Peptide backbone modifications of amyloid β (1–40) impact fibrillation behavior and neuronal toxicity

Fibril formation of amyloid β (Aβ) peptides is one of the key molecular events connected to Alzheimer’s disease. The pathway of formation and mechanism of action of Aβ aggregates in biological systems is still object of very active research. To this end, systematic modifications of the Phe19–Leu34 hydrophobic contact, which has been reported in almost all structural studies of Aβ40 fibrils, helps understanding Aβ folding pathways and the underlying free energy landscape of the amyloid formation process. In our approach, a series of Aβ40 peptide variants with two types of backbone modifications, namely incorporation of (i) a methylene or an ethylene spacer group and (ii) a N-methylation at the amide functional group, of the amino acids at positions 19 or 34 was applied. These mutations are expected to challenge the inter-β-strand side chain contacts as well as intermolecular backbone β-sheet hydrogen bridges. Using a multitude of biophysical methods, it is shown that these backbone modifications lead, in most of the cases, to alterations in the fibril formation kinetics, a higher local structural heterogeneity, and a somewhat modified fibril morphology without generally impairing the fibril formation capacity of the peptides. The toxicological profile found for the variants depend on the type and extent of the modification.

Alzheimer's disease (AD), being one of over 40 maladies caused by protein or peptide aggregation and misfolding, is a progressive and multifactorial disease leading to age-related changes in the brain. The etiology of AD is complex and influenced by genetic, environmental, and/or lifestyle factors 1 . In spite of intense research efforts invested into different aspects of this devastating form of dementia, AD is currently still incurable 2 . Progress in developing therapeutic treatment options is, among others, hampered by the long progression of the disease without any clinical symptoms, while moderate to severe levels of amyloid aggregates and neurofibrillary tangles in the brain result in the loss of neurons, disturbance of neuronal connectivity and in brain atrophy. The key molecular events underlying AD are fibril formation of amyloid β (Aβ) peptides and posttranslational modifications of the tau protein 3,4 .
However, amyloid fibril formation does not follow only a single pathway, but occurs through several routes in the energy landscape 5,6 . Intermediates that are formed on the way include soluble oligomers, to which strong neurotoxicity is ascribed via different molecular mechanisms such as membrane disruption, receptor-binding followed by activation of signaling pathways or glutamatergic neuronal hyperactivation [7][8][9] .
What remains highly enigmatic in the field of protein misfolding in general is the observation that peptides and proteins of no or of very diverse three-dimensional structure can form amyloids, in which the cross-β structure is the only common conformational motif 18,19 . This suggests that very general physical interactions cause these peptides or proteins to aggregate and form fibrils. To investigate the contributions of specific physical interactions systematically, targeted perturbations induced by individual mutations in the Aβ sequence represent a useful strategy [20][21][22][23] . Each amino acid carries specific properties such that exchange of a given residue introduces a new physical force and the response of the system to such a perturbation can be analyzed in detail. Such knowledge can be useful for the development of small molecular weight drugs that feature exactly that property for preventing formation of toxic Aβ structures. In our previous work, a rational library of relatively minor modifications was introduced to modify one of the most crucial molecular contacts in Aβ 40 fibrils between the two opposing β-strands, which provided important insights. In a series of studies, we systematically modified the Phe 19 -Leu 34 hydrophobic contact (Fig. 1), which has been found in almost all structural studies of Aβ 40 fibrils [13][14][15][16] . Overall, these studies showed that the Aβ 40 structure is very robust against almost any of the introduced modifications in the side chains of either Phe 19 or Leu 34 . For example, a double mutation of the hydrophobic contact between Phe 19 and Leu 34 into two Lys residues (Phe 19 Lys and Leu 34 Lys) introduced two positive charges, electrostatically repelling each other, which forced these side chains to move out of the hydrophobic core of the β-cross structure 24 . However, the general morphology of the fibrils was only marginally influenced. An introduction of a proline into position Phe 19 prolonged the fibrillation times (based on ThT fluorescence assay) dramatically 25 , but left the final fibril morphology largely unchanged. The only remarkable exception to the high tolerance of Aβ 40 against any modification of the Phe 19 /Leu 34 contact was that fibrillation was completely abrogated when a salt-bridge (Lys 19 -Glu 34 ) was introduced to replace the hydrophobic contact between Phe 19 and Leu 34 25,26 . Taken together, while local alterations in structure and dynamics of Aβ 40 have been observed, these studies showed that the Aβ 40 cross-β structure is very robust, extremely stable and can tolerate and accommodate many local alterations in the physical interaction fields. However, any specific mutation influenced the toxicity of the Aβ 40 fibrils suggesting that the Phe 19 and Leu 34 is of crucial importance as an early folding contact in Aβ 40 structure formation 27 .
