A unified in vitro to in vivo fluorescence lifetime screening platform yields amyloid β aggregation inhibitors

Inhibiting the aggregation of amyloid β (1-42) is a promising strategy for the development of disease-modifying Alzheimer’s disease therapeutics. To date, however, no sufficiently efficacious inhibitors have been identified, despite the best efforts of >200 advanced drug development campaigns. This failure can be attributed to limitations in current compound screening and in vivo validation assays. Here, we report an in vitro to in vivo screening platform based on the use of a fluorescence lifetime aggregation sensor. The microfluidic “nanoFLIM” assay developed circumvents issues that plague conventional assays, such as lack of reproducibility, high cost and artefactual false read-outs. The fluorescence lifetime sensor can also dynamically monitor peptide aggregation in cellular and Caenorhabditis elegans disease models, providing directly comparable aggregation kinetics, which is not achievable by any other method. The power of this unified system for accelerating hit-to-lead strategies, lowering attrition rates and expediting in vivo screening, was demonstrated with a pilot screening campaign of 445 compounds, revealing a new inhibitor that can inhibit amyloid β self-assembly in vitro as well as in cellular and whole organism disease models.


Introduction 28
Aggregation of the peptide amyloid β (1-42) (Aβ42) is a pathological hallmark of Alzheimer's disease (AD), 29 a multifaceted neurodegenerative disorder for which no preventative measures or disease modifying therapies 30 exist. 1, 2 The development of Aβ42 aggregation inhibitors has long been recognised as a potential strategy for 31 AD treatment, 3, 4, 5, 6 but to date only one inhibitory drug has been approved, subject to an unusual nine-year 32 post-approval trial. 7, 8 Indeed, more than 200 unsuccessful attempts to develop medicines in clinical trials to 33 treat and potentially prevent AD are on record, 9, 10 suggesting that finding therapeutic molecules in this area 34 is much harder than in other drug development campaigns. However, the number of drugs candidates in 35 clinical trials at any time remains above 100. 11 Setbacks in the development of such clinical candidates stem 36 from a variety of issues along the drug development pathway, including the types of assays historically used 37 and the difficulty of integrating biophysical and cell-based studies. 12 The intrinsically disordered nature of the 38 peptide and aggregate heterogeneity has limited structural studies on the peptide, thereby occluding structure-39 based design strategies. 13,14,15 Available compound libraries often lack novelty and scaffold diversity, 16 and 40 current screening approaches are limited by reagent consumption, assay reproducibility and spectral inference 41 from intrinsic properties of the test compounds. A major limitation in the development of AD therapeutics is 42 the absence of methods to test the activity of hit compounds in cellular or whole organism models quickly 43 and reliably: there are currently no high-throughput screening systems to directly monitor cellular Aβ42 44 aggregation in real time, or compare in vitro and cellular anti-aggregation activity. 17 Current techniques rely 45 on fixing cells to stain amyloid deposits or simple cell viability tests, neither of which are dynamic or revealing 46 about the underlying aggregation processes. Filtering and prioritising in vitro hits for those that show the 47 strongest in vivo activity is often too laborious to pursue extensive 'hit-to-lead' strategies, potentially resulting 48 in the advancement of lead compounds whose sub-optimal in vivo activity only becomes apparent in late stage 49 development. 18, 19, 20 50 Herein, we report an approach that addresses these issues by employing a unified assay format to cover in 51 vitro to in vivo compound screening for Aβ42 aggregation inhibitory activity, in order to better identify and 52 validate drug candidates, and to avoid wasting efforts on false positive results that have previously led to 53 failed drug discovery campaigns. This comprehensive assay is based on the use of an amyloid aggregation 54 fluorescence lifetime sensor, whereby Aβ42 aggregation is monitored by changes in the fluorescence lifetime 55 of an attached fluorophore, which is significantly quenched upon self-assembly of the peptide. 