|Home | About | Journals | Submit | Contact Us | Français|
In yeast cells infected with the [PSI+] prion, Sup35p forms aggregates and its activity in translation termination is down-regulated. Transfection experiments have shown that Sup35p filaments assembled in vitro are infectious, suggesting that they reproduce or closely resemble the prion. We have used several EM techniques to study the molecular architecture of filaments, seeking clues as to the mechanism of down-regulation. Sup35p has an N-terminal “prion” domain; a highly charged middle (M-)domain; and a C-terminal domain with the translation termination activity. By negative staining, cryo-EM, and scanning transmission EM (STEM), filaments of full-length Sup35p show a thin backbone fibril surrounded by a diffuse 65nm-wide cloud of globular C-domains. In diameter (~8nm) and appearance, the backbones resemble amyloid fibrils of N-domains alone. STEM mass-per-unit-length data yield ~1 subunit per 0.47 nm for N-fibrils, NM-filaments, and Sup35p, further supporting the fibril backbone model. The 30 nm radial span of decorating C-domains indicates that the M-domains assume highly extended conformations, offering an explanation for the residual Sup35p activity in infected cells, whereby the C-domains remain free enough to interact with ribosomes.
The yeast non-chromosomal genetic element [PSI+] was identified as a prion (infectious protein) form of the cytosolic protein Sup35p (Wickner, 1994). In its normal soluble form, Sup35p associates with Sup45 in a heterodimer that is responsible for translation termination (Stansfield et al., 1995). This activity is markedly reduced in [PSI+] cells - i.e. [PSI+] acts as a global suppressor (Hawthorne and Leupold, 1974). Sup35p's N-domain (residues 1 - 123), which is necessary and sufficient for the prion phenotype, has a marked propensity to form amyloid fibrils (Glover et al., 1997; King et al., 1997; Paushkin et al., 1997; Ter-Avanesyan et al., 1994). This domain is rich in polar uncharged residues, especially Gln and Asn (together, 45%), and also in Gly (17%) and Tyr (16%). In contrast, the M-domain (residues 124 – 253) is highly charged (40% of charged residues). The C-domain (residues 254 – 685), which performs the translation termination function, is closely homologous to the eukaryotic release factor 3 (e.g. > 50% sequence identity to human eRF3). A crystal structure has been determined for the C-domain of Sup35p from S. pombe (Kong et al., 2004). As S. cerevisiae Sup35p shares 65% sequence identity with this protein, it is likely to have a closely similar, globular, structure. As for the N-and M-domains, similar sequences are found only in the homologs of other yeasts and related fungi (Chernoff et al., 2000; Resende et al., 2002; Resende et al., 2003). The eRF3 proteins of higher mammals also have N-terminal extensions but they are rich in Pro, Gly, and Ser and not in Gln or Asn (Jean-Jean et al., 1996; Inge-Vechtomov et al., 2007). The N-terminal extensions of both Sup35p and eRF3 play roles in normal mRNA turnover by interacting with the poly(A)-binding protein PABP (Hoshino et al., 1999; Hosoda et al., 2003). However, only that of Sup35p has been shown to fibrillize.
Much evidence has accumulated in support of the “amyloid fibril* backbone” model of prionogenesis (Speransky et al., 2001; Taylor et al., 1999) as applied both to another yeast prion, Ure2p/[URE3] and to Sup35p/[PSI+] (Baxa et al., 2006; Glover et al., 1997). In both cases, the N-domains are envisaged to switch from unfolded to amyloid conformation (Nelson et al., 2005; Marshall et al., 2010) as they are recruited into the backbones of growing filaments, and these backbones are surrounded by the globular C-domains. The latter domains appear to retain their native folds, as attested by Ure2p's invariant thermal profile (Baxa et al., 2004) and retained glutathione peroxidase activity (Bai et al., 2004), and by Sup35p's retention of its GTPase activity (Krzewska et al., 2007). Accordingly, their mechanisms of inactivation are not based on unfolding of domains with consequent loss of function. Ure2p appears to be inactivated by steric impedance whereby its C-domain is prevented from binding (and thus inhibiting) the transcription factor Gln3p (Baxa et al., 2002). However, the mechanism whereby Sup35p activity is down-regulated has not been evident. Other differences exist between the two prion proteins. Ure2p has no obvious sequence that serves as counterpart to the M-domain of Sup35p, although it has a linker region of about 25 amino acids connecting the Asn/Gln-rich region (residues 1 -65) with the C-domain - see Discussion. And whereas Ure2p is a dispensable protein, Sup35p is essential and must retain some, albeit greatly reduced, activity.