In the current study, we turn our attention to the backbone of Aβ 40 peptides and investigate the impact of two modifications of the Aβ 40 peptide backbone at and around the important positions Phe 19 and Leu 34 . These modifications correspond to (i) an extended backbone structure (by the introduction of a methylene (X+1) or ethylene (X+2) spacer) and (ii) a peptide backbone N-methylation-Me-X) (N-Me-X) pattern (see Fig. 2). The rationale was that an extended backbone should induce a register shift to the standard in-register parallel β-sheet organization of the well-known steric zipper motif 13,28 of Aβ 40 fibrils. Such a shift in register would challenge the inter-β-strand side chain contacts as well as backbone β-sheet hydrogen bridges. The N-methylation of the peptide backbone would disturb the intermolecular hydrogen bonds between neighboring β-sheets [29][30][31] . To achieve that, we used combinations of Phe, Leu, 3-amino-4-phenylbutanoic acid (β-homo-Phe), 3-amino-5-methylhexanoic acid (β-homo-Leu), 4-amino-5-phenylpentanoic acid (γ-Phe), 4-amino-5-methylheptanoic acid (γ-Leu), N-methyl-phenylalanine (N-Me-Phe) and N-methyl-leucine (N-Me-Leu) as summarized in Figs. 1 and 2. With these modifications, the amyloid peptides receive more degrees of freedom by either (i) a more flexible backbone or (ii) a steric constraint, and the mutations are expected to introduce more structural variability into the system. This would (i) influence the register between the two opposing β-strands in the Aβ 40 structure 15,32-34 and (ii) interfere with the hydrogen bonding networks within the parallel intermolecular β-sheet arrangement [29][30][31] .
We used a broad repertoire of analytical techniques for the investigation of the amyloid fibrillation kinetics (thioflavin T (ThT) and crystal violet (CV) fluorescence spectroscopy), fibril morphology (transmission electron microscopy (TEM)), mesoscopic fibril packing parameters (X-ray diffraction), and fibril local structure and dynamics (solid-state nuclear magnetic resonance (NMR) spectroscopy). Further analyses include cellular toxicity studies (lactate dehydrogenase (LDH) release assay, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) conversion assay, activated caspase-3 and neurite length measurements) to determine the biological impact of the modifications and, thus, the relevance for pathologies observed in AD. introduced and the NMR labeling schemes applied. From the N-terminus, the peptide fibrils feature an unstructured N-terminus continue with β-strand 1 which ends at the beginning of the loop region, followed by β-strand 2 and the C-terminus. Both β-strands oppose each other with the side chains of the respective amino acids interacting in a steric-zipper motif. We targeted the hydrophobic contact between Phe 19 and Leu 34 by introduction of backbone modifications, namely methylene or ethylene spacer or N-methylation of the amide, in both residues in different combinations (see Fig. 2). Two isotopic labeling schemes (light and dark gray spheres) were applied for the investigations of local structure and dynamics of Aβ 40 fibrils by solid-state NMR.

Results
Fibrillation kinetics using the ThT fluorescence assays. The fibrillation kinetics of the investigated mutant and wild type Aβ 40 peptides were followed by recording the thioflavin T (ThT) fluorescence intensity over time which indicates formation of amyloid β fibril structures 35 . The characteristic parameters that can be determined from a plot of the fluorescence intensity vs. time (see supplementary Fig. S1) are the lag time (t lag ) and fibrillation time (t fib ), which are presented in Fig. 3 for all peptide variants.
All Aβ 40 variants with extended backbone modifications, except F+1/L+1, show the typical sigmoidal fluorescence intensity curves as a function of time (see Fig. S1, SI) consisting of the three characteristic phases: the lag phase, the fibrillation phase, and the plateau phase. Based on the results, F+2/L+2 (t lag = 3.3 ± 1.0 h; t fib = 9.1 ± 1.9 h) shows a lag time shorter than the WT (t lag = 6.7 ± 1.1 h; t fib = 3.5 ± 1.7 h), but a significantly longer fibrillation time until approaching the plateau phase, indicating that the increased degrees of freedom in the peptide backbone alter the energy landscape of the fibrillation process. The lag times for F+1 (t lag = 2.3 ± 0.7 h; t fib = 2.3 ± 1.1 h), F+2 (t lag = 4.1 ± 0.5 h; t fib = 4.5 ± 1.1 h) and L+1 (t lag = 4.0 ± 0.4 h; t fib = 1.5 ± 0.8 h) are shorter than for the WT, and L+2 (t lag = 6.7 ± 0.6 h; t fib = 2.4 ± 0.5 h) is in the same order of magnitude, showing that most of the mutations provide a stimulus for an earlier start of the fibrillation process. On the other hand, the fibrillation times are very similar.
