Image‐Based Morphological Profiling Identifies a Lysosomotropic, Iron‐Sequestering Autophagy Inhibitor

Abstract Chemical proteomics is widely applied in small‐molecule target identification. However, in general it does not identify non‐protein small‐molecule targets, and thus, alternative methods for target identification are in high demand. We report the discovery of the autophagy inhibitor autoquin and the identification of its molecular mode of action using image‐based morphological profiling in the cell painting assay. A compound‐induced fingerprint representing changes in 579 cellular parameters revealed that autoquin accumulates in lysosomes and inhibits their fusion with autophagosomes. In addition, autoquin sequesters Fe2+ in lysosomes, resulting in an increase of lysosomal reactive oxygen species and ultimately cell death. Such a mechanism of action would have been challenging to unravel by current methods. This work demonstrates the potential of the cell painting assay to deconvolute modes of action of small molecules, warranting wider application in chemical biology.


Introduction
Thei dentification of small molecules to probe biological systems is at the heart of chemical biology.T arget-agnostic phenotypic screens represent arapid way to identify bioactive small molecules in physiologically relevant systems. [1] However,am ajor challenge with this approach is the subsequent elucidation of molecular modes of action (MMOA) and target identification (ID) of bioactive compounds. [2] Widely employed target ID techniques include affinity-based pull-down using immobilised derivatives followed by mass spectrometric protein identification. [3] This typically requires the synthesis of as uitably functionalised probe,w hich might be time consuming or even infeasible if the compound of interest is highly complex. Emerging target identification strategies, which do not rely on modifications of the hit compound, such as thermal proteome profiling,a re powerful additions to the target ID toolkit. [4] However,t hese techniques are restricted to small molecules that mediate their activity through the targeting of proteins.I nc ontrast, various drug classes target DNA, [5] RNA, [6] and lipids, [7] and the discovery of regulatory RNA-targeting small molecules has recently emerged as an ew field. [8] Therefore,t he development of new methods, which enable the delineation of bioactive-small-molecule modes of action not mediated by binding to aprotein target, is in high demand.
Morphological profiling has recently emerged as ac omplementary strategy for small-molecule-target identification. Monitoring changes in cellular morphology induced by ah it molecule and comparing these to changes induced by aset of reference compounds with known modes of action and targets can provide target hypotheses.Morphological profiles can be extracted from simple brightfield images, [9] and obtained from complex fluorescence-based high-content screens in which multiple subcellular compartments are labelled with various fluorophores. [10] Them ultiplexed use of different fluorophores has been established in the "cell painting" assay, [10,11] and has been proposed as an ew strategy for determining whether ac ompound displays bioactivity in av ery broad setting. [12] In light of the promise and the potential of this approach, we explored the use of the cell painting assay for MMOAi dentification where other target identification methods had failed. [13] Recently we identified oxautin-1 (1), acinchona alkaloidderived autophagy inhibitor containing an oxazatwistane scaffold (Scheme 1a). [14] Autophagy is ac ellular recycling process that degrades misfolded, aggregated, and/or superfluous proteins and organelles.The inhibition of autophagy is considered ap otential anti-cancer strategy [15] making the identification of new small-molecule autophagy inhibitors and their targets an intensive area of ongoing research. [14,16] Oxautin-1 was predicted to inhibit both autophagosome biogenesis,and the fusion of autophagosomes and lysosomes, but its MMOAh ad remained elusive.G iven this unknown mode of action, we embarked on the synthesis and biological investigation of more readily accessible and structurally more diverse oxautin analogues.Wenow report the discovery of the cinchona-alkaloid-derived autophagy inhibitor autoquin (2, Scheme 1a). Analysis of morphological changes induced by autoquin in the cell painting assay unraveled that, like oxautin-1, autoquin inhibits autophagy by indirect modulation of the activity of the lysosomal enzymes acid sphingomyelinase and acid ceramidase,r esulting in impaired lysosome-autophagosome fusion. Deeper investigation revealed that autoquin also sequesters Fe 2+ in lysosomes,which results in increased formation of lysosomal reactive oxygen species (ROS) and cell death.