In this study, we used a variety of EM approaches to investigate the architecture of Sup35p filaments assembled in vitro, aiming in particular to elucidate the dispositions of the M- and C-domains. A strong indication that filaments assembled in vitro are essentially the same material as the prion was given by the demonstration that yeast cells may be infected by transfecting them with filaments assembled in vitro from purified recombinant Sup35p-NM (King and Diaz-Avalos, 2004; Tanaka et al., 2004).
The [PSI+] prion exhibits phenotypically distinct strains, so-called variants (Derkatch et al., 1996). The most obvious distinction between variants is according to their residual activity in translation termination, with “strong” variants referring to low residual activity and conversely. We assembled filaments from purified recombinant full-length Sup35p at 4°C, under conditions that give rise, predominantly, to strong [PSI+] variants when the assembly products are transfected into yeast cells (Tanaka et al., 2004; see Methods).
Electron micrographs of freshly assembled filaments, prepared by negative staining, show a core fibril surrounded by irregularly distributed globules (Figs. 1A & 1B). The core fibrils are 8 - 10 nm wide and they do not have a stain-penetrable lumen; the globules are ~ 5 nm in diameter. As the peripheral material is diffuse, the outer limit to which it extends is not a conspicuous feature but, on careful examination of the micrographs, it is readily apparent; moreover, this edge can be enhanced by band-pass filtering (Fig. 1C). These data put the outer diameter of the filaments at ~ 65 nm.
The core fibrils are similar in appearance and diameter to N-domain fibrils, i.e. of Sup35p1-114 (Fig. 1F) and NM-domain filaments of Sup35p1-253 (Fig. 1E), although the latter appear to be somewhat thicker and to have less sharply defined edges (Kishimoto et al., 2004). Based on these observations, we infer that the peripheral globules seen on filaments of full-length Sup35p (and not on NM-fibrils) represent C-domains and their wide displacement from the core fibril implies that the M-domains are in extended conformations.
The core fibrils of all three constructs have either a wavy appearance with an axial repeat of 40 – 60 nm, or appear straight without an evident repeat. Typically, a given fibril maintains the same appearance along its length. These indications of fibril polymorphism are reminiscent of similar features exhibited by fibrils assembled from Ure2p N-domain fusions (Baxa et al., 2002) and of discreet classes of fibril distinguished in STEM analyses of specimens assembled from constructs in which the N-terminal half of the Sup35p N-domain was fused to GFP (Diaz-Avalos et al., 2005).
After protracted storage (2-12 months) at 4°C, filaments of full-length Sup35p are altered in appearance, with a diameter of ~30 nm and rough, irregular, edges (Fig. 1D). They resemble the image published by Glover et al. (1997). We also observed a similar albeit less extreme effect for filaments in the presence of high ionic strength (data not shown). We attribute this change in appearance vis-à-vis freshly prepared filaments to a coagulation of C-domains on to the core fibril.
To confirm that negative staining of freshly prepared Sup35p filaments conveyed a native-like appearance, we also observed vitrified specimens by cryo-EM (Figs. 2A, B). These images depict core fibrils of 8 - 10 nm diameter and, albeit less evidently than in negative stain, a cloud of surrounding globules, giving an overall diameter of ~ 60 nm. Cryomicrographs of NM-filaments (Fig. 2D) and N-fibrils (Fig. 2E) depict rather featureless, non-hollow, fibrils similar in width to each other and to the core fibril of full-length filaments (Fig. 4). These results confirm that in filaments of full-length Sup35p in solution, the M-domains are in extended conformation(s).
