelerate melanin polymerization (10,15,16), and organisms that lack PMEL or carry mutations in the PMEL gene are characterized by various degrees of hypopigmentation (13,(17)(18)(19)(20)(21). Because PMEL is synthesized in the endoplasmic reticulum (ER) as a type I transmembrane protein (22)(23)(24) but only initiates amyloid fibril formation within the lumen of endosomal membrane compartments (25)(26)(27), PMEL must navigate the secretory pathway from the ER to endosomes in a non-amyloid form. It is thus an excellent model system by which to dissect the regulation of amyloid formation within the endomembrane system.
Our current understanding of the processes involved in PMEL biosynthesis and amyloidogenesis is as follows. Following synthesis and addition of four core N-linked glycans within the ER (generating a "P1" precursor form), PMEL is exported to the Golgi where it is modified extensively by O-glycosylation (28,29) and maturation of the N-linked glycans to the complex type (26). This generates a full-length "P2" form that is subsequently cleaved by a proprotein convertase into two disulfidelinked fragments, M␣ and M␤ (30), either in the trans-Golgi network (31) or within endosomes (32) (see Fig. 1A). A subsequent proteolytic cleavage by BACE2 within endosomes (33) or by an ADAM (a disintegrin and a metalloproteinase) protease at the plasma membrane (34) cleaves M␤ into two fragments; the C-terminal fragment (CTF) contains the transmembrane domain and is eventually degraded in a ␥-secretase-dependent manner in lysosomes (35), whereas the other fragment (M␤Ј) is liberated from the membrane covalently bound to M␣ (33,36). No longer anchored to the membrane, most of the M␣ polymerizes into amyloid fibrils within endosomes (30,37) and concomitantly or subsequently undergoes additional proteolytic processing events to form smaller fragments; some of these fragments, called M␣Ј, are derived from the central RPT domain and are detected within preparations of the parallel arrays of functional amyloid fibrils characteristic of early stage melanosomes (27,29,38). Proprotein convertase and BACE2 cleavage are necessary for functional amyloid fibril formation, as interference with either of these two proteolytic processing events results in the formation of disordered PMEL aggregates instead of ordered fibrils (30,33). However, these cleavage events are not sufficient to induce fibril formation because a small fraction of cleaved PMEL is secreted from cells in culture (26,36,39) in a non-amyloid form. 3 Therefore, additional regulatory mechanisms must govern PMEL fibril formation.
Regulated disulfide bond rearrangements have the potential to affect both protein conformation and protein function. For example, interaction of the ␤-barrel protein LptD with the LptE subunit of the LPS translocon in the E. coli outer membrane triggers a disulfide bond rearrangement within LptD that activates the translocon (40). PMEL has 13 cysteine residues (Fig.  1A, black circles), of which 11 are exposed to the lumen of the endomembrane system and thus might participate in disulfide bonds. Indeed, the existence of intermolecular disulfide bonds can be inferred from data showing that P2 and nonfibrillar M␣ and M␤ fragments are incorporated into higher order oligomers when samples are analyzed by non-reducing SDS-PAGE (26). Here we have shown that disulfide bonds maintain PMEL in a tethered dimeric form and that resolution of the dimer requires the integrity of a cysteine-rich Kringle-like domain (KLD) that is outside of the amyloidogenic region. Mutagenesis of this region blocks dimer resolution and impairs amyloidogenesis. Our data support a model in which the generation and destruction of a disulfide-stabilized dimer contributes an additional layer of regulation to the timing of PMEL amyloid formation.

Antibodies and Reagents
Unless otherwise specified, chemicals were obtained from Sigma-Aldrich and tissue culture reagents from Life Technologies. Protease inhibitors were purchased from Roche Diagnostics, gene amplification primers from Integrated DNA Technologies (Coralville, IA), GoTaq DNA polymerase from Promega Corp. (Madison, WI), and restriction enzymes and T4 DNA ligase from New England Biolabs (Ipswich, MA). HMB45, a mouse monoclonal antibody to PMEL, was purchased from Enzo Life Sciences (catalogue No. ENZ-C34930; Farmingdale, NY). Rabbit polyclonal antibodies PMEL-I, PMEL-N, and PMEL-C were affinity-purified as described previously (26,29,41). The mouse monoclonal antibody NKI-beteb was obtained from Lab Vision (Fremont, CA), and H4A3 to LAMP1 (lysosome-associated membrane protein 1) was from the Developmental Studies Hybridoma Bank (University of Iowa, Iowa City). Secondary antibodies conjugated to Alexa Fluor 488 or Alexa Fluor 594 for immunofluorescence microscopy or to alkaline phosphatase for immunoblotting were obtained from Jackson ImmunoResearch Laboratories (West Grove, PA). Secondary antibodies conjugated to IRDye 800CW or IRDye 680LT for immunoblotting were obtained from LI-COR Biotechnology (Lincoln, NE).

DNA Constructs and Cloning
Wild-type hPMEL (long form) and the cleavage site mutant ⌬CS in the pCI vector (Promega) have been described previously (26,30). The cysteine mutants were generated by sitedirected mutagenesis using two-step gene amplification as described (42), and plasmid inserts were verified by DNA sequencing. The sequences of primers used for gene amplification and for the production of all mutants are available upon request. at 19,000 ϫ g for 20 min, 1 volume of 6ϫ SDS sample buffer (0.4 M Tris, pH 6.8, 12% (w/v) SDS, 34% glycerol, 0.02% (w/v) bromphenol blue) with or without 11% ␤-mercaptoethanol was added to 5 volumes of detergent-soluble lysate. Samples were then heated at 60°C for 20 min, fractionated by SDS-PAGE on Tris-glycine gels containing 7, 10, or 12% acrylamide, and transferred electrophoretically to PVDF membranes. Membranes were blocked using blocking buffer (TBS, pH 7.4, and 0.2% Tween 20 with 5% (w/v) nonfat dry milk) or Odyssey blocking buffer (LI-COR Biotechnology), incubated with primary antibodies diluted in blocking buffer, and washed with TBST (TBS, pH 7.4, 0.2% Tween 20). In some experiments, membranes were subsequently incubated with secondary antibodies conjugated to alkaline phosphatase diluted in blocking buffer, washed with TBST, detected using enhanced chemifluorescence (GE Healthcare Life Sciences), and analyzed using a Molecular Dynamics Storm 860 Molecular Imager Phosphor-Imager (GE Healthcare Life Sciences). In other experiments, membranes were probed using secondary antibodies conjugated to IRDye-800CW or IRDye-680LT diluted in blocking buffer, washed with TBST, and analyzed using an Odyssey infrared imaging system (LI-COR Biotechnology). Quantification of bands was performed using the ImageJ gel analyzer tool (National Institutes of Health) or Image Studio Lite (LI-COR).

Sucrose Gradient Sedimentation
Cells treated with PBS, 50 mM NEM were lysed in fractionation buffer (150 mM sodium chloride, 0.02% (w/v) sodium azide, 100 mM Tris, pH 7.4, supplemented with protease inhibitor tablets, 50 mM NEM, and 20 mM iodoacetamide) containing 250 mM n-octylglucoside. The lysate was clarified by centrifugation at 19,000 ϫ g for 15 min, layered on top of 5-20% sucrose gradients in 150 mM sodium chloride, 0.02% (w/v) sodium azide, 25 mM n-octylglucoside, 100 mM Tris, pH 8, and fractionated by sedimentation velocity as described (43). Briefly, gradients were subjected to ultracentrifugation in a Beckman L8 -70 M Ultracentrifuge (Beckman Coulter Inc.) for 38 h at 30,000 rpm in an SW-41 rotor. Fractions were collected from bottom to top using a peristaltic pump, and aliquots of each fraction were analyzed by non-reducing SDS-PAGE and immunoblotting. Standard proteins (ovalbumin, rabbit IgG, bovine liver catalase, and/or bovine thyroid thyroglobulin) were fractionated on parallel gradients; these fractions were analyzed by reducing SDS-PAGE followed by staining with Coomassie Brilliant Blue R-250.

