Structural rearrangement of amyloid-β upon inhibitor binding suppresses formation of Alzheimer’s disease related oligomers

The formation of oligomers of the amyloid-β peptide plays a key role in the onset of Alzheimer's disease. We describe herein the investigation of disease-relevant small amyloid-β oligomers by mass spectrometry and ion mobility spectrometry, revealing functionally relevant structural attributes. In particular, we can show that amyloid-β oligomers develop in two distinct arrangements leading to either neurotoxic oligomers and fibrils or non-toxic amorphous aggregates. Comprehending the key-attributes responsible for those pathways on a molecular level is a pre-requisite to specifically target the peptide's tertiary structure with the aim to promote the emergence of non-toxic aggregates. Here, we show for two fibril inhibiting ligands, an ionic molecular tweezer and a hydrophobic peptide that despite their different interaction mechanisms, the suppression of the fibril pathway can be deduced from the disappearance of the corresponding structure of the first amyloid-β oligomers.


Introduction
Dementia is a widespread condition in the western civilization with Alzheimer's disease (AD) being the most common form (Folch et al., 2018;Hamley, 2012;Kumar and Hamilton, 2017). Up to now, no effective treatment of AD has been developed, and substantial efforts are underway to develop drugs that effectively inhibit AD pathogenesis. The cause of onset and progression of AD is unknown, but evidence points to the increased formation of oligomers of the neurotoxic amyloid-b peptide (Ab 42 ) playing a major role in the development of AD (Hamley, 2012;Anand et al., 2014;Bu et al., 2016;Defelice and Ferreira, 2002;Gilbert, 2014;Lee et al., 2017;Zhao et al., 2012). The soluble Ab 42 oligomers aggregate into fibrils and the detrimental effects of AD seem to be caused by neurotoxicity of the oligomers (Habchi et al., 2017;Dobson, 2003). Investigation of the oligomerization pathway to identify oligomeric states and interactions stabilizing these states is therefore imperative. When isolated in solution, Ab 42 shows characteristic patterns of an unfolded polypeptide chain lacking persistent tertiary structure (Habchi et al., 2016;Chiti et al., 2003). Depending on concentration, temperature, and homogeneity of the starting material, the polypeptide chain rapidly forms fibrils. The significance of the exact peptide structure for the aggregation process was suggested by molecular dynamic simulations (MD) (Barz et al., 2018). Nevertheless, structure determination of the fibrillar state of Ab 42 has long been challenging (Miller et al., 2010) and only recently structures for Ab 42 fibrils were determined using solid-state nuclear magnetic resonance (NMR) and cryo-electron-microscopy (EM) (Colvin et al., 2016;Gremer et al., 2017;Wälti et al., 2016;Xiao et al., 2015). The analysis of Ab 42 structure by solid-state NMR and by cryo-EM consistently revealed an S shaped conformation formed by two branches of Ab 42 as the basis structure within the fibril. The first branch is stabilized via hydrophobic interactions between amino acids L17 to I32 (Colvin et al., 2016;Gremer et al., 2017;Wälti et al., 2016;Xiao et al., 2015), including the hydrophobic KLVFF region (residues 16-20). The second branch is stabilized via an ionic interaction between the K28 sidechain and the C-terminal A42 (Colvin et al., 2016;Wälti et al., 2016;Xiao et al., 2015). Oligomerization of monomeric Ab 42 is conceivable as a parallel stacking in the axial direction of fibril growth. However, recently published fibril structures show fibrillization to occur not via a stacking of a monomeric Ab 42 base (MB), but with a dimeric Ab 42 base (DB) as the basic module for fibrillary stacking. DB consists of two S shaped Ab 42 molecules ordered in a C2 symmetric ying-yang-like fashion (Colvin et al., 2016;Gremer et al., 2017;Wälti et al., 2016). This DB nucleation of Ab 42 leads typically to fibrils which are known as the on-pathway aggregates. At the same time, Ab 42 aggregation might deviate from this oligomerization following a socalled off-pathway aggregation. The aggregates formed via the off-pathway are amorphous and do not show toxicity (Hamley, 2012;Lee et al., 2017;Jiang et al., 2012;Kłoniecki et al., 2011;Sitkiewicz et al., 2014;Taylor et al., 2010;Woods et al., 2013). Until today, several mass spectrometry (MS) studies investigated Ab 42 oligomerization (Bernstein et al., 2009;Pujol-Pina et al., 2015). The combination of MS with ion mobility spectrometry (IMS) has shown differences stemming from different Ab 42 isoforms (Bernstein et al., 2005;Soper et al., 2013) as well as ligand binding (Young et al., 2015;Zheng et al., 2015;Beck et al., 2015). However, no structural information deduced from IMS measurements has yet allowed deciphering the fundamental mechanism of fibril growth and no structural basis for interfering with potential on-and off pathway oligomer formation by inhibitors of low molecular weight could be elucidated. Thus, we applied native MS combined with IMS to investigate the formation of the first Ab 42 oligomers. While oligomers of Ab 42 with same size but different arrangements do not differ in mass, they do in their shapes which makes it possible to separate the different species in an electrospray ionization (ESI) IM experiment Konijnenberg et al., 2013;Young et al., 2014). This allowed us to delineate the structural differences between peptides aggregating via different oligomerization pathways to determine attributes relevant for the toxicity of Ab 42 . Further, for two Ab 42 ligands OR2 (Austen et al., 2008) and CLR01 (Sinha et al., 2011) with different interaction modes, if binding to Ab 42 , we can show structural and kinetic differences in inhibition of the same on-pathway oligomerization and deduce a structural basis for their mode of action. Comprehension of the molecular mechanisms controlling the on-or off-pathway aggregation of Ab 42 is pre-requisite for the design of pharmaceutic agents suppressing Ab 42 oligomerization and thereby neurotoxicity.