21,22 Changes 56 in the aggregation profile in the presence of a small molecule is indicative of a modulatory effect on Aβ42 57 self-assembly, be it to promote, delay or inhibit the process. 58 A medium-throughput microfluidic assay (dubbed the nanoFLIM) was designed to screen compound libraries 59 using fluorescence lifetime imaging microscopy (FLIM), with large sample sizes (>100 assays per 60 experiment) and nanolitre volume requirements (18 nL per test). To demonstrate the potential of the system, 61 a selection of novel chemical libraries -developed using diversity-oriented synthesis (DOS) or rationally 62 targeted drug discovery strategies -was interrogated, yielding MJ040, a novel lead Aβ42 aggregation 63 inhibitor. The fluorescence lifetime sensor protocol was also applied to dynamically screen compounds for 64 Aβ42 anti-aggregation activity in SH-SY5Y neuroblastoma cells, where MJ040 was again shown to display 65 an inhibitory effect. The activity of a rationally designed prodrug version, MJ040X, was further validated by 66 probing in vivo Aβ42 aggregation in the disease model Caenorhabditis elegans, in which treatment was seen 67 to delay or completely inhibit the aggregation process. 68 In addition to an unrivalled economy in reagent consumption, the nanoFLIM platform provides the only 69 unified direct comparison of Aβ42 aggregation propensity in the presence of small molecules in vitro, in live 70 cells and in a whole organism disease model. The successful identification of MJ040X as a novel lead Aβ42 71 aggregation inhibitor, capable of preventing amyloid formation in all three relevant formats, suggests that 72 promising candidates for the development of therapeutically active AD treatment readily result from this 73 approach, thereby providing a route to drug development for a hitherto elusive challenge. 74 75

Small molecule library screening at the nanoliter scale identifies novel inhibitory scaffold 77
To overcome problems that have plagued previous screening campaigns, namely assay variability, data 78 quality and reagent consumption, 12, 23 a novel workflow was designed in which Aβ42 aggregation was 79 monitored within arrayed nanoliter droplets using fluorescence lifetime imaging. 24, 25 The arrayed format 80 enabled the simultaneous imaging of 110 droplets, each corresponding to a distinct experiment. Figure 1a  81 shows an overview of the nanoFLIM assay platform, in which 110 traps sit along a serpentine channel in a 82 microfluidic chip. The droplets (18 nL) lodged in these traps were generated and deposited using droplet-on-83 demand technology, 26, 27 producing defined quantities of partially labelled monomeric peptide (845 pg per 84 droplet) and the respective drug candidate molecule (360 pg per droplet) in a precisely known order (Fig.  85   1a,b). Once droplets were sequentially trapped in the device, reaction progress was imaged over several hours, 86 to give characteristic aggregation profiles (Fig. 1c). 445  with 9 repeats of 9 compounds is shown in Figure 2a. Such screening efforts resulted in the identification of 92 hit compound MJ040 (Fig 2b), which originates from a cinchophen scaffold library and exerts a profound 93 inhibitory effect on Aβ42 aggregation compared to other library members (Fig. 2a). 28 The ability of MJ040 to 94 inhibit Aβ42 aggregation was quantified by nanoFLIM measurement of Aβ42 aggregation kinetics at several 95 concentrations of the compound, giving an IC50 value of 3.8 ± 0.8 μM (Fig. 2c,d). Notably, we multiplexed 96 aggregation assays to overcome batch to batch variability even shown for recombinant Aβ 29 by carrying out 97 repeats. The high number of repeats in this assay made it possible to obtain reliable aggregation data 98 immediately from one set of measurements, with incomparably low reagent consumption. The low 99 micromolar range of the IC50 value is similar to known in vitro inhibitory compounds, e.g. EGCG (IC50 6.4 ± 100 0.7 μM), 30 an inhibitor that has reached phase III clinical trials (NCT00951834) for its Aβ42 anti-aggregation 101 activity. Two compounds, MJ001 and MJ042 (Fig. 2a,b), showed weaker, yet significant inhibitory activity 102 and were carried forward together with MJ040 for further validation. 