CMT-3 targets different α-synuclein aggregates mitigating their toxic and inflammogenic effects

Parkinson's disease (PD) is a neurodegenerative disorder for which only symptomatic treatments are available. Repurposing drugs that target α-synuclein aggregation, considered one of the main drivers of PD progression, could accelerate the development of disease-modifying therapies. In this work, we focused on chemically modified tetracycline 3 (CMT-3), a derivative with reduced antibiotic activity that crosses the blood–brain barrier and is pharmacologically safe. We found that CMT-3 inhibited α-synuclein amyloid aggregation and led to the formation of non-toxic molecular species, unlike minocycline. Furthermore, CMT-3 disassembled preformed α-synuclein amyloid fibrils into smaller fragments that were unable to seed in subsequent aggregation reactions. Most interestingly, disaggregated species were non-toxic and less inflammogenic on brain microglial cells. Finally, we modelled the interactions between CMT-3 and α-synuclein aggregates by molecular simulations. In this way, we propose a mechanism for fibril disassembly. Our results place CMT-3 as a potential disease modifier for PD and possibly other synucleinopathies.


Results
Comparative effect of CMT-3, DOX and MINO on α-synuclein amyloid aggregation. The chemical structures of CMT-3, DOX and MINO are shown in Fig. 1a. The decreased antimicrobial activity of CMT-3 results from the removal of the dimethyl-amino group (DMA) group in the upper peripheral region of tetracyclines, at position 4 of the A ring 30 . Nevertheless, the relevance of these substituents on the inhibition of α-synuclein aggregation is unknown. We compared the impact of CMT-3, DOX and MINO on α-synuclein amyloid aggregation by monitoring fluorescence intensity of the amyloid-specific probe Thioflavin T (ThT) (Fig. 1b). The IC 50 values for the inhibition of α-synuclein aggregation, estimated from the dose-response curves, were 14.91 ± 2 μM and 14.49 ± 3 μM for CMT-3 and DOX, respectively. Since CMT-3 showed optimal inhibition of α-synuclein amyloid aggregation at 100 μM, hereafter all experiments were performed at this concentration. Note that MINO failed to inhibit amyloid aggregation of α-synuclein (Fig. 1b).
In order to confirm these results, we employed transmission electron microscopy (TEM) to visualize the effect of these different tetracyclines on α-synuclein aggregation. In agreement with ThT assays, while CMT-3 and DOX halted fibril growth, MINO failed to inhibit the formation of α-synuclein amyloid structures (Fig. 1c). From these results, we conclude that CMT-3 is as potent as DOX in inhibiting α-synuclein amyloid aggregation, while MINO lacks anti-aggregant properties.

CMT-3 inhibits α-synuclein amyloid fibril formation by binding to aggregated species.
To further characterize the effect of CMT-3 on α-synuclein aggregation, kinetic studies were performed using two complementary techniques, i.e. turbidimetry (Fig. 2a) and ThT fluorescence (Fig. 2b). While the former measures a general state of aggregation, the latter reports only the formation of amyloid aggregates. In the absence of CMT-3, both turbidimetry and ThT fluorescence techniques revealed a sigmoidal behavior for α-synuclein aggregation kinetics, with a latency time of 8 h followed by an exponential phase that reached a stationary state around 44 h, in agreement with the previous results 17 (Fig. 2a,b). However, in the presence of CMT-3, supramolecular arrangements larger than α-synuclein monomers were formed with a lag time of 12 h (Fig. 2a). Nevertheless, aggregates did not lead to the formation of amyloid fibril structures as monitored by ThT fluorescence (Fig. 2b). This observation is in agreement with small aggregates observed in TEM studies (Fig. 1c).
Small molecules can interfere with protein aggregation by binding to different species along the aggregation pathway 31,32 . In order to discover at which stage of the amyloid aggregation process CMT-3 binds to α-synuclein, we monitored changes in fluorescence anisotropy of CMT-3 upon addition of aliquots of an aggregation reaction harvested at different time-points. As seen in Fig. 2c, the fluorescence anisotropy (r) values of CMT-3 increased concomitantly to the appearance of aggregated species of α-synuclein (Fig. 2c). Binding data suggests that CMT-3 only interacts with aggregated forms of α-synuclein, which would therefore leave monomers available to perform their physiological functions.
CMT-3 reshapes α-synuclein early aggregates into non-toxic species. The toxic effects of aggregated α-synuclein have been attributed to oligomeric species (αS oli ) formed in the early stages of protein aggregation 33 . Here, we assessed the impact of CMT-3 on the structure, morphology and toxicity of these oligomeric species by infrared spectroscopy (FTIR), transmission and scanning electron microscopies (TEM and SEM), and cell culture assays.
