Asymmetric-flow field-flow fractionation of prions reveals a strain-specific continuum of quaternary structures with protease resistance developing at a hydrodynamic radius of 15 nm

Prion diseases are transmissible neurodegenerative disorders that affect mammals, including humans. The central molecular event is the conversion of cellular prion glycoprotein, PrPC, into a plethora of assemblies, PrPSc, associated with disease. Distinct phenotypes of disease led to the concept of prion strains, which are associated with distinct PrPSc structures. However, the degree to which intra- and inter-strain PrPSc heterogeneity contributes to disease pathogenesis remains unclear. Addressing this question requires the precise isolation and characterization of all PrPSc subpopulations from the prion-infected brains. Until now, this has been challenging. We used asymmetric-flow field-flow fractionation (AF4) to isolate all PrPSc subpopulations from brains of hamsters infected with three prion strains: Hyper (HY) and 263K, which produce almost identical phenotypes, and Drowsy (DY), a strain with a distinct presentation. In-line dynamic and multi-angle light scattering (DLS/MALS) data provided accurate measurements of particle sizes and estimation of the shape and number of PrPSc particles. We found that each strain had a continuum of PrPSc assemblies, with strong correlation between PrPSc quaternary structure and phenotype. HY and 263K were enriched with large, protease-resistant PrPSc aggregates, whereas DY consisted primarily of smaller, more protease-sensitive aggregates. For all strains, a transition from protease-sensitive to protease-resistant PrPSc took place at a hydrodynamic radius (Rh) of 15 nm and was accompanied by a change in glycosylation and seeding activity. Our results show that the combination of AF4 with in-line MALS/DLS is a powerful tool for analyzing PrPSc subpopulations and demonstrate that while PrPSc quaternary structure is a major contributor to PrPSc structural heterogeneity, a fundamental change, likely in secondary/tertiary structure, prevents PrPSc particles from maintaining proteinase K resistance below an Rh of 15 nm, regardless of strain. This results in two biochemically distinctive subpopulations, the proportion, seeding activity, and stability of which correlate with prion strain phenotype.


Introduction
Transmissible spongiform encephalopathies (TSEs), also known as prion diseases, comprise a group of lethal neurodegenerative disorders that affect several mammalian species including humans [1,2]. The central molecular event in TSEs is the misfolding and aggregation of the host-encoded prion glycoprotein, PrP C , into a wide variety of assemblies associated with disease, collectively called PrP Sc [3,4]. During the course of the disease, a constellation of PrP Sc particles, ranging from small, soluble, protease-sensitive (PrP sen ) oligomers, to large, less soluble, partially protease-resistant (PrP res ) fibrils, progressively accumulate in the brain. Under controlled experimental transmission conditions, a specific disease phenotype, characterized by incubation period, clinical signs of disease, PrP Sc glycotype, and tissue tropism, can be reproduced in hosts with the same genetic background, leading to the operational concept of prion strains [2,5]. It is well established that different prion strains arise from different conformations of PrP Sc , but within a given strain, there is also a diversity of PrP Sc structures that may further contribute to phenotype [6][7][8][9][10][11][12]. However, the molecular basis linking PrP Sc strain-specific structures and disease phenotypes remains elusive, in large part because isolating such structurally and biochemically diverse PrP Sc particles is a challenging task.
One approach to isolating PrP Sc particles has been sedimentation velocity. This technique has generated data that supports a strain-specific size distribution of PrP Sc particles and has provided evidence for an association between PrP Sc quaternary structure and its infectivity [13][14][15][16][17]. However, the limited resolution of this technique, differences in sample preparation and experimental conditions among laboratories, and the challenges of accurately assigning size values to isolated particles, have created inconsistent descriptions of the PrP Sc subpopulations contained in a prion isolate. A higher resolution technique, size exclusion chromatography (SEC), has been also used to investigate PrP Sc structural heterogeneity within prion strains [18][19][20][21], but limitations related to protein-stationary phase interaction and fractionation range have restricted SEC to the study of the smallest PrP Sc oligomers present in a prion isolate, excluding from the analysis a major proportion of the largest and, potentially, most biologically relevant PrP Sc particles.
We have adopted asymmetric-flow field-flow fractionation (AF4) as an ideal technique for isolating and characterizing PrP Sc particles. AF4 is a flow-based separation technique widely used for the isolation and characterization of nanoparticles and polymers in the pharmaceutical industry, and is increasingly being applied to the characterization of biological macromolecules, protein complexes, extracellular vesicles, and viruses [22][23][24]. Two perpendicular flows, forward laminar flow and crossflow, are applied to particles, which are then separated based on their hydrodynamic properties. The technique allows for high resolution and reproducibility, and by adjusting crossflow gradients during the fractionation run, a very wide range of particle sizes can be separated, from a few nanometers to several micrometers [25]. Unlike SEC, the absence of a stationary phase greatly reduces the problems of sample-stationary phase interaction, which is particularly troublesome for "sticky" particles like large PrP Sc aggregates. It also allows for fractionation of complex mixtures such as brain homogenate. The use of inline multi-angle and dynamic light scattering (MALS and DLS) detectors allows for precise measurement of the size and shape of the eluting particles.
Surprisingly, the use of AF4 technology to study neurodegenerative disease-related protein aggregates has been very limited to date [26][27][28][29]. Within the field of prion disease research, AF4-MALS-DLS was first used to determine the size of the most infectious prion particle [26]. In this seminal work, PrP Sc from 263K-infected Syrian hamster brains was first highly purified and digested with proteinase K, then disrupted by detergent, sonication and freeze-thaw before being fractionated and analyzed. Recently, we developed a method for using AF4 on brain homogenates without the need for initial purification or sonication, and with in-line DLS measurements, we were able to ascertain the size distribution of PrP Sc particles in the brains of RML-infected mice [29]. Now we have applied our isolation method and AF4-MALS-DLS analysis to the characterization of inter-and intra-strain differences of subpopulations of PrP Sc particles in the brains of Syrian golden hamsters infected with three different prion strains. All three strains share the same primary PrP structure; two strains (Hyper and 263K) were derived from different sources but produce almost identical phenotypes and share biochemical properties, the third (Drowsy) has a different phenotype and biochemical profile [6,10,30]. Our highresolution fractionation and analysis provides the most detailed description, to date, of PrP Sc particles in the brain at terminal stage of disease. We demonstrate that quaternary structure is a major contributor to PrP Sc heterogeneity, but for all strains studied, a fundamental change, likely in secondary/tertiary structure, prevents PrP Sc particles from maintaining proteinase K resistance below an R h of 15 nm. These distinct populations of PrP sen and PrP res also have strain-specific changes in glycosylation pattern and templating activity.

