The molecular basis for apolipoprotein E 4 as the major risk factor for late onset Alzheimer ' s disease

Apolipoprotein E4 (ApoE4) is one of three (E2, E3 and E4) human isoforms of an alpha-helical, 299-amino acid protein. Homozygosity for the ε4 allele is the major risk factor for developing late onset Alzheimer's disease (AD). ApoE2, ApoE3 and ApoE4 differ at amino acid positions 112 and 158 and these sequence variations may confer conformational differences that underlie their participation in the risk of developing AD. Here, we compared the shape, oligomerisation state, conformation and stability of ApoE isoforms using a range of complementary biophysical methods including small angle X-ray scattering, analytical ultracentrifugation, circular dichroism, X-ray fibre diffraction and transmission electron microscopy We provide an in-depth and definitive study demonstrating that all three proteins are similar in stability and conformation. However, we show that ApoE4 has a propensity to polymerise to form wavy filaments which do not share the characteristics of cross-beta amyloid fibrils. Moreover, we provide evidence for the inhibition of ApoE4 fibril formation by ApoE3. This study shows that recombinant ApoE isoforms show no significant differences at the structural or conformational level. However, selfassembly of the ApoE4 isoform may play a role in pathogenesis and these results open opportunities for uncovering new triggers for AD onset.