The disposition of the M-domains in NM-filaments, i.e. in the absence of C-domains, is less clear although the images are consistent with them also being extended and on the outside of the backbone fibril. An additional observation supports this inference. When NM-fibrils are confined to a thin film for vitrification, adjacent, approximately coplanar, fibrils tend to arrange themselves such that the inter-fibril spacing is roughly uniform (Fig. 2D). This equi-spaced arrangement is to be expected if mutual electrostatic repulsion, dominated by the charged M-domains, were to be a major contributor to the total free energy of the system, as this contribution would be minimized by such an arrangement (see Supp. Materials).
Dark-field STEM microscopy of unstained specimens (Muller and Engel, 2001; Thomas et al., 1994) affords another imaging method in which contrast is generated directly by scattering from the specimen, not indirectly as with staining or shadowing. To preserve native structures, specimens are prepared by freeze-drying although this procedure is less efficacious than vitrification (as used in cryo-EM). There are, however, compensating advantages. Dark-field STEM is operated in-focus and records a signal that is linearly proportional to projected mass. This means that faintly defined edges can be detected without the complication of phase contrast interference fringes, and that images may be quantitated to yield molecular mass measurements. The latter property has been exploited extensively in studies of amyloid fibrils (Goldsbury et al., 2005; Baxa et al., 2006; Goldsbury et al., 2010).
We applied this technique to filaments of full-length Sup35p (Fig. 3A), NM-filaments (Fig. 3B), and N-fibrils (Fig. 3C). Measurements of mass-per-unit-length were performed in each case, and yielded symmetrical unimodal distributions (Figs. 3D-F), whose means and standard deviations were, respectively: 166 ± 17 kDa/nm; 55 ± 7.4 kDa/nm; and 28.7 ± 2.9 kDa/nm. To compare filaments assembled from subunits with different masses, it is instructive to consider the numbers of subunits per axial step of 0.47nm (the cross-beta spacing). Thus we obtained 0.99 ± 0.10 subunits per step for filaments of full-length Sup35p; 0.99 ± 0.13 subunits per step for NM-filaments; and 0.91 ± 0.10 subunits per step for N-fibrils. To within experimental error, the axial packing density was the same in each case, in agreement with models in which the N-domains form a backbone that is decorated with the M- and C-domains, if present. These results are consistent with 1 subunit per 0.47 nm, in agreement with the majority of STEM mass-per-unit-length data recorded for Sup35p1-61-GFP filaments (Diaz-Avalos et al., 2005). The latter study also reported one class of filament that was ~ 20% higher in linear density. While noting this discrepancy, we note that there are differences between the respective systems: the filaments studied here were assembled from constructs containing the full-length N-domain, not a fusion protein involving roughly half of the N-domain; and their assembly was spontaneously nucleated, not seeded with yeast extracts.
As for the radial dimension, filaments of full-length Sup35p are seen to have compact core fibrils surrounded by diffuse peripheral density. NM-filaments have fuzzier edges than N-fibrils but are otherwise similar to them. These observations are quite consistent with those from the other imaging methods employed. From these data, axially averaged profiles of projected density were calculated (Supp. Fig. 1) and from them, radial density profiles (Figs. 3D & E). They show a common core structure surrounded by additional density in the NM-filaments that extends further in the filaments of full-length Sup35p. In comparison, the transverse profiles of projected density from the cryo-EM data show very similar profiles in shape and extent for all three kinds of filament (Fig. 4), consistent with their having similar or identical backbones, and diffuse peripheral material.
Imaged by multiple EM techniques, in vitro-assembled filaments of full-length Sup35p consistently show an “amyloid fibril backbone” architecture. Their core fibrils match in diameter and appearance, amyloid fibrils formed from N-domains alone (Figs. 1--3).3). The surrounding globules are not seen with NM-filaments, indicating that they are contributed by C-domains: moreover, their size (~ 5 nm) is consistent with a globular protein of 45 kDa. As there are no reports of N- or M-domains forming globular particles in this (or any) size range, C-domains are left as the only candidate. It remains to account for the M-domains that must connect the peripheral globules (C-domains) to core fibrils (N-domains). In view of the highly dispersed radial distributions of these globules (the full filament diameter is ~ 65 nm), the M-domains must assume extended conformations.