Interestingly, the double mutations seem to have a rather strong impact on the fibrillation kinetics, as the F+1/L+1 variant shows two fibrillation phases, the first one starting immediately after dissolution in the fibrillation buffer reaching the plateau phase relatively fast and a second slower one starting after about 20 h (see supplementary Fig. S1). This behavior is reminiscent of the studies by Muschol and co-workers who concluded that sigmoidal kinetics indicate basically "oligomer-free" fibril formation, whereas in a biphasic growth the initial phase is dominated by globular oligomers 36,37 . For that purpose, they screened for an oligomer-specific dye and identified crystal violet (CV) as such. Therefore, we conducted fibril formation assays with CV for WT, F+1/ L+1 and F+2/L+2 and found that for F+1/L+1 the CV fluorescence responses are indicative of fast oligomer formation with a constant concentration level up to 20 h which is not the case for the other two investigated peptides (Fig. S2, SI).
Noticeably, the backbone methyl modified N-Me-F and N-Me-F/L variants do not fibrillate at all under the applied conditions and N-Me-L has a very extensive lag time of 135.7 ± 18.5 h and a somewhat longer fibrillation time of 19.4 ± 4.0 h. This reveals that methylation of the amide group in the peptide backbone has a tremendous effect on the formation of fibrils, where disturbances at Phe 19 seem to be more severe. This effect was also www.nature.com/scientificreports/ observed earlier by Meredith and co-workers 30,31 and was related to inhibition of amyloid β toxicity of some short peptides 38,39 , from which a whole strategy of aggregation inhibitors in medicinal chemistry was developed 40,41 . Mesoscopic cross-β structure investigated by X-ray diffraction. For X-ray powder diffraction measurements, samples of the respective Aβ 40 fibrils were mounted to nylon loops using oil and measured using Cu K α X-ray radiation (Cu K α , λ = 1.5406 Å). By applying Braggs law, the detected Debye cones, represented in the diffraction intensity vs 2θ plot (Fig. 5), could be transformed into the inter-sheet and inter-strand distances 42 . Relatively sharp reflections in the range of 2θ = 14-18° and at 21° result from salts present in the amyloid samples.
All the mutants show an intense and relatively well-defined reflection assigned to inter-strand spacing (4.7 Å) which is characteristic for the cross-β structure of amyloid fibrils (along growing direction) 42,43 . For WT, F+1 and F+2 this peak is relatively sharp, while for all the mutants, having a mutation at the Leu 34 site, the peak is particularly broadened indicating more structural heterogeneity. Additionally, the signals for the inter-sheet distances are even more diffuse and less intense compared to WT, F+1 and F+2. Both observations are a hint towards higher structural heterogeneity in the packing when the Leu 34 is mutated. Within the series WT (9.8 Å), F+1 (9.7 Å), and F+2 (9.3 Å) the packing becomes slightly tighter while the overall diffraction pattern remains similar. Interestingly, L+1 shows two separated reflections (11.2 and 8.3 Å), which is unique for this mutant with this particular clarity. In summary, especially Leu 34 site mutations introduce a type of perturbation into the fibril structure, which leads to shorter inter-sheet distances on an average, but with rather diffuse distributions.
Investigation of the local molecular structure of the mutated Aβ 40 fibrils by solid-state NMR spectroscopy. To obtain insights into the secondary structure on the level of individual amino acids, solidstate NMR spectra of the peptide variants that formed fibrils were recorded using two different isotopic ( 13 C/ 15 N) labeling schemes (see Fig. 1). One of them was chosen for investigating the local influence of the mutations in the vicinity of the Phe 19 mutation site (labeling scheme 1: Val 18 , Phe 20 , Ala 21 , Gly 33 ). In the other case, the labeling   www.nature.com/scientificreports/ labeled nuclei are listed in supplementary Tables S1 and S2. Calibration-independent differences of the 13 Cα and 13 Cβ signals are reported for both labeling schemes, which are characteristic for the local secondary structure of the respective amino acids (Fig. 6) 44 . Some of the mutants with labeling scheme 1, namely L+1 (Gly 33 ), F+2 (Ala 21 ), F+1/L+1 (Ala 21 , Gly 33 ) and F+2/L+2 (Gly 33 ), as well as with labeling scheme 2, L+1 (Ile 32 , Val 36 ) and N-Me-L (Ile 32 ), show polymorphisms while WT, L+2 and F+1 do not (supplementary Tables S1 and S2). This highlights again the higher degree of freedom in some modified Aβ peptides allowing diverse modification and possibly multiple interactions 45,46 . However, no conclusive pattern among the investigated mutants could be found in this regard.
