:: разработка системы запретов на использование кинжалов на основе капсаицина для модуляции взаимодействия киназы A/PHB2 и митофагии в раковых клетках.

Introduction

Mitochondria are in constant equilibrium between turnover and biogenesis. To be able to metabolically sustain an uncontrolled cell proliferation in cancer cells, mitochondria are frequently hijacked to produce significant amounts of ATP 1 . In parallel, damaged mitochondria can be eliminated through a process called mitophagy 2 , 3 . Mitophagy is controlled by several proteins, including the scaffold protein Prohibitin-2 (PHB2), which binds to the autophagy mediator LC3 4 . PHB2 and its homolog, Prohibitin-1 (PHB1), form a heterodimeric complex located on the Inner Mitochondrial Membrane (IMM), and constitute a hub for many signaling pathways beyond mitophagy in cancer cells 5 .

Recently, the overexpression of the cancer-related protein Aurora kinase A/AURKA was shown to induce mitophagy by functionally interacting with PHB2 6 . AURKA regulates a wide variety of mitotic and non-mitotic functions at different subcellular compartments, including mitochondrial dynamics and ATP production 7 – 11 . We previously showed that the AURKA-dependent phosphorylation of PHB2 on Ser39 induces the formation of an AURKA-LC3-PHB2 tripartite complex 6 . We also showed that this tripartite complex triggers mitophagy in a PARK2/Parkin-independent manner 6 . AURKA-dependent mitophagy is used to eliminate defective mitochondria in cancer cells, while selecting a population of hyperfunctional organelles with high ATP production levels. We also found that the natural polyphenol xanthohumol acts as a PHB2 ligand on the inner membrane of mitochondria to destabilize the AURKA-LC3-PHB2 tripartite complex. By blocking the interaction between AURKA and LC3, treatment with xanthohumol results in mitophagy inhibition. Under these conditions, the AURKA-dependent mitochondrial ATP overproduction is also abolished, thereby linking effective mitophagy and the capacity of mitochondria to increase their ATP production rates 6 . However, the polyphenolic structure of xanthohumol hampers its development as an anticancer agent.

As part of our ongoing programs on the development of PHB ligands 12 – 16 , we here screen already-available alternative PHB ligands in their capacity to alter the protein-protein interaction between AURKA and PHB2. We then optimize the activity of the most promising hit —capsaicin—to obtain 16 analogs targeting the AURKA/PHB2 interaction. We then illustrate that these ligands differentially affect the capacity of AURKA to interact with PHB2. Using a combination of quantitative fluorescence microscopy and molecular docking, we identify two PHB ligands dramatically reinforcing the interaction between AURKA and PHB2. By binding both to the AURKA active site and the inhibitory pocket of PHB2, we show that compounds 12 and 13 (N-(4-hydroxy-3-methoxybenzyl)butyramide) behave as inhibitors of this complex at mitochondria. We renamed N-(4-hydroxy-3-methoxybenzyl)butyramide HMBB, and we show that it is specific toward the mitochondrial roles of AURKA. While it blocks AURKA-dependent mitophagy, it does not deactivate the kinase at centrosomes. These features make HMBB a promising target to counteract the effects of AURKA overexpression at mitochondria.

Results

Screening of available PHB ligands identifies capsaicin as a compound lowering the interaction between AURKA and PHB2

As an alternative to xanthohumol 6 , we selected three PHB ligands with different chemical structures and with different functional consequences on PHB2 signaling pathways, such as FL3, fluorizoline, and capsaicin 5 (Fig. 1 ).

We then compared the capacity of FL3, capsaicin, and fluorizoline to alter the protein-protein proximity between AURKA and PHB2 using Förster’s Resonance Energy Transfer coupled to Fluorescence Lifetime Imaging Microscopy (FRET/FLIM) in live cells 6 , 17 , 18 . In these experiments, AURKA was fused to GFP and was used as a FRET donor, while PHB2 was fused to mCherry and was used as a FRET acceptor. FRET was represented using ΔLifetime, which corresponds to the net difference between the lifetime of AURKA-GFP in the absence (AURKA-GFP + Empty vector, hereby donor-only condition) and in the presence of the acceptor (AURKA-GFP + PHB2-mCherry) (Fig. 2 ). MCF7 breast cancer cells expressing AURKA-GFP in the presence or absence of PHB2-mCherry were incubated for 6 h with each of the three compounds, or with the vehicle (DMSO) prior to FRET measurements.