Results
Every Ab 42 oligomer has two arrangements Figure 1A depicts an ion mobility (IM) driftscope plot of Ab 42 . Oligomers from monomer to nonamer with several charge states in different intensities can be observed. The IMS driftscope spectrum reveals more than one drift time for every m/z value, either arising from different tertiary or quaternary structures for each of the oligomers or an overlap of the charge distributions of different oligomeric species. In such cases where the origin of the measured peaks is not directly clear, the isotopic pattern of the m/z signals can be used to unambiguously determine the mass of the correlating species. While the isotopic patterns show that some overlap occurs, there is clear evidence that all oligomers adopt several conformations. The insets in Figure 1A show an example for both cases. Inset (ii) depicts for the MS peak of m/z = 2258 several species with different drift times. These show different isotope distributions, by which they can be assigned to the respective oligomers, showing that this mass peak contains signal of several overlapping oligomers. In contrast, inset (i) shows for the MS peak at m/z = 1806 an identical isotope distribution for all three appearing drift time signals, indicating the existence of only one oligomer -the dimer (5-times charged) with three conformations. The two smaller of these three dimer conformations represent structures present in solution (Kaltashov and Eyles, 2002;Ruotolo and Robinson, 2006). The signal with the lowest drift time decreases in intensity with a higher collision energy (CE), while the signal with the highest drift time increases (Figure 2A). This is due to collision induced unfolding (CIU) of the Ab 42 ion in the gas phase into the larger conformation. The same observation holds true for all oligomers. For every oligomer, we observe two differently structured oligomeric states and for higher lab frame energies (higher charge states and/or higher collision energies), two additional extended structures as the result of unfolding of the two compact species (see as well Figure 2C). No dominating larger oligomer, such as a 'magic' hexamer, as suggested previously, was found (further discussion in Fig. Appendix 1-figure 1 and 2; Bernstein et al., 2005). To better understand the oligomerization process, we used the drift times of the structured oligomeric states to calculate their collision cross-section (CCS) (Laszlo and Bush, 2017;Stow et al., 2017;Warnke et al., 2017). Experimental CCS can be compared to theoretical growth models and to theoretically calculated CCS of structural models. As it is under debate whether low mass Ab 42 oligomers already possess a stable structure, rather than having predominantly unstructured features, (Kumar and Hamilton, 2017;Colvin et al., 2016;Abedini and Raleigh, 2009;Rauk, 2009) we first calculated CCS fitting isotropic and linear growth models as published by Bleiholder, et al. (Figure 1B;Bleiholder et al., 2011). Proteins, aggregating in an unstructured fashion would resemble an isotropic growth pattern while fibrillary growth would match a linear fit. We calculated growth models for comparison with both sets of experimentally Signals corresponding to oligomers up to nonamer in several charge states and several conformations can be seen (detailed description of the peak assignment can be found in Appendix 1- figure 1A). The intensity is decoded logarithmically in a heat map where red denotes a high, and blue a low intensity. The observable oligomers are denoted as follows: M = monomer, D = dimer, Tr = trimer, Te = tetramer, Pe = pentamer, Hx = hexamer, Hp = heptamer, Oc = octamer, Nn = nonamer. The two insets show driftscope zooms of the indicated m/z areas. (i) Shows the driftscope of the 5-times charged Ab 42 dimer at m/ z = 1806. The identical isotope distribution of the three peaks unambiguously indicate the presence of three different conformations for the same oligomeric state (blue). The different isotopic distributions in (ii) show that the mass peak of m/z = 2258 consists of overlapping species of several oligomers (the dominating ones are depicted in yellow, orange, blue). For the experimentally determined oligomers in A, the CCS were calculated. (B) depicts the CCS for all oligomers which could be assigned to solution conformations of the respective oligomer (red and blue circles). For comparison, the lines indicate theoretically calculated CCS following an isotropic growth model (dotted lines) or linear growth model (solid lines) (Bleiholder et al., 2011). (C) depicts the proposed process during the first self assembly steps of Ab 42 peptides. The first two monomers can form a planar dimer, which is the base (DB) for further addition of Ab 42 monomers. If no planar dimer is formed, the Ab 42 monomers stack axially onto the monomeric Ab 42 base (MB). This results in two morphologies: A single stranded aggregate with a MB conformation, and a zipper like structure with a DB arrangement for fibril growth. derived CCS. In both cases, we used the respective dimer CCS as a spatially isotropic peptide as reference for the isotropic growth model. We assess the growth behavior separately for both data sets (depicted in red or blue in Figure 1B, respectively). The data set with the lower CCS values (red in Figure 1B) clearly corresponds to the isotropic growth model for the lower oligomers, while unambiguous assignment to either model is not possible for higher oligomers. The experimental data points for each oligomer show a less homogeneous distribution of CCS than the blue dataset, as seen by the larger error