Solubilization step results in a high yield of soluble PrP aggregates for all strains, with no change in PK-resistance patterns compared with starting material
Our goal was to isolate PrP Sc aggregates, defined as all types of PrP aggregates that are specifically associated with prion infection, whether directly infectious themselves and resistant to proteinase K or not. To solubilize PrP from membranes while preserving PrP Sc assemblies close to their "natural" state, we incubated strains 263K, HY or DY brain homogenate (BH) with 2% dodecyl-β-maltoside and 2% sarkosyl sequentially, as optimized by Tixador et al. [14].
Detergent-insoluble particles that could affect the quality of our AF4 fractionations were removed by a short centrifugation. We performed batch mode DLS measurements on the supernatant and pellet and found similar sized particles were isolated for each strain: the supernatant contained particles of 3-300 nm R h compared with 40-600 nm R h in the pellet (Fig 1A and 1B). By immunoblot, the amount of PrP in the supernatant and pellet, before PK digestion, was similar for the three strains ( Fig 1C and 1D), with a supernatant:pellet proportion of 72:28. Treatment of both supernatant and pellet with proteinase-K (PK) yielded the expected PrP res banding patterns, with the unglycosylated isoform migrating at 21 kDa for HY and 263K and 19 kDa for DY [30]. The ratio PrP res :total PrP was higher for HY and 263K than for DY with values of 0.56 (±0.05), 0.59 (±0.04) and 0.38 (±0.08), respectively, for supernatant PrP and 0.58 (±0.05), 0.57 (±0.06) and 0.29 (±0.10) for the pelleted PrP. Assuming all animals have comparable levels of PrP C , this was consistent with the fact that DY PrP Sc is less resistant to PK than 263K and HY PrP Sc [30].

Strains with the same phenotype have the same PrP Sc size distribution
Batch mode DLS measurements of the soluble starting material indicated that particles had a size distribution of 3-300 nm R h (Fig 1A). We, therefore, optimized the AF4 crossflow gradient for separating particles of 3-300 nm R h within a 60 min run (see Methods for details). We injected 250 μg of total soluble protein and collected 200 μL fractions each minute. As expected, all fractograms had similar UV absorbance profiles (Fig 2A), indicating a similar Particle size distributions determined by DLS measurements of detergent-soluble (A) and pellet (B) fractions after sequential incubation of 263K (red), HY (green), and DY (blue) brain homogenates with 2% dodecyl-β-maltoside and 2% sarkosyl, followed by 10 min centrifugation at 20,000 x g. Representative immunoblotting (C) and densitometric analysis (D) of detergent-soluble (S) and pellet (P) fractions of 263K (red), HY (green) and DY (blue). The pellet was resuspended in half of the original volume; the same volumes of S and resuspended P fractions were loaded in the polyacrylamide gel. Twice the volume of PK treated samples (20 μg/mL of PK for 1h at 37˚C) were loaded in the same gel for quantification. SAF83 (dilution 1:10,000) was used as primary antibody. Bars represent average ± SE of three BHs for each prion strain.
https://doi.org/10.1371/journal.ppat.1009703.g001 amount and composition of proteins in the BH of the three strains. The R h of the eluting particles measured by in-line DLS matched that seen in batch mode DLS, 3-300 nm R h .
To determine PrP elution profiles, fractions were immunoblotted for PrP and PrP signal intensity quantified (Fig 2B). The PrP intensity from each fraction was plotted against the average R h of the particles present in that fraction and the area under the curve from the entire run was normalized to 1 (Fig 2C).
Because our samples were not protease-digested before fractionation, the PrP in the sample contained normal PrP C in addition to PrP Sc . To determine the contribution of PrP C to the , and DY (blue) detergentsolubilized BHs. Solid curves: UV signal (protein content) at 280 nm; circles: R h (average ± SD) from thirty DLS measurements taken during the one-minute collection for each AF4 fraction; rhombus: particle length, calculated by fitting MALS signal to a rod-shape model. B) Representative immunoblots for PrP content in collected AF4 fractions; fractions from each brain homogenate were loaded in two polyacrylamide gels (gel 1: fraction 2 to 26/28, gel 2: fraction 28/30 to 50/54). SAF83 (dilution 1:10,000) was used as primary antibody. C) Size distributions of prion particles depicted as relative amount of PrP (from PrP immunoblot intensity; average ± SE) contributing to each particle size (R h (nm)). Non-infected (dotted line), 263K (red), HY (green), DY (blue). Three brains for each prion strain were analyzed and averaged. D) Electron micrographs of fractions #47 (> 70 nm R h ) from HY (left) and DY (right) AF4 fractionation. No fibrillar particles were visible on the same fraction number from NBH fractionation. Bars = 100 nm. E) Density of PrP Sc particles (average ± SE) for each particle length. 263K (red), HY (green), DY (blue). The correlation between particle length and R h is shown as a black dotted line. total PrP in the size distribution curves, we analyzed uninfected normal brain homogenate (NBH) under the same fractionation conditions (Fig 2B and 2C). All PrP C from age-matched uninfected brain homogenates eluted within the first 16 fractions, with the maximum PrP signal intensity corresponding to particles of~4 nm R h , and no signal in particles larger than 7 nm. Conversely, prion-infected samples had PrP particles ranging from 3-300 nm R h .
Interestingly, almost identical PrP elution patterns were seen for 263K and HY, but there was a very different pattern for DY. For 263K and HY, the PrP signal intensity increased by four times from 7 to 40 nm R h particles, and the signal intensity remained almost constant as particle size increased beyond 40 nm R h . In contrast, the PrP distribution for DY showed a constant decrease in PrP signal from 7 to 300 nm R h . At~30 nm R h , the size distribution curves from the three strains intersected, indicating that DY contains more PrP Sc particles smaller than 30 nm, and 263K and HY have more PrP Sc particles larger than this.
For PrP Sc particles increasing in size from 60-200 nm, particle density remains constant for 263K and HY but decreases for DY To determine how many PrP Sc particles were present in each fraction (particle density), we needed information about particle length and PrP content for each fraction. Electron microscopy images of larger fractions from HY and DY indicated that both contain fibrillar material, likely prion fibrils based on appearance ( Fig 2D). In addition, previous light scattering data on purified 263K particles has revealed that assemblies of 12 nm R h and larger are transitioning to a fibrillar form [26]. Therefore, we fit our MALS data to a rod-shaped model to determine particle lengths in fractions greater than 15 nm R h (Fig 2A and Table 1).
Once particle length was determined, we could use MALS data to directly calculate particle density (see Methods for details). However, because our starting material was brain homogenate, particles other than PrP Sc were present in the fractions. To overcome this issue, we analyzed only fractions 46-50, with particles of R h >70nm and highly enriched for PrP Sc ; these The R h of particles was calculated from DLS measurements. The length of the particles was calculated fitting the MALS measurements to a rod-shape model. The number of PrP monomers per PrP Sc particle was estimated from the PrP Sc particle length, assuming a contribution of 0.49 nm in length per PrP monomer based on the recent parallel in-register intermolecular beta sheets (PIRIBS) model for the 263K strain. The PrP normalized intensity was determined by immunoblotting using Sha31 antibody. The relative number of PrP Sc particles was calculated by dividing the PrP normalized intensity by number of PrP monomers per PrP Sc particle. The maximum possible PrP Sc particle density was calculated with ASTRA software (see methods) for fractions 46