Treatment with FL3 did not significantly impact the interaction between AURKA and PHB2 in FRET/FLIM analyses. In contrast, treatment with capsaicin or fluorizoline brought the ΔLifetime of AURKA-GFP in the presence of PHB2-mCherry closer to that of their respective donor-only counterparts (Fig. 2 ). This indicates that both compounds significantly lower the interaction between AURKA and PHB2. However, fluorizoline also displayed a significant green autofluorescence that made image-based analyses difficult. Thus, we selected capsaicin as a promising compound for the development of PHB ligands that could differentially modulate the AURKA/PHB2 interaction. Capsaicin did not promote a nuclear localization of PHB2 in MCF7 cells, in contrast with a previous report in HeLa cells 19 . Instead, our FRET/FLIM analyses corroborate previous evidence, showing that PHB2 could only be located at mitochondria in MCF7 cells 6 .

Our data show that capsaicin mildly lowers the protein-protein interaction between AURKA and PHB2 at mitochondria. We next synthesized a series of capsaicin derivatives, and we evaluated their effects on the AURKA/PHB2 interaction.

Chemical synthesis of capsaicin derivatives

After comparing several of methods of acylation, we found that Schotten-Baumann acylation allows the amine of vanillylamine 1a to be selectively acylated without affecting the phenol as described by Short and coll (Fig. 3A ) 20 . With this process, we synthesized 16 capsaicin derivatives potentially behaving as PHB ligands (Fig. 3B ). After the synthesis was completed, we then screened these compounds to explore whether they could alter the AURKA/PHB2 interaction at mitochondria.

Capsaicin derivatives differentially affect the interaction between AURKA and PHB2

To evaluate the effects of the 16 newly-synthesized capsaicin derivatives (Fig. 3B ) toward the AURKA/PHB2 interaction, MCF7 cells co-expressing AURKA-GFP and PHB2-mCherry were treated for 6 h with each capsaicin derivative or with the DMSO control compound. Their effects on the interaction between AURKA and PHB2 were then evaluated using FRET/FLIM. The dynamic range of a GFP-mCherry chimeric tandem construct shows a maximum of 250 psec of ΔLifetime when compared to a GFP donor-only condition 17 . Given that the AURKA-GFP/PHB2-mCherry pair in the DMSO condition shows a mean ΔLifetime of 120–150 psec per se, we explored whether the 16 capsaicin derivatives could induce variations to these mean ΔLifetime values in cells co-expressing AURKA-GFP and PHB2-mCherry.

To this end, we calculated the mean ΔLifetime difference from the AURKA-GFP/PHB2-mCherry pair in the DMSO condition for each capsaicin derivative syntesized (Table 1 ). Variations were considered mild when a given compound induced changes in the AURKA-GFP ΔLifetime of ±30–50 additional psec compared to what was observed in the AURKA-GFP/PHB2-mCherry pair in the DMSO condition. This corresponds to 12–25% of the maximum ΔLifetime response of a GFP-mCherry tandem 21 . Alternatively, variations were considered significant when a given compound induced changes to the AURKA-GFP ΔLifetime which were superior to ±50 additional psec compared to what was observed in the DMSO condition. This is more than 25% of the maximum ΔLifetime response of a GFP-mCherry tandem 21 .

The replacement of the acyl moiety of capsaicin by various linear and unsaturated fatty acids in compounds 1-7 (6–18 carbons) was detrimental to the FRET effect. Indeed, compounds 1, 4, 6 , and 7 showed no effect on AURKA-GFP ΔLifetime when AURKA-GFP and PHB2-mCherry were co-expressed. Compounds 2, 3 , and 5 mildly lowered the AURKA/PHB2 interaction, but with a standard deviation superior to that of capsaicin (Table 1 ).