Results and Discussion
Theo xautins were synthesised by intramolecular cyclisation of quinidine (3)a nd cinchonine (4)b ut respective analogues cannot be obtained from the cinchona alkaloids quinine (5)a nd cinchonidine (6)o wing to their different configuration at the carbon atom a to the quinuclidine nitrogen atom, precluding cyclisation and limiting further exploration of oxazatwistane analogues as autophagy inhibitors (Scheme 1a). [14] In addition, the cyclisation to yield the oxazatwistane ring system in the oxautins requires somewhat harsh conditions and removes two vectors for further functionalisation, thereby further limiting exploration of SAR. Therefore,w ei nvestigated whether the oxazatwistane core was required for autophagy inhibitory activity.T ot his end, the four major cinchona alkaloids quinidine, quinine,c inchonine,a nd cinchonidine were subjected to Borono-Minisci conditions [17] to selectively functionalise the C2 position and to evaluate the importance of the relative stereochemistry at the quinuclidine ring (Scheme 1b). In addition to the expected C2-functionalised derivatives,s ome reactions also delivered am inor product corresponding to the C5-(compounds 9, from quinidine) or C7-functionalised scaffold (10, from cinchonine,a nd 13,f rom cinchonidine). In addition, functionalisation of the C3 position had not previously been investigated for the oxazatwistanes.Selective halogenation at C3 [18] provided intermediates that could be subjected to metalcatalysed cross coupling reactions (Scheme 1c, [14][15]. Suzuki reactions enabled the synthesis of 13 additional analogues (16 a-l).
Theresulting 49-membered compound collection was investigated for autophagy inhibition in MCF7 cells stably expressing EGFP-tagged LC3, aw idely used autophagy marker. [19] Compounds that were able to reduce EGFP-LC3 puncta formation upon autophagy induction by amino acid starvation using EarlesB alanced Salt Solution (EBSS) were classed as hits. [20] Thed irect oxautin-1 [14] analogue 2 derived from quinidine but lacking the oxazatwistane ring, displayed very similar potencyinthe autophagy assay,suggesting that the oxazatwistane ring was not essential for biological activity (Table 1, Entry 1). This compound, which we named autoquin, provided ab enchmark against which all other compounds were assessed. Small variations at the C2 position Scheme 1. Synthesis of acinchona alkaloid-derived compound library.a)Molecular structures of previously identified autophagyi nhibitor oxautin-1, newly discovered inhibitor autoquin, and the four most abundant cinchona alkaloids. b) Synthesis of C2-functionalised derivatives using the Borono-Minisci reaction. c) Synthesis of C3functionalised derivatives using selective CÀHa ctivation followed by Suzuki coupling. See Table 1f or details of the R-groups investigated. resulted in amodest drop in activity (Table 1, Entries 2and 3), while removal of the p-F substituent completely abolished it (Entry 4). Va rying the position of substitution on the phenyl ring reduced or abolished activity (Entries 5-7). Compounds with aryl substituents at C5 of the quinoline ring were generally inactive (Entries 8, 9, and 11) though a m-Cl substituent returned some activity (Entry 10). In general, cinchonine derivatives,lacking the C6 methoxy group (Entries 12-21), were all less active than the quinidine-derived compounds,c onfirming the importance of this residue for optimal activity.The p-F-C 6 H 4 substituent at C2 retained the highest levels of activity (Entry 12), observed for all the quinidine-derived compounds (Entry 1) and the oxazatwistanes. [14] To assess the importance of the stereochemistry at the quinuclidine ring, as mall collection of C2-substituted derivatives of quinine and cinchonidine was synthesised and evaluated (Entries 22-36). Although three quinine-derived analogues showed appreciable levels of activity,they were all at least 6-fold less active than autoquin and were not pursued further.C rucially,t he p-F-Ph substituted quinine and cinchonidine analogues (Entries 22 and 29) were significantly less active than their quinidine and cinchonine-derived diastereomers (Entries 1and 12). All compounds with substituents at the C3 position on the quinoline ring were either less active than autoquin (Entries 39 and 40) or completely inactive (Entries 37, 38, and 41-49).