PLOS PATHOGENS
Asymmetric-flow field-flow fractionation reveals molecular basis of prion strain variation fractions had the highest ratio of PrP:total protein as determined by PrP immunoblotting intensity and UV absorption at 280 nm. In addition, the light scattering intensity and the corresponding particle density was much lower for NBH than for infected BHs in this region of the fractogram, suggesting that the main contribution to light scattering signal in fractions 46-50 was from large PrP Sc particles (S1 Fig). The PrP Sc particle densities for particles of R h~8 0-90 nm were 4.51x10 8 (± 8.54x10 7 ), 6.43x10 8 (±1.43x10 8 ), and 2.0x10 8 (± 1.61x10 7 ) particles/ mL for 263K, HY, and DY, respectively (Fig 2E and Table 1).
We also wanted to ascertain PrP Sc particle densities for smaller fractions. Having determined the particle lengths in each fraction, we then estimated the number of PrP monomers that would fit into fibrils of a given length, based on a parallel in-register intermolecular beta sheets (PIRIBS) architecture, with one PrP monomer present for every 0.49 nm in fibril height [31]. For particles of R h 15 nm, this equaled particles of length~58 nm, comprised of 116 PrP monomers (see Table 1). Although lateral bundling of PrP Sc fibrils might also occur, its effect on mass per unit length calculations was not considered here.
Using the immunoblot data, we divided the PrP immunoblotting intensity from each fraction by the number of PrP monomers comprising the PrP Sc fibrils at that fraction. These values provided information about the relative amounts of PrP Sc particles in each fraction. Then we used the ratio of calculated particle density:PrP immunoblot intensity from fractions 46-50 to back-calculate the particle densities for the smaller fractions (Table 1).
PrP Sc size distribution plots indicated that HY and 263K had almost a constant PrP Sc particle density, 2.0x10 9 particles/mL, in the range of 60-200 nm length (from~15 to 40 nm R h according to DLS) ( Fig 2E). Above 200 nm in length, the particle density decreased. DY PrP Sc showed a more pronounced and sustained decrease in the number of particles as length increased from~50 to 800 nm.

Glycosylation profile changes with the size of prion particle in a strainspecific manner
The ratio of di-, mono-, and un-glycosylated isoforms of PrP is one criterion for strain typing; however, 263K, HY and DY share the same PrP glycosylation pattern when unfractionated BH is analyzed by immunoblotting [6,32]. Our analysis of unfractionated BH showed the same di: mono:un-glycosylated ratio of 38:36:26 for total PrP (Figs 1A and 3A) and 55:35:10 for PrP res for all three strains. Given the differences in glycosylation ratios between total PrP and PrP res , we questioned whether changes in glycosylation pattern might occur in different subpopulations of PrP Sc aggregates and, if so, whether these changes differed among strains. Indeed, immunoblot analysis of AF4 fractionations revealed strain-specific variation of PrP glycosylation patterns with the particle size.
The contribution of PrP C to the glycoform ratio in the first 16 fractions (particles smaller than 7 nm R h ) was determined by analyzing the NBH fractionation immunoblot (Fig 2B). More than 80% of PrP C eluted in fractions 6-10, where the glycoform ratio was 46:25:29 ( Fig  3B). Between fractions 11 to 16, the percentage of di-glycosylated isoform increased to 75:10:15, as expected, given that extra glycan groups increase the R h of PrP C . No appreciable PrP signal was detected after fraction 20, meaning that the contribution of PrP C to glycoform ratio calculations was limited to particles smaller than 7 nm R h .
Given that the host PrP C is the same in all prion strains studied (Syrian golden hamster), we predicted that the glycoform ratios of PrP within fractions 6-10 would be similar across all strains. However, whereas HY and DY had almost the same glycoform ratios as NBH in these fractions, 263K had a higher proportion of di-glycosylated and less un-glycosylated isoforms ( Fig 2B). This may indicate that there was less PrP C in these fractions, as suggested by the PrP size distribution analysis (Fig 2C), and/or a greater amount of small, highly di-glycosylated PrP Sc oligomers contributing to the PrP signal, that shifted the glycoform ratio of this strain in these fractions.
For particles larger than 7nm R h , 263K and HY had very similar glycosylation patterns. In particles of size range 7-15 nm R h , ratios were 65:25:10 and 60:30:10 respectively; for particles 15-45 nm R h , the ratio gradually shifted to 55:35:10 and 51:36:13 respectively, and then remained almost constant in particles of 45-300 nm R h (Fig 3B and 3C). Both the reductions in di-glycosylated and increases in mono-glycosylated isoforms were statistically significant for 263K and HY (Fig 3C).  Fig 1C). B) Percentage of di-(circles), mono-(triangles), and un-glycosylated (rhombus) PrP isoforms present in each AF4 fraction as calculated from NBH (empty symbols), 263K (red), HY (green), and DY (blue) immunoblots of three BHs for each strain (representative immunoblot shown in Fig 2B). The

PLOS PATHOGENS
Interestingly, DY had a distinct glycoform ratio which remained relatively constant at 45:38:17 for all particles larger than 7 nm. There was a trend towards a reduction in di-glycosylated particles with increasing size, but these changes were not statistically significant (Fig 3C).