bars, suggesting less clearly defined structures. Assessment of the growth behavior of the data set with the higher CCS values (indicated in blue in Figure 1B) shows that the isotropic growth curves describes our data up to the tetramers well, after which the experimental data clearly deviates from the isotropic growth model predictions. The CCS of these higher order oligomers have a different dependence on the oligomeric Ab 42 state. CCS for oligomers larger than a tetramer can be perfectly fit by a linear growth model. Thus, our data show a change in the arrangement of oligomers from a more globular to a linear growth with oligomers larger than tetramers following the linear growth model. This suggests that less than four peptides do not stabilize each other sufficiently to form a stable structure. This finding is especially interesting in the light of predictions based on cryo-EM and solid-state NMR studies, which proposed a minimum of five Ab 42 molecules required to form a regular fibril structure (Gremer et al., 2017). Our observation of linear growth from tetramer onward is encouraging for the aim to correlate our findings with structures determined by solid-state NMR (Colvin et al., 2016;Wälti et al., 2016) or by cryo-EM (Gremer et al., 2017), bearing in mind that structural polymorphism has been reported both with regard to the microscopic structure and morphology of fibrils. The common feature of the reported Figure 2. IM spectra of the five-times charged Ab 42 dimer (i) and of the seven-times charged trimer (ii) for different CE. The structured MB conformation at a drift time of 8 ms shows an unfolding product (drift time above 12 ms), whereas the structured DB conformation (drift time of 11 ms) does not unfold before undergoing collision induced dissociation. (B) Driftscope plots of the five-times charged dimer (i) and the seven-times charged trimer (ii), shown in A. The colors indicate the two dimer structures (purple and red) as well as the unfolding product (green). For the trimer, a second unfolding product is detected (dark blue). (C) Increased lab frame energies can lead to unfolding of both structured conformations for each oligomer: CCS for the four species (two compact and two extended states) for each of the different oligomers, averaged over all charge states. Color code matches with B+D. (D) CIU of the different MB oligomers depending on CE, corrected for the charge states of the oligomers. The stability of the oligomers increases with their size at different rates below and above the tetramer (inset).
structures is a fibrillary structure with S-shaped Ab 42 forming a dimer as base for the fibril growth. These imply that Ab 42 peptides have two different ways of interacting with each other -a planar interaction, forming a planar dimer or a stacking interaction, allowing growth along the fibril growth axis. The two different CCS we observe for each oligomer also suggest more than a single growth mechanism (Bleiholder et al., 2011). Both sets of CCS show a very similar increase of CCS per oligomer (compare the linear fits in Figure 1B), suggesting growth via the same S-shaped Ab 42 monomers. But the different y-interceptions indicate a different growth base leading either to a singlebase oligomer or an oligomer building on a dimeric base, known as b-sheet zippers (Bleiholder et al., 2011). Thus, we assign the oligomers giving the larger set of the experimentally determined CCS to Ab 42 peptides in DB arranged structures. For the second set of CCS, the overall values are smaller than could be explained by structures with a DB Ab 42 -they show less homogeneous structures and therewith less clearly follow a linear growth model. This corroborates the assignment of these oligomers to MB arranged structures, which could be less well defined than the DB arranged counterparts. In summary, we assign the two different CCS to arise from either growth via singular Ab 42 peptides starting from an Ab 42 monomer or alternatively stacking onto a planar Ab 42 dimer ( Figure 1C). Based on these findings, we conclude that two oligomerization pathways exist, one via DB which are on pathway to the stable fibrillary structures as suggested by NMR and cryo-EM measurements (Colvin et al., 2016;Gremer et al., 2017;Wälti et al., 2016). We suggest that the other pathway, where oligomers are formed via MB, can potentially lead to amorphous aggregates but not stable fibrils. This model of two growth pathways is interesting in the light of studies showing that a single mutation of Ab 42 can change the aggregation pathway leading to amorphous aggregates only, rather than fibrils, as was shown for the Ab 42 mutant F19P (de Groot et al., 2006). If our model is correct and the DB conformation is essential for the formation of fibrils, we would expect it to be missing for this mutant. In supplementary data Appendix 1-figure 3A, we show the differences in the IM spectrum of F19P as opposed to wild-type Ab 42 , for the example of (Ab 42 ) 2 5+ (D 5+ ). In support of our model, we can assign the structured and the extended conformation of the MB conformation, but no signal that corresponds to a DB arrangement. The respective TEM images shows amorphous aggregation (Appendix 1- figure 3B). These findings show that we can determine with native MS the structural development of small Ab 42 oligomers to occur via two pathways, MB and DB stacking. The observed growth pattern changes for DB oligomers larger than a tetramer, which represents a landmark in Ab 42 oligomer growth, at which the Ab 42 peptides support each other enough to build structures which follow linear growth.