The Amide I' band, located in-between 1700 and 1600 cm −1 of the infrared spectrum, is widely used to quantify protein structural changes during the aggregation process 34,35 . This band showed significant differences between αS oli samples prepared in the absence or in the presence of CMT-3 (Fig. 3a). Curve-fitting analysis was performed, as described in "Methods", with the purpose of obtaining a relative quantification of these changes.  Table S1). TEM images of α-synuclein samples incubated with and without CMT-3 after 16 h of orbital agitation at 37 °C revealed the formation of αS oli in both samples (Fig. 3b). The average size of the resulting aggregates was calculated from the TEM images. Aggregated species of α-synuclein formed in the presence of CMT-3 were larger, with an average diameter of 70 ± 8 nm, compared to those formed without the drug, which were 30 ± 5 nm. To gain better topological and morphological information of these early aggregates, the same TEM samples were analyzed through SEM (Fig. 3c). In this way, we confirmed that the oligomers formed in the presence of CMT-3 were definitely larger in size than those formed in the absence of CMT-3.
Next, we assessed whether the morphological and structural changes in αS oli induced by CMT-3 affected the toxic potential of oligomeric species. For this purpose, SH-SY5Y cells were incubated with α-synuclein oligomers prepared in the absence (αS oli ) or in the presence of CMT-3 (αS:CMT-3 oli ) for 24 h, and cytotoxicity was measured by an LDH assay. As shown in Fig. 3d, oligomeric species obtained in our standard conditions (αS oli ) caused a significant reduction in cell viability. In contrast, αS:CMT-3 oli species had no significant deleterious effect on SH-SY5Y cells. Importantly, this result was not due to a direct effect of CMT-3 per se, since CMT-3 had www.nature.com/scientificreports/ no protective effect when treated simultaneously with oligomers (αS oli + CMT-3) (Fig. 3d). These results suggest that oligomers formed in the presence of CMT-3 (αS:CMT-3 oli ) were structurally and morphologically distinct species from αS oli , and furthermore were innocuous in a cell culture toxicity assay.
To confirm the capacity of CMT-3 to disassemble αS PFF , we used the amyloid-reporter probe Thioflavin S (ThS) and confocal microscopy imaging. In continuously shaken α-synuclein samples, the amount of fluorescent αS PFF species progressively increased with time, indicating a rise in amyloid fibril formation ( Supplementary  Fig. S1). However, in samples where CMT-3 was added to αS PFF , the amount of amyloid structures within resulting species (αS PFF :CMT-3) diminished over time ( Supplementary Fig. S1), reinforcing data described in Fig. 4a. www.nature.com/scientificreports/ To acquire a deeper insight into the structure of the new species that result from the interaction of CMT-3 with αS PFF (αS PFF :CMT-3), electron and atomic force microscopy were used. As depicted in Fig. 4b, TEM images of αS PFF after 96 h of incubation with or without CMT-3 addition at 44 h, confirmed a significant reduction in both the number and size of fibrils. This property was only observed for CMT-3, as the addition of DOX to αS PFF had no significant impact on fibril formation (Fig. 4b). In the absence of CMT-3, αS PFF reached lengths of 640 ± 64 nm after 96 h of incubation, while αS PFF :CMT-3 in treated samples comprised of shorter fibrils (165 ± 38 nm) (Fig. 4b). These measurements represent the mean ± standard error for 3 independent experiments. In order to uncover a more detailed representation of the αS PFF :CMT-3 fragments, the same TEM samples were analyzed by SEM at different magnifications. In agreement with previous results, CMT-3-induced disruption of αS PFF was evident. At 50,000×, a panoramic view of the grid confirmed that CMT-3 induced a reduction in the total amount of fibrils and aggregated fibrillar clusters (Fig. 4c). In close-up views at 200,000×, the difference in size between αS PFF and αS PFF :CMT-3 species was observed in greater detail.
Next, atomic force microscopy (AFM) was applied in order to gain a sharper and more complete picture of the species resulting from CMT-3-induced disassembly. As shown in Fig. 4d, AFM revealed that untreated αS PFF were abundant and arranged in clusters. However, αS PFF treated with CMT-3 showed a clear reduction in number and length of α-synuclein aggregates and fibril clusters compared to control samples (Fig. 4d). From these results, we conclude that CMT-3, unlike DOX, was able to disrupt αS PFF clusters into smaller aggregates.