Abstract: Apolipoprotein E4 (ApoE4) is one of three (E2, E3 and E4) human isoforms of an alpha-helical, 299-amino acid protein. Homozygosity for the ε4 allele is the major risk factor for developing late onset Alzheimer's disease (AD). ApoE2, ApoE3 and ApoE4 differ at amino acid positions 112 and 158 and these sequence variations may confer conformational differences that underlie their participation in the risk of developing AD. Here, we compared the shape, oligomerisation state, conformation and stability of ApoE isoforms using a range of complementary biophysical methods including small angle X-ray scattering, analytical ultracentrifugation, circular dichroism, X-ray fibre diffraction and transmission electron microscopy We provide an in-depth and definitive study demonstrating that all three proteins are similar in stability and conformation. However, we show that ApoE4 has a propensity to polymerise to form wavy filaments which do not share the characteristics of cross-beta amyloid fibrils. Moreover, we provide evidence for the inhibition of ApoE4 fibril formation by ApoE3. This study shows that recombinant ApoE isoforms show no significant differences at the structural or conformational level. However, selfassembly of the ApoE4 isoform may play a role in pathogenesis and these results open opportunities for uncovering new triggers for AD onset.
I have a number of comments after reading the manuscript and some criticism of the folding data fitting. 1. It was not clear to me initially why the authors were trying to measure the hydrodynamic radius and Rg of the protein. The results are generally listed without a clear hypothesis. It would be a lot clearer to the reader if the introduction to the manuscript and the different results sections clearly indicated what they are testing. I gathered from the discussion that the previous literature on these proteins has suggested that apoE4 has an expanded radius more characteristic of a molten globule or partially disordered state. The data reported here disagree with this hypothesis. It also states at the beginning of the discussion that previous work has measured the stability of these isoforms to be substantially different. Data presented here suggests otherwise. The paper would be considerably strengthened by making this clear from the start.
Response: We thank the reviewer for their comments and we have now added further background in the introduction and results section to explain our aims. In the context of the literature, our work provides an in depth characterisation showing that the recombinant ApoE isoforms differ very little from one another at the level of multimerisation. Page 5.In contrast to previous studies [25,26], we show that the three recombinant isoforms share very similar quaternary, tertiary and secondary structures and thermal and chemical stability Cover Letter and We have developed a method to produce and maintain full length ApoE proteins recombinantly (see methods and [30]) and it was therefore necessary to first fully characterise these proteins.
2. With regards to the data fitting of the GuHCl unfolding data: while there is nothing wrong with the 3-state model proposed and the shape of the unfolding curves of each protein does indeed suggest that there are at least 3 states of the protein, the different free energies for unfolding reported in table 4b are very unlikely to be accurate. The unfolding curves for the three proteins are essentially superimposable so obtaining such hugely different values for the free energies stems from an inaccurate fitting of the transitions. The authors do not report m-values for unfolding and these would provide the reader with an estimate for the accuracy of the fitting. Proteins with similar sizes and shapes will give a similar sum of the m-values because the m value is representative of the total buried hydrocarbon in a folded protein. If the values obtained differ between similar proteins, then it suggests that the model that is fitted is incorrect rather than that the proteins are actually different in terms of amount of buried hydrocarbon. By this I mean that the 3 state model may not be the correct model: the presence of further states of the protein would make the observed transitions look artificially broad, lowering the fitted m-values. Given that the authors have established that all three proteins have the same size and shape, they can happily assume the same m-values for their proteins. I suggest the authors attempt fitting using the same m-values for all 3 apoE isoforms, or if this does not work, they should admit that the data do not fit to a model that assumes 2 unfolding transitions. They would still be justified in reporting the midpoint concentrations of GdnHCl for each apparent transition. The small differences in the values of these observed midpoints for each isoform are a better reflection of any potential differences in stability. As concluded in the text, the stability difference between the isoforms is indeed marginal and does not seem to explain the fibrillisation propensity of the proteins.
Response: We thank the reviewer for their helpful suggestions. We find that the data does not fit to a 2 state model and agree that the data in the original table was extensive and parameters likely covary resulting in inaccurate values. Therefore, we have now replaced this table with a table  showing transition midpoints as they are very well defined by the data and altered the results text to  refer to this new table (now named table 4c). between the amounts of ApoE3, E4 and E2 unfolded at 37°C but these are within experimental error so we do not see any significant difference at physiological temperature. 5. To follow on from point 4., the small amount of GuHCl required to induce a first unfolding transition suggests that these proteins are only marginally stable at the temperature of the experiment. With regards to this temperature, it is unclear what conditions the protein is equilibrated in as the GuHCl titration samples are incubated overnight at 4 degrees then measured in a fluorimeter at 20 degrees. Depending on the folding and unfolding rates of the protein, this may or may not be sufficient to re-equilibrate the samples to the new temperature. Do the titrations relate to the stability of the proteins at 4 degrees or at 20 degrees? Given that assembly is then measured at 37 degrees, it may be useful to understand how the stability of the individual domains changes at this temperature, something which would require titration with GuHCl at 37 degrees if at all possible.
Response. We are grateful for these suggestions. However, our self-assembling protein presents some difficulties in following chemically induced unfolding. The methods explain that the proteins are incubated at 4°C overnight and is then incubated for 5 mins at 20°C during the five scans collected which are then averaged. We do not observe any difference between these graphical outputs. The volume is small and therefore would expect temperature equilibration to take places rapidly. Unfortunately, ApoE proteins are prone to fragmentation and degradation, so it is important to maintain them at a low temperature over time (hence the choice of 4°C overnight). Furthermore, we have observed that at 37°C, the three proteins do not appear to differ in their degree of unfolding and that overtime, ApoE4 assembles. Here we are interested to examine and compare the unfolding of the three isoforms and therefore, we feel that the experimental design is suitable for this work to avoid any deterioration or aggregation of the samples.
Thank you for your consideration.
With best wishes, Louise Dear Sheena Thank you for sending the responses for our submitted manuscript. We have carefully considered all the comments and improved the manuscript following these recommendations. Below we outline our response to each comment.