Much evidence supports the proposition that Ure2p filaments have an amyloid fibril backbone (Baxa et al., 2006). In this system, the full extent of the polypeptide segment linking the N-domain to the C-domain is unclear although it should end by Met95, the first residue of the C-domain (Umland et al., 2001). Met95 is preceded by a run of 25-30 residues in which there is a higher incidence of charged residues than in the preceding residues 1-65 (which suffice for fibrillation), and part or all of this segment may form the linker (Baxa et al., 2003). In either event, this segment is much shorter than the Sup35p M-domain and this disparity correlates with the diameters of the respective filaments (25-30 nm for Ure2p - Baxa et al., 2003, Ranson et al., 2006 - and ~ 65 nm for Sup35p - see above). The latter number is consistent with an end-to-end linker length of ~ 30 nm if it is oriented radially, and longer if it runs in a more tangential direction.
If we take the M-domain to be 130 residues, the observed linker length of 30 nm corresponds to ~ 2.3 Å per residue, on average, suggesting an unfolded conformation. If the polypeptide were to be completely unfolded, the step could be up to 3.5Å per residue, which would correspond to a contour length of ~ 45 nm. In view of the strong likelihood of curvature, the measured end-to-end length of ~ 30nm is quite compatible with such a contour length, although it does not rule out the possibility of short structured regions in the M-domain. The inference of an unfolded M-domain is further supported by results from analyzing the Sup35p sequence with algorithms designed to identify “natively unfolded” regions (Foldindex; (Prilusky et al., 2005); RONN: (Yang et al., 2005)) - data not shown. It is also consistent with data from solid state NMR and H/D exchange data (Shewmaker et al., 2009; Toyama et al., 2007) on NM fibrils in solution which showed that only small parts of the M domain (potentially, the most N-terminal parts) are ordered, and with unpublished circular dichroism data on M-domains cited by Liu et al. (2002). Our results show that in solution only a small part of the M-domain, most likely the most N-terminal stretch adjacent to the prion domain, can be arranged in an ordered, parallel in-register structure as has been suggested by the basis of solid state NMR measurements on dried and wet NM-fibrils (Shewmaker et al., 2009; Shewmaker et al., 2006). Moreover, the high incidence of charged residues in the M-domain tells against a parallel in-register arrangement as it would bring many like residues close together, despite mutual electrostatic repulsion.
Nevertheless, involvement of some of the M-domain in amyloid formation is also indicated by the fact that residues 124-147 (the N-terminal 10%) of the M-domain are required for the faithful maintenance of certain variants in vivo (Bradley and Liebman, 2004), raising the possibility that they may act as an extension of the N-domain in forming amyloid. If the M-domain is, in effect, shorter than 130 residues, the average step per residue will be proportionately longer than 2.3 Å, pointing even more firmly to a conformation that is largely if not wholly unfolded.
To fulfil its function, Sup35p has to bind Sup45p and interact with a translating ribosome to promote termination. As Sup35p is an essential protein, this function cannot be completely lost in [PSI+] cells. In general, it has been suggested that residual soluble Sup35p provides the activity needed for prion-infected cells to survive (Tanaka et al., 2004). However, very little soluble Sup35p protein has been detected, especially in strong strains (loc. cit., their Supp. Fig. S3; also, Zhou et al., 1999). Our current picture of the filament structure suggests another possibility, i.e. the C-domains could still accomplish their task while attached to the backbone fibril via the long flexible M-domain, albeit with reduced efficiency (Fig. 5). The proposition that the C-domains may retain some activity in the prion form of Sup35p raises the possibility that filament formation represents a mechanism for down-regulation, whereby total activity in an infected cell has contributions from filaments and from a small pool of still-soluble protein and is reduced accordingly, relative to an uninfected cell in which all Sup35p is soluble.