Size Exclusion Chromatography
Cells were lysed in fractionation buffer containing either 250 mM n-octylglucoside, 40 mM dodecyl-␤-D-maltoside, or 4% (w/v) Triton X-100. After clarification at 19,000 ϫ g for 15 min, detergent-soluble lysates were analyzed by size exclusion chromatography on an ÄKTAFPLC system (GE Healthcare Life Sciences) using a Superose 6 10/300 GL column (GE Healthcare Life Sciences) in running buffer (150 mM sodium chloride, 0.02% (w/v) sodium azide, 10 mM HEPES, pH 7.4) containing a lower concentration of the same detergent used for lysis, either 25 mM n-octylglucoside, 1 mM dodecyl-␤-D-maltoside, or 0.1% (w/v) Triton X-100. The column was eluted at a flow rate of 0.4 -0.45 ml/min with a fraction volume of 300 or 600 l. Aliquots of each fraction were analyzed by non-reducing SDS-PAGE and immunoblotting. The elution volumes of thyroglobulin (669 kDa), ␣ 2 -macroglobulin (720 kDa), rabbit IgG (150 kDa), ovalbumin (45 kDa), and/or cytochrome C (12 kDa) were also evaluated in separate runs with fractions analyzed by SDS-PAGE followed by staining with Coomassie Brilliant Blue R-250.

Molecular Weight Calculations
Molecular weights were calculated using the formula mass ϭ (6Nas 20,w 20,w )/(1-20,w ), where N is Avogadro's number, a is the Stokes radius as measured by size exclusion chromatography, s 20,w is the sedimentation coefficient at 20°C in water, 20,w is the viscosity of water at 20°C, is the partial specific volume (see below), and 20,w is the density of water at 20°C (44). We assumed that p250 and p160 associate with one detergent micelle each and that M␣ϩM␤Ј and M␣ do not associate with detergent at all. Octylglucoside micelle size was estimated to be 25 kDa (45). The partial specific volume of each species was estimated by summing the partial specific volumes of each amino acid, the N-linked glycan of the high mannose or complex type, 20 kDa of O-linked glycans for each RPT domain, and the octylglucoside detergent micelle for those species in which it was warranted (43).

Metabolic Labeling and Immunoprecipitation
Metabolic labeling with [ 35 S]methionine/cysteine and immunoprecipitation analyses were conducted essentially as described (26). Briefly, cells were harvested by centrifugation following digestion with trypsin-EDTA, washed, and cultured in suspension for 30 min in methionine/cysteine-free DMEM containing 5% dialyzed FBS. Cells were then pelleted and labeled metabolically in suspension in the same medium containing 35 S-Met/Cys Express mix (PerkinElmer Life Sciences) for 10 -15 min (pulse-chase analyses) or 2 h (steady-state labeling for two-dimensional PAGE). For pulse-chase analyses, cells were pelleted again and chased for the indicated periods of time in standard growth medium containing excess methionine/cysteine as described (26). Cells were then lysed in lysis buffer and clarified by centrifugation at 19,000 ϫ g for 20 min. After first preclearing the lysates over unconjugated protein A beads (Life Technologies) and then over protein A beads conjugated to normal rabbit serum, normal mouse serum, or a mouse isotype control antibody, the samples were subjected to immunoprecipitation using the specified antibodies. The immunoprecipitated material was fractionated by SDS-PAGE, and the resulting gels were dried and analyzed on a PhosphorImager as described above.

Two-dimensional Non-reducing/Reducing SDS-PAGE
Triton X-100 lysates from cells metabolically labeled for 2 h were subjected to immunoprecipitation using the PMEL antibodies ␣PMEL-N, NKI-beteb, or ␣PMEL-C after preclearing first with unconjugated protein A beads and then with protein A beads conjugated to either normal rabbit serum or normal mouse serum as described above. The immunoprecipitated material was then fractionated by SDS-PAGE in a 7% polyacryl-amide tube gel under non-reducing conditions. The tube gel was reduced by incubation in 0.5% (w/v) DTT, 0.5% (w/v) SDS, 125 mM Tris, pH 6.8, for 2 h at room temperature and then mounted and sealed (with 1% agarose) on top of a standard SDS-polyacrylamide gel. After fractionation by SDS-PAGE, the gels were dried and analyzed on a PhosphorImager as detailed above.

Immunoprecipitation/Immunoblotting
HeLa cells expressing wild-type hPMEL or the C301S variant were released from plates using PBS, 5 mM EDTA and harvested by centrifugation. The cells were treated with PBS, 50 mM NEM on ice, pelleted, frozen, and subsequently lysed (ϳ3 ϫ 10 6 cells/ ml) in lysis buffer. Following clarification at 19,000 ϫ g, the lysates were precleared over unconjugated protein A beads prior to immunoprecipitation using NKI-beteb, ␣PMEL-C, normal rabbit serum, or a mouse isotype control antibody prebound to protein A beads. The beads were washed four times with wash buffer (0.1% (w/v) Triton X-100, 150 mM sodium chloride, 0.02% (w/v) sodium azide, 10 mM HEPES, pH 7.4) followed by one wash with PBS. The immunoprecipitated material was then eluted by adding 2ϫ SDS sample buffer containing 3.6% ␤-mercaptoethanol and heating the samples for 20 min at 60°C. After fractionation by SDS-PAGE and transfer to PVDF membranes, immunoblots were probed with ␣PMEL-C or HMB45 and analyzed using the Odyssey imaging system detailed above.