Reviewer #1: We thank the reviewer for their positive remarks.
Authors need to add some discussions on the limitations of this work. Specifically, apoE proteins in the physiological context are glycosylated and can be lipidated, but these bacterial produced apoE proteins have none of these. Thus, while this work has provided insights on biophysical properties of recombinant apoE isoforms as research tools, these may not be directly assumed to be the case in human physiological or diseased conditions." Response: We agree with the reviewer on this point and we have added further discussion to explain our results in the context of physiological protein (glycosylated and lipidated). We have also ensured that the use of recombinant protein in aqueous buffer has been made clear. The behaviour of lipidated protein will be important for further studies in the future.
Page 10. In vivo, ApoE is glycosylated and also lipidated [45]. While we have explored the structural conformations adopted by recombinant non-modified ApoE, previous studies have shown that the recombinant ApoE produced in E.coli adopts a very similar conformation and folding to protein produced in adenovirus [46]. Further studies will be necessary to explore the behaviour of lipidated proteins.
Reviewer #2: We thank the reviewer for their supportive comments.
I have a number of comments after reading the manuscript and some criticism of the folding data fitting. 1. It was not clear to me initially why the authors were trying to measure the hydrodynamic radius and Rg of the protein. The results are generally listed without a clear hypothesis. It would be a lot clearer to the reader if the introduction to the manuscript and the different results sections clearly indicated what they are testing. I gathered from the discussion that the previous literature on these proteins has suggested that apoE4 has an expanded radius more characteristic of a molten globule or partially disordered state. The data reported here disagree with this hypothesis. It also states at the beginning of the discussion that previous work has measured the stability of these isoforms to be substantially different. Data presented here suggests otherwise. The paper would be considerably strengthened by making this clear from the start.
Response: We thank the reviewer for their comments and we have now added further background in the introduction and results section to explain our aims. In the context of the literature, our work provides an in depth characterisation showing that the recombinant ApoE isoforms differ very little from one another at the level of multimerisation.
Page 5.In contrast to previous studies [25,26], we show that the three recombinant isoforms share very similar quaternary, tertiary and secondary structures and thermal and chemical stability and We have developed a method to produce and maintain full length ApoE proteins recombinantly (see methods and [30]) and it was therefore necessary to first fully characterise these proteins.
2. With regards to the data fitting of the GuHCl unfolding data: while there is nothing wrong with the 3-state model proposed and the shape of the unfolding curves of each protein does indeed suggest that there are at least 3 states of the protein, the different free energies for unfolding reported in table 4b are very unlikely to be accurate. The unfolding curves for the three proteins are essentially superimposable so obtaining such hugely different values for the free energies stems from an inaccurate fitting of the transitions. The authors do not report m-values for unfolding and these would provide the reader with an estimate for the accuracy of the fitting. Proteins with similar sizes and shapes will give a similar sum of the m-values because the m value is representative of the total buried hydrocarbon in a folded protein. If the values obtained differ between similar proteins, then it suggests that the model that is fitted is incorrect rather than that the proteins are actually different in terms of amount of buried hydrocarbon. By this I mean that the 3 state model may not be the correct model: the presence of further states of the protein would make the observed transitions look artificially broad, lowering the fitted m-values. Given that the authors have established that all three proteins have the same size and shape, they can happily assume the same m-values for their proteins. I suggest the authors attempt fitting using the same m-values for all 3 apoE isoforms, or if this does not work, they should admit that the data do not fit to a model that assumes 2 unfolding transitions. They would still be justified in reporting the midpoint concentrations of GdnHCl for each apparent transition. The small differences in the values of these observed midpoints for each isoform are a better reflection 4. Top of page 11, the authors state that all proteins have the same level of unfolding at 37 degrees. While this is apparent from the CD data, the stability data provides a measure of how many unfolded (U) or partially unfolded (I) molecules are present at any temperature. While the proportion of U or I is of course very small at 37 degrees, it will still be different to the same extent for each isoform, such that a sample of ApoE4 will potentially have 10x more U or I molecules than ApoE2. This may lower the barrier to any assembly state as each molecule of ApoE4 will then be 10x more likely to unfold, even if these events are rare at 37 degrees.
Response: This is an interesting point and thank the reviewer for their insightful comments. We have now calculated the % of protein unfolded at 37°C based on the best fit unfolding curves to our data and included this information in a new table 4b. This shows that there are minor differences between the amounts of ApoE3, E4 and E2 unfolded at 37°C but these are within experimental error so we do not see any significant difference at physiological temperature. 5. To follow on from point 4., the small amount of GuHCl required to induce a first unfolding transition suggests that these proteins are only marginally stable at the temperature of the experiment. With regards to this temperature, it is unclear what conditions the protein is equilibrated in as the GuHCl titration samples are incubated overnight at 4 degrees then measured in a fluorimeter at 20 degrees. Depending on the folding and unfolding rates of the protein, this may or may not be sufficient to re-equilibrate the samples to the new temperature. Do the titrations relate to the stability of the proteins at 4 degrees or at 20 degrees? Given that assembly is then measured at 37 degrees, it may be useful to understand how the stability of the individual domains changes at this temperature, something which would require titration with GuHCl at 37 degrees if at all possible.
Response. We are grateful for these suggestions. However, our self-assembling protein presents some difficulties in following chemically induced unfolding. The methods explain that the proteins are incubated at 4°C overnight and is then incubated for 5 mins at 20°C during the five scans collected which are then averaged. We do not observe any difference between these graphical outputs. The volume is small and therefore would expect temperature equilibration to take places rapidly. Unfortunately, ApoE proteins are prone to fragmentation and degradation, so it is important to maintain them at a low temperature over time (hence the choice of 4°C overnight). Furthermore, we have observed that at 37°C, the three proteins do not appear to differ in their degree of unfolding and that overtime, ApoE4 assembles. Here we are interested to examine and compare the unfolding of the three isoforms and therefore, we feel that the experimental design is suitable for this work to avoid any deterioration or aggregation of the samples.