The phenomenon of highly charged, extended, arms protruding from a protein filament backbone is reminiscent of the neurofilament subunits NF-M and NF-H, whose long C-terminal extensions become heavily phosphorylated and serve as spacers that appear to keep adjacent neurofilaments much further apart than the nominal diameter of their backbones (~ 10nm; Zhu et al., 1997), either by a bridging mechanism (Hisanaga and Hirokawa, 1989) or an “entropic brush” mechanism (Brown and Hoh, 1997).
Several models have been proposed for the amyloid backbone formed by polymerized N-domains. These include stackings of β-helices (Kishimoto et al., 2004; Krishnan and Lindquist, 2005). The STEM mass-per-unit-length data have implications for this proposal, in light of the properties of known β-helices (more generally, β-solenoids). If the fibril backbone were to consist of a single protofibril, it would have to have only a single coil (with 4.7 Å axial rise) per subunit. Coils with various cross-sectional shapes have been observed in crystal structures of β-solenoid proteins and they typically have 19-30 residues per coil (Kajava and Steven, 2006). As the N-domain has 123 residues, only 15-24% of it would then be in the cross-β backbone and the remaining 76-85% presumably in appended loops. There are no precedents for β-helices with such a small content of cross-β structure. Moreover, known β-helices have diameters of 40-45 Å, about half the N-domain fibril diameter. Alternatively, the fibril might consist of 2, 3 or 4 bundled protofibrils with, respectively, 2, 3 or 4 coils per subunit in each protofibril. However, in documented cases of amyloid fibrils in which multiple protofibrils coil around a central axis, electron micrographs show pronounced periodic thickening and thinning (e.g. β-amyloid - Goldsbury et al., 2000; IAPP - Goldsbury et al., 1997). Het-s fibrils afford a well authenticated example of a β-helical amyloid whose single protofibril has 2 coils per subunit (Wasmer et al., 2008). In fibrils with three protofibrils, the axial periodicity is conspicuous (Sen et al., 2007). Sup35p N-domain fibrils do not exhibit such variations in width (Figs. 1--3).3). Moreover, only the one coil-per-subunit scheme would be consistent with a parallel in-register arrangement (Shewmaker et al., 2006). Taken together, these considerations make this kind of model very unlikely unless the putative β-helices are quite different from known examples of this structure.
Another class of fibril model is “parallel superpleated β-structures” which envisage planar serpentine folds for the subunit and axial stacking in the fibril. This topology was suggested for Sup35p N-domain fibrils by Kajava et al. (2004) and subsequently advocated by Shewmaker et al. (2006). Kajava et al. derived a model of this kind for Ure2p N-domain fibrils to account for cross-β-type diffraction data, STEM mass-per-unit-length data, core fibril dimensions and its lack of a hollow lumen, among other constraints. Its extension to Sup 35p N-domains was based on their having similar amino acid sequences (Asn/Gln-rich, low on hydrophobic and charged residues). Shewmaker et al. (2006 & 2009) presented solid-state NMR data that support a “parallel in-register” arrangement. All parallel superpleated β-structures structures have one subunit per 4.7 Å axial step and are therefore consistent with the STEM mass measurements reported here. However, they can vary in the shape of the serpentine fold. In this context, we agree with Shewmaker et al. (2006) that the number of β-sheets (in this context, also the number of β-strands per subunit) and the positions of the turns have yet to be demonstrated.
The complete SUP35 gene as well as genes coding for Sup35p1-114 and Sup35p1-253 were cloned into a pET-17b expression vector by PCR using NdeI and XhoI restriction sites and pQE10-SUP35 (kindly gifted by V. Kushnirov) as template. All genes encode an additional RGS-His tag at the N-terminus with the sequence MRGSH6TDLAT. The cloned constructs resulted in proteins with a molecular mass of 78.2 kDa, 26.1 kDa, and 14.8 kDa for Sup35p, Sup35p NM-domain, and Sup35p N-domain, respectively. The expression vectors were transformed into BL21 Rosetta (DE3) pLysS (Novagen) and overexpression was induced by adding 1mM IPTG to growing cells at an OD of ~0.8 at 600 nm in LB medium containing 0.1 mg/ml ampicillin at 37°C, and continued for 4 hours.