Oligomer stability is size-dependent
The IM spectra in Figure 2A show the two structured species of D 5+ at drift times around 8 ms and 11 ms, representing the DB and MB arrangements, respectively. At higher CEs, the IM spectra additionally reveal unfolding for the Ab 42 oligomers whereby, despite peak overlap, the oligomeric state can be unambiguously assigned due to isotopic resolution of the corresponding driftscope plots ( Figure 2B). To further investigate this effect, we submitted the Ab 42 oligomers to increasing collisional activation in the collision cell, leading to CIU of the peptide ions. This unfolding process reports on the intramolecular stability within the polypeptide chain and can vary if these interactions are changed by ligand binding, mutations or changes in pH (Dixit et al., 2018). Only one species of the two Ab 42 dimer conformations of D 5+ shows unfolding upon collision with inert gas molecules. A signal correlating to the unfolding product of the DB conformation is missing for this charge state. Thus, the CE applied is sufficient to unfold the MB but not the DB conformation before collisioninduced dissociation (CID) of the dimer at 125 eV (Figure 2A(i)). (For signal intensities in dependence of different CE see as well Figure 4D(i)). For all oligomers, the higher charge states show unfolding for both species (Figure 2A(ii) as example for Tr 7+ ). Figure 2C shows the CCS for the two observed structured complexes as well as two unfolding products for all oligomers. Supplementary data of Appendix 1-figure 4 depicts this in more detail for the example of the nine-times charged pentamer. The intermolecular stability shifts with oligomeric size. Observing these shifts within the MB conformations allows us to monitor the stability increase of oligomer growth along the growth axis. For the dimer, 50% unfolding is reached at 85 eV ( Figure 2D). The CE needed for unfolding increases for every oligomer with the heptamer unfolding at 450 eV. Interestingly, the energy gaps between the CE sufficient to unfold the different oligomers are equidistant from dimer to tetramer and again from pentamer, but with a decreased energy gap. In line with this, the red line in the inset in Figure 2D shows equidistance of the energy gaps for oligomers larger than a tetramer. Compared to this trend, which reflects the stability against unfolding for the larger oligomers, the CEs needed to unfold oligomers smaller than a tetramer (gray line in inset of Figure 2D) are surprisingly low. This supports our earlier notion that an ordered stable fibrillary structure evolves only after the tetramer, in line with previously proposed theoretical models (Gremer et al., 2017).

MB and DB structures
After establishing that Ab 42 can oligomerize via two pathways with an MB or DB base, we attempted to shed more light on the resulting structures. We made use of protein database (PDB) entries of Ab 42 fibrils to calculate theoretical CCS for the different oligomers. To distinguish between fibrillary growth via stacking based on an Ab 42 monomer and a growth structure with a planar arranged DB Ab 42 , we dissected the PDB structure and constructed two sets of Ab 42 oligomers. Each structure was subjected to MD simulation for up to 1 ms to allow rearrangement of the fibril fragments, taking into account the effects of the charges carried due to the ESI process. The resulting structures support our earlier notion regarding development from more isotropic structures toward more stable fibril-like structures for the higher oligomeric states (example structures shown in supplementary data Appendix 1- figure 5 and 6). We analyzed the obtained structures for different structural elements, such as coiled or b-sheet structure (see supplementary data Appendix 1-figure 6). The most striking feature is the increase of the percentage of b-sheet motive, the bigger the oligomers. This is the case for all tested PDB structures as well as both types of oligomers: the MB and the DB based structures. The higher b-sheet content correspond to the higher stability, seen for the higher oligomers. These findings correlate well with insights from infra-red spectroscopic analysis of IMS-MS separated gas phase ions of smaller amyloidic peptides. As well an increase in b-sheet content was revealed for higher oligomers, showing that stable structural features remain in the gas phase . Using the MD structures, theoretical CCS were calculated (see supplementary information for detailed description). While it should be noted that we are comparing gas phase structures here, which stem from different solution structures, the found correlations are noteworthy. The best correlation was found with CCS obtained from PDB structure 5OQV (see supplementary data Appendix 1- figure 7). The theoretical CCS correlate well with the experimentally determined values. They show the same increase of CCS per additional Ab 42 monomer for both sets of values, supporting the S-structured Ab 42 as the building block in each case. Additionally, the theoretical CCS for the fibrillary DB oligomers correspond very well to the overall values of the experimentally determined CCS.

CLR01 inhibits Ab 42 oligomerization
Our MS investigations of the different aggregation pathways of Ab 42 oligomers might provide a rationale for aggregation inhibition by low-molecular-weight inhibitors. We therefore investigated aggregation inhibition with respect to the inhibitor CLR01 (molecule depicted in Figure 4A (ii) and supplementary data Appendix 1-figure 8). As ESI only allows detection of the first few oligomers and does not report on time-dependent changes, we performed time-resolved laser-induced liquid bead ion desorption mass spectrometry (LILBID-MS) measurements. For this purpose, aliquots of Ab 42 were incubated at 22˚C in the presence and absence of inhibitors. Figure 3 and supplementary data Appendix 1- figure 9 show the development of the aggregation process of Ab 42 without and with a 4-fold excess of CLR01. For Ab 42 alone, aggregation development suggests a constant progression from monomers up to dodecamers after 200 min. In comparison, the highest oligomeric state present 200 min after addition of CLR01 is a hexamer. The kinetics of Ab 42 aggregation, show the inhibitory effect of CLR01 (Figure 3). The inhibiting effect is comparable to that observed for OR2 (Stark et al., 2017; Appendix 1-figure 8).