46
Alzheimer's disease (AD) is the most prevalent dementia, and the sporadic, late onset form comprises 47 95% of all AD cases. Diagnosis is based on the observation of specific brain pathology of 48 extracellular amyloid plaques composed of Amyloid-beta (Aβ) peptide and intracellular 49 neurofibrillary tangles formed by tau [1]. The biggest risk factor for developing AD is age, but the ε4 50 variant of the apolipoprotein E (APOE) gene is the strongest genetic risk factor for the development of 51 late onset AD. The human APOE gene encodes three major protein isoforms, ApoE2, ApoE3 or 52 ApoE4 that differ from one another at amino acid positions 112 and 158. While E2 contains two 53 cysteine residues, E3 contains a cysteine and an arginine, and E4 contains two arginine residues at 54 both sites respectively [2]. ApoE4 presents a risk in a gene dose-dependent manner, with ε3/ε4 55 heterozygotes having a risk increased by three times and ε4/ε4 homozygotes having up to 15 times 56 more chance of developing AD [3]. ApoE3 is the most common isoform, associated with a neutral 57 risk. ApoE2 is under-represented in the population but is thought to be associated with a lower 58 propensity for AD [4].

60
ApoE is a predominantly α-helical protein with 299 amino acids that is mainly produced in the liver 61 and by astrocytes in the brain, as well as neurons particularly under stress conditions [5,6] Table 1). A hydrodynamic radius of 5.8 -6.5 nm was 121 calculated for each isoform and a frictional ratio f/f 0 above 1.7 suggests elongated shape for the 122 homotetramers (Table 1, Figure S2). Analytical ultracentrifugation was used to investigate the size 123 and shape in solution in further detail. As we extensively dialyzed or buffer exchanged ApoE 124 isoforms into phosphate buffer prior to stability and aggregation studies, we were interested if dialysis 125 affected ApoE oligomerisation in solution. We therefore compared sedimentation in the original size 126 exclusion buffer and in phosphate buffer after dialysis. AUC revealed no differences in sedimentation 127 velocity between the three isoforms in either buffer ( Figure S3, Figure 1 C and D, Table 2a) and the 128 major species was characterised with a molecular mass of 130 kDa (Table 2a).

