Research Article

Biochemistry and Chemical Biology

SDR enzymes oxidize specific lipidic alkynylcarbinols into cytotoxic protein-reactive species

Institut de Pharmacologie et de Biologie Structurale, IPBS, CNRS, Université de Toulouse, France
Leiden Institute of Chemistry, Leiden University, Netherlands
Department of Chemistry, Technical University of Munich, Germany
SPCMIB, UMR5068, CNRS, Université de Toulouse, UPS, France
LCC-CNRS, Université de Toulouse, CNRS, UPS, France
INRAE, UR 875 Unité de Mathématique et Informatique Appliquées, Genotoul Bioinfo Auzeville, France
Equipe labellisée la Ligue contre le Cancer 2018, France
Department of Organic and Inorganic Chemistry, Science Center, Federal University of Ceará, Brazil
INRAE, US 1426 GeT-PlaGe, F-31326, France

May 10, 2022

Open access
Copyright information

Abstract
Editor's evaluation
Introduction
Results
Discussion
Materials and methods
Data availability
References
Article and author information
Metrics

Abstract

Hundreds of cytotoxic natural or synthetic lipidic compounds contain chiral alkynylcarbinol motifs, but the mechanism of action of those potential therapeutic agents remains unknown. Using a genetic screen in haploid human cells, we discovered that the enantiospecific cytotoxicity of numerous terminal alkynylcarbinols, including the highly cytotoxic dialkynylcarbinols, involves a bioactivation by HSD17B11, a short-chain dehydrogenase/reductase (SDR) known to oxidize the C-17 carbinol center of androstan-3-alpha,17-beta-diol to the corresponding ketone. A similar oxidation of dialkynylcarbinols generates dialkynylketones, that we characterize as highly protein-reactive electrophiles. We established that, once bioactivated in cells, the dialkynylcarbinols covalently modify several proteins involved in protein-quality control mechanisms, resulting in their lipoxidation on cysteines and lysines through Michael addition. For some proteins, this triggers their association to cellular membranes and results in endoplasmic reticulum stress, unfolded protein response activation, ubiquitin-proteasome system inhibition and cell death by apoptosis. Finally, as a proof-of-concept, we show that generic lipidic alkynylcarbinols can be devised to be bioactivated by other SDRs, including human RDH11 and HPGD/15-PGDH. Given that the SDR superfamily is one of the largest and most ubiquitous, this unique cytotoxic mechanism-of-action could be widely exploited to treat diseases, in particular cancer, through the design of tailored prodrugs.

Editor's evaluation

This manuscript describes an elegant chemical-genetic strategy to discover that human oxidoreductase HSD17B11 is a major contributor to the bioactivation of the dialkinylcarbinol class of cytotoxic natural products. Mechanistic work further revealed that the reactive metabolites generated by HSD17B11 modify lysine and cysteine side-chains on proteins, leading to an unfolded protein response and apoptotic cell death. This study thus provides a plausible mechanism to explain how dialkynylcarbinol compounds exert their cytotoxic properties and identifies enzyme targets for controlling this process in human cells.

https://doi.org/10.7554/eLife.73913.sa0

Introduction

Nature is a rich source of bioactive compounds, some of which can be directly exploited to treat diseases. Some of them reveal sophisticated mechanisms of action which can be mimicked by designing synthetic molecules with specific features (Newman and Cragg, 2020). Marine sponges have attracted pharmaceutical interest since the discovery in the 1950s of C-nucleosides in Cryptotethia crypta that led to the development of cytosine arabinoside (ara-C or cytarabine) and analogues as anticancer treatments for acute myelogenous leukemia (Bergmann, 1950; Ellison et al., 1968). In a different structural series, several cytotoxic acetylenic lipids bearing a terminal alkenylalkynylcarbinol (AAC) pharmacophore have since been isolated from marine sponges, such as petrocortyne A (Figure 1—figure supplement 1), isolated from Petrosia sp. (Seo et al., 1998) and fulvinol isolated from Haliclona fulva (Ortega et al., 1996). The simplest cytotoxic AAC representative, (S)-eicos-(4E)-en-1-yn-3-ol ((S)–1, Figure 1—figure supplement 1), was isolated from the marine sponge Cribrochalina vasculum (Gunasekera and Faircloth, 1990). It demonstrated high cytotoxic activity selectively towards non-small cell lung carcinoma cells as compared to normal lung fibroblasts (Zovko et al., 2014). Starting from (S)–1, an extensive structure-activity relationship study in human cancer cell lines established that (Figure 1—figure supplement 1): (i) the non-natural enantiomer (R)–1 has higher cytotoxic activity, (ii) homologues with shorter lipidic tails are more cytotoxic, with an optimum total aliphatic backbone of 17 carbon atoms (e.g. (R)–2), and (iii) replacement of the internal C=C bond by a C≡C bond, giving rise to a terminal dialkynylcarbinol (DAC) pharmacophore, further increases cytotoxicity, to reach an IC₅₀ down to 90 nM for the DAC (S)–3 (El Arfaoui et al., 2013; Listunov et al., 2015a; Listunov et al., 2018b). However, despite this significant level of activity, the mode of action of this family of molecules, including the natural compound (S)–1, remains elusive (Zovko et al., 2014).

Here, we use functional genomics and chemoproteomics to decipher how cytotoxic DACs and related molecules mediate their biological effect. We discover that they behave as prodrugs enantiospecifically bioactivated by a member of the Short-chain Dehydrogenase/Reductase (SDR) family. Finally, we design new SDR-bioactivated DACs derivatives, establishing this family of lipidic alkynylcarbinols as a large and untapped reservoir of cytotoxic prodrugs.

Results

The SDR HSD17B11 governs (S)-DACs cytotoxicity

To determine how cytotoxic DACs mediate their effect on human cells, we applied a genetic approach using the pseudo-haploid human cell line HAP-1 (Carette et al., 2009). Given that (S)–3 had the greatest cytotoxic activity of all the DACs previously tested (El Arfaoui et al., 2013; Listunov et al., 2015a; Listunov et al., 2018b), we screened for mutations that could render HAP-1 cells resistant to (S)–3. We first confirmed in HAP-1 that (S)–3, but not (R)–3 (Figure 1A), exhibits nanomolar cytotoxic activity (Figure 1B, IC₅₀ 62.4 nM), in agreement with previous results on HCT116 colon cancer cells (El Arfaoui et al., 2013). We used Ethyl-Methane Sulfonate (EMS) to generate a mutagenized HAP-1 population and selected resistant clones using a lethal 250 nM (S)–3 concentration. Ten individual (S)–3-resistant clones (DACR) were isolated, displaying a 38- to 62-fold resistance to (S)–3 (Figure 1C) but similar sensitivity as parental cells to two unrelated compounds, bortezomib and doxorubicin (Figure 1—figure supplement 2A,B). Based on previous work (Wacker et al., 2012; Bossaert et al., 2021), and considering that EMS induces mainly point mutations under these conditions (Forment et al., 2017), we selected four DACR clones for RNA-seq analysis, to identify mis- or non-sense mutations accounting for the resistance. Around nine mutated genes were identified per clone (Figure 1—figure supplement 2C), with KCTD5 and HSD17B11 being the only mutated genes shared by more than two clones (Figure 1—figure supplement 2D). KCTD5 encodes for an E3-ubiquitin ligase substrate adaptor identified in a genetic screen as a negative regulator of the Akt pathway (Brockmann et al., 2017). However, while KCTD5 mRNA was expressed in all DACR clones, HSD17B11 mRNA levels were strongly reduced in the only clone without HSD17B11 coding mutations (#A5, Figure 1—figure supplement 2E). This suggested that mutations or lack of expression of HSD17B11 were responsible for DACR clone resistance. To confirm this, we sequenced HSD17B11 cDNAs from six other DACR clones, and detected non-synonymous HSD17B11 mutations in five, and no HSD17B11 cDNA in the sixth, suggesting loss of expression (Figure 1D, E). These data strongly supported a role for HSD17B11 in mediating (S)–3 cytotoxicity.

Figure 1 with 5 supplements see all

Download asset Open asset

HSD17B11 is necessary for DAC (S)–3 cytotoxic activity.

(A) DAC (S)–3 and (R)–3 structures. (B) Cell viability analysis of HAP-1 or U2OS cells treated for 72 h with the indicated concentrations of (S)- or (R)–3. (C) Cell viability analysis of individual DAC-resistant clones or wild-type HAP-1 treated for 72 hr with the indicated concentrations of (S)–3. (D) List of mutations identified by RNA-seq or targeted sequencing of HSD17B11 in individual DAC-resistant clones. (E) Schematic representation of HSD17B11 functional domains. The positions of the identified mutations are indicated in red. The Y185, K189 (indicated in black), and S172 amino acids are critical for catalysis. (F) Analysis by immunoblotting of HSD17B11 levels in wild-type HAP-1 and DAC-resistant clones. Ku80 was used as a loading control. The black arrow indicates HSD17B11 position. (G) Analysis by immunoblotting of HSD17B11-GFP levels in individual clones of DAC-resistant clone A5 complemented with GFP, wild-type or S172L mutant HSD17B11-GFP. SAF-A and total H2AX were used as loading controls. (H) Cell viability analysis of individual clones of DAC-resistant clone A5 complemented with GFP, wild-type or S172L mutant HSD17B11-GFP treated for 72 h with the indicated concentrations of (S)–3.