HY is comprised of more stable PrP Sc particles than 263K
PrP Sc structural stability is another feature used for prion strain typing. Incubation of brain homogenate with increasing concentrations of denaturant has previously demonstrated that 263K and HY are structurally more stable than DY [10,12,33,34]. Additionally, differences in the exposure of N-terminal regions in a narrow range of denaturant concentrations revealed a higher stability of HY over 263K at tertiary and secondary structure [10]. Instead of measuring the structural stability of the whole population of PrP Sc particles present in a BH, we explored the dissociating effect of SDS micelles over the different sub-populations of PrP Sc particles. Using the same fractionation conditions as before, we increased the concentration of sodium chloride from 20 to 150 mM in the AF4 running buffer, promoting SDS micellar formation (S2 Fig). Under SDS micellar conditions, all prion strains underwent a change in size distribution of PrP Sc particles (Figs 4A, 4C and S3). As expected, the proportion of particles with R h4 nm, corresponding to monomeric (and/or dimeric) PrP, significantly increased in all cases. For DY, the amount of monomerization was the largest, increasing from 10% to 50% of the total population. In contrast, HY and 263K particles of R h~4 nm increased from 10% to only 25% of the population, indicating fewer particles were monomerized. Interestingly, AF4 fractionation also revealed subtle but statistical differences between 263K and HY. SDS micelles increased the population of particles with R h 4-10 nm more for 263K than for HY as measured by subtracting the cumulative curves from AF4 fractionation in the presence and absence of SDS micelles (Fig 4D). These results indicate a rank stability of PrP Sc quaternary structure of HY > 263K >>> DY.

PrP res size distribution is independent of the strain phenotype
To investigate the association between PrP size and protease resistance, we PK-treated the AF4 fractions depicted in Fig 2B and quantified the amount of PrP res by immunoblotting (Fig 5A). We used Sha31, an antibody that recognizes the epitope 145-152 of the hamster PrP, which is within the PK-resistant core of PrP Sc . The loss of epitope detection is interpreted as at least partial digestion of this epitope and the associated core. We therefore define PrP res as those particles whose Sha31 epitope remains stable under conditions of PK digestion. As expected, the AF4 fractions containing the smallest particles showed a lower PrP signal after PK digestion [13,15,18]. Surprisingly, despite the differences in size distribution of total PrP between HY, 263K and DY, immunoblotting analysis showed a very similar pattern of the PK resistant 20 kDa fragment and its mono-and di-glycosylated isoforms, with a signal increase starting at fractions 26-28 in the three strains ( Fig 5B). We plotted PrP res size distribution by correlating the R h measurements of AF4 fractionation, before PK digestion, with the immunoblotting intensity of the PK-treated fractions. PrP res signal appeared rather abruptly at~15 nm R h , and the intensity of PrP res signal reached a maximum at~40-50 nm R h , and was sustained until >200 nm R h (Fig 5C and 5D). This result strongly suggests that PrP Sc particles cannot maintain PK-resistance below 15 nm R h , regardless of strain.

Within each strain, PrP res particles have similar stability regardless of size
Next, we analyzed the structural stability of PrP res particles, defined here as the presence of epitope 145-152 under PK digestion conditions, as a function of size by incubating fractions larger than 25 nm R h with increasing concentrations of SDS followed by digestion with PK. As expected, immunoblotting analysis showed that PrP res particles were more stable for 263K than for DY (Fig 6). Under our experimental conditions, 263K PrP res intensities could be fitted to a sigmoidal curve, with 50% signal reached at 0.09-0.11% SDS. Conversely, DY PrP res signal was better fitted to a single exponential decay model with its 50% intensity occurring at 0.05-0.08% SDS (Fig 6). Regardless of inter-strains differences, the structural stability of PrP res particles within a given strain remained constant despite increasing size, at least in the range from 25 to 65 nm (Fig 6C). These results suggest that after acquiring PK resistance, no major conformational changes take place in PrP structure, at least around this epitope, other than the addition of more PrP units.

PrP Sc particle seeding activity varies with particle size in a strain-specific manner
It has been previously demonstrated that HY has a higher templating activity than DY in vitro and in vivo [12,35,36]. Here we used RT-QuIC assay to compare the templating activity of the AF4 fractions containing PrP particles of 10, 20, 40 and 70 nm R h from 263K, HY and DY (Fig  7). For 263K and HY, the aggregation of recombinant hamster-prion protein (SHaPrP) had a shorter lag phase when seeded with fractions containing larger particles (40 and 70nm R h ) than when seeded with the same volume of fractions containing smaller particles (10 and 20nm R h ) ( Fig 7A, 7B and 7D). Interestingly, all tested DY fractions had the same seeding activity, with lag phases comparable to the slower HY and 263K fractions (Fig 7C and 7D). We then evaluated the ability of these fractions to convert PrP C into PrP res in CAD5-PrP−/− (HaPrP) cells, a murine catecholaminergic cell line lacking endogenous mouse PrP expression and expressing SHaPrP [37]. After seven passages of CAD5-PrP−/− (HaPrP) cells post-exposure to particles of 8, 10, 30 and 70 nm R h from HY, the highest PrP res signal was found in cells treated with 30 and 70 nm R h particles (Fig 7E). Interestingly, we also detected PrP res in cells treated with the 8 nm R h particles, which are PrP sen particles as assessed by Sha31 antibody. While there are no reports of successful infection of CAD5-PrP−/− (HaPrP) cells with DY, surprisingly, we were able to detect PrP res signal for cells treated with 30 and 70nm R h particles from DY, although with much less intensity than those from HY. The lack of infectivity of PrP sen DY particles, despite their intrinsic seeding activity, could be influenced by their lower structural stability, and consequent higher clearance, than the HY particles of the same size. To more directly compare the seeding activity of particles of the same size among prion strains, we seeded the RT-QuIC reaction with the same amount of total PrP monomeric units, as measured by immunoblotting. For

PLOS PATHOGENS
particles smaller than 20 nm R h , HY and 263K showed higher seeding activity than DY, whereas for particles 40 nm R h and larger, DY particles were more efficient (Fig 7F). Of note, there was a statistically significant difference in lag phase between HY and 263K for particles bigger than 40 nm. For all these reactions, the amount of PrP added as seed was based on monomeric PrP content, not the number of PrP Sc particles. Thus, a direct correlation between seeding activity and size of PrP Sc was evident in the case of DY, with a shortening of lag phase as R h increases, but for HY and 236K, this correlation was harder to establish, since both templating activity and number of PrP Sc particles decreased with the increment of particle size.