103 Dropix, Dolomite 26 ) droplets containing monomeric peptide together with test compounds are formed and 106 filled into a droplet array chip in a predefined order. The brightfield image insert shows the filled microfluidic 107 device, in which 110 droplets, with 18 nL of aqueous solution per droplet, are trapped within square grids 108 above a serpentine channel (Scale bar = 2 mm). Fluorescence lifetime imaging is used to monitor the 109 aggregation kinetics of the peptide, which is partially labelled with a reporter dye, and the effect that different 110 extrinsic factors or inhibitory compounds can have on the process. The fluorescence intensity of monomeric 111 or aggregated peptide droplets are similar. However, the fluorescence lifetime sensor provides a quantitative 112 measure of the aggregation state of the peptide, which are either monomeric (blue, lifetime ~3200 ps) or 113 aggregated (yellow, lifetime ~2700 ps). The ability of the system to monitor the process of Aβ42 aggregation 114 with different peptide labelling densities, under various pH, temperature, shearing conditions and with a 115  several hours for imaging. iv) Subsequent droplets flow under previously trapped droplets until v) they reach 120 an empty grid where they become trapped. c) Aβ42 aggregation profiles for 10 droplets formed from the same 121 10 μL stock solution of peptide. A great discrepancy between the initiation of aggregation in the first droplet 122 and the last is observed, highlighting the need for many repeats when monitoring self-assembly of this highly 123 aggregation prone peptide. 20 μM Aβ42-488, 20% labelled. Thus far, screening for Aβ42 aggregation inhibitors has mainly employed Thioflavin T (ThT) fluorescence 144 assays, whereby a red shift and enhancement in the fluorescent signal of the ThT dye upon binding to 145 structures rich in β-sheets is correlated with increasing protein aggregation. 32 The anti-aggregation activity 146 observed using the ThT fluorescence assay and the nanoFLIM was compared, and the data for four notable 147 compounds are shown in Figure 3. Both the nanoFLIM (Fig. 3a) and ThT (Fig. 3b) assays indicated that 148 MJ040 exerts a strong inhibitory effect, with calculated IC50 values of 4.3 ± 1.3 μM and 5.8 ± 1.7 μM, 149 respectively (Fig 2d). The mode of interaction of MJ040 was investigated by a combination of NMR 150 spectroscopy and molecular modelling, allowing us to locate the MJ040 binding site in a hydrophobic cleft 151 near the C-terminus of monomeric Aβ42 (Extended Data Figure 1). 152 With compound MJ036, conflicting results were obtained in the two assay formats. In the fluorescence 153 lifetime sensor measurements, no inhibitory activity was observed with MJ036 ( Fig. 3a), while the ThT 154 fluorescence assay identified MJ036 as a potent aggregation inhibitor (Fig. 3b). To resolve this apparent 155 inconsistency, morphological analysis of the aggregation products by transmission electron microscopy 156 (TEM) (Fig. 3e) and atomic force microscopy (AFM) (Supplementary Fig 7) was performed. These 157 orthogonal techniques rule out non-spectral and fluorescence interference, providing a method of assessing 158 fibril formation independent of the presence of extrinsic dyes. Many aggregated species were observed 159 following 7-day incubation with MJ036 ( Fig. 3c-iii), and only very few with MJ040 ( Fig. 3c-iv). This 160 suggests that MJ036 had been incorrectly assigned as a strong inhibitor by ThT fluorescence. Based on 161 nanoFLIM it can be correctly discounted as a false positive. Dot blot assays, with the toxic oligomer-specific 162 A11 antibody, were also used to investigate the formation of aggregated species Aβ42 in the presence of the 163 compounds (Fig. 3d). 33 A11-sensitive species were not detected in the MJ040 treated sample, suggesting that 164 the small structures observed by the scanning microscopy techniques were innocuous aggregates. In contrast, 165 A11-immunoreactive structures were detected in the MJ036 treated sample. Thus, MJ036 neither prevents 166 the formation of Aβ42 fibrils nor potentially toxic oligomeric Aβ42 species, and was therefore incorrectly 167 assigned as an inhibitor by ThT fluorescence analysis. 