Figure 1—source data 1 Source data related to Figure 1F. The tiff files correspond to uncropped pictures of the chemiluminescent signal acquired on a BioRad Chemidoc. The regions used to generate the figure are highlighted by back squares in the jpg file, which also contains at the bottom an overlay with a picture of the membrane to locate the protein ladder positions.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig1-data1-v1.zip
Download elife-73913-fig1-data1-v1.zip
Figure 1—source data 2 Source data related to Figure 1G. The tiff files correspond to uncropped pictures of the chemiluminescent signal acquired on a BioRad Chemidoc. The regions used to generate the figure are highlighted by back squares in the jpg file, which also contains at the bottom an overlay with a picture of the membrane to locate the protein ladder positions.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig1-data2-v1.zip
Download elife-73913-fig1-data2-v1.zip

HSD17B11 encodes for the estradiol 17-beta-dehydrogenase 11, a member of the SDR super-family. HSD17B11, also called SDR16C2 [Persson et al., 2009], PAN1B, DHRS8, or retSDR2, localizes to the endoplasmic reticulum (ER) and lipid droplets (LD) via a N-terminal targeting domain (Figure 1E), where it uses NAD+ to catalyze oxidation of the C17 carbinol center of androstan-3-alpha,17-beta-diol to generate androsterone, a weak androgen (Brereton et al., 2001; Horiguchi et al., 2008a) (see Figure 2A). The HSD17B11 protein was barely detectable in all the DACR clones (Figure 1F, lower band), suggesting that the mutations result in protein instability. Using the DACR#A5 clone, in which HSD17B11 RNA was strongly down-regulated (~200 fold, Figure 1—figure supplement 2E), we performed complementation experiments with plasmids coding for GFP alone, or wild-type (WT) or S172L HSD17B11-GFP. This mutation was selected because the S172 residue is critical for catalysis (Filling et al., 2002; Gao et al., 2021), and the DACR#A4 clone, which carried S172L mutations, was the only one in which traces of full-length HSD17B11 could be detected (Figure 1F). Complemented DACR#A5 cells stably expressing WT and S172L HSD17B11-GFP at similar levels were successfully isolated (Figure 1G), and (S)–3 was ~50 times more active against cells expressing WT HSD17B11 compared to control GFP-complemented cells or cells expressing S172L HSD17B11 (Figure 1H). This supports that HSD17B11 catalytic activity is critical for (S)-DAC cytotoxicity. Notably, the DACR#A4 clone (S172L mutation) was also resistant to six other cytotoxic AACs: the naturally occurring AAC (S)–1 (Figure 1—figure supplement 3A), its synthetic enantiomer (R)–1, its shorter homologue (R)–2, the synthetic AAC (S)–4 with an internal C≡C bond and an external C=C bond (Figure 1—figure supplement 3B), the allenylalkynylcarbinol (AllAC) (R,S_a)–5 (Listunov et al., 2018a) and the more cytotoxic butadiynylalkynylcarbinol (BAC) (S)–6 (Bourkhis et al., 2018; Figure 1—figure supplement 3C). Thus, HSD17B11 functionality governs the enantiospecific cytotoxicity of the natural compound (S)–1 but also of all the more cytotoxic synthetic derivatives tested. In addition, HSD17B11 has been recently identified as mediating, through an unknown mechanism, the cytotoxic effect of dehydrofalcarinol, a polyacetylenic compound with a terminal butadiynylalkenylcarbinol motif isolated from several plants of the Asteraceae family (Grant et al., 2020).

Figure 2 with 4 supplements see all

Download asset Open asset

DACones are protein reactive species.

(A) Reaction catalyzed by HSD17B11. (B) Clickable DACs and DACones used in the study. (C) Viability analysis of U2OS cells treated in PBS for 1 h with (S)–3 or DACones and incubated for an additional 72 h after drug washout. (D) FBS or purified BSA were incubated 40 min at 30 °C with clickable DAC (S)–9 or clickable DACone 10. After reaction, CuAAC was used to ligate an azido-AlexaFluor647 to clickable molecules. Modified proteins were detected by scanning membrane fluorescence after SDS-PAGE and transfer. Ponceau S stains total proteins. (E) BSA or BLG were incubated with the indicated DACs or DACones, as in (D). After reaction, modified proteins were detected as in (D). Coomassie stains total proteins. (F) WT or S172L HSD17B11-GFP were immunoprecipitated from complemented U2OS KO HSD17B11 cells and incubated with clickable DAC 9 and BLG. After reaction, modified proteins were detected in the supernatant (BLG) or on the beads (HSD17B11-GFP) as in (D). GFP immunoblotting confirmed that equal amounts of WT and S172L proteins were used. (G) Analysis by direct-infusion mass spectrometry of purified BLG (mixture of isoform A and B) modified or not by DACone 10. Cyan and green arrows indicate the formation of a first and second adduct, respectively. (H) % of each amino acid detected as modified by DACones 10 or 11 in U2OS extracts as determined using an isoDTB-ABPP-based framework. (I) Proposed reactions of DACones with cysteine and lysine side chains in proteins.

Figure 2—source data 1 Source data related to Figure 2D. The tiff files correspond to an uncropped picture of the AlexaFluor647 fluorescence signal, acquired on an Odyssey LI-COR, and of a scan of the membrane stained with Ponceau S. The jpg file combines both pictures and can be used to locate the protein ladders.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig2-data1-v1.zip
Download elife-73913-fig2-data1-v1.zip
Figure 2—source data 2 Source data related to Figure 2E. The tiff files correspond to uncropped pictures of the AlexaFluor647 fluorescence, acquired in gel on an Odyssey LI-COR, and of the gels after staining with Coomassie blue scanned with the BioRad Chemidoc. Two different gels were used (respectively labeled upper and lower). Each jpg file combines the two pictures used to generate upper and lower parts of the figure.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig2-data2-v1.zip
Download elife-73913-fig2-data2-v1.zip
Figure 2—source data 3 Source data related to Figure 2F. The tiff files correspond to uncropped pictures of the AlexaFluor647 fluorescence signal, acquired on an Odyssey LI-COR, of a scan of the membrane stained with Ponceau S, for the upper part (supernatant), and of the chemiluminescent signal acquired using autoradiographic films for the lower part (beads). The jpg file combines the pictures used to generate the figure and can be used to locate the protein ladders.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig2-data3-v1.zip
Download elife-73913-fig2-data3-v1.zip

We next tested the cytotoxic activity of (S)–3 on a panel of 15 cancer cell lines. This revealed that the osteosarcoma U2OS cell line was the most sensitive to (S)–3 while the breast cancer cell line T47D was highly resistant (Figure 1—figure supplement 4A). Accordingly, HSD17B11 protein was undetectable in T47D, while U2OS displayed the highest levels (Figure 1—figure supplement 4B), in agreement with reported mRNA levels (The Cancer Cell Line Encyclopedia dataset [Barretina et al., 2012]). In addition, (S)–3 was particularly cytotoxic toward four other osteosarcoma cell lines as compared to normal cell lines or primary osteoblasts (Figure 1—figure supplement 4C). CRISPR/Cas9-mediated inactivation of HSD17B11 also conferred significant (S)–3 resistance to U2OS cells, which was suppressed by wild-type HSD17B11-GFP but not by the S172L mutant or GFP alone (Figure 1—figure supplement 5A, B). In contrast, complementation with HSD17B11 carrying the L14P or V16D mutations, lying outside of the catalytic domain and identified in the DACR clones #B1 and #A1/#A2, respectively, restored (S)–3 cytotoxic activity, in agreement with these mutations affecting HSD17B11 protein stability and not its catalytic activity (Figure 1—figure supplement 5C, D). These data also support that the C-terminal FLAG-GFP tag and/or the CMV promoter-based overexpression partly overcome the impact of these mutations on HSD17B11 expression level. The role of HSD17B11 in (S)–3 cytotoxic activity was further confirmed using two different small-interfering RNAs (siRNA) to down-regulate HSD17B11 in U2OS (Figure 1—figure supplement 5E, F) and in the non-small cell lung carcinoma cell line A549, in which CRISPR/Cas9-mediated HSD17B11 inactivation also conferred (S)–3 resistance (Figure 1—figure supplement 5G,H). Altogether, these data establish that HSD17B11 is critical in multiple cell lines for (S)–3 cytotoxic activity, and suggest that (S)–3 behaves as an HSD17B11-bioactivated prodrug. In addition, the acute toxicity of (S)–3 towards osteosarcoma cell lines suggests that DACs could be developed into a targeted anticancer therapy, but this would need to be further investigated, especially in vivo.