Fig 6. Conformational stability of 263K and DY PrP Sc particles ranging from 25 to 65 nm R h . A) Representative immunoblots of 263K (top) and DY (bottom) fractions incubated with increasing concentrations of SDS and digested
with PK. 3F4 (dilution 1:3,000) was used as primary antibody. B) Densitometric analysis of 263K (top) and DY (bottom) of remaining PrP res after incubation of AF4 fractions with increasing concentrations of SDS. 263K data was fitted to a sigmoidal function whereas DY data was fitted to a single decay hyperbolic function. Samples from three BH for each strain were analyzed. C) SDS concentration at which PrP res reaches half of the maximum intensity as a function of particle size. Data represents average of 3 replicates and error bars represent SE. https://doi.org/10.1371/journal.ppat.1009703.g006

PLOS PATHOGENS
Asymmetric-flow field-flow fractionation reveals molecular basis of prion strain variation

Discussion
Prion strains consist of a spectrum of conformationally distinct PrP Sc particles that encode the information for a specific disease phenotype. However, the level of intra-and inter-strain heterogeneity of these PrP Sc particles, as well as the biological relevance of this heterogeneity, is poorly understood. Using an AF4-DLS-MALS configuration, we isolated and precisely determined the size distribution of PrP Sc particles in the brains of Syrian golden hamsters infected with 263K, HY or DY strains at end stage of disease. In addition, we provided a detailed

PLOS PATHOGENS
Asymmetric-flow field-flow fractionation reveals molecular basis of prion strain variation description of changes in glycoform profiles, resistance to protease, structural stability, and replication activity as a function of PrP Sc quaternary structure. Our data allow us to propose a model that explains PrP Sc heterogeneity.

PrP Sc quaternary structure is associated with disease phenotype
Analysis of unfractionated PrP Sc from HY and 263K brain homogenate previously demonstrated that they have very similar secondary structures, structural stabilities, replication activities, PK digestion profiles, and glycosylation patterns [6,12,38,39]. Here we show that these two prion strains with almost identical phenotype also have almost identical size distributions of PrP Sc particles, further supporting the association between PrP Sc quaternary structure and strain phenotype. Conversely, DY, a strain with different clinical signs, incubation period, brain titre, and brain lesion profile, has a substantially different PrP Sc size distribution, with a larger population of small PrP sen particles and smaller population of large PrP res particles than HY and 263K. The PrP sen :PrP res ratio was previously correlated with the incubation period in hamster-adapted prion strains, in which "fast" strains like 263K and HY have smaller PrP sen : PrP res ratio than "slow" strains like DY [10,40]. Our results not only confirm this association between PrP sen :PrP res ratio and incubation period, but also provide a detailed description of the size distribution and biochemical features of the PrP Sc particles comprising each of these two populations at end-stage of disease.

The combination of particle density, structural stability, and replication activity of PrP Sc particles contributes to prion strain incubation period
In addition to the PrP sen :PrP res ratio, PrP Sc chemical stability has been cited as an explanation for variable incubation periods. When unfractionated brain homogenate was treated with increasing concentrations of SDS or GdnHCl, an inverse correlation between PrP Sc chemical stability and incubation period was found for hamster-adapted prion strains [12,33,34]. Interestingly, the opposite situation was described for mouse-adapted prion strains, in which accumulation of less stable PrP Sc particles was found in strains with shorter incubation periods [41][42][43]. Two different hypotheses have been proposed to explain these converse correlations: a) the more stable hamster-PrP Sc particles have a lower rate of clearance in the brain, thus more time to replicate and to exert their neurotoxic effects, shortening the incubation period; b) the less stable mouse-PrP Sc particles have a higher rate of fragmentation which generates more replicative nuclei, causing accelerated PrP Sc propagation and shortening the incubation period [10,12,36,[42][43][44]. Our study shows that both structural stability and templating activity change with the size of PrP Sc particles and do so in a strain-specific manner. Since the particle density of PrP Sc subpopulations is also strain-specific, we propose that the incubation period of a prion strain is a consequence of the combination of the stability, the templating activity, and particle density of each PrP Sc subpopulation accumulated in the brain. For example, highly replicating PrP Sc particles, if only present in low amounts, will not exert much effect on the overall replication activity of the strain. This is the case for DY. We found that PrP res particles from DY have higher replication activities than PrP res from HY and 263K, possibly because their lower stability creates more replication nuclei; however, the total amount of these DY PrP res particles is much lower than the amount of the less replicative HY and 263K PrP res particles. Furthermore, the predominant population in DY is PrP sen , which has a lower replication activity than the PrP sen from HY and 263K. In addition, the lower stability of DY PrP Sc particles, which favours replication by creating new seeds, could also reduce replication capacity by promoting clearance. Thus, the combination of more PrP sen particles with low replication efficiency and structural stability and fewer, less stable PrP res particles with higher replication efficiency, leads to an overall combined effect that DY is less efficient at replicating, resulting in a longer incubation period.