Having established that the oxazatwistane scaffold was not essential for autophagy inhibition and that the p-F-C 6 H 4 substituent at C2 of the quinoline ring was best for autophagy inhibition, we proceeded to validate autoquin as an autophagy inhibitor.A sd escribed above, autoquin showed ad ose-dependent inhibition of EGFP-LC3 puncta after 3hours upon autophagy induction by amino acid starvation in the primary screening assay (Figure 1a,b). Additionally,itincreased the stability of the chaperone p62 to autophagosome-mediated degradation as assessed by western blot (Figure 1c), suggesting that it is an inhibitor of autophagic flux. However, autoquin also showed ad ose-dependent increase in LC3-II levels,suggesting that it is also an inhibitor Table 1: Structure activity relationships of cinchona alkaloid-derived autophagy inhibitors. IC 50 data represents the ability to inhibit autophagy induced by amino acid starvation using EBSS and is mean AE SD of three independente xperiments. 2 = Autoquin. of autophagosome maturation, similarly to oxautin-1 (Figure 1c). This finding was further strengthened by using at andem mCherry-EGFP-LC3 expressing cell line,w hich enables the simultaneous monitoring of autophagosomes (green and red fluorescence) and autolysosomes (red fluorescence only). Exposing fed cells to autoquin for 24 hours resulted in amarked increase in autophagosomes,confirming its inhibitory effect on autophagosome maturation (Figure 1d).
Having validated autoquin as ab ona fide autophagy inhibitor and confirmed its effect on autophagosome maturation, we focused our attention on making smaller, more targeted modifications to the autoquin scaffold to identify as uitable immobilisation point for affinity-based target enrichment. Reduction of the alkene resulted in two derivatives with retained biological activity (17 a,b), suggesting that this may be as uitable position for further functionalisation (Scheme 2a and Figure 1c). C5-substituted analogues (18 a,b)w ere less active,a so bserved with the unsaturated analogues (Scheme 2a and Table 1). Arylation of the vinyl group via Heck coupling produced three derivatives (19 a-c)t hat, though less active than autoquin, retained good potencyl evels.O xidation of the hydroxy group to aketone led to an inactive compound (20)w hile,i nterestingly,methylation of the hydroxy group produced av ery active compound (21), suggesting that this position is critical for potency.T oa ccess ad erivative suitable for immobilisation and affinity-based target enrichment, autoquin was subjected to an ene reaction with 2-(Boc-amino)ethanethiol to yield ac ompound (23), which was further elaborated into ap ull-down probe (25, Scheme 2b). Ac orresponding negative probe (24) lacking the p-fluorophenyl group was also synthesized from 22.W hile the negative probe and its precursors were all inactive,the final positive probe 25 was as well. Thep henomenon by which the introduction of an unprotected amino-PEG linker leads to loss of activity has been observed by us on several occasions and can be ascribed to ap resumable loss of cell permeability.Inthis case,the presence of two basic amines makes it more likely that ad oubly charged molecule would interact with the cell membrane.A s the pull-down experiment was carried out in cell lysates and the intermediate 23 retained appreciable levels of activity,wec ontinued with 25 as apositive probe.
To identify autoquin target proteins by affinity enrichment, the probes were immobilised on NHS-activated magnetic beads and incubated with cell lysate.Proteins selectively enriched with the autoquin-derived probe compared to the negative probe were considered to be hits.T he only hit that was conclusively identified in all replicates was ferrochelatase (Supplementary Figure 1). Ferrochelatase (FECH) is located in the mitochondria and catalyses the insertion of iron into protoporphyrin IX, the last step in heme biosynthesis. [21] No known link between FECH and autophagy has previously been reported. However,a utophagy has been shown to play ar ole in mediating iron homeostasis through the selective mm. e) Autophagosomes (AP;y ellow puncta) and autolysosomes (AL;red puncta) from (d) were quantified and data represented as percentageo fcell area. Bar graphs show mean AE SD from three biologically independent experiments. Data points represent individual cells pooled from the three independent experiments (n ! 23 cells per replicate).S ignificance was determined from biologicalreplicates using atwo-tailed, unpaired ttest. ns = not significant, **p = 0.0064.

Angewandte Chemie
Research Articles 5724 www.angewandte.org degradation of ferritin. [22] In an attempt to validate FECH as at arget of autoquin, the pull-down was repeated with nonimmobilised autoquin as ac ompetitor and analysed by western blot. However,d espite FECH enrichment by the positive probe (25)c ompared to the negative probe (24), no competition was observed (Supporting Information, Figure S2 a). Additionally,a utoquin was not able to stabilise FECH to thermal denaturation as assessed by isothermal dose-response fingerprinting,e ven at very high concentrations (Figures S2 b,c). Thus,FECH was not considered adirect functional target of autoquin.