In almost all the mutants, except for L+1, the signal for the carbonyl group (CO) could not be detected in the 13 C NMR spectra for Gly 33 (supplementary Table S1), so that it cannot be used as a parameter for the local impact of the mutation on Gly 33 . The 15 N NMR signals for Ala 21 in the double mutants, F+1/L+1 and F+2/L+2, are relatively strongly upfield shifted (119.4 and 118.7 ppm, respectively), which indicates effects on hydrogenbonding between β-strands (supplementary Table S1). In labeling scheme 1, all investigated mutants show β-sheet secondary structures, suggesting that the induced perturbation by extended backbones does not influence the local secondary structure around the Phe 19 mutation site compared to the WT. However, L+1 seems to be special in the sense that its 13 Cα- 13 Cβ chemical shift differences, at least for Phe 20 and Ala 21 , is significantly smaller than for WT. Therefore, L+1 was chosen to be investigated in labeling scheme 2 together with N-Me-L. Their 13 Cα- 13 13 Cβ chemical shift differences of the labeled amino acids subdivided into the two applied labeling schemes: Val 18 , Phe 20 , Ala 21 , Gly 33 (labeling scheme 1) and Phe 19 , Ala 21 , Ile 32 , Val 36 (labeling scheme 2) derived from 13 C-13 C DARR NMR experiments. Gly 33 is not shown because it is lacking a Cβ atom. The bars represent the reference values of α-helix (black), random coil (gray) and β-sheet (light gray) secondary structure as reported in the literature 44 . Experimental data is given as varying symbols in the usual color code according to the legend. The standard error for determination of chemical shifts from 13 C-13 C DARR NMR experiments is ± 0.5 ppm. Other conformations, if applicable, are shown as filled forms in the respective color and shape (polymorphs of Val 18 in F+1/L+1 (labeling scheme 1) are identical). www.nature.com/scientificreports/ even vary between predominantly random coil and β-sheet structure. Polymorphism or structural variation on the molecular level occurs frequently in amyloid fibril formation and may even correlate with diverging progress of the diseases among patients [47][48][49] . In conclusion, the local secondary structure around the Leu 34 mutation site is more sensitive towards backbone mutations of these types than Phe 19 . In addition to amyloid structure determination, NMR spectroscopy is a very sensitive tool to obtain information on molecular dynamics. The order parameter (S) derived from the measurement of motional averaged 1 H-13 C dipolar couplings using DipShift experiments 50,51 yields information on the motional amplitude of a given C-H bond vector. The molecular order parameters of the labeled residues in the Aβ 40 variants are shown in supplementary Fig. S4. Overall, no drastic changes in the order parameters throughout the variants using labeling scheme 1 were observed. For amino acid Gly 33 the Cα-Hα bond vector became a bit more rigid compared to the WT, when any type of mutation was introduced, but interestingly the conformers for Gly 33 for F+1/L+1 or L+1 differ in mobility. Worth noticing, in F+1/L+1 the phenyl ring of Phe 20 is slightly more rigid compared to WT and the other mutants. For labeling scheme 2, the observations are relatively similar, no drastic changes were observed. The only detail worth highlighting is that for L+1 all carbons show a tendency for higher rigidity, except for the phenyl ring in Phe 19 , where the amplitude for the ring motions increased a bit, which might give a hint towards the influence of a mutation at Leu 34 onto its crucial contact partner, the Phe 19 .
In order to examine a potential register-shift between the Phe 19 and Leu 34 contact in the opposing β-strands ( Fig. 1) induced by either backbone extension or backbone N-methylation, we exemplarily chose L+1 and N-Me-L for further NMR analysis. To this end, the amino acids Phe 19 , Ala 21 , Ile 32 , Val 36 (labeling scheme 2) were 13 C and 15 N isotopically labeled, which only allows for mutations at Leu 34 . 13 C-13 C DARR NMR experiments were performed with mixing times of 600 ms to probe the long distance and 50 ms for short distance contacts. For L+1 the cross peaks at around 129.6/12-25 ppm, which are assigned to side chain contacts between the labeled side chains of Ile 32 and Phe 19 , are only detected in the DARR NMR spectrum recorded with long 600 ms mixing time (Fig. 7, center). The WT, on the other hand, also shows this contact (Fig. 7, right), although the chemical shift of the Phe ring system is slightly lower. However, comparing the recorded NMR spectra for L+1 and N-Me-L, the same contact between Ile 32 and Phe 19 was detected also for N-Me-L (Fig. 7, center). This leads to the conclusion that no significant alterations in the sidechain zipper motif or a register shift are induced by the backbone modifications. But we note that the ring of Phe 19 is also in contact with the side chain of Ile 32 which was not reported before.