168 To further assess the inadequacy of the conventional ThT assay 34 for screening the cinchophen library, the 169 fluorescence emission spectra of ThT with previously formed Aβ42 fibrils in the presence of the compounds 170 was investigated (Fig. 3c). The addition of either MJ036 or MJ040 was shown to reduce the ThT emission 171 signal (with 440 nm excitation, as typically used in the real time assay), indicating that the compounds 172 interfere with the ThT assay readout as a result of their intrinsic fluorescence properties or competitive binding 173 interactions with the peptide or ThT dye itself. 35 As such, the inhibitory activity of these compounds cannot are also believed to contribute to incorrect assignments in the ThT fluorescence assay, in this case false 178 negative results. When added to the ThT-Aβ42 fibril sample, the emission at 488 nm (the emission wavelength 179 used in the conventional ThT assay) is higher than that of the Aβ42 fibrils and dye alone, thereby potentially 180 masking the compounds inhibitory activity against Aβ42 aggregation (Fig. 3b,d), which may explain why they 181 will not be found in a screen based on ThT fluorescence. The inhibitory activity of MJ001 and MJ042 was 182 validated using TEM, where in both cases only small aggregates were observed following compound 183 treatment ( Fig. 3c-ii,-iv). NanoFLIM screening, therefore, is less susceptible to misleading readouts that 184 produce false positive and negative results than the conventional ThT assay when screening spectroscopically 185 active small molecule libraries. (~33%, measured as a function of impaired cellular metabolic activity), and the addition of MJ040 to the 222 extracellular medium was not capable of significantly rescuing the cells (Fig. 4a). In a bid to improve cellular 223 activity, the carboxylic acid of the compounds was masked as a methyl ester, to generate the prodrug MJ040X 224 (Fig. 4b, Supplementary Synthetic Procedures). This compound displayed poor inhibitory activity in vitro 225 ( Supplementary Fig. 12), suggesting the carboxylate is necessary for interaction with the peptide. In the cell-226 based assays, however, the compound resulted in a significant increase in cell vitality (Fig. 4a). The rescuing 227 effect suggests that the lack of activity observed for the original compound MJ040 is caused, at least in part, 228 by poor cellular uptake as a consequence of the anionic group. The rationally designed prodrug MJ040X 229 successfully permeates into the cells, where the ester is then hydrolysed to generate the free active drug, which 230 inhibits the aggregation process in the cellular environment. 231 There is a severe shortage of methods to monitor Aβ42 self-assembly in cellular models in real time. Amyloid 247 aggregates in fixed cells can be imaged with immunological staining, which suffers cross reactivity issues 248 with the amyloid precursor protein and its derivatives, 38, 39 or with the use of amyloid sensitive dyes such as 249 ThT and Congo Red. These effectively stain fixed aggregates, but are generally unsuitable in dynamic live 250 cell studies, as they cannot detect small aggregates low in β sheet content and can induce a mild inhibitory 251 effect on the aggregation process. after 12 h (Fig. 5a). 25 To test if the system could be used for small molecule inhibitor screening, the previously 267 reported inhibitor EGCG was employed. 45 It was found that addition of EGCG to the extracellular medium at 268 the same time as Aβ42 addition had little inhibitory effect on intracellular peptide aggregation, with both 269 conditions reaching the same fluorescence lifetime value (Supplementary Fig. 15). However, pre-incubating 270 the cells in a drug solution for one hour prior to the addition of the peptide, was shown to significantly inhibit 271 the aggregation in the cells, as measured at 12, 24 and 48 h (Fig. 5a). The measurable change in the 272 fluorescence lifetime observed here is directly comparable to that monitored in the in vitro format, permitting 273 a comparative analysis of compound activity in both formats, which is not possible by any other method. This 