Dialkynylketones are protein-reactive species

We next investigated the downstream mechanism of cytotoxic action of the DAC (S)–3. The C17 carbinol center of androstan-3-alpha,17-beta-diol, which is naturally oxidized by HSD17B11 (Figure 2; Brereton et al., 2001), has the same spatial orientation as the (S)–3 carbinol when its lipidic chain is superimposed with the C13(C18) side of the steroid skeleton (Figure 2B). This suggested that HSD17B11 enantiospecifically recognizes and oxidizes (S)–3 into a “dialkynylketone” 7 (DACone), a diynone that could be the cytotoxic species. However, when the DACone 7 was previously synthesized and tested, no cytotoxic activity was found (Listunov et al., 2015a). Given the high in vitro electrophilic reactivity of ynones as Michael acceptors of thiols and amines (Worch et al., 2021), we considered that medium components such as serum albumin may rapidly react with and inactivate DACones. To test this, we synthesized the DACone 7, as well as a homologue with a shorter alkyl chain 8, and treated U2OS cells in a protein-free medium (PBS containing CaCl₂ and MgCl₂ to maintain cellular adhesion). Both the DACones 7 and 8 were indeed cytotoxic in the absence of serum, with 8 (short chain) being even more active than (S)–3 (Figure 2C). While the cytotoxicity of (S)–3 was strongly reduced by inactivation of HSD17B11, the cytotoxicity of the DACones 7 and 8 was not affected, supporting the notion that the DACones are the cytotoxic products generated from DACs by HSD17B11.

To further analyze the interaction between DACones and proteins, we synthesized ‘clickable’ analogues, that is bearing a terminal C≡CH tag, for each DAC enantiomer ((S)–9 and (R)–9), and for long and short DACones (10 and 11, Figure 2B), and used them to monitor the formation of covalent bonds between DACones and serum proteins by copper-catalyzed azide-alkyne cycloaddition (CuAAC ‘click chemistry’, [Tornøe et al., 2002; Rostovtsev et al., 2002]). The clickable DACone 10, or clickable DAC (S)–9 as control, were incubated with fetal bovine serum (FBS) or purified bovine serum albumin (BSA), followed by CuAAC-mediated ligation of an AlexaFluor647-azido fluorophore to the free C≡CH tag. The proteins were separated by SDS-PAGE and scanned for fluorescence (Speers and Cravatt, 2004). Covalent adducts were formed on BSA with DACone 10 but not with (S)–9 (Figure 2D). Moreover, the DACone 10 also reacted with several other model proteins, including the bovine beta-lactoglobuline (BLG) (Figure 2—figure supplement 1A). Using BSA and BLG, we established that DACone adducts are produced only when using the clickable DACones 10 or 11 (Figure 2E), suggesting that the terminal triple bond of the DACone pharmacophore is modified or masked after reaction. Finally, we could recapitulate the activation of (S)–9, but not of (R)–9, into protein-reactive species by immunopurified WT HSD17B11, but not by the S172L mutant (Figure 2F), supporting an enantiospecific bioactivation of (S)–9 into the BLG-reactive DACone 10 by HSD17B11. Considering that HSD17B11 known activity is the NAD+-dependent oxidation of a secondary carbinol into a ketone and that the only hydroxyl group on (S)–3 is the one occurring in the dialkynylcarbinol pharmacophore, this experiment strongly supports the notion that HSD17B11 enantiospecifically oxidizes (S)–3 into the DACone 7, which immediately reacts with nearby proteins, including HSD17B11-GFP itself as observed in Figure 2F. This high level of reactivity unfortunately precludes isolating the HSD17B11-produced DACones. To understand the basis for this enantiospecific bioactivation, we used AlphaFold2 (Jumper et al., 2021; Evans et al., 2021; Mirdita et al., 2021) to generate a structural model of HSD17B11 (Figure 2—figure supplement 2A) and performed molecular docking of (S)–3 and (R)–3 into its catalytic core. Both (S)–3 and (R)–3 docked into the catalytic cavity (Figure 2—figure supplement 2B,C), but (R)–3 had a lower computed affinity than (S)–3 (155 nM vs 15 nM) and only (S)–3 had its hydroxyl group properly positioned to engage hydrogen bonds with the S172 and Y185 catalytic amino acids (Figure 2—figure supplement 2D, E), which is critical for further carbinol oxidation via hydride transfer to NAD+ (Filling et al., 2002; Gao et al., 2021). Combining docking of (S)- and (R)–3 on AlphaFold2 models of the 12 other human 17β-hydroxysteroid dehydrogenase (17β-HSDs) SDRs, with filtering to select the most stringent interactions, was used to identify other SDRs that might be able to bioactivate (S)–3 (Figure 2—figure supplement 2F). This filtering revealed that, beyond HSD17B11, (S)–3 also docked onto the catalytic domains of only two 17β-HSDs, HSD17B13 and HSD17B3, while (R)–3 docked onto two different 17β-HSDs, HSD17B9 and HSD17B6. HSD17B3 (aka EDH17B3 or SDR12C2 [Persson et al., 2009]) is a reductive SDR involved in testosterone biosynthesisso it was not tested further. In contrast, since HSD17B13 (aka SCDR9 or SDR16C3 [Persson et al., 2009]), whose expression is restricted to the liver, is an oxidative SDR, the closest homologue of HSD17B11 and is also localized at the ER (Horiguchi et al., 2008b), we tested whether it could complement U2OS KO for HSD17B11. This revealed that HSD17B13 is also able to bioactivate (S)–3 into cytotoxic compounds, albeit in a less efficient manner as compared to HSD17B11 (IC₅₀ of 30 nM vs 12 nM for HSD17B11 with similar complementation levels, Figure 2—figure supplement 2G, H).

Reaction of DACones with proteins

To further decipher the reaction of DACones with proteins, we used direct-infusion mass spectrometry to analyze BLG modified with the clickable DACone 10. Purified BLG contains two isoforms (A and B, differing by 86.0 Da) and, when incubated with DACone, both BLG isoforms were completely modified with the formation of one or two adducts of ~+242 Da (Figure 2G), which corresponds to the mass of the clickable DACone 10. Monitoring the absorbance spectra of modified BLG revealed that BLG gains an absorption band at ~323 nm upon modification by DACone (Figure 2—figure supplement 1B). Using this, we confirmed that both BLG and BSA are modified by the DACones 7 and 8 or their clickable analogues 10 and 11 (Figure 2—figure supplement 1B, C). The shorter DACone 8 proved to be even more reactive, in line with its greater cytotoxicity (Figure 2C, Figure 2—figure supplement 1B, C). Next, we assessed the selectivity of DACones towards amino acid residues in the whole proteome in an unbiased fashion. For this purpose, we incubated the DACones 10 and 11 with U2OS total cell extracts in PBS. We then used residue-specific chemoproteomics with isotopically labeled desthiobiotin azide (isoDTB) tags (Backus et al., 2016; Weerapana et al., 2010; Zanon et al., 2020) coupled to a novel MSFragger-based FragPipe computational platform (Zanon et al., 2021) to detect the modified amino acids on the enriched peptides. This revealed that both DACones reacted with cysteine and lysine side chains, with the expected modification being detected ( + 729.4498/ + 723.4408 Da (Heavy/Light) for DACone 10, + 631.3404/ + 625.3332 Da (Heavy/Light) for DACone 11, Figure 2H, Figure 2—figure supplement 1D, E, F, Supplementary file 1A, B). We also detected many unmodified peptides (Supplementary file 1D, E), most of them with at least one missed lysine trypsin cleavage site (~92% of sequences), suggesting that these were still modified during digest and the modification lost during the subsequent workflow, potentially during the final trifluoroacetic acid elution (0.1%, pH~2 [Zanon et al., 2021]). We cannot fully exclude that other amino acids were also modified by the probe to some degree and that this modification was also lost during the workflow, but this data points to the fact that lysines and cysteines are the main modification sites. We next confirmed the reactivity of DACones with cysteine and lysine side chains by monitoring the appearance of the ~323 nm absorbance band after reaction of the DACone 8 with isolated amino acids, using N-acetylated versions to prevent reactions with the N-terminal amino group. At neutral pH, DACones only reacted with N-acetyl-L-cysteine (NAC) but not with N_α-acetyl-L-lysine (NAK) (Figure 2—figure supplement 1G, left spectrum), whereas at higher pH they reacted with both NAC and NAK (Figure 2—figure supplement 1G, right spectrum), in agreement with the nucleophilic reactivity of the non-protonated ε−NH₂ group of the lysine chain. No reaction was observed with N-acetyl glycine (NAG), supporting that the reaction involves the side chain. By monitoring the adducts absorbance, we confirmed that the DACone 8-NAK linkage is progressively lost when incubated in 0.1% TFA (38% reduction after 1 h incubation), while the DACone 8-NAC linkage remains unaffected in these conditions (Figure 2—figure supplement 1H, I). These data suggest that the reactivity with lysine side chains is underestimated by our isoDTB-ABPP experiment due to preferential loss of the lysine-DACone species. The reaction with lysine side chains is compatible with the pK_a value for the lysine ε-amino group that can be as low as ~5 in hydrophobic domains in proteins (Isom et al., 2011). Accordingly, analysis of the sequence context of the amino acids identified as modified by the DACone 10 revealed an enrichment of hydrophobic amino acids around the modified lysines, which was not observed for the modified cysteines (Figure 2—figure supplement 3A, B, Supplementary file 2). Using nuclear magnetic resonance (NMR), we characterized the products of the reaction of the short DACone 8 with NAC (Figure 2—figure supplement 4A, B) or NAK (Figure 2—figure supplement 4C, D). This revealed that a covalent bond forms by addition of the thiol (NAC) or amino (NAK) group onto the terminal alkyne of the DACone head (Figure 2—figure supplement 4E). A similar reaction probably occurs with the cysteine and lysine residues of proteins, which is supported by their gain of a similar absorbance band upon treatment with DACone (Figure 2I). This additional absorption band can be accounted for by the donor-acceptor extension of π-electron delocalization in the enone adducts (S–CH=CH–C=O for NAC, N–CH=CH–C=O for NAK). Altogether, our data show that DACones are highly reactive with proteins in vitro.