Cellular toxicity assays of mutated Aβ fibrils.
To probe the biological activity of the individual Aβ 40 variants, four different cell culture-based experiments were conducted (Fig. 8). In a first assay, the MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) conversion for mitochondrial activity was performed which provides information about the proportion of metabolically active cells (upper left). Noticeably, F+1/L+1 has the clearest negative impact on the cell viability compared to control and WT among all the mutants. Same is true for the second type of cell viability testing, being the lactate dehydrogenase (LDH) assay. It is based on the release of the cytosolic enzyme LDH into the cell culture medium upon damage to the plasma membrane (indicating cell death). For that, F+1/L+1 exhibit a considerably destructive effect on the cell membrane integrity, while only L+2 is also significantly different to the control and WT, but not as pronounced compared to F+1/L+1.
The outcome of the immunocytochemical staining for activated caspase-3, which is a protein involved in apoptotic neuronal cell death program, is interesting due to the apparently protective activity of the N-methylated  Table S1), N-Me-L (supplementary Table S1) and N-Me-F/L (supplementary Table S1), by suppressing the caspase-3 activation compared to control. Neurite length (in µm) as a marker of neuronal viability was measured after incubation with the mutant peptides. However, no significant differences were detected (Fig. 8) indicating no impact of the individual Aβ 40 variants on this aspect of neuronal wellbeing of the surviving neurons.

Discussion
The formation of amyloid fibrils, which is the process of conversion of soluble peptide monomers into insoluble aggregates with fibrillar structure, follows the widely accepted nucleation-dependent polymerization model 18,19,52,53 . Recent research has shown that the individual aggregation pathways of proteins and peptides can be very complex and there are several detours in addition to the sigmoidal kinetic fibrillation behavior, indicating that local structural minima exist on the free energy map 36,54 . This is an intrinsic property of the funnel-shaped free energy landscape of amyloid formation 55,56 . Nevertheless, various folding pathways lead to a highly stable www.nature.com/scientificreports/ low energy structure that is dominated by the cross-β motif. In our approach, we probe how the folding of Aβ 40 peptides is influenced by point mutations at Phe 19 and Leu 34 , which form an important early hydrophobic contact in the Aβ 40 folding process. These rational modifications can provide information on key points on the energy landscape of the fibril formation process. Our former activities involved, among others, alterations of the Phe 19 and/or Leu 34 side chains 25,[57][58][59] . In the current study, we focus our attention to the peptide backbone, which we modified by (i) extension in the backbone structure by introduction of methylene (X+1) or ethylene (X+2) spacers and (ii) N-methylation of the backbone amide groups (N-Me-X (supplementary Table S1)) 60 (Fig. 2). In these peptide variants, the backbone extension increases the degrees of freedom in the peptide backbone and could perturb the ordered in-register hydrophobic zipper structure motif. The N-methylation of the amide abolishes the formation of the intermolecular hydrogen bond between neighboring β-strands at the respective mutation sites. Our previous work on fibril formation of side chain modified Aβ 40 peptides led us to conclude that the amyloid structure is extremely robust against a multitude of perturbations introduced by individual natural and non-natural side chain modifications 25,[57][58][59] . This conclusion also holds for the Aβ 40 peptides with backbone extensions as applied here. While moderate alterations in the fibrillation kinetics (Fig. 3), fibril morphology (Fig. 4) and some heterogeneity of the cross-β structure (Fig. 5) are observed, the general β-sheet backbone secondary structure and the cross-β conformation are essentially not influenced by these mutations. The only exception is the L+1 variant, where Ile 32 shows some local random coil structure. Overall, the steric-zipper motif of the side chains 33,60 that describes the interlocking of the amino acid residues along with the intermolecular hydrogen bond patterns remain the dominating structure forming principle also in fibrils of backbone extended Aβ 40 peptides. Thereby, the side chains remain in a tightly packed conformation as confirmed by high order parameters determined in DipShift NMR experiments indicating low conformational flexibility. Although backbone extensions provide additional degrees of freedom, the interlocking of the side chains is the dominating motif