274 protocol was validated by testing a range of other known inhibitory small molecules, which were each seen 275 to inhibit the aggregation to different degrees (Supplementary Fig. 16). Treatment with MJ040 reduced the 276 extent of Aβ42 aggregation relative to that of peptide alone, and the modified MJ040X provided an even 277 stronger inhibitory effect (Fig. 5b,d). This supports the working idea that MJ040 displays limited 278 permeability as a result of its anionic centre, and that masking this functionality confers more desirable 279 pharmacokinetic properties. In order to show that the aggregation inhibition did not occur during the pre-280 incubation of the drug with Aβ42 prior to their uptake into cells but indeed intracellularly, we also tested 281

Hit compound MJ040X inhibits Aβ42 aggregation in whole organism disease model 307
Whole organism studies using a C. elegans disease model were carried out to demonstrate the ability of our 308 unified fluorescence lifetime sensor assay to report on aggregation in matched in vitro and in vivo formats.  with a statistically significant reduction observed at day 12 (Fig. 6a,b,d). It was found that treatment with the 320 MJ040X from adulthood delayed this process. A statistically significant difference in fluorescence lifetime 321 between the treated and untreated control was observed at day 12 of adulthood, but by day 15 peptide 322 aggregation was evident in the treated worms also (Fig. 6d), suggesting a narrow, but significant therapeutic 323 window. Treatment from larval stage, however, prevented Aβ42 aggregation until day 15, the last day of 324 measurement for adult worms (Fig. 6e). each. Data were analysed by a two-way Anova and a Bonferroni post-test (see Supplementary Table S4). e) 335 Bar diagram displaying a comparison of the mean fluorescence lifetime values observed when the C. elegans 336 were treated from day 1 of the larval stage. Drug treatment from the larval stage prevents aggregation until 337 day 15, based on two biological repeats with 8-18 worms analysed per repeat. All data are reported as mean 338 fluorescence lifetime + SEM and the statistical analysis was performed using a one-way ANOVA with Sidak's 339 multiple comparison test (see Supplementary Table S5). A typical time trace and exponential decay fits for 340 the data can be found in Supplementary Figure 17. 341

Implications and Conclusions 343
Recent failures in clinical trials suggest that current aggregation screening strategies are limited in their ability 344 to provide therapeutically viable hit compounds. 12, 19, 51 Limitations stem from issues in reproducibility 345 (partially due to peptide quality and solubility in a heterogeneous assay), 52 fluorescence interference caused 346 by intrinsic properties of the test compounds, 35 and inability to efficiently validate and prioritise hit 347 compounds in vivo. There is currently no single method that permits protein aggregation and its inhibition as  To illustrate the potential of this system, a pilot screening campaign was performed with 445 compounds 395 from medicinally-relevant chemical libraries, yielding a total hit rate of 13% (>30% Aβ42 aggregation 396 inhibition). The lead inhibitor identified, MJ040, and rationally designed prodrug MJ040X, were shown to 397 exert strong inhibitory effects in vitro, in live cells and in disease model C. elegans, emphasising that 398 biologically active inhibitors can be identified through this comprehensive assay platform. We also believe 399 that it could be easily adapted to screen for aggregation inhibitors of other amyloidogenic proteins, including 400 functional bacterial amyloids, thereby holding potential for the identification of hit compounds for the 401 treatment of a range of amyloid disorders and bacterial pathogenesis. 59 402 The breadth of the fluorescence lifetime screening platform and the potential high-throughput afforded by its 403 use, will expedite the rate at which hits are identified, validated in vivo and prioritised for future hit 404 development strategies, with the attrition rate in moving through these stages minimised by the unified 405 analysis. The efficiency afforded by this approach has already yielded a lead in MJ040X, but the high hit

52.