130
Small angle X-ray scattering (SAXS) was used to explore whether the tetramers of the ApoE isoforms 131 differ in dimensions and shape. All three proteins exhibited identical scattering profiles indicating that 132 they are very similar in shape (Figure 2A, B, C). A radius of Gyration (R G ) of 5.6 nm (56 Å) was 133 calculated using the Guinier approximation and comes close to the hydrodynamic radius determined 134 in the other techniques (

139
Together AUC, SEC-MALS and SAXS show that the three ApoE isoforms are similar in size, shape 140 and multimerization, and all form tetrameric species in solution.

143
Circular dichroism (CD) spectroscopy and tryptophan fluorescence were used to probe potential 144 differences in secondary and tertiary structure of the three isoforms. Negligible spectral differences 145 were observed, suggesting no major conformational differences between the three proteins ( Figure 3A 146 and B). Analysis of CD spectra using Dichroweb [31 2002, 32, 33] showed that the three isoforms all 147 possessed around 58% α-helical content (Table 3), with no significant differences between them (one-148 way ANOVA: F(2,9)=4.197, p>0.05).

150
Folding stability of ApoE proteins was examined using thermal and chemical denaturation and 151 monitored using CD and tryptophan fluorescence. Thermal denaturation monitored using CD at 152 222nm showed a sigmoidal unfolding curve for all three isoforms. A phenomenological Boltzman 153 sigmoidal curve was fitted to the data and revealed that ApoE3 and ApoE4 have similar melting 154 temperature of 52.39° C and 51.32°C respectively, compared to the apparent higher melting 155 temperature for ApoE2 of 60.28°C ( Figure 3C(i)). However, despite these slight differences in the 156 apparent mid-point melting temperature, no differences were observed at physiological temperature of 157 37°C. One-way ANOVA provided evidence of no significant difference in the mean ellipticity at 222 158 nm between the three isoforms at 37°C (F(2,9)= 0.9426, p=0.4249; Table 4a

161
showing similar thermal denaturation profile for all three isoforms. Furthermore, the fraction unfolded 162 at 37°C showed there were no significant differences between the three isoforms at physiological 163 temperature, consistent with the mean ellipticity at 222 nm at the same temperature (Table 4b).

172
GuHCl concentrations corresponding to the midpoint of each transition were calculated (Table 4c).

173
Similar to the thermal denaturation study, fraction unfolded curves were calculated for each isoform 174 in order to directly compare the shape of the chemical denaturation profile ( Figure 3D(ii)). The  ApoE3 and E4 infers that ApoE2 has a slight increased resistance to denaturation compared to E3 and 181 E4 although the difference in apparent stability is marginal (Table 4c).

185
To further investigate the behaviour of the ApoE isoforms, the three proteins were incubated in 20 186 mM PB at 37°C for 24 h. Native PAGE showed that ApoE2 and E3 ran consistently at a similar 187 mobility at 0 h and after 24 h incubation. However, ApoE4 also formed higher MW species prior to, 188 and following incubation. By 24 h, the majority of the ApoE4 protein formed higher oligomers that 189 did not run through the gel indicating oligomerisation of the ApoE4 protein ( Figure 4A). Thioflavin T

217
CD and X-ray fibre diffraction experiments were conducted to investigate the nature of ApoE4 fibrils 218 and to investigate whether these assemblies are amyloid-like in structure and undergo the expected 219 conformational change to -sheet [29]. CD spectra collected for ApoE4 fibrils incubated for 24 h 220 showed a spectrum with minima at 208 nm and 222 nm consistent with a predominantly α-helical 221 secondary structure content ( Figure 5D), of comparable shape and intensities to that of non-assembled 222 ApoE4 ( Figure 3A). To ensure that any remaining soluble protein does not dominate the spectra, the 223 sample was centrifuged at high speed to separate supernatant and pellet fractions. The CD data for the 224 resolubilised pellet fraction show E4 retains its α-helical structure after incubation at 37°C for 24 h, 225 and that there is almost no soluble protein left given the very small CD signal in the supernatant 226 fraction ( Figure 5D). The fibre pellet was sonicated to ensure that protein was resuspended 227 sufficiently and the spectrum continued to demonstrate -helical content.