Inhibitors stop origin of dimer-based oligomers
We further measured changes in IM of Ab 42 in the presence of fourfold excess of CLR01 and OR2 (supplementary data Appendix 1- figure 8). For both ligands, we detect IM peaks for the MB Ab 42 dimer binding to a single ligand molecule. This peak appears at higher drift times compared to the Ab 42 dimer. In sharp contrast, no signal can be observed for the DB Ab 42 dimer, showing the suppression of this conformer by both inhibitors. (Figure 4A(i) and (ii) and Appendix 1- figure  9A). Both inhibitors having the same effect (inhibition of DB dimer formation) is especially interesting in view of the differences of these inhibitors. The inhibitor OR2 is designed to interfere with the central hydrophobic KLVFF region in Ab 42 (Austen et al., 2008), while CLR01 was shown in the monomeric unfolded state of Ab 42 to interact in the area of R5, K16, and K28 (supplementary data Appendix 1- figure 10; Zheng et al., 2015;Sinha et al., 2011;Schrader et al., 2016;Sinha et al., 2012a;Sinha et al., 2012b), suggesting a different mechanism of action. The disappearance of one of the basic modules for Ab 42 oligomerization (the DB module) upon binding of ligands shows the ligands' influence on the quaternary structure of Ab 42 , which must be achieved via different changes in tertiary structure: Such changes in tertiary structure can go along with a change in a protein's stability, which can be observed as a different reaction toward collisional activation. Figure 4 depicts the effect of increasing CE on free and CLR01-bound D 5+ . (The respective experiments with OR2 are shown in supplementary data Appendix 1- figure 12). Both species, the bound and the unbound one, dissociate into Ab 42 monomers (M 2+ and M 3+ ) ( Figure 4B(i) and (ii)). CLR01 can remain bound to one Ab 42 monomer. To look at this in more detail, CIU heat maps ( Figure 4C) of free and CLR01 bound D 5+ and the respective intensity plots ( Figure 4D) show the effect that increasing CE has on the dimer. An enhanced CE leads to CIU, as seen by an increase of drift time in Figure 4C, and CID observable by the disappearance of the dimer signal. For easy comparison significant CEs (leading to 50% CIU or CID) are summarized in the graphs in Figure 4E(i) and (ii). For the different species, these CEs deviate noticeably, revealing differences regarding the dimers' stabilities toward unfolding or dissociation (Figure 4(i) and (ii) as well as supplementary data Appendix 1-figure 11), which will be explained in detail in the following: For unbound Ab 42 , we observe differences between the MB and the DB structures. The increase of CE to 85 eV causes 50% unfolding of the MB dimer ( Figure 4D(i) and supplementary data Appendix 1figure 11B), and higher energies are needed for dissociation. In contrast, the same 85 eV is already sufficient for 50% dissociation of the DB dimer, which interestingly dissociates without any prior unfolding ( Figure 4C(i) and D(i)). For ligand bound Ab 42 , we can only observe the effects of CE for the MB dimers, as both ligands prevent the formation of the DB dimer. Both ligands increase the CE needed for 50% CID slightly (for CLR01 shown in Figure 4B(i) and supplementary data Appendix 1- figure 11C). The unfolding process is not hampered upon OR2 binding (a similar amount of CE leads to 50% CIU with or without OR2 bound) while CLR01 interaction stabilizes the intramolecular structure, so 50% CIU cannot be achieved as maximally 15% of the proteins unfold prior to complete D 5+ dissociation ( Figure 4C(ii) and D(ii)). While OR2 does not alter the unfolding tendencies of Ab 42 , CLR01 strongly stabilizes intramolecular interactions within an Ab 42 peptide, while leaving intermolecular interactions in terms of aggregation axis growth unaltered. These results show that both ligands have a different mechanism by which they hinder the planar DB interaction and therewith suppress DB formation, leading to the formation of amorphous aggregates instead of fibrils, as seen in TEM images ( Figure 5). This is especially interesting, as it shows that drugs, which inhibit the fibril formation of Ab 42 might interact via different mechanisms, but can still be assessed by simply monitoring their ability to suppress the DB dimer.

Discussion
Our data support the formation of two arrangements for the first step in fibril formation of Ab 42 , the major cause for progression of the AD. We can characterize stability and formation kinetics of the two fundamentally important Ab 42 dimers which form the basis for Ab 42 oligomers. According to previously characterized fibril structures, (Colvin et al., 2016;Gremer et al., 2017;Wälti et al., 2016;Xiao et al., 2015) dimers can either form C2 symmetric Ab 42 structures (DB form) or exhibit only translational symmetry along the aggregation axis (MB). These dimers serve as base for further aggregation. Our IM data reveals that only oligomers larger than a tetramer support each other sufficiently to adopt a stable structure. The structures of smaller oligomers are less defined, tending to adjust to a more globular overall shape. Nevertheless, well-defined inter-and intramolecular interactions are already formed, as seen by CIU and CID data. This is relevant, as it shows that the foundation for the fibril formation is already laid in the first dimer. Inhibitors or mutations that influence the formation of the aggregation base will lead to different oligomerization processes. Figure 5 summarizes our findings for Ab 42 . Despite different binding sites of the two different inhibitors (ionic/hydrophobic), it is remarkable that both disturb the formation of the S-shaped Ab 42 structure enough to hinder the formation of the planar DB Ab 42 , which is a pre-requisite for the formation of stable fibrils. Without this DB dimer the remaining Ab 42 monomers stack axially in a less stable manner. This leads to the inhibition of the on-pathway fibrillary growth as we could show with time resolved measurements. We conclude that an undisturbed S-shaped structure of the Ab 42 monomer is relevant for an orderly evolvement of large oligomers and fibrils.