Since chemical proteomics had not validated at arget, autoquin was investigated in the multiparametric imagebased cell painting assay,w hich enables interrogation of ac ellular system in its entirety. [10][11][12]13] Cell painting involves staining cells with markers for specific cellular compartments following compound treatment. Av ast array of phenotypes including cell shape,m orphology,s ize,a nd fluorescence intensity and distribution can be assessed simultaneously for each treatment condition, generating as et of fingerprints characteristic to ap articular mode of action or target. [10][11]23] Thep ower of this assay becomes apparent when al ibrary of reference compounds of known bioactivity is included in the screen and their fingerprints are compared with profiles recorded for novel compounds.I np rinciple,t his can enable the discovery of bioactivity profiles that are different, and thus novel, compared to ar eference set, but also known modes of action can be revealed by similarity assessment. [13] Autoquin, oxautin-1, and several analogues were characterised in the cell painting assay,i nw hich their effect on 579 parameters (see the Supporting Information for the delineation of the parameters) was compared with the results obtained for ar eference compound set comprising 3000 compounds with known bioactivity (see the Supporting Information for details). To assess the similarity in the bioactivity of the fingerprint profiles, "biological similarity" was employed (BioSim; see the Supporting Information for determination of similarity). Furthermore an "induction" value (the fraction of parameters (in %) that underwent significant changes (median absolute deviation (MAD) value upon compound treatment of at least + /À three-fold of the median determined for the DMSO controls;s ee the Supporting Information)) was determined as measure for compound bioactivity.Compounds with an induction value of > 10 %w ere considered bioactive in the cell painting assay.T his analysis resulted in the discovery of three annotated compounds with high similarity (> 80 %) in their bioactivity fingerprints to autoquin (Figure 2a and Figure S3). Although at first glance,p erphenazine (reportedly an onselective G-protein-coupled receptor ligand), loperamide (an opioid receptor agonist), and toremifene (an estrogen receptor ligand) do not display obvious biological or indeed chemical similarity,a ll three have been reported to be lysosomotropic compounds. [24] Lysosomotropic compounds are typically hydrophobic with at least one basic nitrogen atom that, upon protonation, enables them to be trapped in the lysosomes. [25] Ac omparison of their physicochemical properties revealed that all compounds are likely to be fully protonated at pH 4-5, typically found in the lysosome (Supporting Information, Table S1). This would favour am odel in which they are able to pass cellular and lysosomal membranes before being protonated and trapped in the lysosome.Apotential lysosomotropic profile had previously been suggested in the cell painting assay for structurally different compounds,b ut was not further investigated. [11a] To confirm whether autoquin and oxautin are indeed lysosomotropic,weassessed their ability to inhibit the accumulation of the lysosomal tracer Lysotracker Red (LR) DND-99. Ad ecrease in fluorescence intensity is often characteristic of alysosomotropic phenotype,ifthe lysosomal pH is increased. Both autoquin and oxautin-1 showed adosedependent decrease in lysosomal accumulation of LR after a3hour treatment, similarly to the known lysosomotropes chloroquine and chlorpromazine (Figure 2b,c). Lysosomotropic compounds are also often functional inhibitors of acid sphingomyelinase (FIASMAs), and other sphingolipid hydrolases including acid ceramidase. [26] They do not directly interact with the hydrolases (therefore,t hey are qualified as functional inhibitors) but rather affect the inner lysosomal membrane localisation of acid sphingomyelinase and other sphingolipid hydrolases through ad irect interaction with the negatively charged lipid bis(monoacylglycerol)phosphate (BMP), resulting in the degradation of the hydrolases. [27] Both autoquin and oxautin-1 were tested in af luorescencebased assay to monitor both acid sphingomyelinase and acid ceramidase activity in ac ell-based and ac ell-free system. [28] FIASMAs characteristically inhibit hydrolase activity in cellbased assays,inwhich intact lysosomes are present, but not in cell-free systems.Both autoquin and oxautin-1 inhibited acid sphingomyelinase and ceramidase activity in intact cells (Figure 2d)b ut not in lysates (Figure 2e), similarly to the control compound desipramine.