Disassembled α-synuclein PFF (αS PFF :CMT-3) are non-toxic and unable to either disrupt membrane integrity or promote seeding. Since CMT-3 induced the disassembly of αS PFF into smaller aggregated species (αS PFF :CMT-3), we analyzed the seeding capacity as well as the biological impact of such disrupted . Cytotoxicity values were normalized by the signal observed after addition of Triton X-100, which induced complete disruption of the cells. Data represents the mean ± S.E.M (n = 11). One-way ANOVA followed by Holm-Sidak's multiple comparisons test. ****p < 0.0001 vs NT.
Scientific Reports | (2020) 10:20258 | https://doi.org/10.1038/s41598-020-76927-0 www.nature.com/scientificreports/ fibrils. To address this, we carried out seeding experiments in which αS PFF and αS PFF :CMT-3 species were each challenged with fresh αS m in order to comparatively evaluate their seeding capacity by ThT fluorescence assays. As depicted in Fig. 5a, incubation of αS m in the absence of aggregate seeds presented a characteristic sigmoidal curve including the lag phase reported for amyloid aggregation kinetics. In contrast, αS PFF acted as seeds for the reaction, abolishing the lag phase and triggering rapid aggregation of αS m into amyloid fibrils, in agreement with previous reports 37 . However, αS PFF :CMT-3 species were unable to elicit this type of response, indicating their inability to act as polymerization seeds. In addition, these data suggest that CMT-3 not only disassembles amyloid fibrils, but also might remodel the resulting amyloid fragments (Fig. 5a). Interestingly, αS PFF :CMT-3 species even hindered the aggregation process of native αS m , implying that amyloid fibrils disassembled by CMT-3 could per se act as anti-aggregant agents.
To test the impact of αS PFF :CMT-3 species on membrane integrity, content leakage assays with synthetic membranes were performed. For this, we monitored the release of calcein entrapped in DOPC:DOPA (1:1) lipid vesicles upon the addition of αS m , αS oli, αS PFF and αS PFF :CMT-3. As shown in Fig. 5b, monomeric species and αS PFF did not induce significant changes in membrane permeability. On the contrary, αS oli induced an increase in the release of the fluorescent probe from liposomes. Interestingly, αS PFF :CMT-3 species were incapable of altering membrane permeability (Fig. 5b). The cytotoxicity signal was normalized to values obtained after addition of Triton X-100. Data represents the mean ± S.E.M (n = 11). One-way ANOVA followed by Holm-Sidak's multiple comparisons test. ****p < 0.0001 vs NT.
Scientific Reports | (2020) 10:20258 | https://doi.org/10.1038/s41598-020-76927-0 www.nature.com/scientificreports/ In order to determine if αS PFF :CMT-3 species were neurotoxic we measured the release of cytosolic enzyme lactate dehydrogenase (LDH) in SH-SY5Y human neuroblastoma cells, an in vitro model of dopaminergic neurons. In accordance with membrane permeability assays, αS oli triggered cytotoxicity while αS PFF did not, as previously reported 38 . Consistent with membrane permeability assays, αS PFF :CMT-3 species failed to induce LDH release in the culture medium (Fig. 5c). The fact that αS PFF :CMT-3 species were unable to act as polymerization seeds, alter lipid membrane permeability, nor exhibit toxicity in a cell culture model of dopaminergic neurons, has important implications from a therapeutic perspective.

Disassembled α-synuclein PFF (αS PFF :CMT-3) are less inflammogenic for microglial cells.
Fibrillary aggregates of α-synuclein possess the capacity of triggering brain neuroinflammation, a process thought to constitute a core component of PD 11,39 . For this reason, we compared the impact of 24 h-treatments with αS PFF or αS PFF :CMT-3 on microglial cell activation by measuring levels of TNF-α a pro-inflammatory cytokine, and glutamate, a non-cytokine inflammation marker 40,41 . Considering the reported anti-inflammatory effects of CMT-3 27 , we used αS PFF :CMT-3 samples that were dialyzed prior to addition to microglial cell cultures to remove the unbound fraction of tetracycline (see "Methods" for details).
In accordance with our previous reports 11,42 , treating microglial cells with 70 μg/ml of αS PFF , a concentration known to induce activation in this model system, robustly stimulated TNF-α release (Fig. 6a). The inflammatory potential of αS PFF :CMT-3 species appeared, however, significantly reduced compared to intact αS PFF (Fig. 6a). Considering its known anti-inflammatory effects 27 , unbound CMT-3 was removed by dialysis before treatment (described in "Methods"). Nevertheless, the anti-inflammatory properties of CMT-3 were not responsible for the difference observed, since microglia cell cultures pre-incubated with CMT-3 and then treated with αS PFF (CMT-3 preinc + αS PFF ) showed no statistical differences in TNF-α release compared to αS PFF (Fig. 6a). www.nature.com/scientificreports/ Following, glutamate release was measured as an alternative inflammation marker for microglial cells (Fig. 6b). As expected, a treatment of microglial cells with αS PFF substantially increased the release of the neurotransmitter 11 , while αS PFF :CMT-3 produced an attenuated response compared to that of intact αS PFF , (Fig. 6b). Again, cells pre-incubated with CMT-3 and later treated with αS PFF (CMT-3 preinc + αS PFF ) showed no difference with respect to αS PFF -treated conditions. Monomeric forms of α-synuclein, in contrast to αS PFF, had no or limited influence on the production of these inflammation markers (Supplementary Fig. S2).