Bioactivated (S)-DACs lipoxidize multiple proteins in cells

Protein modification by lipidic DACs equates to their lipoxidation (a term used to designate the covalent modification of a protein by a reactive lipid [Viedma-Poyatos et al., 2021]) by one or several C17 hydrophobic chain(s) (Figure 3—figure supplement 1). Considering that protein palmitoylation (addition of a C16 lipidic chain) can trigger membrane tethering of proteins, we hypothesized that lipoxidation by DACs could affect protein localization and/or function and account for the cytotoxicity of bioactivated DACs in cells as described for other reactive lipids (Viedma-Poyatos et al., 2021). To challenge this hypothesis in cells, we took advantage of the clickable DAC 9 (Figure 2B, [Listunov et al., 2015b]). As observed for the DAC 3, the cytotoxicity of the clickable DAC 9 was enantiospecific, biased toward (S)–9, and dependent on bioactivation by HSD17B11 (Figure 3—figure supplement 2A). Cells were treated with clickable (S)- or (R)–9 DACs, extracts prepared and click chemistry used to detect the covalent adducts of DACs onto proteins (Speers and Cravatt, 2004). Multiple modified proteins were detected in extracts from (S)–9-treated cells, while no adduct with (R)–9 was detected (Figure 3A).

Figure 3 with 5 supplements see all

Download asset Open asset

(S)-DACs lipoxidize multiple cellular proteins, triggering their association with cellular membranes.

(A) U2OS cells were incubated for 2 h with 2 µM (S)- or (R)–9, proteins were extracted and DAC-modified proteins were detected by CuAAC-mediated ligation of azido-AlexaFluor-647 to clickable molecules, separation by SDS-PAGE, transfer to a membrane which was scanned for fluorescence. (B) Landscape of proteins modified in U2OS cells by clickable DAC (S)–9 computed from three independent experiments. Fold enrichment (FC) as compared to the clickable (R)–9 is computed and color-coded as depicted. Box size corresponds to -log(p) computed as described in the materials and methods section. (C) PSMD2 or control immunoprecipitations (Ctrl) were performed from extracts of U2OS cells treated 2 h with 2 µM clickable DAC (S)- or (R)–9. DAC-modified proteins were detected by CuAAC-mediated ligation of azido-AlexaFluor-647 to clickable molecules, separation by SDS-PAGE, transfer to a membrane, which was scanned for fluorescence. PSMD2 was subsequently visualized by immunoblotting. (D) PSMD2 immunoprecipitations were performed from extracts of wild-type or DAC-resistant HAP-1 cells treated or not for 2 h with 2 µM clickable DAC (S)–9. DAC-modified proteins were detected by CuAAC-mediated ligation as in (C). PSMD2 was subsequently visualized by immunoblotting. (E) U2OS cells were treated 2 h with 0.5 µM DAC, fixed, permeabilized, and clickable molecules were detected by click with AlexaFluor488 azide. (F) U2OS expressing GFP-BRAT1 were treated 2 h with 1 µM (S)–3, pre-extracted, fixed and processed for analysis by fluorescence microscopy.

Figure 3—source data 1 Source data related to Figure 3A. The tiff files correspond to an uncropped picture of the AlexaFluor647 fluorescence signal, acquired on an Odyssey LI-COR, and of a scan of the membrane stained with Ponceau S. The regions used to generate the figure are highlighted by back squares in the jpg file.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig3-data1-v1.zip
Download elife-73913-fig3-data1-v1.zip
Figure 3—source data 2 Source data related to Figure 3C. The tiff files correspond to uncropped pictures of the AlexaFluor647 and PSMD2 fluorescence signal, both detected on a LI-COR Odyssey. The jpg file combines both pictures and can be used to locate the protein ladders.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig3-data2-v1.zip
Download elife-73913-fig3-data2-v1.zip
Figure 3—source data 3 Source data related to Figure 3D. The tiff files correspond to uncropped pictures of the AlexaFluor647 and PSMD2 fluorescence signal, both detected on a LI-COR Odyssey. The jpg file combines both pictures and can be used to locate the protein ladders.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig3-data3-v1.zip
Download elife-73913-fig3-data3-v1.zip

To identify the proteins lipoxidized by DACs upon bioactivation, and inspired by previous studies (Larrieu et al., 2014; Vila et al., 2008), we used streptavidin pull-down to isolate (S)–9-modified proteins after click chemistry-mediated ligation of a biotin handle in extracts of treated cells and identified them using bottom-up proteomics. Forty-two proteins were significantly enriched more than 2-fold in the (S)–9 condition as compared to (R)–9 (Figure 3B, Supplementary file 3), with three proteins being enriched more than 60-fold: BRAT1, PLIN3, and PSMD2. In the MS/MS data, we could not detect peptides with the DACone-triazole-PEG-biotin modification, suggesting either that the biotin-carrying peptides remained associated with the beads or that the modification was lost during the LC-MS/MS workflow. We also attempted to adapt the isoDTB tag-based chemoproteomics workflow to identify the modified peptides directly from U2OS cells treated for 2 h with 2 µM (S)–9. Under these conditions, we detected a few peptides modified on a cysteine by the expected adduct plus the mass of a dithiothreitol group (DTT, Supplementary file 4A), evocative of a DTT reaction with the internal alkyne functionality (see Supplementary file 4B for the potential adduct structure, Lei et al., 2021), a type of adduct which was also detected in the isoDTB-based analysis presented in Figure 2H, albeit in a smaller proportion (~30% of all quantified sites with DACone 10 & 11). This experiment supports that the reaction between DACones and N-acetyl-amino acids established in vitro also takes place in cells when DACones are produced by HSD17B11. The small number of modified sites identified in this experiment can be explained by the instability of the adduct on lysines during the isoDTB-ABPP workflow, the lower DAC concentration used for cell treatment (2 µM for (S)–9) as compared to the in vitro experiment (100 µM of 10 or 11) and the need for bioactivation of (S)–9 in contrast to the constantly active 10 or 11.

To validate that the three most highly enriched hits are modified in cells, we overexpressed BRAT1, PSMD2 and PLIN3 individually (in addition to TK1) as GFP fusions in U2OS cells (Figure 3—figure supplement 2B), and used GFP pull-down to determine whether they were modified by clickable DAC 9 in cells. These proteins were found robustly modified by (S)–9 but not by (R)–9, especially BRAT1 and PSMD2, while GFP alone was not modified (Figure 3—figure supplement 2C). The fact that only 42 proteins were found robustly and reproducibly enriched with (S)–9 over (R)–9 in cells indicates a certain degree of selectivity for protein modification by DACones that could result from its production and reactivity being restricted to the vicinity of HSD17B11. In agreement, while the thiol group of reduced glutathione can be modified in vitro by DACones at pH7 with the formation of an adduct similar to the one with NAC (Figure 3—figure supplement 3A–C), suggesting that it could provide some protection against the reactivity of DACones in cells, the toxicity of (S)–3 was not significantly decreased by co-treatment with a cell-permeable glutathione or increased by a glutathione-S-transferase inhibitor (GSTi, Figure 3—figure supplement 3D, E). These data suggest that DACones rapidly and selectively react with a specific set of proteins in the vicinity of HSD17B11 as shown for another in situ generated electrophile (Paxman et al., 2018).