In light of these findings,were-evaluated the affinity pulldown data and also took into consideration the recently reported biological activity of the natural product salinomycin, which sequesters iron to the lysosomes,i nhibiting autophagy and causing ferroptosis. [29] As imilar phenotype has also been reported for the lipophilic iron chelators di-2pyridylketone 4,4-dimethyl-3-thiosemicarbazone (Dp44mT) and, to al ower extent, desferrioxamine (DFO). [30] We speculated that the ability of autoquin to pull down FECH might have been iron-dependent and not specific to FECH. This hypothesis was strengthened by reports that quinine and related cinchona alkaloids are able to chelate iron, contributing to their anti-malarial effect. [31] To assess whether autoquin displayed asimilar mechanism of action to salinomycin, we evaluated its impact on lysosomal mass, lysosomal Fe 2+ ,a nd ROS production. Autoquin significantly increased lysosomal mass after 24 hours,a s assessed by Lysotracker deep red (DR) staining (Figure 3a,b). Ad ecrease in lysotracker staining at early time points (< 4hours) owing to increased pH is generally observed for lysosomotropic compounds including autoquin (Figure 2b,c);h owever,t his effect is reversed at later time points (> 24 hours) as cells adapt to protect themselves from lysosomal stress. [32] Autoquin also significantly enhanced lysosomal Fe 2+ levels (Figure 3a,c), as assessed by the turn-on fluorescent probe RhoNox-M, and overall Fe 2+ levels,a s assessed by RhoNox-1 fluorescence ( Figure 3d). The concomitant increase in lysosomal mass and Fe 2+ levels also resulted in ah ighly significant increase of lysosomal reactive oxygen species (ROS), as assessed by CellRox DR (Figure 3e and Figure S4 a,b). Similarly to salinomycin, autoquin also induced lipid peroxidation, as assessed by aBODIPY 581/591 undecanoic acid (C11) probe [33] (Figure S4 d) and caused lipid membrane permeabilization (Figure S4 e). As salinomycin had shown promising results in the selective targeting of breast cancer stem cells, [29,34] atumorigenic cell subpopulation typically associated with resistance to chemotherapy and sustained tumor growth, it was speculated that autoquin may display as imilar profile and potential. Autoquin was selectively cytotoxic against transformed human mammary epithelial HMLER CD44 high /CD24 low cells (HMLER CD24 low ), an established model of human breast cancer stem cells, Figure 2. Autoquin is alysosomotropic compound that acts as af unctional inhibitor of acid sphingomyelinase. a) Cell painting profiles of autoquin and its most biosimilar compounds oxautin-1, perphenazine, loperamide, and toremifene. See Figure S3 for representative images. b) Representative fluorescence microscopyi mages of MC7 cells treated with lysosomotropic compounds for 3hand stained with Lysotracker DND-99;scale bar = 110 mm. c) Quantification of (b), n = 3, data is mean AE SD. d) Product/substrate ratio of the acid sphingomyelinase (left) and acid ceramidase (right) reaction in intact cells. n = 3, data is mean AE SD. e) Product/substrate ratio of the acid ceramidase reaction in cell lysates. n = 3, data is mean AE SD. Statistical significance comparingtreated samples to the DMSO control for (d) and (e) was assessed using the Student's t-test. *p < 0.05, **p < 0.01, ***p < 0.001. compared to ac ontrol isogenic cell line (HMLER CD24 high ) ( Figure S4 f). As such, autoquin represents an ew lysosomotropic compound that selectively targets breast cancer stem cells,and may hold promise in the study of anti-cancer agents that target the lysosomes, [35] af ield that has been growing steadily in recent years,with multiple clinical trials focused on the approved anti-malarials chloroquine and hydroxychloroquine.

Conclusion
We have employed the cell painting assay to identify the MMOAo fa utoquin, ac inchona-alkaloid derived autophagy inhibitor.While affinity-based proteomic experiments proved inconclusive,i mage-based profiling identified similarities in phenotypic profiles determined for autoquin and known lysosomotropic compounds.T his MMOAd oes not require adirect binding event between autoquin and atarget protein, which explains why MS-based proteomic approaches were inconclusive.However,pull-down experiments suggested that perturbation of iron homeostasis may contribute to the activity of autoquin, which was confirmed using markers for lysosomal ROSa nd lipid peroxidation. This work showcases image-based profiling as an excellent complementary tool for mode-of-action and target identification. Given that the lack of identification strategies for non-protein targets has been amajor drawback of MS-based approaches,weenvisage and indeed encourage the chemical biology community to embrace the cell painting assay as an additional, alternative technique for the discovery of modes of action and novel bioactivity.C rucially,t his technique obviates the need for functionalisation of the active molecule,m aking it particularly suitable for the identifying the MMOAofcomplex, NPlike compounds.T ofurther improve cell painting and increase its adoption, significantly larger libraries of reference compounds will be required. Although truly novel modes of action will still require proteomic experiments including pulldowns and/or thermal proteome profiling to identify the molecular targets underlying an observed phenotype,the cell painting assay is astrong addition to the chemical biologists tool-kit for tackling the challenge of target identification.