Proteomic Analysis of p250 and p160
Affinity Purification-Pellets of 2.6 -7.8 ϫ 10 8 NEM-treated MNT-1 cells were either homogenized with a Dounce homogenizer and enriched for the total membrane fraction as described previously (30) or lysed directly in lysis buffer and clarified by centrifugation at 19,000 ϫ g for 15 min. The detergent-soluble fraction was then loaded onto three columns in tandem containing Sepharose 4B (Sigma), rabbit IgG-conjugated agarose (Sigma), and ␣PMEL-C-conjugated Sepharose (generated by combining 3 ml of 1 mg/ml ␣PMEL-C with 3 ml of packed CNBr-activated Sepharose 4B from GE Healthcare Life Sciences according to manufacturer's instructions). The ␣PMEL-C column was subsequently washed with 600 ml of wash buffer and eluted using 0.63 g/ml PMEL-C peptide (CPI-GENSPLLSGQQV, Peptide 2.0 Inc., Chantilly, VA). Eluate was collected in 0.5-1-ml fractions, and aliquots were analyzed by immunoblotting for PMEL content. Peak fractions were precipitated using a 5-fold volume of either methanol or acetone at Ϫ20°C. The concentrated eluate was then fractionated on a 3-8% Tris acetate protein gel (Life Technologies) and stained with Coomassie Brilliant Blue G-250.
In-gel Digestion-Each sample was excised from the gel and cut into 1-mm cubes, destained in a solution containing 50% methanol, 1.25% acetic acid, reduced with 5 mM DTT (Thermo Fisher Scientific), and alkylated with 20 mM iodoacetamide (Sigma). Gel pieces were then washed with 20 mM ammonium bicarbonate (Sigma) and dehydrated with acetonitrile (Thermo Fisher Scientific). Trypsin (Promega; 5 ng/ml in 20 mM ammonium bicarbonate) was added to the gel pieces, and proteolysis was allowed to proceed overnight at 37°C. Peptides were extracted with 0.3% trifluoroacetic acid (J. T. Baker, Avantor Performance Materials, Center Valley, PA) followed by 50% acetonitrile. The extracts were combined, and the volume was reduced by vacuum centrifugation.
Mass Spectrometry Analysis-Tryptic digests were analyzed by LC-MS/MS on a hybrid LTQ Orbitrap Elite mass spectrometer (Thermo Fisher Scientific) coupled with a nanoLC Ultra (Eksigent Technologies Inc., Dublin, CA). Peptides were separated by reverse phase-HPLC on a nanocapillary column (75 m ϫ 15 cm Reprosil-pur 3 m, 120 Å (Dr. Maisch, Ammerbuch, Germany)) in a Nanoflex chip system (Eksigent Technologies). Mobile phase A consisted of 1% methanol (Thermo Fisher Scientific) and 0.1% formic acid (Thermo Fisher Scientific) and mobile phase B of 1% methanol, 0.1% formic acid, and 80% acetonitrile. Peptides were eluted into the mass spectrometer at 300 nl/min with each reversed phase-HPLC run comprising a 90-min gradient from 10 to 25% B in 65 min and then 25 to 40% B in 25 min. The mass spectrometer was set to repetitively scan m/z from 300 to 1800 (r ϭ 240,000 for LTQ-Orbitrap Elite) followed by data-dependent MS/MS scans on the 20 most abundant ions with a minimum signal of 1500, dynamic exclusion with a repeat count of 1, repeat duration of 30 s, exclusion size of 500, and duration of 60 s, isolation width of 2.0, normalized collision energy of 33, and waveform injection and dynamic exclusion enabled. The Fourier transform mass spectroscopy full-scan automatic gain control target value was 1E6, and the MSn automatic gain control was 1E4. Fourier transform mass spectroscopy full-scan maximum fill time was 500 ms, and ion trap MSn fill time was 50 ms; microscans were set at 1. The FT preview mode, charge state screening, and monoisotopic precursor selection were all enabled with rejection of unassigned and 1ϩ charge states.
Protein Identification-MaxQuant/Andromeda version 1.5.2.8 was used to identify proteins by searching against a human complete proteome sequence database downloaded from Uni-Prot on July 14, 2014. The MaxQuant search parameters are detailed in the supplemental Table S1 (see parameters and summary tabs), and the identified protein groups and their estimated MS1 summed spectral intensities for each gel slice are listed in the proteinGroups tab of Supplemental Table S1.

Immunofluorescence Microscopy
Cells grown on coverslips were fixed with 2% formaldehyde in PBS, permeabilized and labeled with primary antibodies and fluorochrome-conjugated secondary antibodies in PBS, 0.2% (w/v) saponin, 0.1% (w/v) BSA as described previously (26), and analyzed on either a DM IRBE microscope (Leica Microsystems, Wetzlar, Germany) equipped with a ϫ100 Plan Apochromat objective lens (1.4 N.A., Leica), a Retiga Exi Fast 1394 digital camera (QImaging, Surrey, Canada), and OpenLab software (Improvision, Lexington, MA) or a DMI 6000B microscope (Leica Microsystems) equipped with the same ϫ100 objective, an Orca Flash 4.0 V2 camera (Hamamatsu, Bridgewater, NJ), and Leica Application Suite X software (Leica Microsystems). Images captured in sequential z-planes separated by 0.2 m were deconvolved and manipulated using the Volume Deconvolution module in OpenLab or Gold's three-dimensional deconvolution algorithm with three iterations in the Leica Application Suite. Insets were magnified and further adjustments in brightness and contrast were performed using Adobe Photoshop (Adobe Systems, Mountain View, CA).

Electron Microscopy
For electron microscopy, HeLa cells transfected with pCI-PMEL wild type, C301S, or C566S with or without pEGFP-C1 (Clontech Laboratories) in a 9:1 ratio were grown in 10-cm dishes and either fixed in situ, high-pressure frozen, or sorted for GFP expression prior to fixation in suspension. For fixation in situ, cells were incubated in Karnovsky's fixative (4% paraformaldehyde, 4 mM calcium chloride, 72 mM sodium cacodylate, pH 7.4) containing 0.5% glutaraldehyde for 1-2 h. This solution was then removed and replaced with Karnovsky's fixative containing 2% glutaraldehyde, and the cells were fixed overnight at room temperature. After the cells were scraped using a rubber policeman, the overnight fixative was replaced with Karnovsky's fixative containing 0.5% glutaraldehyde and the cells stored at 4°C. For high-pressure freezing, cells were released from plates using PBS, 5 mM EDTA, pelleted by centrifugation, and frozen under high pressure in an HPM 010 machine (ABRA Fluid, Widnow, Switzerland) at the University of Pennsylvania Electron Microscopy Resource Laboratory. Sample dehydration by freeze substitution was performed in a Leica AFSII (Leica Microsystems) over a period of 72 h at Ϫ90°C in glassdistilled acetone supplemented with 0.1% uranyl acetate and 2% OsO 4 . Cells were then embedded in EPON resin at room temperature followed by polymerization for 48 h at 60°C. HeLa cells sorted prior to fixation were co-transfected with pEGFP, released from plates using PBS and 5 mM EDTA, and sorted for GFP expression on a Moflo Astrios EQ (Beckman Coulter Inc.) by the Flow Cytometry Core Laboratory (Children's Hospital of Philadelphia). Similar to the cells sorted in situ, these sorted cells were fixed for 1-2 h in Karnovsky's fixative containing 0.5% glutaraldehyde, overnight in Karnovsky's fixative containing 2% glutaraldehyde, and stored in Karnovsky's fixative with 0.5% glutaraldehyde at 4°C. Both the cells fixed in situ and the cells sorted prior to fixation were subsequently dehydrated and embedded in EMbed-812 (Electron Microscopy Sciences, Fort Washington, PA) at the Electron Microscopy Resource Laboratory. Ultrathin sections were cut on either a Reichert-Jung Ultracut E or a Reichert Ultracut S microtome (now Leica Microsystems), placed on either uncoated copper mesh grids or Formvar-coated copper slot grids, and stained with 50% ethanol, 1% uranyl acetate, 1% lead citrate as described previously (46). The specimens were then examined with a JEOL 1010 electron microscope fitted with a Hamamatsu C4742-95 digital camera (Hamamatsu, Bridgewater, NJ) and AMT Advantage image capture software (Advanced Microscopy Techniques, Woburn, MA). For quantification of fibril-containing organelles, a combined total of at least 30 cells from the three independent experiments were analyzed for each transfected cell type. The number of multivesicular bodies that lacked fibrils and the number of multivesicular bodies that contained fibrils (or fibrillar compartments lacking other internal structure) were quantified, and the percentage of fibril-containing multivesicular bodies was calculated for each cell. These data were analyzed and plotted using Prism 6 (GraphPad Prism 6, La Jolla, CA), and statistical analyses were performed using a two-tailed unpaired t test.