Protease resistance develops at a hydrodynamic radius of 15 nm, regardless of strain
Surprisingly, despite the marked differences in size distribution of total PrP (PrP sen plus PrP res ) and in the ratio of PrP sen /PrP res between the "fast" and "slow" strains studied here, the size distributions of PrP res are almost identical. In all cases, we observed a transition from PrP sen to PrP res starting at R h~1 5nm. This result strongly suggests a strain-independent conformational change in PrP Sc structure once particles reach 15 nm R h . Alternately, if the smallest particles with partial resistance to PK digestion simply came from the fragmentation of larger PrP res particles in the brain or during sample preparation, the 15 nm R h PrP res particles might not reflect the point at which conformational change occurs, but rather the smallest size at which a PK-resistant conformation can be maintained. Interestingly, we recently described the evolution of PrP Sc particles in the brain of RML-infected mice over the course of disease, observing that PrP res particles were also always 15 nm R h or larger [29].
To estimate how many PrP units would comprise a particle of 15 nm R h , we made two assumptions, one assuming a fibrillar shape [26], and one assuming a parallel in-register intermolecular beta-sheet (PIRIBS) architectures [31]. This, combined with our MALS measurements, allowed us to estimate that fibrillar PrP particles of R h~1 5nm will have a length of~58 nm and consist of 116 PrP monomers giving a final mass of 3944 kDa, if we use an average M w = 34 kDa per monomer (based on the glycoform stoichiometry shown for 263K and HY, Fig  3). This size estimate is in line with other studies; sucrose gradient centrifugation and SEC studies of 263K and mouse-adapted scrapie identified PK-resistant particles starting at a M w higher than 2000 kDa [13,18]. We further validated our assumptions by comparing our M w calculation of larger particles to those reported by Silveira et al., on AF4-fractionated 263K. Silveira et al. found that 263K PrP res particles of R h~3 7 nm are fibrillar particles with a calculated M W of 7770 (± 4270) kDa (by MALS) and corresponding to 362 (± 198) PrP units considering a PrP molecules averaging 21.5 kDa for these PK treated particles [26], comparable to our 39 nm R h particles with calculated 399 PrP monomers (Table 1).
We also did these calculations based on a four-rung β-solenoid structure containing two protofilaments, with two PrP monomers present for every 1.92 nm in fibril height, and a rod radius of 5.5 nm [45][46][47][48]. S1 table depicts the results for these calculations.

AF4 fractionation reveals new strain differences
Although previous studies have not detected differences in glycosylation patterns between 263K, HY and DY [30,32], our data show that the transition from PrP sen to PrP res correlates with changes in glycosylation pattern and the degree of this change is strain-specific. Thus, we observed a more pronounced variation in glycotype stoichiometry for 263K and HY than for DY, which shows a lower and almost constant percentage of di-glycosylated isoform for all PrP Sc particles. Interestingly, lower levels of the di-glycosylated isoform have been associated with the accumulation of smaller and less stable PrP Sc particles in rodent prion strains [42,49,50]. Recent studies have demonstrated the relevance of PrP Sc glycotype stoichiometry in prion replication and faithful transmission of strain information [49,50]. Our data show that this stoichiometry either has an influence on, or is influenced by, the size of the PrP Sc assemblies.
AF4 fractionation also revealed subtle differences between the almost identical prion strains HY and 263K [33,38]. Using antibodies against the N-terminal region, Safar et al. reported a higher epitope exposure for 263K than for HY, when crude brain homogenates were incubated with 2 to 3M GdnHCl [10]. Here we show that 263K PrP Sc particles dissociate in SDS at a higher rate than HY particles, demonstrating a difference in stability of PrP Sc assemblies between these two strains. In addition, RT-QuIC assays for the fractionated PrP Sc populations revealed a higher replication activity for HY particles with R h > 40nm.

A model for PrP Sc heterogeneity
Collectively, our data suggest that the main contributor to PrP Sc intra-strain heterogeneity is PrP Sc quaternary structure. PrP Sc particles grow in size with quasi-equivalent biochemical properties. Only one major change in the core structure (secondary/tertiary structure) takes place, and this change manifests as a switch from protease-sensitive to protease-resistant species, accompanied by a slight change in glycosylation stoichiometry, at a strain-independent size of 15 nm R h (Fig 8). Once PrP Sc becomes partially resistant to PK digestion, its conformational stability remains almost constant up to at least R h~7 0 nm (or~500 nm fibril length), suggesting that no other major conformational changes take place in PrP Sc core structure. This results in a continuum of quaternary structures, with the amount of PrP Sc particles comprising each size subpopulation, along with their specific stability and templating activity, defining the incubation period, and potentially the clinical phenotype, of a particular prion strain. Thus, we A continuum of PrP Sc quaternary structures accumulate in the brain of prion-infected animals. Whereas the size of PrP Sc particles is the main contributor to PrP Sc heterogeneity, a major change in the core structure occurs when 15 nm R h is reached, generating two distinctive subpopulations that define the properties of the prion strain. At this size, there is a substantial increase in resistance to proteolysis and a reduction in the incorporation of di-glycosylated PrP isoforms into the PrP Sc particles. Once the mature PrP res particle is formed, the structural stability and glycosylation patterns remain constant despite further addition of PrP monomeric units. The PrP Sc templating activity is also affected by the size of the particle, increasing as the particle size increases, as seen in the case of DY. where k is a PrP Sc subpopulation. The structural stability of PrP Sc particles will influence incubation period through 1) rate of clearance, and 2) rate of templating nuclei generation by fragmentation.
The above considerations regarding PrP Sc particle sizes assume that the isolated PrP Sc particles contain only prion protein, but it is likely that that other proteins and macromolecules could be interacting with PrP Sc . If the degree of interaction varies with strain, this could affect the particle size distribution. Additionally, although we intentionally chose sample solubilization conditions that minimize the dissociation of PrP aggregates, the necessary presence of detergents in the solubilization and AF4 running buffers could still affect the actual size distribution of native PrP particles present in the prion infected brains, especially in the case of DY. Finally, although we divided PrP Sc particles into PrP sen and PrP res populations, this was based on Sha31 immunoblot detection, which is specifically probing for the resistance of epitope 145-152, so we cannot comment on the existence of assemblies with smaller resistant cores more C-terminal to this epitope.

Conclusion
Our results show that PrP Sc quaternary structure is a significant contributor towards PrP Sc structural heterogeneity, resulting in biochemically distinctive PrP Sc subpopulations, and the proportion of these subpopulations correlates with prion strain phenotype. Our AF4-MALS approach to separating and analyzing prion aggregates from different prion strains has implications for other protein misfolding-related neurodegenerative disorders where conformational strains have been proposed [51][52][53][54][55][56][57]. The structural heterogeneity of α-synuclein, amyloid β, and tau aggregates could form the molecular basis for the broad spectrum of phenotypes seen in synucleinopathies, Alzheimer's disease, and other tauopathies. By applying our technique to these conditions, we may not only promote an understanding of strain-phenotype relationships across neurodegenerative diseases but help direct therapies towards the most relevant aggregate conformations.