229
To examine the molecular structure of the mature ApoE4 filaments, 100 µM ApoE4 was incubated       ApoE4 is less well tethered to the C-terminal domain than for ApoE3. Others have pointed to 260 differences in stability and conformation [29,44] and have suggested that ApoE4 is less stable and 261 more prone to aggregation to form amyloid-like fibrils [29]. In vivo, ApoE is glycosylated and also

299
The loss of stability of ApoE4 as measured by spectroscopic methods, albeit under extreme conditions 300 such as high temperature or under denaturing conditions, may be related to its propensity to 301 polymerise into filamentous structures. We showed here that ApoE4 forms filamentous structures 302 after only 24 h incubation at physiological temperature, pH 7.4, while ApoE2 and ApoE3 remained 303 soluble and globular under the same conditions. TEM reveals that the ApoE4 filaments have a 304 polymeric appearance (beads on a string) and there is a clear hierarchical assembly of small spherical 305 species to small, elongated fibrils and finally fibrils, which show identical diameters. CD shows that 306 following incubation, the ApoE4 fibres retain their α-helical secondary structural content. The CD for 307 the whole fraction is almost identical to ApoE4 prior to incubation. To probe whether the -helical 308 intensity arises from soluble, unassembled protein, we examined the CD from a sedimented sample.

309
Both pellet and supernatant fractions show a spectrum consistent with high -helical content.

310
Furthermore, the intensity of the spectrum in the supernatant was very low suggesting that the 311 majority of the ApoE4 protein is found within the fibre containing pellet. This result contrasts with a 312 small shift from α-helix to β-sheet previously described by Hatters et al [29]. Thioflavine T

339
The majority of individuals are heterozygous ε3/ε4, and this confers an increased risk for AD 340 development, albeit lower than for ε4/ε4 individuals [61]. Here we investigated the aggregation 341 potential for mixed population of ApoE3 and ApoE4 and revealed that E3 inhibits E4 fibril formation 342 in vitro, leading to a mid-range aggregation kinetic for assembly. Therefore the increased dose of 343 ApoE4 in ε4/ε4 individuals leads to increased assembly compared to half dose of ApoE4 in ε3/ε4 344 people.

347
The detailed studies conducted have revealed that recombinant ApoE2, ApoE3 and ApoE4 resemble 348 one another at a quaternary, tertiary and secondary structural level. All three ApoE isoforms form 349 elongated tetramers in solution and each monomer is rich in α-helical conformation. There is no 350 evidence that the aqueous proteins differ at a structural level. Furthermore, chemical and thermal 351 denaturation studies reveal that the three proteins are similarly stable and follow comparable 352 unfolding mechanisms by melting and by chemical denaturation. However, we have shown that 353 ApoE4 has a higher propensity for polymerisation at physiological temperature and pH, to form 354 elongated fibrous structures which retain the native -helical conformation. The propensity of ApoE4 355 to self-assemble may play an important role in the mechanism by which it increases susceptibility to 356 develop AD which impacts on ε4/ε4 individuals more severely than those with ε3/ε4.

360
All materials were purchased from Sigma-Aldrich or Fisher Scientific at the highest purity available.

388
The frictional ratio f/f 0 = R S /R min was calculated by assuming the minimal radius (R min ) of a sphere

443
FarUV CD data were collected using a Jasco-715 CD-spectrometer (Jasco, Goh-Umstadt, Germany) 444 at different time points. Temperature was maintained at 21°C using a Peltier controlled cell holder.

458
Results represent a mean of values from a minimum of three spectra per isoform. One-way ANOVA 459 was performed to compare the α-helical content between the isoforms, using GraphPad Prism.