p250 and p160 Are Disulfide-bonded Species Composed of
Golgi-modified PMEL-The Golgi-matured P2 form of PMEL and the products of proprotein convertase cleavage, M␣ and M␤ (Fig. 1A), migrate with M r of ϳ120,000, 100,000, and 28,000, respectively, when analyzed by reducing SDS-PAGE (26,38). However, when analyzed by non-reducing SDS-PAGE, most detergent-soluble forms of Golgi-matured PMEL migrate with a higher M r (26), indicating that the vast majority of preamyloid P2, M␣, and M␤ isoforms exist as components of disulfide-bonded species. To further investigate the nature of these species, immunoblots were performed using antibodies that can distinguish between immature and mature PMEL isoforms. MNT-1 human melanoma cells (which endogenously express PMEL) or transfected HeLa cells transiently expressing full-length PMEL (isoform 1) were treated with NEM prior to and during lysis in TritonX-100 to prevent the formation of artifactual disulfide bonds. Detergent-soluble lysates were then fractionated by SDS-PAGE on low percentage polyacrylamide gels under non-reducing or reducing conditions and analyzed by immunoblotting using the HMB45, ␣PMEL-I, ␣PMEL-N, or ␣PMEL-C antibody. The HMB45 monoclonal antibody recognizes the O-glycosylated RPT region of mature PMEL isoforms present only in post-Golgi compartments (29,47). The ␣PMEL-I antibody recognizes the same domain in its non-glycosylated form as is characteristic of immature PMEL isoforms located in the ER (29). The ␣PMEL-N and ␣PMEL-C antibodies recognize the N terminus and C terminus of PMEL, respectively, regardless of maturation status (26,41).
Under non-reducing conditions, HMB45 detected two major species that migrate with M r of ϳ160,000 (p160) and ϳ250,000 (p250). Also detected were larger PMEL species retained at the boundary between the stacking and separating gel and minor amounts of p160*, M␣ϩM␤Ј, and free M␣ (Fig. 1B, lanes 1 and  2). The p160* species is a slightly faster migrating version of p160 resulting from an alternatively spliced PMEL variant expressed in melanocytic cells but not transfected HeLa cells (41). The M␣ϩM␤Ј species is composed of M␣ covalently bound to M␤Ј based on the following criteria: (i) it is detected by the HMB45 and ␣PMEL-N antibodies but not the ␣PMEL-C antibody, indicating that it lacks the PMEL C terminus; (ii) it comigrates with the major form of secreted PMEL (see Fig. 5A), which consists predominantly of disulfide-bonded M␣ϩM␤Ј complexes (36); and (iii) it is depleted following inhibition of BACE2 (see Fig. 6A), the enzyme that cleaves M␤ into CTF and M␤Ј fragments. As only a single band corresponding to M␣ was observed under reducing conditions (Fig. 1B, lanes 3 and 4; note that M␣ migrates with a higher M r when derived from MNT-1 cells than when derived from HeLa cells, due to differences in terminal glycosylation (26)), disulfide bonds must be necessary to maintain the integrity of p250 and p160 as well as that of p160* and M␣ϩM␤Ј. Differences in the relative abundance of these species in lysates of MNT-1 cells versus PMEL-expressing HeLa cells likely reflect cell type-specific differences in the rates at which these species are formed and consumed by intercon-version, secretion, or incorporation into insoluble amyloid fibrils.
The p250 and p160 bands were not labeled by ␣PMEL-I under non-reducing conditions; instead, this antibody detected a band that corresponds to monomeric P1 as well as a series of high M r bands ranging from ϳ220,000 to the top of the separating gel (Fig. 1C, lanes 1 and 2). Because the high M r bands collapse into P1 under reducing conditions (Fig. 1C, lanes 3 and  4; note that the slightly faster migrating band (*) corresponds to the P1 form of the alternatively spliced PMEL variant, PMEL-ss (41)), these species most likely represent partially folded disulfide-bonded conformers of PMEL in the ER (Fig. 1C, cfmrs). The ␣PMEL-N and ␣PMEL-C antibodies recognized p250 and p160 as well as the immature PMEL conformers detected by ␣PMEL-I, indicating that all of these high M r species contain either full-length PMEL or M␣ disulfide-bonded to M␤ (Fig. 1,  D and E). Therefore, only Golgi-matured forms of full-length or proprotein convertase-cleaved PMEL are present in p250 and  4) expressing wild-type PMEL were lysed in buffer containing 1% (w/v) Triton X-100, and the soluble material was fractionated by SDS-PAGE under non-reducing (NR) or reducing (R) conditions. Proteins transferred to membranes were then probed with the PMEL antibodies HMB45 (B), ␣PMEL-I (C), ␣PMEL-N (D), or ␣PMEL-C (E). The migration of molecular weight standards is indicated to the right of each blot. Bands corresponding to p250, p160, p160*, M␣ϩM␤Ј, M␣, immature PMEL conformers (cfmrs), and P1 are indicated by arrows. Note that M␣ and P1 comigrate in MNT-1 cells and that free M␤ was run off the end of these gels. *, an alternatively spliced PMEL product (PMEL-ss). The lane numbers are indicated at the bottom.
p160. Furthermore, the presence of p250 and p160 both in MNT-1 cells that endogenously express PMEL and in non-pigment HeLa cells expressing a PMEL transgene suggests that no additional pigment cell-specific proteins are required for the formation of mature disulfide-bonded PMEL species.
p250 and p160 Represent Distinct Physiological Complexes-To test whether the disulfide-bonded p250 and p160 species observed by non-reducing SDS-PAGE correspond to physiological PMEL-containing complexes, we assessed the biophysical properties of p250 and p160 under non-denaturing conditions. First, we examined the PMEL species by sedimentation velocity analysis. Lysates of MNT-1 cells were prepared using n-octylglucoside, a non-ionic detergent with a small micelle size and a relatively high partial specific volume (45). After fractionation through sucrose gradients containing n-octylglucoside, mature PMEL species were identified by immunoblot analysis of gradient fractions separated by non-reducing SDS-PAGE ( Fig. 2A). Sedimentation coefficients were then calculated by comparing the peak position of mature PMEL species, detected using the HMB45 antibody, with the peak position of globular protein standards (Fig. 2, B and C). Distinct sedimentation coefficients were calculated for p250, p160, M␣ϩM␤Ј, and "free" M␣. Consistent with the relative order of their migration by non-reducing SDS-PAGE, p250 had the highest calculated s value followed by p160, M␣ϩM␤Ј, and lastly free M␣. This suggests that p250 and p160 are not artifacts of non-re-ducing SDS-PAGE but distinct native disulfide-bonded PMEL complexes. The sedimentation coefficients of all four PMEL species were substantially smaller than expected given the magnitude of their migration by non-reducing SDS-PAGE ( Table  1), suggesting that these species might have extended glycoprotein structures.
To further characterize the PMEL species under native conditions, MNT-1 cell lysates were fractionated by size exclusion chromatography in n-octylglucoside. Aliquots of each fraction were separated by non-reducing SDS-PAGE, and mature PMEL species were detected by immunoblotting with HMB45. As observed in the sedimentation analyses, the four PMEL species, p250, p160, M␣ϩM␤Ј, and M␣, migrated with distinct elution volumes by size exclusion chromatography (Fig. 3, A  and B). However, when the peak elution volume of each species was compared with that of globular protein standards, the mature PMEL species had peak elution volumes much smaller, and therefore Stokes radii much larger, than expected for globular proteins of similar molecular weight ( Fig. 3C and Table 1). For example, the Stokes radius of p250 (10.4 nm) was calculated to be larger than that of ␣ 2 -macroglobulin (8.8 nm), a 720-kDa tetrameric protein (49). To test whether the exceptionally large Stokes radii reflected association with multiple detergent micelles, cell lysates were prepared and analyzed by size exclusion chromatography using detergents with different micelle sizes: n-octylglucoside (average micelle size of 8 kDa (50)), dodecyl-␤-D-maltoside (average micelle size of 50 -76 kDa (50)) and Triton X-100 (average micelle size of 81 kDa (51)). If any of the PMEL species had been associated with multiple detergent micelles, those species would have eluted with distinct volumes in each of the three detergents. However, the elution volume of each PMEL species was roughly similar in all three detergents (Fig. 3 and Table 1), suggesting that detergent micelles do not contribute substantially to the large Stokes radii of the mature PMEL species. Therefore, the unusually high Stokes radii and low sedimentation coefficients of p250, p160, M␣, and M␣ϩM␤Ј most likely reflect extended glycoprotein conformations resulting from the heavily glycosylated RPT domain of mature PMEL. Remarkably, the estimated molecular weight calculated using the sedimentation and size exclusion chromatography properties of each of the four PMEL species closely approximated the molecular weight of each species as determined by SDS-PAGE (Table 1).