In summary, molecular species resulting from the disassembly of αS PFF by CMT-3, were less effective at activating microglia than intact αS PFF , and this ability was unrelated to intrinsic anti-inflammatory properties of this tetracycline.
Binding mode of CMT-3 to α-synuclein fibrils and subsequent disassembly mechanism. In order to shed light on the binding mode of CMT-3 to aggregated α-synuclein, we performed high throughput unbiased molecular dynamic simulations. Briefly, in these simulations the ligand was randomly placed in a water box around the protein and the binding events were analyzed using the Markov state model framework 43 . This framework allows the identification of long timescale phenomena by performing multiple shorter simulations in parallel. In this way, five binding poses were identified with the first binding pose having the largest implied time scale (Supplementary Fig. S3). As seen in Fig. 7a, this binding pose involves the interaction of CMT-3 with the folded monomer at the edge of the α-synuclein fibril and perpendicular to the fibril elongation axis. An analysis of the contact frequency between the protein and the ligand revealed that CMT-3 interacts mainly with residues Ala91 (98%) and Val55 (92%), followed by Phe94 (52%), Lys97 (50%), Ala56 (46%) and Lys96 (44%) (Fig. 7b). As we suggested previously 26 , the interaction occurred mainly between carbonyl and amide atoms of the protein main chain and the hydrogen bond donor and acceptor groups present in the lower peripheral region of the tetracycline core structure (Fig. 7c). The interaction is further stabilized by hydrophobic contacts between the side chains of Val55, Ile88 and Phe94 and the carbocyclic rings in CMT-3. In this way, CMT-3 would interfere with the fibril's capacity to recruit new monomers by blocking the residues with capability to form cross-β hydrogen bonds at the edge of the protein scaffold. Moreover, CMT-3 seems to disrupt the native contacts present in the fibrillary structure.
Upon analysis of the contact map in the solid-state NMR (ssNMR) structure deposited in the Protein Databank, interchain contacts between residues in neighboring monomers were observed as an offset diagonal in the contact matrix ( Supplementary Fig. S4). These lines in the contact map correspond to in-register parallel cross-β sheets. It should be pointed out that this contact map corresponds to the idealized structure deposited in the database. On the other hand, the dynamic picture gained from these simulations shows that while the contacts in the hydrophobic core of the Greek key motif of the fibril are quite conserved (residues 46-54 and 63-96), there are certain regions of the structure with increased flexibility (residues 1-45, 55-62 and 97-120). These flexible segments can be seen as regions with a large fluctuation over their average position in the structure (see Root mean square fluctuation (RMSF) plot in Supplementary Fig. S5), or alternatively, as contacts with lower frequency on the contact map (marked with a blue square in Fig. 7d). These observations are in agreement with experimental data showing that the loop in the segment 51-67 is more disordered 44 . In our simulation studies, we found that upon binding of CMT-3 to residues 91-95, the flexible region comprised between residues 55-62 detaches from the protein scaffold, as revealed by the disappearance of the native contact on the contact map (Fig. 7e). Instead, this flexible region folds back into CMT-3 keeping a strong interaction between CMT-3 and the backbone atoms of Val55 and Ala56. We hypothesize that this could represent an initial step in fibril disassembly, which could then progress upon binding of additional CMT-3 molecules into adjacent regions.

Discussion
Despite the enormous efforts poured into developing treatments for Parkinson's disease (PD), a definitive solution is yet to be found. Several promising molecules that target PD pathogenic pathways such as α-synuclein aggregation, oxidative stress, and neuroinflammation, had limited success in clinical trials 45 . Due to the multifactorial basis of the disease, a multitarget drug with efficient activity against all these processes could open new opportunities for effective treatments.