Ethics statement
All animal studies described herein were performed in accordance with Canadian Council on Animal Care (CCAC) guidelines, under protocol AUP914 approved by the animal care use committee for Health Sciences Laboratory Animal Services at the University of Alberta.

Hamster-adapted prion strains / brain tissues
Brain tissues were taken from clinically affected Syrian golden hamsters (Envigo) inoculated with brain homogenates (BH) of hamster-adapted strains of different origin: transmissible mink encephalopathy-derived Hyper (HY) and Drowsy (DY), and Scrapie-derived 263K. Three 263K-infected brains were obtained from hamsters inoculated with 10% 263K-stock BH and euthanized upon presenting with overt clinical signs at 74 days post infection (dpi). Seven HY-infected brains were generated by inoculation of 10% HY-stock BH, resulting in advanced clinical disease at 70 dpi. The DY samples included two brain homogenate pools; pool 1 consisted of three brains from hamsters inoculated with a 10 −5 BH dilution of a DY-stock that were euthanized with clinical signs at 240 dpi. The second drowsy pool was composed of seven DY-infected brains from hamsters inoculated with a 10 −5 dilution of the pool 1 and collected after >200 dpi. Two individual DY-infected brains from hamsters euthanized at 168 and 170 dpi were generated by inoculation of a 10 −2 dilution pool 2 into naïve hamsters.

Brain homogenization and sample blinding
Individual brains were weighed and homogenized in ultra-pure water (Sigma) with ceramic beads in an Omni Bead Ruptor to make 10% brain homogenates (w/v). The HY and DY pools were initially homogenized in ultrapure water using a dounce glass homogenizer. The resulting pool homogenate was further passaged through needles of different diameters, 17-21G. All samples were aliquoted in sets of three for each strain (individual or pool homogenate) and assigned a sample number for blinded analysis. Aliquots were kept at -80˚C until further analysis.

Batch mode dynamic light scattering (DLS)
DLS measurements of solubilized brain homogenate and pellets resuspended in solubilization buffer (50 mM HEPES pH 7.4; 300 mM NaCl; 10 mM EDTA; 4% (w/v) dodecyl-β-D-maltoside (Sigma)) were performed with a Malvern Zetasizer-Nano S. A 633 nm wavelength HeNe laser was used to detect backscattered light at a fixed angle of 173˚. Measurements were performed at 20˚C and the solution viscosity and refractive index of water was assumed for calculation purposes. The data was collected without attenuation and a minimum number of 10 consecutive runs of 10 seconds each was averaged to obtain the autocorrelation function. Particle size was calculated by the manufacturer's software through the Stokes-Einstein equation assuming spherical shapes of the particles.

Asymmetric-flow field-flow fractionation (AF4)
Brain homogenates (10% w/v) were solubilized by adding an equal volume of solubilization buffer (50 mM HEPES pH 7.4; 300 mM NaCl; 10 mM EDTA; 4% (w/v) dodecyl-β-D-maltoside (Sigma)) and incubated for 45 min on ice. Sarkosyl (N-lauryl sarcosine; Fluka) was added to a final concentration of 2% and incubated on ice for 30 min. The sample was centrifuged (20,000xg, 10 min) at 4˚C. The supernatant was collected and 350 μg total protein were subjected to asymmetrical-flow field-flow fractionation on an AF2000 Postnova system using 50 mM HEPES pH 7.4 (containing 50 mM sodium chloride and 0.05% sodium dodecyl sulfate (SDS) to minimize interaction of proteins with the channel membrane) as the running buffer. The channel was 26.5 cm in length and 350 μm in height, constructed with a trapezoidal spacer of maximal width 21 mm at the inlet, and lined with a 10 kDa cutoff polyethersulfone membrane at the accumulation wall. Samples were focused for 4 min and then eluted at a channel flow of 0.5 mL/min with constant cross-flow for the first 10 min, decreasing from 1.5 to 0.35 mL/min in the following 15 min, from 3.5 to 0 mL/min in the next 30 min, and run with no cross-flow for the last 10 min. A slot pump was run at 0.3 mL/min to concentrate the samples before they passed through the detectors. Fractions of 0.2 mL were collected. Multi-angle light scattering (MALS) and dynamic light scattering (DLS) were collected simultaneously in the inline DAWN HELEOS II detector (Wyatt Technology), operating at a wavelength of 662 nm. Data analysis was performed with ASTRA 6.1.7.17 software (Wyatt Technology).

Hydrodynamic radius measurements from in-line DLS
DLS data were collected every 2 seconds at 140.1˚scattering angle for determination of hydrodynamic radius (R h ) of the AF4 eluting particles. Fitting the data to an autocorrelation function was performed using the cumulants model. The injection of 350 μg of protein in each AF4 run ensures sufficient sample to scatter at least three times more light than the solvent, providing good signal/noise ratios for all size ranges.

Calculation of particle length using in-line MALS
Static light scattering data were collected at 16 detector angles simultaneously every 2 seconds across the fractogram. For particles with sizes above the angular variation detection limit (R g 10 nm), we used the MALS data to determine particle length assuming a rod-shaped particle. This assumption is supported by previous AF4-MALS-DLS and electron microscopy analysis of highly purified 263K strain PrP Sc particles where it was shown that PrP Sc particles became ellipsoid-spherical at R h = 12.4 nm, and larger particles had a more fibrillary structure [26]. We also detected fibrillar structures in our larger fractions where PrP was enriched (Fig 2D). The excess Rayleigh ratio, R(θ), as a function of scattering angle, was fit to the Rayleigh-Gans approximation for an infinitely thin rod using the ASTRA software, in which the form factor, P(θ), obeys the following relationship [58]: where u = [(π n ο /λ ο ) L sin(θ/2)], L is the rod length and is assumed to be much greater than the rod diameter, n ο is the refractive index of the solvent at the incident radiation (vacuum) wavelength, and λ ο is the incident radiation (vacuum) wavelength.