PMEL Is the Sole Component of Both p250 and p160 -To determine whether p250 and p160 contain additional polypeptides besides PMEL, we analyzed the contents of PMEL immunoprecipitates by two-dimensional non-reducing/reducing SDS-PAGE. We used transfected HeLa cells transiently expressing wild-type PMEL in this experiment because M␣ is more clearly identified in HeLa cell lysates analyzed by reducing SDS-PAGE. This is because, unlike melanocytic cells, transfected HeLa cells typically do not express the alternatively spliced form of PMEL (PMEL-ss) that comigrates with M␣ by reducing SDS-PAGE (see Fig. 1, B-D). In addition, M␣ comigrates with P1 in MNT-1 cells but not in transfected HeLa cells as a result of differences in terminal glycosylation. After extensively labeling biosynthesized proteins with [ 35 S]methionine/ cysteine, PMEL species in detergent cell lysates were immuno-  Table 1. precipitated using ␣PMEL-N, ␣PMEL-C, or NKI-beteb (52), a monoclonal antibody to the PKD (a region with homology to a repeated domain in polycystin 1 (29)) and then fractionated by two-dimensional PAGE and analyzed by phosphorimaging. Using this technique, the higher M r bands observed by standard SDS-PAGE (Fig. 1) were resolved into their individual components. Similar results were obtained using all three PMEL antibodies (Fig. 4). Spots corresponding to the core-glycosylated P1 and the Golgi-modified P2 forms of PMEL were identified close to the diagonal and thus represent the monomeric forms of these PMEL species. The p160 band typically observed by onedimensional non-reducing SDS-PAGE resolved into two species by two-dimensional PAGE; p160a contained the M␣ and M␤ products of proprotein convertase cleavage, whereas p160b contained uncleaved, full-length, Golgi-modified P2. The p250 band also resolved into both the cleaved and uncleaved forms of Golgi-modified PMEL. A number of other bands present in the non-reducing dimension resolved into P1 under reducing conditions and thus most likely represent disulfide-bonded ER folding intermediates. Importantly, no other radiolabeled polypeptide was reproducibly detected within the p250 or p160 bands. Furthermore, three independent attempts to identify additional components by tandem mass spectrometry analysis of affinity-purified PMEL from MNT-1 cells yielded no reasonable candidates other than chaperones and protein disulfide isomerases, which were present in vastly substoichiometric    Table 1.
amounts ( Table 2 and supplemental Table S1) and were associated only with immature forms of PMEL. 4 Other identified peptides either derived from common contaminants such as keratins, derived from proteins with molecular weights that were inconsistent with that of p250 and p160, or were very poorly recovered relative to PMEL-derived peptides. We therefore concluded that p250 and p160 are composed exclusively of PMEL or PMEL fragments. As detailed above, the molecular weight of p250 and p160 calculated from the combined size exclusion chromatography and velocity sedimentation data are ϳ241,000 and ϳ176,000, respectively, closely correlating with their migration by SDS-PAGE. Given these data and the absence of any additional protein components, p250 most likely represents a PMEL dimer, and p160b and p160a most likely represent extended disulfidedependent conformations of a single PMEL monomer and a single M␣/M␤ complex, respectively.
p250 and p160 Are Intermediates of PMEL Fibril Formation-The reactivity of p250 and p160 with HMB45 but not with ␣PMEL-I suggests that these two species are primarily present in post-ER compartments. To define the kinetics of their formation, we performed a metabolic pulse-chase and immunoprecipitation analysis. MNT-1 cells were metabolically labeled for 15 min with [ 35 S]methionine/cysteine and then chased in medium containing excess unlabeled amino acids for 0 -120 min. PMEL isoforms in detergent-soluble cell lysates and the culture medium from each time point were immunoprecipitated using NKI-beteb and analyzed by non-reducing or reducing SDS-PAGE. At the pulse, the primary PMEL species detected under non-reducing conditions was monomeric P1, which migrated with a M r of ϳ110,000 (Fig. 5A, lane 1). The p250 band first appeared after 5 min of chase and peaked at 30 min (Fig. 5A, lanes 2-4). This resembles the kinetics of P2 formation/accumulation, as observed under reducing conditions (Fig. 5B, lanes 1-6), suggesting that dimer formation occurs just prior to ER exit or in the early Golgi. The p160 band was first observed after 15 min of chase and peaked at 30 -60 min (Fig. 5A, lanes 3-5). This resembles the kinetics of proprotein convertase cleavage, as indicated by the appearance of the M␤ cleavage product observed under reducing conditions (Fig. 5B, lanes 1-6). Essentially similar results were obtained upon pulse-chase analysis of PMEL expressed in transiently transfected HeLa cells (Fig. 5C). Thus, p160 is likely generated in the trans-Golgi network or in a post-Golgi compartment. Importantly, at no point did monomeric forms of P2 or M␣ appear in cell lysates analyzed under non-reducing conditions. This indicates that the vast majority of post-ER/prefibrillar PMEL must be present in disulfide bond-dependent intermediates.
As observed previously (24,26,36,52), a small fraction of PMEL was secreted into the medium of MNT-1 cells starting at 30 min of chase and peaking at 120 min (Fig. 5, A and B, lanes  7-12). Consistent with those reports (36), PMEL secreted in this manner comprised primarily M␣ϩM␤Ј with lesser amounts of free M␣.
To verify band identity and to probe for potential precursorproduct relationships, MNT-1 cells were treated with small molecule inhibitors of trafficking and processing steps. Detergent-soluble lysates were then analyzed by non-reducing SDS-PAGE and immunoblotted with HMB45 or ␣PMEL-C. Treatment of MNT-1 cells with monensin, an inhibitor of intra-Golgi transport (53), led to the accumulation of p250 and higher M r oligomers as well as the concomitant depletion of p160 (Fig. 6,  A and B, lane 2). This supports the conclusion that p250 forms during ER exit or in the early Golgi and that p160 forms in the trans-Golgi network or later. It also suggests that p250 might be the precursor of p160. Treatment with ␤-secretase inhibitor IV at concentrations that inhibit BACE2 cleavage (54) resulted in the substantial accumulation of p160 and the minor accumulation of p250 and higher M r oligomers (Fig. 6, A and B, lane 3). This result indicates that the resolution of p160, and perhaps a fraction of p250, requires BACE2 cleavage, which is known to cleave M␤ to M␤Ј and the CTF (33). Because BACE2 inhibition has been shown to impede PMEL fibril formation (33), this also suggests that p160 cannot directly assemble into functional amyloid fibrils. Lastly, treatment with the cysteine protease inhibitor E-64 resulted in the accumulation of free M␣ and M␣ϩM␤Ј (as detected by HMB45 but not by ␣PMEL-C) but had no effect on p250 or p160 (Fig. 6,  A and B, lane 4). Prolonged treatment with E-64 also resulted in the accumulation of M␣ and the depletion of proteolytic fragments of M␣ containing the RPT domain (M␣Ј) in the detergent-insoluble fraction of MNT-1 cells (Fig. 6C), suggesting that cysteine protease inhibitors, like ADAM prote-4 T. Ho, unpublished data. ase inhibitors (55), interfere with the processing of M␣ and M␣ϩM␤Ј into smaller fragments.
Importantly, treatment of these cells for an additional 2 h in the presence of both E-64 and the protein synthesis inhibitor cycloheximide to allow for further processing in the absence of new PMEL biosynthesis led to the depletion of p160 but not of M␣ or M␣ϩM␤Ј (Fig. 6, A and B, lane 5). This suggests that p160 is a true intermediate in PMEL processing that precedes the formation of free M␣ and M␣ϩM␤Ј. Finally, none of the inhibitor treatments (except for cycloheximide) affected the level of P1 detected by ␣PMEL-C (Fig. 6B), which is consistent with the notion that immature PMEL monomers are located predominantly in pre-Golgi compartments (i.e. the ER).