Among the three main hits, PSMD2 drew our attention as an essential protein in HAP-1 cells (Blomen et al., 2015). PSMD2, also called Rpn1, is a critical non-catalytic subunit of the 19 S regulatory particle of the 26 S proteasome, a large complex responsible for the ubiquitin-dependent degradation of cellular proteins. PSMD2 is essential for 19 S assembly and for docking of ubiquitin, ubiquitin receptors and the deubiquitinase USP14 (Shi et al., 2016). Immunoprecipitation of endogenous PSMD2 from U2OS cells treated with clickable DAC confirmed that PSMD2 is covalently modified after treatment with the DAC (S)–9 but not with (R)–9 (Figure 3C). Moreover, the clickable DACone 10 efficiently modified PSMD2 in vitro (Figure 3—figure supplement 2D). Using the (S)-DAC-resistant HAP-1 clones A1 and A4 (expressing V16D and S172L HSD17B11 mutants, respectively), we also confirmed that the modification of cellular proteins by (S)–9, including PSMD2, was dependent on HSD17B11 (Figure 3D and Figure 3—figure supplement 2E). Our proteomics approach also revealed that, in addition to PSMD2, a cluster of proteins involved in protein quality control (PQC) was also modified by the DAC (S)–9 (Figure 3B), including several protein disulfide isomerases (P4HB/PDIA1, PDIA6 and TMX1), thioredoxin reductases (TXNDC5 and TXNRD1) and protein chaperones (the ER-resident HSP70, HSPA5/GRP78/BiP; and HSP90AB1), the alteration of which likely also contributes to the DAC cytotoxic effect. Altogether, these data show that (S)-DACs are bioactivated by HSD7B11 into highly reactive DACones that covalently lipoxidize nearby proteins, including essential proteins involved in PQC such as PSMD2, a critical subunit of the ubiquitin-proteasome system (UPS).

We then used click-based imaging to monitor the localization of DAC-modified proteins. (S)–9 gave a strong nuclear and cytoplasmic staining, the latter being evocative of ER and mitochondrial membranes (Figure 3E). In agreement, we observed a co-occurrence of (S)–9-click staining with markers of ER (Figure 3—figure supplement 3A) and mitochondria (Figure 3—figure supplement 3B, C). The lack of staining in cells treated with the inactive (R)–9 (Figure 3E) supported that the staining corresponds to DAC-modified proteins. No staining was observed with (S)–3, supporting that the -C≡CH group of the dialkynylcarbinol head is modified after bioactivation and subsequent reaction with proteins (as shown in Figure 2I). Since GFP-BRAT1 is the protein that was the most robustly modified by DACs in cells (Figure 3—figure supplement 2C), we pre-extracted soluble proteins with mild nonionic detergent to assess GFP-BRAT1 association to subcellular compartments. Under these conditions, most GFP-BRAT1 was removed by the pre-extraction in untreated cells, as expected for a soluble protein, while it was retained to subcellular compartments evocative of ER and nucleus after treatment with (S)–3 (Figure 3F). These data suggest that protein lipoxidation by bioactivated DACs results, at least for some of them, into their relocalization to cellular membranes.

(S)-DACs trigger ER-stress, inhibition of ubiquitin-proteasome system (UPS) and apoptosis.

In agreement with (S)-DAC impairing PQC, we observed that treatment of cells by (S)–3 triggers ER swelling as shown by the appearance of a large cytoplasmic vacuole between 4 and 8 h of treatment that preceded cell death (Figure 4A). The use of a GFP variant targeted and retained into the ER confirmed that these vacuoles derived from this compartment (Figure 4B), while a mCherry protein addressed to mitochondria in the same cells showed that (S)-DAC treatment concomitantly triggered mitochondrial fission, a hallmark of cell stress (Figure 3—figure supplement 4, see also Video 1). ER swelling is a feature of ER-stress, which can result from the accumulation of unfolded proteins within the ER and can be triggered by various defects of PQC such as inhibition of the UPS, as seen with the UPS inhibitor MG132 (Figure 4—figure supplement 1A). In agreement with (S)–3 cytotoxicity being mediated by the accumulation of unfolded proteins, vacuolization and cell death induced by (S)–3 could be blocked by inhibition of protein synthesis by cycloheximide (CHX, Figure 4—figure supplement 1B). (S)–3 also triggered a strong accumulation of the chaperone HSP70 and of the cell stress response protein p21, similarly to MG132 (Figure 4C) or PSMD2 depletion using siRNA (Figure 4—figure supplement 1C). It is noteworthy that ER-stress itself can trigger UPS inhibition, likely through consuming free ubiquitin (Menéndez-Benito et al., 2005). In agreement with this hypothesis, we observed that (S)–3 induces the accumulation of poly-ubiquitinated proteins (Figure 4D), a hallmark of UPS inhibition, and blocked the degradation of an artificial substrate of the UPS system in a manner similar to MG132 (Figure 4E and F). Similarly to MG132, (S)–3 also blocked the assembly of the protein 53BP1 into foci at sites of DNA double-strand breaks (Figure 4—figure supplement 1D), a process that depends on the local de novo DNA damage-induced ubiquitination of histones (Mailand et al., 2007), thereby confirming depletion of free ubiquitin by treatment with (S)–3. It is noteworthy that (S)–3 treatment itself did not induce DNA damage nor blocked ATM activation, an important component of the DNA damage response (Figure 4—figure supplement 1D). Accumulation of unfolded proteins in the ER activates three ER-resident transmembrane signal transducers: the IRE1α kinase/endoribonuclease, the PERK kinase and the ATF6 transcription factor, which represent the three arms of the Unfolded Protein Response (UPR). The UPR aims at restoring ER protein homeostasis by reducing the influx of proteins into the ER and by increasing the activity of ER protein quality control mechanisms (Preissler and Ron, 2019; Szegezdi et al., 2006). IRE1α activation is marked by its autophosphorylation on S724 and the processing of the XBP-1 mRNA resulting in the production of the XBP1-s transcription factor. PERK activation is marked by its autophosphorylation on T980 and the phosphorylation of eIF2α S51, which reduces translation initiation and triggers the accumulation of the ATF4 transcription factor. ATF6 activation relies on its cleavage which releases a short-lived transcriptionally active fragment. Persistent UPR activation can promote apoptosis in a JNK-dependent manner (Szegezdi et al., 2006). In agreement with (S)–3 triggering unfolded proteins accumulation and UPR activation, we observed that treatment with (S)–3 induced the activation of the three UPR pathways with the rapid phosphorylation of IRE1α (S724), PERK (T980) and eIF2α (S51), a strong decrease of full-length ATF6, preceding the accumulation of cytoplasmic HSP70 and the transient accumulation of XBP-1s and ATF4 (Figure 4G). Of note, treatment with inhibitors of either IRE1α endoribonuclease, IRE1α kinase or PERK did not modify the toxicity of (S)–3, suggesting that the ER stress induced by (S)–3 is too acute to be resolved in a timely manner by the UPR (Figure 4—figure supplement 1E). In contrast, JNK inhibition provided a small degree of resistance to (S)–3 (IC₅₀ shift from 33 to 70 nM), suggesting a contribution of the JNK pathway to cell death (Figure 4—figure supplement 1F). Ultimately, (S)–3 treatment resulted in apoptosis, as marked by PARP-1 cleavage (Figure 4G), in a caspase inhibitor (z-VAD-fmk)-sensitive manner (Figure 4H). Caspase inhibition blocked cell death but not the vacuolization process (Figure 4—figure supplement 1G). Altogether, these results show that (S)–3, once bioactivated by HSD17B11, covalently modifies multiple proteins including critical components of the PQC, resulting in ER-stress, activation of the UPR and inhibition of the UPS, ultimately leading to apoptotic cell death.

Figure 4 with 1 supplement see all

Download asset Open asset

**DAC** (S)–3 triggers ER-stress, ubiquitin-proteasome system inhibition and apoptosis.

(A) U2OS stably co-expressing a GFP variant addressed and retained in the endoplasmic reticulum were treated with 1 µM (S)–3 and monitored by live imaging. Representative pictures of U2OS cells, either untreated (NT) or treated with 1 µM (S)–3 for the indicated time. (B) Representative pictures of U2OS with stably GFP-labeled endoplasmic reticulum and either untreated or treated for 8 h with 1 µM (S)–3. (C) Immunoblotting of extracts from U2OS untreated or treated with 1 µM (S)–3 or 20 µM MG132 for 8 h. (D) Immunoblotting of ubiquitin in extracts from U2OS untreated or treated with 1 µM (S)–3 or 20 µM MG132 for 2 h. High-molecular-weight ubiquitin conjugates are indicated by a vertical bar on the right. (E) Representative pictures of U2OS stably expressing Ub-G76V-YFP and either untreated or treated for 4 h with 1 µM (S)–3 or 20 µM MG132. (F) Analysis of YFP fluorescence by flow cytometry of U2OS Ub-G76V-YFP treated as described in (E) ,% of cells scored as positives using the vertical green bar as a threshold are indicated. (G) Immunoblotting using extracts from U2OS cells treated with 1 µM (S)–3 for increasing times, indicated in hours. To probe for all the UPR markers, the same extracts were analyzed on different immunoblots, each one with its loading control (see the source data), which were grouped in logical order on the figure to facilitate the interpretation. (H) Immunoblotting using extracts from U2OS cells treated with 1 µM (S)–3 for 12 h with or without 50 µM z-VAD-fmk.