We selected CMT-3 (also named COL-3 or Incyclinide), because of its close structural analogy to DOX and MINO, two other tetracycline derivatives reported to exert neuroprotective effects in PD models. In addition, CMT-3 is a tetracycline derivative with reduced antibiotic activity that is ready to enter in phase II clinical trials. Here, we show that CMT-3 inhibited α-synuclein amyloid aggregation in a concentration dependent manner as efficiently as DOX 17 , remodeled α-synuclein oligomers into non-toxic species, and even disaggregated preformed α-synuclein fibrils (αS PFF ). On the contrary, DOX did not induce any changes in αS PFF structures. Moreover, we unveiled that MINO, also failed to inhibit α-synuclein amyloid fibril formation. Importantly, α-synuclein species disaggregated by CMT-3 treatment (αS PFF :CMT-3) were non-toxic, less inflammogenic and did not serve as seeds for further amyloid aggregation. Considering that α-synuclein aggregates have been shown to spread from the brain of patients developing PD to grafted neurons 46 , loss of seeding capacity might be therapeutically relevant. Furthermore, disaggregated fragments (αS PFF :CMT-3) inhibited the protein aggregation reaction, acting as anti-aggregants themselves. Other compounds have been shown to disassemble α-synuclein amyloid fibrils, including epigallocatechin gallate (EGCG), baicalein, rifampicin and oleuropein derivatives [47][48][49][50] . However, to our knowledge this is the first study to evaluate the inflammogenic potential of the disassembled species. A main difference between CMT-3 and other amyloid-disrupting compounds is that this tetracycline did not bind to monomeric α-synuclein, and it would not interfere with the monomer's physiological functions 51 . These attributes, combined with the relative innocuousness, weakened antibacterial activity 27 www.nature.com/scientificreports/ antioxidant and anti-inflammogenic properties 28,29 , render CMT-3 as an attractive candidate for repurposing in PD and other synucleopathies. Tetracyclines, mainly DOX and MINO, have a number of neuroprotective properties in different PD animal models in which neurodegeneration and/or inflammation are induced by 6-hydroxydopamine (6-OHDA), 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), Paraquat or lipopolysaccharide [18][19][20]53,54 . Promising results, mainly attributed to anti-inflammatory and antioxidant capabilities, prompted a clinical trial to test the efficacy of MINO against early PD in 2008. Unfortunately, MINO failed to show any beneficial effects in this study. This trial seemed to bury the hopes of repurposing tetracyclines for PD. However, it is widely accepted that toxic amyloid aggregation of α-synuclein is the key molecular event that provokes the vicious cycle of aggregation, neuroinflammation, and other pathological hallmarks in PD. Here, we demonstrated that MINO was incapable of inhibiting this toxic α-synuclein aggregation in vitro, which may possibly explain retrospectively the absence of clinical improvement in PD patients. On the contrary, DOX and CMT-3 did efficiently prevent www.nature.com/scientificreports/ formation of toxic α-synuclein amyloid aggregates in vitro, with the latter even possessing ability to disaggregate α-synuclein amyloid fibrils. This implies that structural differences in tetracyclines conceal properties that justify re-examining their therapeutic potential for PD. CMT-3, like many other tetracyclines, has a potent inhibitory effect on the activity, activation, and production of matrix metalloproteinases (MMP) 55,56 . For this reason, it has been developed as an anti-metastatic and has been tested in clinical settings, especially for advanced soft tissue and AIDS-related Kaposi sarcomas, and high-grade gliomas 57,58 . CMT-3 has also been shown to decrease lipopolysaccharide-induced microglial activation and cytokine release in the brain, which has been attributed to an inhibitory effect on protein translation 27 . Recently, a molecular mechanism has been provided for the effect of tetracyclines (including CMT-3) at attenuating eukaryotic protein synthesis 59 . This effect also seems to be responsible for enhancing longevity and proteostasis in a C. elegans-ageing model [59][60][61] . Therefore, while the effects we describe for CMT-3 on α-synuclein amyloid aggregation result purely from ex vivo and in vitro assays, this additional mechanism could indirectly help prevent protein aggregation in vivo. Ideally, CMT-3 could attack amyloid aggregation on many fronts by tempering translation, inhibiting aggregation of what has already been translated; disrupting what has already been aggregated and reducing previously triggered inflammation and oxidation.
Considering the pleiotropic nature of tetracyclines, and the fact that chemical modification of substituents on the ring skeleton can bestow different properties, results presented herein highlight how DMA groups that endow antimicrobial activity 30 , interfere with the anti-aggregant properties of the molecule. Specifically, the DMA present in MINO and absent in both DOX and CMT-3 (on C-7) seemed to obstruct the ability of the tetracycline to inhibit α-synuclein amyloid aggregation, while the DMA present in MINO and DOX (on C-4), prevented the ability to disassemble αS PFF . This suggests that substituents on the upper peripheral region of tetracyclines could influence the anti-aggregant and disassembling properties of these molecules, a structure-activity insight that may represent precious information for future drug design.