Cysteine 301 Participates in a Cross-subunit Disulfide Bond in p250 and a Disulfide Bond between M␣ and M␤ Derived from the Same PMEL Precursor in p160 -
To determine which cysteine residues participate in the disulfide bonds that stabilize p250 and p160, we evaluated the effect of single cysteine mutations on the disulfide-bonded PMEL species. The 11 luminal cysteine residues within PMEL (Fig. 7A) were each individually mutagenized to serine and the single point mutants expressed in HeLa cells by transient transfection. Like wild-type PMEL, all mutants were able to exit the ER and localize to a subset of late endosomes and lysosomes labeled with LAMP1 by immunofluorescence microscopy (Fig. 8). The mutants were also all appropriately cleaved in post-Golgi compartments by a proprotein convertase, as each one produced M␤ fragments when analyzed by reducing SDS-PAGE followed by immunoblotting with ␣PMEL-C (Fig. 7B). These data indicate that no single disulfide bond is necessary for PMEL to pass ER quality control checkpoints or to be appropriately modified by Golgi enzymes and proprotein convertase. To assess p250 and p160 formation, lysates were then fractionated by non-reducing SDS-PAGE and immunoblotted with HMB45 (Fig. 7, C-E). Of the four cysteine residues within the region of PMEL that corresponds to M␣, individual mutations in three of them had little to no effect on p250 and p160. Mutagenesis of Cys-60 resulted in disulfidebonded PMEL species that were undistinguishable from those produced by wild-type PMEL (Fig. 7C, lane 2), and mutagenesis of Cys-130 or Cys-138 produced only slight increases in the M r of p250 and p160 (Fig. 7C, lanes 3 and 4). This suggests that although Cys-130 and Cys-138 likely participate in an intramolecular disulfide bond, neither Cys-60, Cys-130, nor Cys-138 is critical for maintaining the integrity of p250 or p160. By contrast, mutagenesis of Cys-301 ablated the formation of both p250 and p160, causing most mature PMEL to migrate as free M␣ by non-reducing SDS-PAGE (Fig. 7C, lane 5). Because Cys-301 was the only cysteine residue within the M␣ region of PMEL required to stabilize both disulfide-bonded PMEL species, Cys-301 must participate in a disulfide bond with a cysteine residue in the M␤ region of PMEL to generate p250 and p160.
To generate p250, a disulfide bond must connect the two PMEL monomers within the dimer. Because p250 contains both M␣ and M␤ products of proprotein convertase cleavage (Fig. 4), a disulfide bond must also form a bridge between the M␣ and M␤ fragments. If Cys-301 is the only cysteine residue in the M␣ region of PMEL that actively functions to fulfill both of these requirements, then Cys-301 of one PMEL molecule must form an intermolecular disulfide bond with the M␤ region of the other PMEL molecule within the p250 dimer. Alternatively, the other cysteine residues in the M␣ region of PMEL could participate in a disulfide bond with M␤ in a redundant manner such that mutagenesis of any one is insufficient to disrupt disulfide bond formation. We therefore tested whether combined mutagenesis of the other M␣ cysteine residues would impact p250 or p160 formation. Relative to single point mutations in Cys-130 or Cys-138, simultaneous mutation of two (Cys-130 and Cys-138) or all three (Cys-60, Cys-130, and Cys-138) of these cysteine residues had no additional impact on the migration or the formation of p250 (Fig. 7D, lanes 2 and 3). Therefore,