TEM images show an off pathway amorphous aggregation instead of an on-pathway fibrillary one in all the cases where we found the DB arrangement to be missing (Ab 42 with OR2 or CLR01, as well as for the Ab 42 mutant F19P). Those amorphous aggregates are known to be non-toxic (Lee et al.,  (ii) show an intensity plot of the peaks visible in the CIU experiment of C. (E) CE which lead to 50% CIU or CID for D 5+ with or without one bound ligand for the MB (i) and DB (ii) conformation. In case of CLR01, the maximally observed amount of CIU product is 15% due to prior CID. Errors given are a conservative estimate of three repeats. Data shown in supplementary data Appendix 1-figure 11. 2017) which explains the detoxifying effect of CLR01 (Sinha et al., 2012a). In summary, we could show by combining our MS and IMS results that the DB conformation is the seed for the toxic on-pathway Ab 42 oligomers. This aggregation module is suppressed by the inhibitors used herein. This makes the suppression of the DB Ab 42 dimer conformation a prime target for future drug development.

Sample preparation
Beta-amyloid: For the mass spectrometry (MS) experiments both recombinant and synthetic Ab 42 was analyzed. Cloning, Expression, and Purification of recombinant Ab 42 : The Ab 42 DNA sequence, codon-optimized for E. coli, used to construct the synthetic gene was the same as that described by Weber et al., 2014, which was based on the previously published construct by Walsh et al., 2009 The gene was constructed with BSaI and XbaI restriction sites and cloned into a pE-SUMOpro expression vector which yielded a construct consisting of a His-6 tagged N-terminal SUMO fusion protein attached to the N-terminus of Ab 42 . SUMO-Ab 42 plasmid DNA was transformed into competent E. coli DH5a subcloning cells (Invitrogen, Carlsbad, CA, USA) by heat shock, amplified and extracted via miniprep according to the manufacturer's guidelines (Qiagen, Hilden, Germany). Expression of uniformly labeled 15 N-or 13 C 15 N-SUMO-Ab 42 involved selecting a single colony of freshly transformed E. coli BL21 DE3 expression cells to inoculate 5 mL LBmedium with ampicillin, from which 100 mL was used to inoculate 100 mL LB-medium with ampicillin. The cells from the 100 mL overnight starter culture were harvested by centrifugation (6000 g, 10 min, 25˚C), suspended in M9 minimal medium and used to inoculate 1 L M9 minimal medium enriched with 13 C-glucose and/or ( 15 NH 4 ) 2 SO 4 (1 g/L). The cultures were grown at 37˚C until an OD 600 of 0.6 was reached, whereupon expression of SUMO-Ab 42 was induced with 1 mM isopropyl b-D-1-thiogalacto-pyranoside. The cells were harvested by centrifugation (4000 g, 30 min, 4˚C) after 5 hr of incubation at 37˚C (5.5 g/L wet cell mass) and stored at À80˚C. Purification of 15 N-or 13 C 15 N-Ab 42 was performed in part as previously described with some modifications (Weber et al., 2014;Walsh et al., 2009). The cell pellet was resuspended in IB buffer -10 mM Tris, 175 mM NaCl, 1 mM DTT, 1 mM EDTA, pH 8.0, sonicated (5 Â 30 s at 50% power) on ice and subsequently centrifuged (6000 g, 10 min, 4˚C). The supernatant was decanted, the pellet resuspended in IB buffer and the sonication and centrifugation procedure was repeated a further time. The pellet was resuspended and washed (IB buffer) using a dounce homogenizer, then centrifuged (6000 g, 10 min, 4˚C). This procedure was repeated three times. The (white IB) pellet was solubilized and sonicated (3 Â 30 s at 50% power) in lysis buffer (either 6 M GdnCl, 100 mM NaPi, 20 mM Tris, pH 8.0 or 8 M Urea, 100 mM NaPi, 10 mM Tris, pH 8.0), then centrifuged (6000 g, 10 min, 4˚C). The supernatant was diluted to a final urea or GdnHCl concentration of 0.5 M, after which SUMO protease Ubiquitin like protein-1 (Ulp1) was directly added and allowed to cleave over 48 hr at 4˚C or 2 hr at room temperature. Cleavage of SUMO-Ab 42 was verified by SDS-PAGE. The cleavage mixture was EtOH precipitated and the pellet was dissolved in 70% formic acid, passed through a 0.45 mm PTFE filter and subjected to HPLC chromatography (Perfectsil RP4 column, 4.6 Â 250 mm, 300 Å C4 5 mM, 60˚C, 1.3 mL/min; 230 nm fixed wavelength detection). A 30-80% gradient was run where the solvent was 400 mM HFIP/TEA, pH 7.6 and the eluent MeOH. Fractions were collected on ice, analyzed by SDS-PAGE and MALDI and the Ab 42 -containing fractions dried at 4˚C in a concentrator plus (Eppendorf, Hamburg, Germany) to produce a peptide film, which was stored at À80˚C. The final yield of Ab 42 after HPLC purification was approximately 3.5 mg/L. Expression and purification of the SUMO-protease Ulp1: Plasmid DNA encoding the residues 403-621 of Saccharomyces cerevisiae Ulp1 with a Histagged N-terminal was used. The pET-28b plasmid encoding Ulp1 was transformed into E. coli DH5a subcloning cells for the amplification of the plasmid DNA, and into BL21 DE3 cells for expression. For expression, 1 L LB medium selective for ampicillin (100 mg/mL ampicillin) was inoculated with cells to an OD 600 of 0.1 and grown at 37˚C and 180 rpm. Expression was induced at an OD 600 of 0.6 by the addition of 1 mM IPTG and growth was continued for a further 4-5 hr. The cells