A sharper picture of the interaction between CMT-3 and αS PFF was gained through molecular modelling. Although existing computational power limits the simulations to only the first steps of this process, we think that a model based on high-throughput molecular dynamic simulations can provide a hint on the mechanism of ligand binding and fibril disassembly. According to our model, CMT-3 would prevent aggregation by blocking specific amino acid interaction in the aggregation-prone segment A 88 AATGFVK 96 in α-synuclein, in a mechanism similar to the one proposed for EGCG 62 . In particular, the presence of hydrogen bond donors and acceptors in the lower peripheral region of CMT-3 seems to be critical for the interaction with carbonyl and amide groups present in the main chain of the protein, impairing its ability to recruit subsequent α-synuclein monomers. Also, the hydrophobic nature of the cyclic rings in CMT-3 would serve to target the hydrophobic core of the fibril. The binding mode is similar to the one recently proposed for the binding of DOX to Aβ amyloid fibers 63 . In this way, the binding mode seems to be dependent on the conformation of the aggregation-prone segment rather than on its sequence, and would explain the capability of tetracyclines to inhibit the amyloid aggregation of diverse proteins such as Aβ 64 , PrP 65 , α-synuclein 17,66 or β2-microglobulin 67 . Furthermore, we propose that CMT-3 might act as a molecular zipper, with the flexible loop folding back into the aggregation-prone segment, disrupting the Greek key motif and initiating fibril disassembly. It must be stressed that based on current simulations, we are not able to explain the different capabilities of CMT-3, DOX and MINO, since the basic structural motif is shared among all of them. Nevertheless, we hypothesize that the DMA groups on the upper peripheral region might play a role by introducing steric impediments that would affect the molecule's anti-aggregant activity. Further studies are needed to validate this hypothesis.
Considering the feasibility of a CMT-3 based treatment, long-term therapy with antibiotic-related compounds could impose a selective pressure and therefore increase the risk of microbial resistance development and native microbiota disturbance. However, several clinical studies indicate that at low doses, below the minimal inhibitory concentration (MIC), antibiotic therapy is safe, does not kill the native microbial communities nor perturb the microbia/human interaction even after following years of treatment 26,[68][69][70] . In addition, previous reports showed that the antimicrobial action of CMT-3 was dramatically reduced compared to that of doxycycline 27 . Therefore, considering its very low antimicrobial activity and the concentration required for its neuroprotective action, long-term therapy with CMT-3 should not negatively affect human native microbiota. Nevertheless, more experimental evidence is required to support this hypothesis.
Whether CMT-3 administration is able to prevent α-synuclein deposition into Lewy bodies or even disassemble α-synuclein fibrils in vivo, either in α-synuclein overexpressing PD mice models or in animals injected with αS PFF is unknown, and remains beyond the scope of this report. Also undetermined are the effects of CMT-3 on different strains of αS PFF or those that harbor mutations that predispose to PD. Although the data that support repositioning tetracycline derivatives for PD stem from either in vitro or toxin-based rodent models, a recent large retrospective study revealed that tetracycline therapy is associated with reduced risk of PD 25 , hinting that there is still as much to be learned from these compounds as there is hope for their repurposing in neurodegenerative diseases.
In conclusion, although a number of compounds have been shown to inhibit α-synuclein amyloid aggregation, few are already being tested in clinical trials, traverse the BBB, and disrupt mature αS PFF like CMT-3. These properties add to the previously known anti-inflammatory and antioxidant abilities of tetracyclines. Therefore, CMT-3 represents an ideal multi-target compound to be considered for clinical studies and potential repurposing in PD.

Methods
Preparation of α-synuclein. Recombinant wild-type human α-synuclein was expressed in Escherichia coli using plasmid pT7-7 encoding for the protein sequence. Purification was performed as previously described 71 . Protein purity was assessed by electrophoresis in polyacrylamide gels under denaturing conditions (SDS-PAGE).
Protein aggregation assays. The aggregation protocol was adapted from previous studies 72  where I F is the normalized fluorescence intensity, [Ligand] is the tetracycline concentration, and IC 50 is the concentration at which aggregation is inhibited at a 50%.
Turbidity measurements. Turbidity caused by α-synuclein aggregation in the presence or in absence of CMT-3 was examined with absorbance at 500 nm. The absorbance of each solution was monitored in a Beckman DU 7500 spectrophotometer.