that remains largely unchanged. There are some alterations worth mentioning in this context, though: first, there is an increasing amount of structural polymorphism observed in the 13 C NMR spectra of some of the Aβ 40 mutants, which demonstrates increased structural heterogeneity of the fibrils of mutated Aβ 40 peptides 46,49 . This is confirmed by noticeably highfield-shifted 15 N NMR signals of Ala 21 in the double-mutants F+1/L+1 and F+2/L+2 (119.4 ppm, 118.7 ppm, respectively, vs. 126.1 ppm in WT; NMR labeling scheme 1). Also, variant L+1 showed some structural alterations towards random coil structure of Ile 32 . For this variant, also a somewhat altered side chain packing was observed. Inversely, the lag times for the backbone extension variants in the ThT fibrillation kinetics are overall shorter as compared to the WT, while the fibrillation times are in the same order of magnitude. The backbone extension modification might thus even lower the activation energy barrier for early nucleation. In the fluorescence fibrillation plots, the double mutants, F+1/L +1 and F+2/L+2, show fibrillation kinetics diverging from normal sigmoidal curves under the applied conditions. For F+2/L+2, it is reminiscent of kinetics based on catalyzed nucleation (e.g. seeding) 61 , where the "seed" is formed during the lag time made from F+2/L+2 itself, which is apparently attributed to this specific combination of backbone modifications. ThT fibrillation curves of F+1/L+1 look like two step fibrillation kinetics resembling curves from the literature on oligomer and fibril formations 37 . Kinetic fluorescence assays with an oligomer-selective dye (crystal violet, CV) revealed the presence of oligomers in the first 20 h of fibrillation 37 (Fig. S2, SI). A hint towards understanding this abnormal behavior has been noticed in the difference of the 15 N chemical shift for Ala 21 for both the mutants with respect to the other variants and the WT. The potential existence of oligomeric species could be associated with the very effective membrane disruption ability of this mutant (Fig. 8), as oligomeric amyloid species are discussed as being the villains in cell toxicity [62][63][64][65][66] . Indirectly, an attenuating effect of F+1/L+1 on the MTT conversion by mitochondria is probably also related to the membrane disruption ability rather than mechanisms activating mitochondrial apoptotic pathways such as caspase-3 activation.
In contrast to the moderate influence of the backbone extensions on Aβ 40 fibril formation and structure, the N-methylation modification of the peptide backbone causes a more severe interference with Aβ 40 structure formation. N-methylation is one of the strategies to stabilize peptides against proteolysis and makes them more water soluble 31,40 . Likewise, methylation that interferes with intermolecular hydrogen bond formation is used for the design of amyloid-β aggregation inhibitors 67,68 . There are at least three identified ways how N-methylation impairs protein aggregation: (i) replacement of the amide proton cuts the hydrogen-bonding possibility between two neighboring β-strands; (ii) prevention of regular packing through steric hindrance (higher steric demand) and (iii) tertiary amides favoring trans-conformation leading to β-strand conformation, which, in this case, does not disturb the secondary structure 69 . Thus, the current results fit into the picture that fibrillation kinetics are either completely blocked (N-Me-F, N-Me-F/L) or greatly slowed down (N-Me-L). In the literature, examples by Meredith and co-workers, who used two times two N-methylated amino acids (2NMe(NTerm): residues 17 and 19; 2NMe(CTerm): residues 37 and 39), fibrils could still be formed, although significantly slower than in absence of the modification 29 . Only four N-methylated amino acids in the same peptide could prevent the fibrils from their formation. Noticeably, they could show that the β-sheet secondary structure for the respective strands is disrupted independently from each other according to their modification pattern, while N-methylation at position 17 and 19 has a higher impact on the fibril formation process, as it is true also for N-Me-F and N-Me-F/L compared to N-Me-L. However, the striking difference is that in our case N-methylation at only one (Phe 19 ) or two residues (Phe 19 /Leu 34 ), which were identified to be a critical early folding contact in fibril extension 27 , are sufficient to prevent from fibril formation. The protective behavior of N-Me-F, N-Me-L and N-Me-F/L counteracting cytotoxicity is expressed in the assay measuring caspase-3 activity, where they significantly decrease caspase-3 activation compared to the WT and partly to the control. This is in alignment with other N-methylated peptides in the literature 70 .