were harvested by centrifugation (4000 g, 20 min, 4˚C), frozen and stored at À80˚C (wet cell mass 5.2 g/ L). Ulp1 SUMO-protease cell pellets were suspended in native buffer (50 mM NaPi, 300 mM NaCl, 10 mM Imidazole, 10 mM b-ME, pH 8.0) and subjected to sonification (15 cycles, 50% power) to lyse the cells. The lysate was centrifuged and the clear supernatant loaded to a pre-equilibrated 5 mL Ni-NTA HisTrap column (Qiagen) whereby an elution gradient with increasing imidazole concentration resulted in the elution of the Ulp1 at 300 mM Imidazole. A yield of 65 mg/L culture was determined using the theoretical extinction coefficient of 28590 M À1 cm À1 at 280 nm. Synthetic Ab 42 was prepared as published before (Stark et al., 2017). Briefly, lyophilized Ab 42 was purchased from AnaSpec, USA (#24224). The peptide was solvated in HFIP containing 3% concentrated NH 3 . After incubation for 5 min, the solution was aliquoted in protein low-bind Eppendorf tubes and the solvent was evaporated using a SpeedVac RVC 2-18 (Christ, Osterode, Germany) concentration system. The remaining peptide film was stored at À80˚C. For usage, the peptide film was solvated using DMSO (1%) and 50 mM NH 4 OAc buffer at pH 7.4 (if not stated otherwise) to a final peptide concentration of 50 mM. It was stored on ice until beginning the kinetic measurements. Incubation of the Ab 42 sample was done at room temperature of 22˚C in its respective Eppendorf tube. Both Ab 42 species, recombinant and synthetic, behave identically regarding their structure as detected by ESI ion mobility spectrometry (IMS) as well as in oligomerization detected by LILBID-MS (supplementary data Appendix 1- figure 13). CLR01 was synthesized and then solvated at a concentration of 2 mM in deionized water (Fokkens et al., 2005). For testing the influence of CLR01 on Ab 42 , the molecule was added in fourfold excess to the peptide. OR2 was synthesized as published before via solidphase synthesis (Cernescu et al., 2012;Matharu et al., 2010). For storage, the molecule was solvated at a concentration of 20 mM in DMSO which was diluted to fourfold excess regarding Ab 42 for MS analysis.

Mass spectrometry analysis
Electrospray ionization mass spectrometry (ESI-MS) was performed on a Synapt G2S (Waters Corpn., Wilmslow, Manchester, UK) equipped with a high-mass quadrupole upgrade. Pd/Pt sputtered nESI tips were pulled in house from borosilicate glass capillaries on a Flaming/Brown Micropipette Puller (P-1000; Sutter Instrument Co.). Ab 42 was analyzed in positive ion mode using a capillary voltage of 1.9 kV. The rest of the settings for MS analysis were adjusted as following: cone voltage 100 V at an offset of 80 V, 20˚C source temperature. The instrument was calibrated by a conventional CsI solution. All experiments were as well performed in negative ion mode. The observed spectra support our conclusions (Appendix 1-figure 2) but show lower signal intensity/quality. Therefore, in this publication we present the data achieved in positive ion mode. IM experiments were done on the same instrument using a traveling wave setup with a wave height of 40 V, a travelling wave velocity of 700 m/s, a nitrogen gas flow of 90 mL/min, drift cell pressure of 3.5 mbar. To calculate CCS values, the instrument was calibrated using cytochrome c, apo-myoglobin and ubiquitin under denaturing conditions (Appendix 1-figure 14). Collision-induced unfolding (CIU) and -dissociation (CID) experiments were done by ramping the trap collision energy (CE) in steps of 5 V from 5 to 50 V. MS-MS was performed to detect the appearance of the dissociation products of the CID experiment. Thereby, the five-times charged dimer peak, which is discussed in the manuscript, was selected at m/z = 1806 using an LM resolution of 12 and a HM resolution of 15. Laser-induced liquid bead ion desorption MS (LILBID-MS) measurements were performed as previously published (Stark et al., 2017;Cernescu et al., 2012). Thereby, for each time-point 4 mL of the Ab 42 solution described above was injected into the droplet generator (MD-K130 from microdrop Technologies GmbH, Norderstedt, Germany) separately. The ions produced by the LILBID process were analyzed as negative ions by time-of-flight. Four spectra were recorded at laser intensities below the damage threshold of the Ab 42 oligomers. Each of those spectra is an average of the analysis of 500 droplets. Mass-calibration was achieved by recording spectra of bovine serum albumin. The oligomeric state of the Ab 42 sample was determined in terms of the monomer-to-oligomer (M/O) ratio by calculating the ratio of the intensities of the Ab 42 peaks: The values of those four spectra were averaged to obtain a time depending M/O value. Data analysis of ESI-MS and IMS experiments was done using the software MassLynx V4.1 and UniDec (Marty et al., 2015). CCS calibration was performed according to the protocol by Ruotolo et al., 2008. To process the LILBID-MS spectra, the software Massign was used (Morgner and Robinson, 2012). Using this software, the raw spectra were calibrated, smoothed and background subtracted. Visualization of the results was performed using (Origin 2018 OriginLab Corporation, USA).