Transmission electron microscopy (TEM) and scanning electron microscopy (SEM). Samples
(50 μl) of a 70 μM α-synuclein solution were adsorbed onto glow-discharged 200 mesh carbon film coated copper grids (Electron Microscopy Sciences) and stained with uranyl acetate (2%). Excess liquid was removed and grids were allowed to air dry. The grids prepared in this way were analyzed using TEM and SEM. For the former, images were captured using a Philips 301 transmission electron microscope, while for SEM, the images were taken at 3 kV with a working distance between 3 and 3.8 mm and high vacuum in a Zeiss Supra FE-SEM microscope. In order to assess the results in an unbiased manner, 30 representative fields were selected from each grid.
Atomic force microscopy (AFM). Aliquots (10 μL) of amyloid samples were deposited on the surface of freshly cleaved mica, washed three times with Milli-Q water, and dried with dry nitrogen. AFM images were acquired with a Multimode 8 AFM (NanoScope V Controller, Bruker, Santa Barbara, CA, USA). Tapping mode imaging was carried out in a dry nitrogen atmosphere using silicon rectangular cantilevers with rotated pyramidal tips. The etched silicon probe with rotated tip (RTESP) probe (Bruker) specifications include a nominal tip radius of 8 nm curvature, nominal cantilever spring constant k of 40 N/m, and nominal resonant frequency of 300 kHz. The AFM images were processed using commercial NanoScope Analysis software (Bruker) to remove the background slope.
Fluorescence confocal microscopy. α-Synuclein amyloid fibrils formed at different time-points were centrifuged 30 min at 12,000 × g. The pellets were resuspended and incubated with 0.2 mg/ml Thioflavin S for 2 h at room temperature under stirring. Then, fibrils were washed by centrifugation and resuspended in freshly prepared buffer three times to remove excess of Thioflavin S. The fibrils obtained were finally resuspended in buffer 20 mM HEPES, pH 7.4, to be observed under a Zeiss LSM 800 confocal microscope (Germany).

Fluorescence anisotropy. The interaction between α-synuclein aggregates and CMT-3 was evaluated by
measuring the fluorescence anisotropy of the intrinsic fluorescence of CMT-3. Briefly, aliquots (100 μl) were taken at different times of the α-synuclein aggregation reaction, CMT-3 was added, and then fluorescence anisotropy was measured. The fluorescence anisotropy was determined in an ISS PC1 photon counting spectrofluorometer with thermostated cuvette support, and equipped with polarizers for both the excitation beam and the emission beam, adjusting the excitation and emission wavelengths at 460 nm and 520 nm, respectively. The fluorescence anisotropy was calculated according to: where g is an instrumental correction factor that normalizes the sensitivity of the photomultiplier, while I VV and l VH are the intensities of the fluorescence with the emission polarizer in parallel or perpendicular, respectively to the excitation polarizer. www.nature.com/scientificreports/ length of 50 nm. The sample chamber was constantly purged with dry air. The spectra were obtained by averaging 256 interferograms collected with a nominal resolution of 2 cm −1 and apodized with a Happ-Genzel function in a Nicolet 5700 spectrometer equipped with a DTGS detector (Thermo Nicolet, Madison, WI) as previously described 74 . Quantitative information on protein structure was obtained through decomposition of the Amide I' band into its constituents as previously described 75,76 . Structural analyses, either in the absence or in the presence of CMT-3 were repeated three times to test the reproducibility of the measurements. In each case, the differences among the experiments was around 3% 77 .
Lipid membrane permeability. Calcein-loaded LUVs of PA/PC (molar ratio of 1:1) were prepared by hydrating lipid films in the presence of calcein and separated from free dye using a Sephadex G-75 column (Sigma). Different aggregates of α-synuclein (140 μM) were added to 50 μM of lipid vesicles. During incubation, changes in the fluorescence intensity of the different samples were monitored at λ exc = 490 nm and λ em = 510 nm in a ISS PC1 spectrofluorometer (Champaign, IL, USA). Total dye release was completed by the addition of 0.2 vol% Triton X-100. The percentage of probe release was calculated as follows: where I F , I T , and I B are the fluorescence intensity of the dye released by the protein, total dye released, and control blank, respectively. Microglial cell isolation and seeding. Coating procedure was performed as described previously 11,42 .

Dialysis.
Briefly, polycation coating solutions containing 1 mg/ml polyethyleimine (PEI) were applied to culture vessels for 2 h at 37 °C, then washed with PBS and used for cell seeding. Newborn mice were sacrificed and had their brain rapidly dissected. Two mouse brains were plated onto PEI-coated Corning T-75 flasks (Sigma-Aldrich) with 12 ml of complete medium (DMEM + 10% FBS and antibiotics) and incubated at 37 °C in an atmosphere of 95% air and 5% CO 2 . 10 ml of culture medium were removed 48 h after plating to eliminate floating debris and 10 ml of fresh medium was added. No additional medium was added until the total disappearance of other cell types. To produce subcultures, microglial cells were recovered by trypsin proteolysis using an EDTA (2 mM)trypsin (0.05%) solution and seeded onto no-coated Nunc 96-well plates at density of 20,000 cells per well for 24 h using the N5B medium (N5 medium + 5% HS + 0.5% FBS + 5 µM glucose) added with glycine 100 µM.