TABLE 2 Analysis of p250 and p160 by LC-MS/MS
Affinity-purified PMEL was fractioned on a non-reducing SDS-polyacrylamide gel, relevant bands excised and digested using trypsin or AspN, and the samples analyzed by mass spectrometry. This table is a brief summary of the top 50 "hits" obtained. Summed intensity refers to the sum of the extracted ion currents from three independent experiments analyzing higher M r disulfide-bonded PMEL species. The proteins listed are of human origin unless otherwise noted. More detailed results, with each band in each of the three experiments analyzed, are provided in supplemental Table S1.
To generate p160a, a disulfide bond must connect Cys-301 of M␣ to a cysteine residue of M␤. However, this disulfide bond could form between M␣ and M␤ fragments derived from the same PMEL molecule or from different PMEL molecules. To distinguish between these possibilities, we analyzed a mutagenized form of PMEL lacking the proprotein convertase cleavage site (⌬CS (30)) by non-reducing SDS-PAGE. As predicted by the migration of uncleaved P2 in p160b by two-dimensional SDS-PAGE (Fig. 4), the ⌬CS variant generated both p250 and p160 with only small shifts in migration (Fig. 7D, lane  4). Thus, the M␣ and M␤ fragments in p160a likely originate from the same PMEL molecule. Given that Cys-301 is required for the formation of both the interchain disulfide bond in p250 and the intrachain disulfide bond in p160b, a disulfide bond exchange must occur during the maturation of p250 to p160.
In an attempt to identify the cysteine residue within the M␤ region of PMEL that forms a disulfide bond with Cys-301, the seven cysteine residues in this region were mutagenized individually to serine and the single point mutants analyzed for their effects on the stability and migration of p250 and p160 by non-reducing SDS-PAGE. Mutagenesis of Cys-475 resulted in disulfide-bonded PMEL species undistinguishable from those produced by wildtype PMEL, but mutagenesis of any one cysteine residue within the cysteine-rich KLD (Cys-516, Cys-525, Cys-533, Cys-541, Cys-550, or Cys-566) completely ablated formation of p160 without disrupting that of p250 (Fig. 7E). Although these data fail to isolate which cysteine residue within the M␤ region of PMEL participates in the disulfide bond with Cys-301, they suggest that an intact KLD is critical for the formation of p160.
KLD Mutagenesis Impairs PMEL Fibril Formation-The ability of KLD cysteine mutants to preferentially stabilize p250 provides a useful tool with which to investigate the function of this disulfide-bonded PMEL intermediate. Previous studies  have shown that, when overexpressed in non-pigmented cells such as HeLa or Mel220, PMEL forms melanosome-like fibrillar structures within multivesicular endosomes that can be visualized and distinguished from non-fibrillar aggregates by electron microscopy (30,31,56,57). In our experiments, transient expression of wild-type PMEL led to the formation of fibrillar sheets in 36 Ϯ 4.6% (mean Ϯ S.E.) of the multivesicular bodies present within a given HeLa cell (Fig. 9). However, following comparable expression of the C566S PMEL mutant, which forms p250 but not p160 (Fig. 7E, lane 8), many fewer multivesicular bodies (18 Ϯ 3.5%, p Ͻ 0.01) showed evidence of fibril formation (Fig. 9). This result suggests that stabilization of p250 can impair the formation of functional amyloid fibrils and that resolution of this disulfide-bonded intermediate may be necessary for fibril formation to ensue.
Surprisingly, expression of the C301S variant, which fails to form both p250 and p160 (Fig. 7C, lane 5), resulted in levels of fibril formation (34 Ϯ 3.8% of multivesicular bodies) that were not significantly different from wild-type PMEL (Fig. 9). This could have been interpreted to suggest that p250 and p160 are not required for fibril formation. However, given that p250 and p160 were detected as distinct entities under non-denaturing conditions by sedimentation and size exclusion chromatography analyses (Figs. 2 and 3), we considered the possibility that similar complexes might form non-covalently in the absence of Cys-301. Indeed, co-immunoprecipitation analyses show that M␣ and M␤ remained associated in cells expressing the C301S mutant of PMEL (Fig. 10, A and B). Moreover, when transfected HeLa cells were analyzed by size exclusion chromatography, the C301S mutant generated PMEL complexes having elution volumes identical to those of wild-type PMEL (Fig. 10, C-H). Thus, although Cys-301-dependent disulfide bonds may provide additional stability to p250 and p160, the data suggest that formation of the corresponding non-covalent complexes is sufficient for function.