Full-length Sup35p was purified by resuspending pelleted expressing cells in 30 mM Tris, 300 mM KCl, 20 mM imidazole, 10% glycerol, 5 mM MgCl2, 2mM β-mercaptoethanol, pH 8.0 containing protease inhibitors (Complete EDTA-free, Roche Applied Science) and lysing them by sonication. Insoluble material was removed by centrifugation (50000g, 1 h, 4°C), and the soluble fraction loaded onto a Ni-NTA column (Qiagen) and eluted with a gradient of 20 mM to 250 mM imidazole in the same buffer. The cleanest fractions were pooled and concentrated (Centricon, 50 KDa MWCO) and further purified on a Superdex 200 column (Amersham Biosciences) in the same buffer. Filaments were prepared by incubating purified protein at concentrations of ~2 mg/ml at 4°C for more than 16 hours. For cryo-EM, the original buffer was exchanged to 20 mM sodium phosphate, 200 mM NaCl, pH 7.4 shortly before vitrification.
As both Sup35p1-114 and Sup35p1-253 were found mostly in the insoluble fraction of the cell lysate, purification was performed on a Ni-NTA column under denaturing conditions. Filaments were assembled by incubating purified protein at concentrations of ~2 mg/ml at 4°C for > 16 hours in 20 mM sodium phosphate, 200 mM NaCl, pH 7.4.
For negative staining, samples were absorbed onto freshly glow-discharged carbon-coated grids, rinsed with water, and stained with 1% uranyl acetate. Specimens were examined on a CM120 TEM (FEI, Hillsboro, OR). For cryo-EM, drops were absorbed to holey carbon films, blotted, vitrified, and imaged on a CM200-FEG microscope (FEI, Hillsboro, OR), as described by Cheng et al. (1999).
Freshly prepared fibrils in 30 mM Tris/HCl pH 8.0, 300 mM KCl, 10% glycerol, 5 mM MgCl2, were freeze-dried according to standard procedures of the Brookhaven STEM facility (Sen et al., 2007). Tobacco mosaic virus (TMV) was included to serve as an internal mass-per-unit length standard (131.4 kDa/nm). After a 2 μl drop of specimen was adsorbed for 1 min, the grid was washed 10 times in water, then blotted and plunged into liquid nitrogen slush. The frozen grid was freeze-dried overnight with gradual warming to -80°C and transferred under vacuum to the STEM. Grids were scanned at -150°C on a custom built STEM at 40 kV with a probe focused to 0.25 nm with a dwell time of 30 μs per pixel.
Dark-field micrographs of 512×512 pixels were recorded at raster steps of 1.0 or 2.0 nm per pixel. The images were analyzed using PCMass (available from the Brookhaven STEM resource) and Bsoft (Heymann, 2001; Heymann and Belnap, 2007). Boxes of 40 nm length and appropriate width (80 nm for full length Sup35p and Sup35p NM and 20 nm for Sup35p N) were used for the measurements. Histograms were calculated with appropriate bin-widths (2.0 kDa/nm for Sup35p1-114 and Sup35p1-253 and 5.0 kDa for full-length Sup35p). A Gaussian distribution was then fitted to the main peak using the Marquardt-Levenberg algorithm, as implemented in SigmaPlot (Systat Software).
Projected densities from dark-field STEM and cryo-EM images were calculated by extracting and computationally straightening filaments which were then projected along the filament axis using Bsoft (Heymann, 2001; Heymann and Belnap, 2007). Radial density profiles were calculated from the STEM data, as described (Steven et al., 1984).
Supplementary Figure 1: Averaged profiles of projected density calculated from STEM micrographs of filaments of full-length Sup35p, Sup35p-N, and Sup35p-NM.
We thank Ms B. Lin for STEM sample preparation and Dr M. Simon for STEM data collection, Drs B. Heymann and N. Mizuno for help with image analysis, and Dr D.C. Masison for helpful discussions. This work was supported by the Intramural Research Program of NIAMS. The BNL STEM is supported by USDOE-OHER.
*In the present context, we use the term “fibril” to denote a protein polymer whose structure is all or almost all amyloid, and “filament” for a structure that has both amyloid and non-amyloid components or whose status in this respect has not been determined.