Primary microglial cell treatments. Primary microglial cells were treated with αS PFF or αS PFF :CMT-3 at 70 µg/ml for 24 h to study inflammation-induced cell response. The monomeric form of α-synuclein was administered at the same concentration. To confirm that a direct action of CMT-3 was not responsible for the possible reduction of inflammation in αS PFF :CMT-3-treated cells, 0.5 µM of this drug was added (the concentration of CMT-3 that would be in wells before dialysis) for 2 h before stimulation with αS PFF . α-Synuclein stock solutions used contained less than 0.1 endotoxin unit (EU)/mg protein and were prepared in 20 mM HEPES, 150 mM NaCl, pH 7.4. The residual endotoxins were quantified using the Limulus amebocyte lysate assay (Thermo Fisher #88282). The experiments were performed twice with n = 5-6/group for each experiment.
Molecular dynamics simulations of α-synuclein fibrils. The systems were built and equilibrated using the software package HTMD 78 . The coordinates for the α-synuclein fibril were taken from the solid-state NMR model deposited on PDB (PDB ID: 2N0A) 44 . To limit computational cost, and in accordance with previous simulations of the system 79,80 , we modelled the segment 27-100 containing the Greek motif of the hydrophobic core. The N and C terminal residues were capped with ACE and CT3. The ligand structure was taken from the ChemSpider database 81 (ChemSpider ID: 28530521) and distributed randomly in the simulation box around the fibril at a minimal distance of 3 Å. The system was solvated using a TIP3P water box and neutralized with 150 mM NaCl for a total of approximately 80 K atoms. Parameters for the protein, solvent and ions were taken from the CHARMm 36 forcefield 82 . Parameters for the ligand were adapted from the CHARMm forcefield for tetracyclines and its analogs developed by Aleksandrov and Simonson 83,84 . Twelve different systems were built and equilibrated independently. The systems were energy minimized for 1000 steps followed by 5 ns equilibration with a 10 kcal mol −1 Å −2 harmonical constraint on C-α and 5 ns equilibration without constraints under CPT conditions. Pressure and temperature were set at 1 atm and 300 K using a Berendsen thermostat and barostat, respectively. The systems were simulated under periodic boundary conditions with a 9 Å cutoff for nonbonded interactions and a Particle www.nature.com/scientificreports/ Mesh Ewald algorithm was applied for the treatment of long-range electrostatic interactions. The integration timestep was set to 4 fs using the hydrogen mass repartition scheme implemented in AceMD 85 . Production simulations were performed in the NVT ensemble using Amber18 program 86 with the target temperature controlled via Langevin dynamics 87 . Bonds involving hydrogen atoms were handled through the SHAKE algorithm 88 , allowing for an integration timestep of 2 fs. Each system was simulated for 600 ns and the trajectories fragmented into 100 ns segments. In order to improve sampling of the system, new 100 ns simulations were run from relevant frames identified in the MSM model, a technique known as respawning 89 . Model building and respawning simulations were performed until the Markov model was converged.
Simulation data analysis. For the analysis of the trajectories we used the pyEMMA package 90 as implemented in the HTMD software package which allows for the construction of a Markov state model (MSM) for the protein-ligand binding process. Conformationally relevant states of the model were built by projecting the simulation on a lower dimensional metric space such as the protein-ligand contact map. A contact is defined whenever a heavy atom of the ligand is within a minimum distance of 5 Å from the carbon alphas of the protein. The data was further projected into the 3 slowest order parameters using the time-lagged independent component analysis (TICA) with a 20 ns lag time and separated into 1000 cluster using the mini-batch k-means clustering algorithm 91 . The MSM was built using a time-lag of 25 ns and the microstates were lumped into 5 macrostates using the PCCA algorithm 92 . Error estimation was performed by conducting the analysis in several independent runs, where each run was bootstrapped leaving 20% of the dataset out randomly.
Statistical analyses. All data was obtained from at least three independent experiments and expressed as mean ± SEM. Multiple-group comparisons were performed with one-way ANOVA. Differences were considered as statistically significant at p < 0.05. Statistical analyses were carried out with GraphPad Prism 5 (San Diego, California, USA).