Number of PrP monomers per particle
Knowing particle size allowed us to calculate the number of PrP monomers that would "fit" into fibrils of different lengths. Based on the recently published model for 263K PrP Sc fibrils [31], we assumed that each monomer contributes with a height of 0.49 nm, giving the following calculation: PrP monomers ¼ fibril length ðnmÞ 0:49 nm monomer À �

Calculation of PrP Sc particle density
Based on the number of monomers present in each sized particle, we could use our immunoblot data of PrP intensities (representing relative amount of total PrP monomers in each fraction) to further calculate the relative number of PrP Sc particles present in each fraction. The absolute number of PrP Sc particles present in each fraction could not be directly measured because there were proteins other than PrP Sc particles present; these other proteins contributed to the MALS signal that would normally be used to calculate particle density. However, we were able to determine the maximum number of PrP Sc particles/mL that could be present, by analyzing MALS data from fractions 46-50. These fractions have much less protein, as measured by UV signal, but have the highest level of PrP, based on immunoblot, making them the purest PrP fraction. By assuming all MALS signal came from PrP Sc particles in these fractions, we calculated the theoretical maximum number of PrP Sc particles/mL (PrP Sc particle density) present in these fractions. To determine the theoretical maximum PrP Sc particle density for the smaller fractions, we took the ratio of PrP Sc particle density:relative number of PrP Sc particles from fractions 46-50 and applied it to the smaller fractions (see Table 1 for details). For particles of uniform density and volume V, the number of particles per mL in the i th slice, n i , is proportional to the zero-angle Rayleigh ratio divided by the square of the particle´s volume: The value of R(0) was determined by fitting the light scattering intensity as a function of angle to the rod model. The refractive index of the prion particles was set to 1.587, as this is a typical average value for protein refractive index [59]. For estimation of particle volume, we used the particle lengths calculated from MALS data and assumed a rod radius of 4.25 nm [31]. From this, the ASTRA software (Wyatt Technology) was able to calculate the number of PrP Sc particles per mL for each eluting data slice across the fractogram [60].

Proteinase-K digestion
For all Proteinase-K (PK) digestions, 0.5 μg/μL of protein was incubated with 20 μg/mL of PK at 37˚C for 1h and 300 rpm orbital shaking. For digestion of AF4 fractions, 10 μL of each fraction were incubated with 20 μg/mL of PK in the presence of 0.5 μg/μL bovine serum albumin (BSA) as the PK substrate to account for the low mass of total protein in eluted fractions. The reaction was terminated by addition of Pefabloc (Sigma) and incubation at 4˚C for 10 min, followed by boiling with SDS-sample buffer and electrophoresis on 4-12% NuPAGE Bis-Tris gels (Invitrogen).

Conformational stability assay
The conformational stability assay was performed as described previously with slight modifications [12]. Briefly, 10 and 20 μL from 263K and DY AF4 fractions were incubated for 30 min. with increasing concentration of SDS, from 0.025 to 0.635% (w/v), at 22˚C and shaken at 450 rpm. The samples were then digested with 1 μg/mL of PK for 1 h at 37˚C and 450 rpm orbital shaking in the presence of 15 μg of bovine serum. The reaction was terminated by the addition of 4 mM PMSF. For 263K samples, equal volumes of 2x loading buffer (125 mM Tris HCl pH 6.8, 20% glycerol, 4 mM EDTA, 6%SDS, and 10% 2-mercapto ethanol) was added, boiled for 10 min., loaded in a 4-12% NuPAGE Bis-Tris precast gels, transferred to PVDF membranes, and probed with the 3F4 antibody (dil. 1:3,000; Sigma). To improve immunoblot PrP res signal, DY samples were methanol precipitated by adding 5x volumes chilled methanol and incubated for 2h at -30˚C. The samples were centrifuged at 18,200 x g for 30 min., and the pellets were dried, resuspended in 1X loading buffer, boiled for 10 min. and subjected to SDS-PAGE and immunoblotting with 3F4 (dil. 1:3,000; Sigma).
PrP signal was analyzed by Image Quant software and PrP res intensity was fitted to sigmoidal and one phase decay models with Graph Pad Prism 8.0.

Expression and Purification of recombinant hamster PrP
Syrian hamster prion protein (SHaPrP) containing residues 90-231 (which form the proteaseresistant fragment in PrP Sc ) was cloned into the pET-15b plasmid between the XhoI and EcoRI sites as described previously [61]. Cys residues were introduced at each terminus by mutating Ser residues in the thrombin cleavage site and the prion site S232. The 19-kDa, Nterminal His-tagged SHaPrP was expressed in Escherichia coli BL21 (DE3) and purified by FPLC (GE Healthcare) using a nickel-nitriloacetic acid (Ni-NTA column). PrP was refolded on the Ni-NTA column, with native folding confirmed by circular dichroism spectroscopy, and the purity and identity of the protein verified by SDS-PAGE and immunoblotting (3F4 antibody, Millipore). intensities from each fraction were first plotted against the average R h for that fraction, then the area under the curve for the entire run was calculated and normalized to 100% for each strain.

Statistical considerations
One-way analysis of variance (ANOVA) was used to identify group-wise differences and posthoc Tukey's test was used to identify pairwise differences (p < 0.05 considered significant). All analyses were performed using Graphpad Prism (v5).  Table. Particle density estimation for HY assuming a four-rung β-solenoid structure. The R h of particles was calculated from DLS measurements. The length of the particles was calculated fitting the MALS measurements to a rod-shape model. The number of PrP monomers per PrP Sc particle was estimated from the PrP Sc particle length, assuming a contribution of 1.92 nm in length per PrP monomer and a PrP Sc fibril formed by two protofilaments. The PrP normalized intensity was determined by immunoblotting using Sha31 antibody. The relative number of PrP Sc particles was calculated by dividing the PrP normalized intensity by number of PrP monomers per PrP Sc particle. The maximum possible PrP Sc particle density was calculated with ASTRA software (see methods) for fractions 46 and 48, and extrapolated to the rest of the fractions based on the relative number of PrP Sc particles. ( � ) these values of particle density were calculated from MALS measurements directly, fitting the data to a rod-shape model, and assuming a rod radius of 5.5 nm. This calculation was performed for all samples; the specific values obtained for brain E are shown in the table.