With the experimental techniques applied, no conclusive statement about the underlying Aβ 40 structural model, be it the U-shape (β arcades) 32 , C-shape (e.g. ex vivo Aβ 40 fibril, PDB: 6SHS) 71 , or the recently found extended shape model 72  www.nature.com/scientificreports/ symmetric Aβ 40 structure (PDB: 2M4J) by Lu et al. 74 , however, can be excluded, since Ala 21 and Phe 20 show Cα-Cβ chemical shift differences characteristic for β-sheet structure. Nevertheless, for L+1 and N-Me-L contacts between Phe 19 and Ile 36 (see NMR labeling scheme 2 and Fig. 7) could be found, which points towards a U-shape model, which we assume also for the other variants. In this context, it seems likely that the smallest fibril unit is not significantly impaired by the register-shift mutations at Phe 19 or Leu 34 and the N-methylation at Leu 34 . Regardless of the individual backbone modifications, parallel hydrogen-bonded β-strands are formed along the fibrillation axis (4.8 Å) verified by X-ray diffraction measurements 33. Interestingly, when in general a mutation is introduced at Leu 34 the inter-β-sheet distance (around the distinct reflection for 4.8 Å) is more diffuse compared to WT, F+1 and F+2. This implies that the 3D space, required by the side chain of Leu 34 , introduces more structural inhomogeneity during packing along the fibrillation axis in the mutants, than the planar phenyl ring does. This is also true for the N-methylated variant meaning that the extended backbone modification and the N-methylation are in terms of local packing density/arrangement more similar than initially expected. This hypothesis is supported by the findings from 13 C-13 C DARR NMR experiments (NMR labeling scheme 2) on L+1, N-Me-L and WT, where the inter-β-strand contact between Phe 19 and Ile 32 could be found for both the mutants due to the same steric demand of the modification (see Fig. 7). All these observations lead to the interpretation that backbone extensions have more impact on Leu 34 , while N-methylations do on Phe 19 .

Conclusion
In summary, in this study, a series of two types of Aβ 40 peptide backbone modification, namely incorporation of (i) a methylene or ethylene spacer group and (ii) a N-methylation at the amide functional group, at the amino acids at positions 19 or 34 were investigated. This approach was designed to probe the response of the Aβ 40 system to the introduced backbone perturbations to identify key points in the free energy landscape of fibril formation. In all peptide variants that formed fibrils, both types of modification at either position 19 and 34 do not disturb the cross-β structure and the hydrophobic zipper interdigitation of the side chains and their dynamic behavior in the β-sheet parts of the putative U-shaped fibrils. However, the modifications lead, in most of the cases, to alterations in fibril formation kinetics, higher local structural heterogeneity, and somewhat modified fibril morphology without generally impairing the fibril formation capacities of the peptides. Both backbone extensions provide additional degrees of freedom, allowing for a more efficient structure formation process as seen in faster fibrillation kinetics. In contrast, N-methylation significantly delays or even completely abolishes fibrillation of the Aβ 40 variants. This concludes a series of systematic studies where the influence of local perturbations in the Aβ 40 sequence were introduced. While most modifications only lead to moderate alterations in the fibril formation process and structure, inter-sheet hydrogen bond formation turns out to be the most important contribution to Aβ 40 fibril formation. Even the lack of only one or two inter-strand hydrogen bonds can completely abolish fibril formation stressing its importance for Aβ40 fibril formation. However, essentially all modifications introduced in the Aβ 40 sequence led to alterations in one way or another in biological activity profiles as concluded from toxicity assays. This hints towards a more pronounced impact of these modifications on the transient oligomeric states of the Aβ 40 peptide variants. For fibrillation, the sample was transferred into 1.5 mL reaction tubes and incubated for 14 days at 37 °C and 450 rpm in a thermoshaker. Fibrils prepared by this procedure were used for X-ray diffraction and NMR measurements. For transmission electron microscopy measurements, fibrils were taken out of the 96-well plate from thioflavin T (ThT) fluorescence fibrillation assay after completion of the fibrillation process.

Methods
Data representation. Experimental data derived from thioflavin T (ThT) fluorescence, transmission electron microscopy (TEM) and toxicological assays and stainings are summarized in box plot representations. The box marks the interquartile range (IQR) delimiting 50% of the data; the empty square is the mean, while the notch represents the 95% confidence interval (CI) of the median; the whiskers show 1.5 times IQR (not the standard deviation) demarcating statistical outliers and crosses indicate 1% or 99% of the data, respectively. Significances were tested applying a heteroscedastic Student's t-Test using two-tailed distributions, *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001 vs control; # vs WT. Crystal violet (CV) fluorescence measurements. Same procedure as for CV fluorescence measurements but using 5 µM CV (6.68 μL mL −1 of a 0.75 mM ThT stock solution in ddH 2 O) and 584 nm excitation wavelength, as well as 620 nm emission detection instead. Triplicates of WT, F + 1/L + 1 and F + 2/L + 2 were measured, but only qualitatively evaluated.