557
X-ray diffraction images were collected using a Rigaku rotating anode source (CuKα) and a Saturn 558 CCD+ detector. Partially aligned fibres were placed in the X-ray beam and exposed for 30 s or 60 s at 559 specimen to detector distances of 50 mm and 100 mm. Diffraction patterns were converted to TIFF

129
Small angle X-ray scattering (SAXS) was used to explore whether the tetramers of the ApoE isoforms 130 differ in dimensions and shape. All three proteins exhibited identical scattering profiles indicating that 131 they are very similar in shape (Figure 2A, B, C). A radius of Gyration (R G ) of 5.6 nm (56 Å) was 132 calculated using the Guinier approximation and comes close to the hydrodynamic radius determined 133 in the other techniques (

ApoE2, E3 and E4 conformationally similar and show only marginal differences in stability
142 Circular dichroism (CD) spectroscopy and tryptophan fluorescence were used to probe potential 143 differences in secondary and tertiary structure of the three isoforms. Negligible spectral differences 144 were observed, suggesting no major conformational differences between the three proteins ( Figure 3A   145 and B). Analysis of CD spectra using Dichroweb [31 2002, 32, 33] showed that the three isoforms all 146 possessed around 58% α-helical content (Table 3), with no significant differences between them (one-147 way ANOVA: F(2,9)=4.197, p>0.05).

160
showing similar thermal denaturation profile for all three isoforms. Furthermore, the fraction unfolded 161 at 37°C showed there were no significant differences between the three isoforms at physiological 162 temperature, consistent with the mean ellipticity at 222 nm at the same temperature (Table 4b).

171
GuHCl concentrations corresponding to the midpoint of each transition were calculated (Table 4c).

219
CD and X-ray fibre diffraction experiments were conducted to investigate the nature of ApoE4 fibrils 220 and to investigate whether these assemblies are amyloid-like in structure and undergo the expected 221 conformational change to -sheet [29]. CD spectra collected for ApoE4 fibrils incubated for 24 h 222 showed a spectrum with minima at 208 nm and 222 nm consistent with a predominantly α-helical 223 secondary structure content ( Figure 5D), of comparable shape and intensities to that of non-assembled 224 ApoE4 ( Figure 3A). To ensure that any remaining soluble protein does not dominate the spectra, the 225 sample was centrifuged at high speed to separate supernatant and pellet fractions. The CD data for the 226 resolubilised pellet fraction show E4 retains its α-helical structure after incubation at 37°C for 24 h, 227 and that there is almost no soluble protein left given the very small CD signal in the supernatant 228 fraction ( Figure 5D). The fibre pellet was sonicated to ensure that protein was resuspended 229 sufficiently and the spectrum continued to demonstrate -helical content.

301
The loss of stability of ApoE4 as measured by spectroscopic methods, albeit under extreme conditions 302 such as high temperature or under denaturing conditions, may be related to its propensity to 303 polymerise into filamentous structures. We showed here that ApoE4 forms filamentous structures 304 after only 24 h incubation at physiological temperature, pH 7.4, while ApoE2 and ApoE3 remained 305 soluble and globular under the same conditions. TEM reveals that the ApoE4 filaments have a 306 polymeric appearance (beads on a string) and there is a clear hierarchical assembly of small spherical 307 species to small, elongated fibrils and finally fibrils, which show identical diameters. CD shows that 308 following incubation, the ApoE4 fibres retain their α-helical secondary structural content. The CD for 309 the whole fraction is almost identical to ApoE4 prior to incubation. To probe whether the -helical 310 intensity arises from soluble, unassembled protein, we examined the CD from a sedimented sample.

311
Both pellet and supernatant fractions show a spectrum consistent with high -helical content.

534
Adjusted fluorescence was plotted against time, and averaged traces for each isoform was obtained on 535 GraphPad Prism. A minimum of 3 different production batches per isoform was used.

43
Fraction unfolded of ApoE at 37°C were calculated from the fraction unfolded curves given 44 by equation 7 (Figure 3c). Non-linear best-fit parameters were determined using GraphPad 45