MD simulations
To correlate the two sets of experimental CCS values we obtained for the different Ab 42 oligomers with potential structures, we derived structures for the different oligomers from existing Ab 42 PDB entries. We selected the PDB entries 5OQV and 2NAO which represent Ab 42 fibrillar structures with a dimer base, deduced from cryo-EM and solid-state NMR experiments, respectively (Gremer et al., 2017;Wälti et al., 2016). Cropping these allowed us to obtain two sets of structures (dimer based (DB) as well as monomer based (MB)) of the respective small oligomers observed in the ESI-MS experiments. The fibril PDB structure 5OQV is a 9-mer with dimeric base, permitting to extract oligomers until 9-mer for the DB structures and oligomers until 5-mer for the MB structures. 2NAO is a 6-mer, accordingly allowing for DB 6-mer and MB 3-mer. Experience has shown that CCS values computed straight from X-Ray, NMR or cryo-EM structures are usually not identical to experimental CCS as received from ESI gas phase ions (Heo et al., 2018). We took potential alterations of the protein structures in the gas phase such as charge driven distortion or compaction due to self-solvation during the ESI process into account by performing vacuum MD simulation using Gromacs 5.0.7 (Berendsen et al., 1995;Pronk et al., 2013;Van Der Spoel et al., 2005). Preceding equilibration of the cropped PDB structures in water were performed based on Pujol-Pina et al., but with the simulation time doubled to 10 ns (Pujol-Pina et al., 2015). Prior to vacuum MD simulations charge effects were taken into account by placing charges onto the isolated structures, according to the most abundant experimental gas phase charge state. Individual charge states were adjusted by manually protonating basic residues according to the experimentally observed species. For each oligomer/charge state combination, a set of structures was generated with the respective number of protons randomly distributed to different basic residues to reflect an ensemble of structures. These structures were then submitted to computation for simulation for 1 ms or 200 ns, respectively, as detailed in supplementary data Appendix 1-figure 1B, using the AMBER99SB-ILDN force field. The resulting structures were used to perform CCS calculations using the ImoS software (Larriba and Hogan, 2013). Nitrogen was selected as a buffer gas, and pressure as well as temperature were set according to experimental conditions. The DSSP algorithm was used to assign the content of secondary structure of the resulting structures (Kabsch and Sander, 1983 The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. peaks, and only very low intensity higher oligomers in the mass spectra. For comparison we investigated insulin, a protein, which is known to aggregate to a hexamer in the presence of zinc ions (Dodson et al., 1979) B shows LILBID results and C shows nESI results of insulin (10 mM insulin in 50 mM ammonium acetate buffer), both without (i) and with (ii) 1.5-fold excess of zinc respectively. Without zinc some aggregation can be seen with both methods, albeit more of the higher oligomers are retained in the LILBID spectrum, which is similar to our observations for Ab 42 (compare Appendix 1- figure 1 and 9). Upon addition of zinc the oligomer distribution shifts to a dominant hexamer, which is clearly visible in both mass spectra. In contrast the hexamer appears in low intensity only for Ab 42 .

Author contributions
Appendix 1-figure 3. Comparision of Ab 42 wt and mutant F19P. (A) Comparison of the IM spectrum of the 5-times charged dimeric wildtype Ab 42 (i) with that of the Ab 42 mutant F19P (ii) shows loss of the DB structure for the later one. (B) TEM images of both species show that this goes along with the loss of the characteristic fibrillary structures, which can only be seen for Ab 42 wt.
oligomers. The structures above show these motives in gold for the different representative Ab 42 oligomers.
Appendix 1-figure 7.  (Gremer et al., 2017;Wälti et al., 2016). Linear fits for oligomers from tetramer on, demonstrate the difference between experimental (solid) and theoretical (dashed) CCS of the DB conformation. The structure of PDB 5OQV (upper plots) matches our experimentally determined CCS values best.
Appendix 1-figure 8. Driftscope of Ab 42 interacting with the two ligands CLR01 (A) and OR2 (B). The marker indicates an interaction of an Ab 42 oligomer with up to four copies of the respective molecule. Higher binding events are not indicated (ligand interaction with: circle = monomer, square = dimer, triangle = trimer).