Figure 4—source data 1 Source data related to Figure 4C. The tiff files correspond to uncropped pictures of the IRDye800CW fluorescence signal acquired on a LI-COR Odyssey. The regions used to generate the figure are highlighted by back squares in the jpg file.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig4-data1-v1.zip
Download elife-73913-fig4-data1-v1.zip
Figure 4—source data 2 Source data related to Figure 4D. The tiff files correspond to an uncropped picture of the IRDye800CW fluorescence signal, acquired on an Odyssey LI-COR, and of a scan of the membrane stained with Ponceau S. The regions used to generate the figure are highlighted by back squares in the jpg file.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig4-data2-v1.zip
Download elife-73913-fig4-data2-v1.zip
Figure 4—source data 3 Source data related to Figure 4G. The tiff files correspond to uncropped pictures of the chemiluminescent signal acquired on a BioRad Chemidoc, of the IRDye800CW fluorescence signal acquired on a LI-COR Odyssey or of the chemiluminescent signal acquired using autoradiographic films. Several immunoblotting, labeled from a to f, of the same extracts were used to generate this figure, all with a loading control. The regions used to generate the figure are highlighted for each immunoblot by back squares in the jpg files.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig4-data3-v1.zip
Download elife-73913-fig4-data3-v1.zip
Figure 4—source data 4 Source data related to Figure 4H. The tiff files correspond to uncropped pictures of the IRDye800CW fluorescence signal acquired on a LI-COR Odyssey. The regions used to generate the figure are highlighted by back squares in the jpg file.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig4-data4-v1.zip
Download elife-73913-fig4-data4-v1.zip

Video 1

Download asset

posterframe for video — This video corresponds to a time lapse of pictures of U2OS cells stably expressing ER-moxGFP and Mito-mCherry and treated with 1 µM (S)-.

Pictures were acquired every 40 min for 1040 min.

Identification of new SDR-specific prodrugs

Our discovery that an SDR, HSD17B11, bioactivates a secondary alcohol prodrug by oxidation into the corresponding ketone is particularly interesting given that the SDR superfamily is one of the largest protein superfamilies, with over 500,000 members found in most forms of life. This means that this mechanism of prodrug activation could be exploited to develop an extensive range of new drugs and compounds to kill cells or organisms expressing specific SDRs with high selectivity. SDRs use NAD(H) or NADP(H) cofactors to perform oxidoreductase, lyase or isomerase activities on a large variety of substrates including steroids, retinoic acids, lipids, polyols, sugars, nucleotides and xenobiotics (Kavanagh et al., 2008). In vitro, SDR enzymes can frequently catalyze both oxidation and reduction, depending on the supplied co-factor. In cells, however, they show directionality, which depends on their sub-cellular localization and cofactor availability. We first determined whether other SDRs with the proper in cellulo directional polarity could activate other secondary alkynylcarbinol-containing compounds into cytotoxic species. We found that the AAC (S)–4 (with a terminal C=CH₂ group, Figure 5A), despite its activity being strongly reduced in DAC-resistant clone A4 (~17-fold, Figure 1—figure supplement 3B), still retained some cytotoxicity (IC₅₀ ~2.6 µM) on these cells, possibly through bioactivation by a second dehydrogenase. To identify such a dehydrogenase, EMS was used to mutagenize the DACR clone #A4 once again. This mutagenized population was selected with a lethal concentration of 10 µM (S)–4 and over twenty resistant clones (AACR clones) were isolated. Among those, we selected seven, which all showed similar increased resistance to (S)–4 (Figure 5A), and submitted them to RNA-seq analysis. This allowed the identification of an average of six coding mutations in each clone (Figure 5—figure supplement 1A), and analysis revealed that the RDH11 gene carried point mutations in five of these seven clones (Figure 5—figure supplement 1A,B). Further analysis revealed that the clone A6 contained two alleles of RDH11, each with a different point mutation (Figure 5B), and RDH11 expression was severely downregulated in the clone B4 (Figure 5—figure supplement 1C). In conclusion, the RDH11 gene was mutated (Figure 5C) or downregulated in all the AACR clones, implicating it as an SDR, with similar prodrug bioactivation capacity as HSD17B11, that mediates AAC (S)–4 toxicity.

Figure 5 with 2 supplements see all

Download asset Open asset

Bioactivation of other lipidic alkynylcarbinols by specific SDRs.

(A) Cell viability analysis of wild-type HAP-1, DACR clone A4 and AACR clones treated with AAC (S)–4. (B) List of mutations identified on RDH11 by RNA-seq of individual AACR clones. (C) Schematic representation of RDH11 with, in red, the positions of the mutations identified and, in black, the three amino acids critical for catalysis. TM = single-pass transmembrane domain. (D) Structure of all-*trans*-retinol, a substrate for RDH11. (E) Analysis by immunoblotting of RDH11 levels in wild-type HAP-1, in DACR clone A4 and in the different AACR clones. (F) Analysis by immunoblotting of RDH11 and HSD17B11 levels in wild-type U2OS or clones inactivated for either HSD17B11, RDH11 or both. (G) Cell viability analysis of wild-type U2OS or U2OS clones inactivated for HSD17B11, RDH11, or both and treated with AADC 12. (H) Structure of prostaglandin E2, a substrate of HPGD. (I) Analysis by immunoblotting of GFP and HPGD levels in WT U2OS or U2OS KO HSD17B11 stably complemented with GFP or HPGD-GFP. (J) Cell viability analysis of U2OS or U2OS inactivated for HSD17B11, stably complemented with either HSD17B11-GFP or HPGD-GFP and treated for 72 h with AllAC (S,S_a)- or (R,S_a)–5.

Figure 5—source data 1 Source data related to Figure 5E. The tiff files correspond to uncropped pictures of the chemiluminescent signal acquired on a BioRad Chemidoc. The regions used to generate the figure are highlighted by back squares in the jpg file, which also contains at the bottom an overlay with a picture of the membrane to locate the protein ladder positions.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig5-data1-v1.zip
Download elife-73913-fig5-data1-v1.zip
Figure 5—source data 2 Source data related to Figure 5F. The tiff files correspond to uncropped pictures of the chemiluminescent signal acquired using autoradiographic films. Two different immunoblotting of the same extracts were used for this figure (respectively labeled upper and lower). The regions used to generate the figure are highlighted for each immunoblot by back squares in the jpg files.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig5-data2-v1.zip
Download elife-73913-fig5-data2-v1.zip
Figure 5—source data 3 Source data related to Figure 5I. The tiff files correspond to uncropped pictures of the chemiluminescent signal acquired on a BioRad Chemidoc. The same extracts were loaded twice (left and right part of the membrane) and the membrane sliced to simultaneously probe GFP and HPGD. The regions used to generate the figure are highlighted by back squares in the jpg file, which also contains overlays of the chemiluminescent signal with a picture of the membrane to locate the protein ladder positions.: https://cdn.elifesciences.org/articles/73913/elife-73913-fig5-data3-v1.zip
Download elife-73913-fig5-data3-v1.zip

The human RDH11 gene codes for retinol dehydrogenase 11 (also called SDR7C1 [Persson et al., 2009], PSDR1 or HSD17B15), a member of the SDR superfamily which also localizes to the ER (Kedishvili et al., 2002) where it uses NADP+ to catalyze the conversion of retinol (Figure 5D, preferentially the 11-cis, all-trans and 9-cis isomers) into retinal through oxidation of the C-15 carbinol center, with a pro-R hydrogen specificity (Haeseleer et al., 2002). Full-length RDH11 protein was not detectable in any of the AACR clones, suggesting that all the identified mis-sense mutations decrease RDH11 stability in a manner similar to HSD17B11 mutations (Figure 5E). To validate the ability of RDH11 to mediate some of the (S)–4 cytotoxic effects, we compared its activity on U2OS, U2OS KO HSD17B11, and U2OS KO HSD17B11 cells in which RDH11 was inactivated using CRISPR/Cas9 (Figure 5—figure supplement 2A). In this panel, we confirmed that, while HSD17B11 was responsible for the nM activity of the AAC (S)–4 (48-fold IC₅₀ increase as a result of HSD17B11 inactivation), the remaining toxicity of (S)–4 in HSD17B11 KO cells was further decreased by inactivating RDH11 (sixfold IC₅₀ increase by RDH11 inactivation, Figure 5—figure supplement 2B). Comparison of the activity of (S)–3 and (S)–4 on U2OS KO [HSD17B11 + RDH11] complemented either by HSD17B11-GFP or RDH11 (Figure 5—figure supplement 2C) revealed that RDH11 bioactivates preferentially (S)–4 (IC₅₀~0.28 µM, with a terminal C=CH₂ group) over (S)–3 (IC₅₀~1.92 µM, with a terminal C≡CH group, Figure 5—figure supplement 2D), but also established that HSD17B11 was much more efficient than RDH11 for the bioactivation of either molecule (IC₅₀ < 40 nM).