Fluorescence intensity data evaluation. Fluorescence intensity data were normalized by I = (It −I min) I max where I is the normalized intensity, I t is the measured intensity value at the corresponding time point, and I min and I max are the minimal and maximal intensities measured for the corresponding well, respectively. Normalized data were fitted using a simple sigmoidal curve model 75 , where I is the normalized intensity, t is the time, t 0 is the time at half maximal intensity, and τ is a measure for the fibrillation time. The lag time was derived by t lag = t 0 − 2τ and the fibrillation time by t fib = 4τ. Reported mean values and standard deviations were calculated from the different fit values. Values discussed in the text for fibril formation kinetics are given as mean ± one standard deviation.

Transmission electron microscopy (TEM).
A volume of 2 μL of a diluted fibril solution from the final state of ThT measurements (1:20 (v/v) with ddH 2 O) was transferred onto a formvar film-coated copper grid. After evaporation of the solvent, the sample was stained with 1% uranyl acetate. Images were recorded on an electron microscope (Zeiss SIGMA) equipped with a STEM detector and operated with Atlas software (Zeiss NTS, Oberkochen, Germany). The diameters of the fibrils were measured (n = 100) with the free of charge program ImageJ 1.53e 76 . Values discussed in the text for fibril diameters are given as mean ± one standard deviation.
X-ray diffraction measurements. Mature fibrils from solid-state NMR samples were taken out of the MAS rotor after measurement. Oil was used to stick the powdered/crystalline samples to nylon loops which were mounted on a Rigaku XtaLAB Synergy Custom X-ray crystallography system (Rigaku, Tokyo, Japan) equipped with a copper MicroMax-007 HF microfocus rotating anode source (Cu K α radiation) and a Hybrid Photon Counting (HPC) X-ray detector (HyPix-Arc 150°). Diffraction images were recorded with 120 s exposure time at room temperature (+ 18 °C). The collected data was analyzed with the powder diffraction tool of the CrysAlis Pro software package. Inter-β-strand and inter-β-sheet distances were calculated using Bragg's law (λ = 1.5406 Å): d = n· 2·sinθ . www.nature.com/scientificreports/ Cell culture. Primary neuronal cultures were established from fetal mouse brains of C57Bl/6 mice at gestation day 16 and grown under standard conditions 84,85 . Briefly, fetal brains were prepared, and neurons were dissociated into single cells by triturating the brains by means of pipets and passing the cell suspensions through sterile nylon meshes (150 µm and 20 μm). Suspensions were then grown in seeding medium (DMEM/Ham's F-12 supplemented with 5% (v/v) fetal horse serum (FHS) and 1% (v/v) penicillin-streptomycin-neomycin (PSN) antibiotic mixture) in 96-well culture plates (for MTT and LDH assays) or on poly-l-lysine-coated glass coverslips in 24-well culture plates (for immunocytochemistry), respectively. The cells were cultured at 37 °C in a humidified atmosphere containing 5% (v/v) CO 2 . On the following day, the seeding medium was exchanged by neuronal medium (Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/Ham's F-12) supplemented with 1% (v/v) PSN, 1% (v/v) N-2 supplement and 20% (v/v) astrocyte conditioned medium). After 3 days in vitro, cultures were stimulated with Aβ 40 peptides at concentrations of 10 μM for 18 h. Neuronal survival was analyzed using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay which measures glycolytic activity. Briefly, MTT was dissolved in cell culture media without serum and added to the primary neurons for 2 h. After addition of dimethyl sulfoxide (DMSO)/EtOH (1:1, v/v) and shaking for 10 min, the absorbance was measured at 592 nm (Mithras LB 940, Berthold Technologies).

Solid
Neuronal cell death was measured in cell culture supernatant by lactate dehydrogenase (LDH) assay (Promega, CytoTox96) according to the manufacturer's protocol. LDH is a cytosolic protein, which is not detectable extracellularly under normal conditions, but only in the case of membrane leakage.
Stained coverslips were examined with an AxioScan.Z1 microscope (Carl Zeiss, Göttingen, Germany). Emission filters were selected for Cy2 (524 nm), for Cy3 (595 nm) and Hoechst 33342 (425 nm). Images were taken by means of Axiocam 506 and a Plan-Apochromat objective (20×/0.8) and digitized using ZEN 3.4 software. For quantification of neurite lengths, regions of interest were digitized by means of Keyence microscope BZ9000 and lengths were measured using BZ II Analyzer 2.1 software.
Control experiments were performed by omitting the primary antibody. All controls were negative, indicating that the observed fluorescence signals were specific. For quantification, ten regions of interest from each coverslip were randomly selected and mean intensity values of activated caspase-3 and Hoechst 33342 signals were determined.

Funding
Open Access funding enabled and organized by Projekt DEAL.