Hellstrand E, Boland B, Walsh DM, Linse S. Amyloid beta-protein aggregation produces highly 543 reproducible kinetic data and occurs by a two-phase process. was dissolved in ice cold trifluroacetic acid (200 mL), sonicated at 0 °C for 60 s, then lyophilised overnight. 574 Ice cold 1,1,1,3,3,3-hexafluro-2-propanol (1 mL) was added, sonicated at 0 °C for 60 s and aliquoted into 20 575 µL portions. The samples were lyophilised overnight and were stored at -80 °C until use. The concentration 576 of the aliquots was determined using amino acid mass spectrometry analysis. The required concentration of 577 unlabelled Aβ42 was prepared by dissolving the solution in dimethyl sulfoxide (DMSO) (5% of total solvent 578 volume), then adding sodium phosphate buffer (NaPi, 50 mM, pH 7.4). Prior to use, the solution was sonicated 579 at 0 °C for 3 min, centrifuged at 13,400 rpm at 0 °C for 30 min to remove preformed aggregates. Lyophilised 580 Aβ42 Hilyte™ Fluor 488 peptide (0.1 mg) was dissolved in 1% NH4OH (200 µL) and sonicated for 60 s at 0 581 °C. The sample was aliquoted into 5 µL units, snap frozen in liquid N2, then stored at -80 °C. Before use, the 582 sample was thawed on ice and NaPi buffer was added to bring the solution to the required concentration. 583 For studies with partially labelled peptide, each peptide was prepared as above then mixed at the appropriate 584 ratios before each set of experiments. This was aliquoted into small units, then snap frozen and stored at -80 585 °C until use. 586 587 Microfluidic Device Fabrication. The silicon master mould was fabricated by MicroLiquid (Gipuzkoa, 588 Spain) using a two layer soft lithographic technique, as previously described. 60, 61 The depth of the first layer 589 with a serpentine channel was 175 μm, while the depth of the square traps were 250 μm. PDMS replicates of 590 these devices were bonded to a thin glass coverslip (thickness 130 µm) using oxygen plasma. The devices 591 were silanized by pipetting a fresh solution of 2% Trichloro(1H,1H,2H,2H-perfluorooctyl)silane in HFE-592 7500 (3M) into the chips. Subsequently, PTFE tubing (diameter: 200 μm) was manually inserted in designed 593 entrance channels and glued in place by curing PDMS over. This ensured air-tight connection as well as 594 preserving the order of the droplets as they transited from tubing to chip. One tubing was then connected to a 595 syringe pump (Chemyx Fusion 200) operating in withdrawal mode, while the other tubing was inserted and 596 clamped into the stainless steel hook of a Mitos Dropix. A gas-tight glass syringe (100 μL) was used to fill 597 the device with HFE-7500 oil containing 1% Pico-Surf surfactant (Dolomite Microfluidics) and the chip was 598 inspected to confirm the absence of air bubbles. Next, 10 μL of each compound was pipetted into the loading 599 strip of the Dropix. The droplet sequence was programmed to obtain 18 nL droplets with 36 nL oil spacing 600 between each drop. Typical flow rate for producing the droplets was 2 μL/min. Before reaching the device, 601 the droplets were slowed down to 1 μL/min and the filling process was monitored with a bright-field camera 602 of an inverted microscope (Olympus, IX71). After completion of the filling process, the flow was stopped, 603 unless specified in shearing experiments.   Figure S17). Briefly, a supercontinuum source (SC390, Fianium) operating at a 40 617 MHz repetition rate was used for excitation. The excitation light was filtered using an acousto-optic tunable 618 filter (AOTFnC-400.650, QuantaTech) centered at 480 nm to excite GFP and AF488 or at 585nm to excite 619 mCherry. Fluorescence emission from the sample passed through a band-pass filter (FF01-525/39-25 or 620 FF01-624/40-25, Semrock) before reaching the detector (PMC-100, Becker & Hickl GmbH). The photon 621 detection rate for each pixel was kept below 1% of the laser repetition rate in order to avoid photon pile-up. 622 Air objectives (PlanApo 2x and 40x, Olympus) were used for imaging the microfluidic chip and C. elegans, 623 respectively. An oil objective (PlanApo 60x BFP1 C2, Olympus) was used to image cells. All FLIM data 624 were analysed using either commercial software SPCImage (Becker & Hickle GmbH) or open-source FLIM-625 fit software. 62 A biexponential fit was used for nanoFLIM and for FLIM data from AF488 containing cells. 626 A single exponential fit was used for FLIM data from C elegans data and from mCherry containing cells 627 (Extended Data Figure 2). The phasor plot analysis 62 feature in SPCImage was used to validate the use of a 628 biexponential fit on certain FLIM data (Extended Data Figure 4). 629 630 AFM. A freshly cleaved mica surface was prepared by sequential treatment with potassium hydroxide and 631 0.1% poly-lysine solution. After the specified time of incubation (20 μM Aβ42, 100 μM compound), samples 632 were transferred directly to the slides and were allowed to dry for 30mins. Samples were then rinsed with 633 Milli-Q water and dried in the air. AFM images were acquired on a commercial system (Bioscope 634 RESOLVE, Bruker) and Nanoscope software (Bruker). The instrument was operated in tapping mode in air.