As a proof-of-principle of the potential value of these discoveries for designing novel prodrugs with controlled cytotoxic activity, we attempted to design a new prodrug whose cytotoxic effects would depend more on the RDH11 SDR. We used WT U2OS or U2OS inactivated for HSD17B11, RDH11 or both (Figure 5F) to test several AAC analogues. Introduction of a second secondary carbinol function in the AAC structure, generating the AlkenylAlkynylDiCarbinol 12 (AADC), reduced its bioactivation by HSD17B11, while leaving intact its RDH11-dependent activation and thereby generating a new prodrug equally bioactivated by HSD17B11 and RDH11 (Figure 5G). As mixtures of diastereoisomers were used for this experiment, lower IC₅₀ can be expected with selected AADC enantiomers, especially with the external carbinol center in the RDH11-preferred (S) configuration.

Another SDR, HPGD, also called 15-PGDH, PGDH1 or SDR36C1 (Persson et al., 2009), is responsible for the inactivation of prostaglandins through oxidation of the (S)-C-15 carbinol center in a NAD+-dependent manner (Figure 5H; Tai et al., 2006). HPGD expression is tissue-restricted (prostate and bladder) and null in U2OS cells (Broad Institute Cancer Cell Line Encyclopedia RNA-seq dataset). To determine whether this SDR could also bioactivate potential prodrugs, we used U2OS inactivated for HSD17B11 and overexpressing GFP alone or HPGD-GFP (Figure 5I), and screened a small collection of lipidic alkynylcarbinols. We found that the AllAC (S,S_a)–5 was selectively bioactivated by HPGD, resulting in an IC₅₀ of ~147 nM in HPGD-overexpressing cells vs normal U2OS (Figure 5J). Thus, this mechanism of action is likely a general property of SDRs that could be exploited to develop a wide range of tailored prodrugs to cause selective cytotoxicity.

Discussion

This study highlights the strengths of the genomic approach implemented here, relying on chemical mutagenesis of human haploid cells, selection with a toxic concentration and the use of RNA-seq to identify simultaneously changes in expression levels and mutations that could confer resistance to cytotoxic drugs. RNA-seq analysis of only four DAC (S)–3-resistant clones was sufficient to identify HSD17B11 as the DAC (S)–3-activating enzyme. Three clones carried non-sense mutations, including one on a key catalytic residue (S172, [Gao et al., 2021]), while the other ones showed lack of HSD17B11 expression. The strength of the approach is also reminiscent of the use of chemical mutagenesis. Indeed, while selection of drug-resistant clones based on the appearance of spontaneous mutations is frequently used in bacteria to identify the mechanism of action of small molecules, exploiting their high rate of mutations, human cells are less prone to spontaneous mutations. In human cells, screens based on spontaneous mutations most frequently use DNA repair (mismatch)-deficient cells, such as HCT-116, to increase the mutation rate (see for example Girdler et al., 2008 and Wacker et al., 2012). However, spontaneous mutations can be diverse in nature and in that regard, the use of chemical mutagenesis with ethyl methanesulfonate (EMS), in addition to increasing the rate of resistant clones formation, simplifies the identification of the mutations of interest. Indeed, EMS does not induce deletions or insertion, but principally single nucleotide changes, most frequently transitions (Forment et al., 2017) which can be selected during the analysis, restricting the number of false positives. Another strength of this approach is the possibility to readily perform multiple rounds of screening to decipher primary and secondary mechanisms of action for a drug. Here, the DAC (S)–3-resistant clones could be mutagenized and selected a second time to identify a second SDR, RDH11, as the AAC (S)–4-bioactivating enzyme. It is noteworthy that the AAC (S)–4 still shows some remaining toxicity on the DAC + AAC-resistant clones (HSD17B11 mutation +RDH11 mutation), which indicates that a third round of mutagenesis, selection and RNA-seq would most likely lead to identify another dehydrogenase, probably in the SDR superfamily, as responsible for the remaining bioactivation. This also illustrates how alkynylcarbinols could constitute a rich reservoir of SDR-specific prodrugs, whose selectivity can be improved through parallel structure-activity relationship studies. Finally, this approach is remarkably easier and cheaper to implement than loss of function CRIPSR/Cas9 screens and can lead to directly identify essential genes as mediators of drug cytotoxic effect. We recently exemplified this by using this approach to directly identify the DNA topoisomerase II alpha (coded by the essential TOP2A gene) as the main driver of the cytotoxic effects of CX-5461, a G-quadruplex ligand and inhibitor of RNA polymerase I (Bossaert et al., 2021). The identified non-sense mutations on TOP2A dissociate its essential function from its function in generating DNA double-strand breaks upon G-quadruplex stabilization by CX-5461.

Through this powerful framework, our study reveals the original mode of action of a large family of natural and synthetic cytotoxic lipids characterized by a chiral terminal functional alkynylcarbinol pharmacophore. We show that these molecules are oxidized in an enantiospecific manner by a specific SDR, HSD17B11, converting them into an alkynylketone species (ynones). Oxidation of DACs produces dialkynylketones (DACones) that proved to be highly protein-reactive electrophiles, forming Michael adducts with cysteines and lysines. Consequently, bioactivated DACs modify several proteins in cells, resulting in their lipoxidation by a C17 lipidic chain (Figure 3—figure supplement 1). Lipoxidation of a protein can modify its solubility, folding, interactions, activity and/or localization (Viedma-Poyatos et al., 2021). In agreement, DAC treatment triggers the association of BRAT1, one of the most DAC-modified proteins, to ER and nuclear membranes (Figure 3F). Covalent modification of BRAT1 by the natural diterpene curcusone D was recently reported and resulted in BRAT1 degradation and reduced DNA damage response (Cui et al., 2021). Regarding (S)-DAC, in addition to BRAT1, several proteins involved in mechanisms of PQC are also lipoxidized. Considering that lipoxidation of a single protein can inhibit the UPS (Shringarpure et al., 2000), the simultaneous modification of several critical actors of PQC is likely to result in an acute proteotoxic stress. In agreement with the general effects of protein lipoxidation, DAC treatment triggers ER-stress and inhibition of the UPS (each one fueling the other, [Menéndez-Benito et al., 2005]). This leads to the early activation of the three UPR pathways (Figure 4G). In addition to unfolded proteins accumulation, UPR activation could also be the result of HSPA5/GRP78/BiP lipoxidation and/or tethering of multiple proteins to ER membranes. Finally, (S)-DACs also quickly induce mitochondrial fission and caspase-dependent apoptosis (Figure 6, see Figure 6—figure supplement 1 for a graphical abstract). Hundreds of cytotoxic natural compounds have one or several alkynylcarbinol motifs, therefore the mechanism of action identified here could be shared in its principles by these molecules. This is supported by the fact that HSD17B11 was recently identified as mediating the toxicity of the natural compound dehydrofalcarindiol (Grant et al., 2020), produced in several plants. It is also noteworthy that two other related lipidic natural products, falcarinol and callyspongynic acid, isolated respectively from several Apiaceae plant species and the marine sponge Callyspongia truncata, have been identified as covalently binding to cellular proteins (Heydenreuter et al., 2015; Nickel et al., 2015), suggesting that they could also be bioactivated into protein-reactive species by a yet unidentified mechanism. Consequently, we propose that alkynylcarbinol-containing natural molecules provide a defense mechanism through bioactivation by specific SDRs in the body of predators, pathogens or parasites. In agreement, fulvindione, the AAC-oxidized form of the cytotoxic Haliclona fulva-produced fulvinol, was found in the body of the dorid nudibranch Peltodoris atromaculata feeding on Haliclona (Ciavatta et al., 2014). The presence of multiple alkynylcarbinol motifs in some of these natural compounds could provide modular pro-cytotoxic agents bioactivable by SDRs in different organisms. The fact that HSD17B11 mediates the toxicity of several natural compounds of unrelated origins questions the reasons for it being the common bioactivating enzyme of these prodrugs. The literature on HSD17B11 is currently limited. It was shown in vitro to promote androgen inactivation by converting the potent androstan-3-alpha,17-beta-diol into a weaker androgen (Brereton et al., 2001). In human, the HSD17B11 variant (rs9991501; Arg283Gln) was found associated with lean body mass (Zillikens et al., 2017), supporting that its activity controls muscle physiology, a process regulated by androgens. However, which HSD17B11-generated metabolites are involved and how the Arg283Gln variant impacts on HSD17B11 activity is currently unknown. HSD17B11 is nearly ubiquitously expressed (Brereton et al., 2001; Chai et al., 2003), but is also overexpressed in some human malignancies, including advanced prostate cancer and non-small cell lung cancer cell lines, as compared to normal tissues or cell lines (Grundner-Culemann et al., 2016; Nakamura et al., 2009). However, the potential role for HSD17B11 in cancer progression remains to be established. The HSD17B11-bioactivated clickable prodrugs described here could represent valuable tools to investigate HSD17B11 enzymatic activity in tissues through imaging and to decipher its physiological and therapeutic relevance.