Discussion
Very little is known about how functional amyloid formation is regulated within cells, especially in mammalian systems. Here we have shown that the pigment cell-specific amyloid protein PMEL forms a disulfide-bonded dimer in the ER or early Golgi. Resolution of this dimer to a disulfide-bonded monomer requires an intact KLD, and mutagenesis of any cysteine residue within the KLD inhibits this process and interferes with the assembly of functional amyloid fibrils. Our data are consistent with a model in which the formation of a disulfide bond-stabilized PMEL dimer in the ER or early Golgi impairs premature fibril formation early in the secretory pathway (Fig. 11). KLDdependent disulfide bond rearrangements in the late Golgi or in a post-Golgi compartment would then resolve the dimer to monomeric forms that could be incorporated into growing amyloid fibrils.
The discovery of higher molecular weight disulfide-bonded p250 and p160 PMEL species that react with an antibody known to detect mature, but not immature, PMEL isoforms originally suggested to us that Golgi-modified forms of PMEL might be covalently bound to one or more unidentified partner proteins. However, several observations led us to conclude that these species contain only PMEL. First, p250 and p160 were detected in both melanocytic cells and transfected HeLa cells. Because pigment cell-specific proteins are not expressed in HeLa cells, this eliminated the possibility of covalent association with a pigment cell-specific binding partner. Second, no polypeptides other than PMEL-derived P2, M␣, and M␤ were consistently detected as components of p250 or p160 by twodimensional PAGE of PMEL immunoprecipitated from cells metabolically labeled with [ 35 S]methionine/cysteine for an extended period of time. This was surprising given that a binding partner would necessarily be stoichiometrically represented within the disulfide-bonded PMEL complexes. Third, no candidate binding partners of the appropriate molecular weight, relative abundance to PMEL, and subcellular localization were identified by mass spectrometry analysis of trypsin-or AspNdigested p250 and p160 bands (see Table 2 and Supplemental  Table S1). Several ER chaperones were detected in moderate abundance; however, these are common contaminants of protein purification by affinity chromatography, and detailed analyses of two abundant protein disulfide isomerase family members revealed interactions with immature ER forms of PMEL but not p250 or p160. 4 The mass spectrometry data effectively ruled out the possibility that a stoichiometric constituent of p250 or p160 was missed by two-dimensional PAGE due to poor labeling with [ 35 S]methionine/cysteine. Thus, although it remains remotely possible that a poorly radiolabeled, ubiquitous binding partner might fail to generate peptides detectable by mass spectrometry using either of two proteases, we favor the more likely interpretation that p250 and p160 are composed entirely of PMEL.
With a molecular weight approximately twice that of a PMEL monomer, p250 is most likely a PMEL dimer. The exact composition of p160, however, was less obvious at first. The migration of p160 by non-reducing SDS-PAGE is considerably slower than that of a completely reduced, Golgi-modified, PMEL monomer. We initially interpreted this to suggest that p160 is composed of a full-length PMEL monomer, or linked M␣/M␤ cleavage products, in complex with an additional PMEL fragment such as M␤. However, the migration of p160 by nonreducing SDS-PAGE was only minimally altered in the absence of M␤ production as tested using the proprotein convertase cleavage site mutant ⌬CS. In addition, the naturally occurring p160b band comprising uncleaved, full-length, Golgi-modified P2 contained no additional PMEL fragments when analyzed by two-dimensional PAGE. As such, we interpreted the data to indicate that P2 migrates with a M r of ϳ160,000 when constrained by disulfide bonds but a M r of ϳ120,000 when fully reduced. We speculated that the disulfide-bonded structure of p160 limits accessibility to SDS, thus reducing the migration of this species by non-reducing SDS-PAGE.
Our results are consistent with a model in which the p250 dimer is an obligate precursor of the p160 monomer, because (i) the formation of p250 occurs prior to that of p160 by metabolic pulse-chase analysis, (ii) inhibition of intra-Golgi transport in cells treated with monensin leads to the accumulation of p250 and depletion of p160, and (iii) none of the cysteine mutants produce p160 independently of p250. Complicating this interpretation, however, was the finding that both p250 and p160 contain the full-length Golgi-modified P2 form of PMEL and the M␣/M␤ form cleaved by proprotein convertase when analyzed by two-dimensional PAGE. Because M␣ and M␤ are generated by proprotein convertase cleavage of P2 in the trans-Golgi network (31) or endosomes (32), ideally p250 would contain only uncleaved P2, and p160 would contain only cleaved M␣/M␤. Although these data raise the possibility that p250 represents an independent "dead-end" pathway for PMEL that never generates fibrils, we consider it more likely that p250 is converted to p160 at the same intracellular site but independently of proprotein convertase cleavage such that cleavage can occur before, during, or after resolution of the p250 dimer (Fig. 11).
Our data further suggest that a disulfide bond rearrangement is required to convert p250 to p160. This is because Cys-301, a cysteine residue conserved throughout vertebrate evolution (58), participates in a cross-subunit disulfide bond between two different PMEL molecules within p250 but in a disulfide bond between M␣ and M␤ derived from the same PMEL precursor within p160. Although proprotein convertase cleavage of both monomers within the p250 dimer would immediately resolve p250 into a "monomeric" form in which M␣ derived from one protomer is linked to M␤ from the other protomer, the observation that p160b can form without proprotein convertase cleavage suggests that the cross-subunit disulfide bond of p250 is broken to form a new disulfide bond connecting M␣ and M␤ derived from the same PMEL protomer. The different relative intensity of p250 to p160 in PMEL-expressing HeLa cells as compared with MNT-1 cells further suggests that this process may occur at different rates in the two cell types. Although disulfide bonds are typically formed and remodeled in the ER (59), disulfide bond rearrangements outside the ER have been reported (40,60). Interestingly, if our model is correct, the disulfide exchange likely occurs without the enzymatic activity of a protein disulfide isomerase; typical ER protein-disulfide isomerases such as PDIA1 and PDIA3 associate exclusively with ER forms of PMEL, 4 and melanocytes do not express the only known protein disulfide isomerase of the endocytic path- FIGURE 9. Cells expressing the C566S PMEL mutant, but not the C301S mutant, exhibit decreased fibril formation relative to those expressing wild type PMEL. HeLa cells transiently transfected with wild-type PMEL, the C301S mutant, or the C566S mutant were analyzed by transmission electron microscopy. A, example of a multivesicular body (MVB) with intraluminal vesicles (arrowheads). B, example of a multivesicular body with fibrils, intraluminal vesicles (arrowheads), and a small multilamellar structure commonly found in late endosomes and lysosomes (asterisk). C, fibril-containing organelles and multivesicular bodies without any evidence of fibril formation were quantified in at least one field of view/cell, and the percentage of fibril-containing multivesicular bodies was calculated for each cell. Box plots show the combined results of three experiments with the line in the center representing the median, the box representing the 25th and 75th percentiles, and the whiskers denoting the minimum and maximum values. Statistics were performed using a two-tailed unpaired t test comparing each mutant to wild-type PMEL. Scale bars are 500 nm.
way, GILT (␥-interferon-inducible lysosomal thiol reductase) (61), unless they are stimulated with ␥-interferon (62). 5 In addition, failure to treat cells with fresh NEM prior to lysis resulted in poor preservation of the disulfide-bonded PMEL species, 4 suggesting that nearby cysteine residues rapidly reduce p250 and p160 when not blocked by alkylation. Because the integrity of the PMEL KLD is required for the resolution of p250 to p160, we speculated that cysteine residues within the KLD initiate this disulfide exchange in an autocatalytic manner. Assessment of this model would require the purification of properly folded recombinant PMEL isoforms, which are not currently available.
Many layers of regulation must exist to protect organisms against the dire consequences of inappropriate amyloid formation. Our data suggest that the KLD protects against premature PMEL fibril formation by regulating the resolution of a disulfide-stabilized PMEL dimer. The importance of an intact KLD 5 M. S. Marks, unpublished data.  1 and 4) were analyzed by reducing SDS-PAGE (R) and immunoblotted (IB) with antibodies to the opposite fragment: HMB45 to M␣ (A) or ␣PMEL-C to M␤ (B). The migration of molecular weight standards are shown to the right; bands corresponding to P2, M␣, P1, and M␤ are indicated with arrows. C-F, HeLa cells transfected with wild-type PMEL (C and E) or the C301S PMEL variant (D and F) were lysed in 250 mM n-octylglucoside lysis buffer and the detergent extract fractionated by size exclusion chromatography in 25 mM n-octylglucoside running buffer. Eluted fractions were analyzed by SDS-PAGE under non-reducing (NR; C and D) or reducing (R; E and F) conditions and immunoblotted using the antibodies indicated. The migration of molecular weight standards is indicated to the right of each blot, and bands corresponding to p250, p160, M␣, and M␤ are indicated by arrows. Lanes are labeled with fraction numbers as presented in Fig. 3A. G and H, HMB45 and ␣PMEL-C immunoreactivity was quantified to determine the amount of M␣ (G) and M␤ (H) present in each fraction. The lowest value quantified for each species was set to 0%, and the highest value was set to 100% for each experiment. Data represent the mean Ϯ S.D. of three independent experiments.
is supported by the finding that fibril formation is impaired in HeLa cells expressing the C566S mutant, a variant of PMEL that generates p250 dimers but not p160 monomers. Given that mutagenesis of any single cysteine residue within the KLD, and even a small deletion, 6 results in the same biochemical effect on p250 and p160, we predict that any disruption to the cysteine arrangement within the KLD will interfere with fibril formation. Surprisingly, all of the individual cysteine mutants pass quality control checkpoints in the ER, are cleaved by proprotein convertases in the trans-Golgi network or endosomes, and localize appropriately to late endosomes and lysosomes. This suggests that the KLD acts primarily in a post-Golgi compartment to regulate p160 formation and promote the assembly of functional amyloid fibrils. Whether the KLD can promote fibril formation independent of its effect on the disulfide-bonded PMEL species cannot be determined with the present data, but the progressive disulfide bond changes during PMEL maturation suggest that a primary function of the KLD is to regulate these changes.
Intriguingly, mutagenesis of Cys-301, the only cysteine residue integral to the formation of both p250 and p160, does not impair PMEL fibril formation. As shown by size exclusion chromatography, this is most likely because the C301S PMEL variant forms non-covalent complexes similar to the disulfidebonded species formed by wild-type PMEL. This indicates that Cys-301-dependent disulfide bonds are not required for amyloid fibril formation but rather stabilize energetically favorable conformations at distinct stages of the PMEL biosynthetic pathway. We speculated that loss of this additional stability under physiological expression conditions might lead to premature fibril formation and consequent pathology that is not detectable in the short-term assays used here, as has been observed for several dominant PMEL mutants (63).
PMEL is not the only amyloid protein influenced by disulfide bonds. Reduction of the single disulfide bond in the functional amyloid protein somatostatin-14 in vitro results in fibrils that form faster,aremoreresistanttodenaturation,andhavedecreasedmonomer release (64). Reduction of the intrachain and/or one of two interchain disulfide bonds in porcine insulin also accelerates fibril formation (65). Interestingly, the cysteine residues of Tau can participate in intramolecular or intermolecular disulfide bonds; the intramolecular disulfide bonds retard fibril formation by restricting the protein to conformations incompatible with the amyloid fold, whereas the intermolecular disulfide bonds accelerate amyloid fibril formation (66). However, many proteins, including insulin, have the ability to form amyloid aggregates in vitro but do not typically do so in vivo (48). Because many proteins of the secretory pathway contain disulfide bonds, we suggest that the use of disulfide bonds to stabilize pro-or anti-amyloidogenic folding intermediates might be a common mode of regulation for amyloid proteins within the endomembrane system.  . Model of PMEL fibril formation. The schematic diagram shows the maturation of p250 into functional amyloid fibrils. The p250 PMEL dimer is resolved to p160 via disulfide bond rearrangement with proprotein convertase cleavage occurring before, during, or after this process. p160 is then cleaved by BACE2 to M␣ϩM␤Ј, and M␤Ј is removed to produce free M␣ by an as yet unknown process. M␣ then assembles into fibrils and is concomitantly or subsequently cleaved by cysteine proteases into smaller fragments including M␣Ј. Note that the CTF is trafficked to lysosomes and is not incorporated into functional amyloid fibrils. The fate of M␤Ј is not known. For simplicity, only one subunit of the original dimer is shown after p160. The red lines represent disulfide bonds between Cys-301 and the KLD.