Figure 6 with 1 supplement see all

Download asset Open asset

Model depicting the stereospecific bioactivation of alkynylcarbinol-containing compounds by specific SDRs into cytotoxic protein-reactive species.

The protein reactive species generated upon bioactivation modify several proteins including the essential 26 S proteasome subunit PSMD2, thereby triggering Ub-proteasome system (UPS) inhibition, ER-stress, activation of the Unfolded Protein Response (UPR) and cell death mediated by apoptosis. See Figure 6—figure supplement 1 for a graphical abstract depicting the main findings and techniques used.

Our work also provides a proof-of-concept that new pro-cytotoxic agents can be designed to be bioactivated through an enantiospecific oxidation catalyzed by selected SDRs. We exemplified this with three different human dehydrogenases, HSD17B11, RDH11 and HPGD (Figure 6). We also show here that multiple human cell lines derived from osteosarcoma, a rare pediatric cancer, were particularly sensitive to (S)-DACs (Figure 1—figure supplement 4C), suggesting that HSD17B11- and more generally SDR-bioactivable prodrugs could find anticancer applications, especially considering that 71 different SDRs are found in humans (Bray et al., 2009). For example, HSD3B1-specific pro-drugs could prove useful to treat castration-resistant prostate cancers, since the stabilized N367T variant of this SDR, found in ~30% of the population (rs1047303), has been associated to resistance to androgen deprivation therapies by allowing self-sufficient production of pro-proliferative androgens by prostate cancer cells (Chang et al., 2013; Hearn et al., 2020; Thomas and Sharifi, 2020). Exploring these potential applications in vivo will be the focus of future studies.

Of note, another family of diyne molecules, the enediynes, are already used in anticancer treatments in particular its best known representative, calicheamicin-gamma1 (cali), extracted from the soil bacteria Micromonospora echinospora (Lee et al., 2002). Cali shows a cytotoxic activity in the picomolar range that is exploited by two anticancer treatments, gemtuzumab ozogamicine (Mylotarg) and inotuzumab ozogamicin (Besponsa), consisting respectively of a CD33-cali or CD22-cali antibody-drug conjugates used for the treatment of acute myeloid leukemia (Williams et al., 2019). In the enediyne motif, the equivalent position of the DAC carbinol unit >CHOH is replaced by an ethylene unit -CH=CH-. Cali is also a prodrug which is activated by the intracellular reducing environment triggering the reduction, mainly carried by cellular glutathione, of a trisulfide unit to a thiol group, initiating a cascade process by an intramolecular Michael addition leading to the DNA-damaging biradical species generated by Bergman cyclization. This, together with the binding of cali to the minor groove of DNA through its aryloligosaccharide moiety, allows the reaction of cali with both DNA strands by promoting hydrogen abstraction on the deoxyribose part of DNA, leading to cytotoxic DNA double-strand breaks (Zein et al., 1988).

The SDR superfamily is one of the largest, with representatives in all known life forms, except viruses: 507 673 SDRs are currently identified in the Uniprot database (in 7546 species, Pfam ID: PF00106). A bioinformatic classification of SDRs has revealed that among the 314 SDR subfamilies identified (Kallberg et al., 2010), half are specific to bacteria. This offers the prospect of designing antibiotics with novel mechanisms of action. In addition, other oxidoreductases, especially among the Medium-Chain dehydrogenase/Reductase (MDR) superfamily, could in principle carry out similar oxidation of secondary alcohol-containing substrates (Persson et al., 2008), which could widen the field of application of our findings. Conversely, some carbonyl reductases, especially among the Aldo-Keto Reductase (AKR) superfamily (Oppermann, 2007; Penning et al., 2021), could theoretically antagonize DACones cytotoxicity by reverting the reactive ketone into the corresponding alcohol, which could be further conjugated and eliminated. However, the fact that modulating glutathione levels or GST activity in cells only had a very small impact on (S)-DAC cytotoxicity (Figure 3—figure supplement 3D,E) indicates that, once produced, DACones immediately react with nearby proteins which limits the activity of general detoxification mechanisms.

Finally, our findings extend the toolbox of protein-reactive warheads with the identification of the DACone motif as a potent reactive group toward the thiol and amino groups of the side chains of cysteine and lysine, respectively. This reaction is quantitative, without co-product and operating in aqueous buffer at room temperature, making it well suited for many applications. In our study, multiple proteins could be functionalized by a terminal alkyne motif using clickable DACone 10 or 11 and the resulting linkage was found to be highly stable at neutral pH, while a selective cleavage of the DACone-amine linkage was observed at acidic pH (Figure 2—figure supplement 1H,I), a property that could be exploited. The reactive species could also be readily produced in situ by using the appropriate SDR with its co-factor (see Figure 2F). Our study also provides the first insights into the proteome-wide reactivity of these novel electrophiles using the isoDTB-ABPP approach (Zanon et al., 2021), which was also applied to demonstrate the formation of the DACone-protein adducts in cells (Supplementary file 4A). Reactive lysines and cysteines have been mapped on functional domains of human proteins (Abbasov et al., 2021; Hacker et al., 2017; Weerapana et al., 2010) and could be targeted through exploiting this new reactive entities supplied exogenously or generated in situ, thereby offering new avenues for protein functionalization or covalent inhibitors design.

Materials and methods

Plasmids

Detailed information regarding the plasmids generated in this study are provided in Supplementary file 5A. Plasmids generated in this study have been produced (i) by ligating annealed primers into BbsI digested plasmids (for cloning of sgRNAs) or (ii) by ligating digested PCR-amplified cDNAs into dephosphorylated plasmids. Phusion DNA polymerase, FastDigest restrictions enzymes and FastAP phosphatase were used (Thermo Fisher Scientific). For all plasmids, DNA sequencing (Mix2Seq, Eurofins Genomics) was used with the specified primers, to confirm that the desired sequence was inserted. All new plasmids have been deposited on Addgene. The plasmids ERmox-GFP (Addgene #68072, Costantini et al., 2015), Ub-G76V-YFP (Addgene #11949, Menéndez-Benito et al., 2005), mCherry-Mito-7 (Addgene #55102, Olenych et al., 2007), pEGFP-N1-ATG-FLAGC (Addgene #60360, Britton et al., 2014), pEGFP-C1-FLAGN (Addgene #46956, Britton et al., 2013), pICE-EGFP-FLAG-Ku70siR-WT (Addgene #46961, Britton et al., 2013), and pCAG-eSpCas9-2A-GFP (Addgene #79145) were provided by Addgene thanks to Addgene contributors.

Oligonucleotides

DNA oligonucleotides used in the study are described in the Supplementary file 5B and were ordered from Eurofins Genomics.

Share this article

Cite this article

HSD17B11 is necessary for DAC (S)–3 cytotoxic activity.

Figure 1—source data 1

Figure 1—source data 2

DACones are protein reactive species.

Figure 2—source data 1

Figure 2—source data 2

Figure 2—source data 3

(S)-DACs lipoxidize multiple cellular proteins, triggering their association with cellular membranes.

Figure 3—source data 1

Figure 3—source data 2

Figure 3—source data 3

DAC (S)–3 triggers ER-stress, ubiquitin-proteasome system inhibition and apoptosis.

Figure 4—source data 1

Figure 4—source data 2

Figure 4—source data 3

Figure 4—source data 4

This video corresponds to a time lapse of pictures of U2OS cells stably expressing ER-moxGFP and Mito-mCherry and treated with 1 µM (S)-.

Bioactivation of other lipidic alkynylcarbinols by specific SDRs.

Figure 5—source data 1

Figure 5—source data 2

Figure 5—source data 3

Model depicting the stereospecific bioactivation of alkynylcarbinol-containing compounds by specific SDRs into cytotoxic protein-reactive species.

Author details

Pascal Demange

Contribution

Contributed equally with

Competing interests

Etienne Joly

Contribution

Contributed equally with

Competing interests

Julien Marcoux

Contribution

Contributed equally with

Competing interests

Patrick RA Zanon

Contribution

Competing interests

Dymytrii Listunov

Contribution

Competing interests

Pauline Rullière

Contribution

Competing interests

Cécile Barthes

Contribution

Competing interests

Céline Noirot

Contribution

Competing interests

Jean-Baptiste Izquierdo

Contribution

Competing interests

Alexandrine Rozié

Contribution

Competing interests

Karen Pradines

Contribution

Competing interests

Romain Hee

Contribution

Competing interests

Maria Vieira de Brito

Contribution

Competing interests

Marlène Marcellin

Contribution

Competing interests

Remy-Felix Serre

Contribution

Competing interests

Olivier Bouchez

Contribution

Competing interests

Odile Burlet-Schiltz

Contribution

Competing interests

Maria Conceição Ferreira Oliveira