Unexpectedly, the introduction of a 5 or 6-membered ring in the hydrophobic acyl chain resulted in an opposite pharmacological activity. Indeed, the constrained analogs 8, 9 , and 10 mildly increased the interaction between AURKA and PHB2. Of note, these compounds also show a lower degree of experimental variability than the previously-tested compounds in our FRET/FLIM analyses (Table 1 ). Similar to compounds 8-10, N -vanillylgeranamide/compound 11 displayed some rigidification as well. However, this compound promoted AURKA/PHB2 interaction to a greater extent than the previous analogs (Table 1 ). Shortening the hydrophobic chain into a crotonic group, as in compound 12 also induced significant changes of more than 50 psec in the AURKA-GFP ΔLifetime, thus increasing the interaction between AURKA and PHB2 compared to that observed in the DMSO condition (Table 1 ). Nevertheless, both compounds 11 and 12 displayed a high standard deviation. Switching to the more flexible butyryl group, as in compound 13 , was another suitable strategy to significantly increase the AURKA/PHB2 interaction. Furthermore, compound 13 displayed a lower standard deviation than compounds 11 or 12 (Table 1 ). Thus, we chose to modify this analog to explore the vanillyl moiety. The methylation of the phenolic moiety, leading to compounds 14 and 15 left the interaction between AURKA and PHB2 unaltered (Table 1 ). On the contrary, its replacement by an aniline as in compound 16 increased the interaction between the two proteins (Table 1 ). This indicates the requirement for a H-bond donor in this position.

To exclude that the capsaicin derivatives could significantly perturb the lifetime of AURKA-GFP per se, each compound was evaluated in the donor-only condition where AURKA-GFP was co-expressed with an empty vector (Supplementary Table 1 ). Previous reports demonstrated that a ΔLifetime difference of around 50 psec from the mean GFP value corresponds to the variability of lifetime differences calculated with our FLIM setup 17 , 18 . Therefore, a variability in the intrinsic AURKA-GFP ΔLifetime values above 50 psec from the mean AURKA-GFP value may indicate that the tested ligand intrinsically affects the lifetime of AURKA-GFP. Globally, the majority of the compounds tested showed a non-significant variation of the AURKA-GFP ΔLifetime in the donor-only condition. Nevertheless, we observed that compound 16 shows a slightly higher variability than the other derivatives under these conditions. This suggests that the differences in ΔLifetime observed upon the co-expression of AURKA-GFP and PHB2-mCherry (Table 1 ) in cells treated with compound 16 could be partially independent of the FRET effect, and could be explained by an intrinsic effect of the compound 16 on the donor lifetime. Therefore, we decided not to use compound 16 for follow-up analyses in living cells.

Overall, our FRET/FLIM results show that the newly-generated capsaicin derivatives differentially modulate the interaction between AURKA and PHB2 at mitochondria in live cells. While the majority had either no effect or a mild effect on this interaction, we determined that compounds 11, 12 , and 13 increase the AURKA/PHB2 protein-protein proximity (Table 1 ). Among these, compound 13 significantly increases the AURKA/PHB2 interaction with a low standard deviation (Table 1 and Fig. 4A–C ). This makes this capsaicin derivative the most promising hit among the 16 compounds generated here. Therefore, we further explored its capacity to modulate the AURKA/PHB2 interaction. Washout analyses revealed that compound 13 strongly stabilizes the interaction between the two proteins even after the drug has been removed. Indeed, 3 h washout of the drug still resulted in an increased interaction when compared to control cells co-expressing AURKA-GFP and a mutated, non-fluorescent PHB2-mCherry control construct known as Dark mCherry (mCherry Y72C 22 , Fig. 4D and Supplementary Fig. 1A–C ). While AURKA-GFP and PHB2-mCherry show ∼7% FRET efficiency in control conditions, compound 13 almost doubles this efficiency (∼13%) both before and after washout (Supplementary Fig. 1D, E ). This further corroborates the role of compound 13 in strengthening this protein/protein interaction. We also observed that compound 13 is equally effective at 50 µM and 10 µM, while its capacity to stabilize the AURKA/PHB2 interaction significantly decreases at 5 µM (Fig. 4E ).

Altogether, compound 13 stabilizes the AURKA/PHB2 interaction in MCF7 cells using FRET/FLIM approaches. Therefore, insights at the molecular level are needed to elucidate the binding site of compound 13 to the two proteins.

Capsaicin and compound 5 induce a molecular clash lowering the AURKA/PHB2 interaction

After observing that compound 13 increases the AURKA/PHB2 interaction, we sought to uncover its binding mechanism to AURKA and PHB2. To this end, we used molecular docking approaches and compared a total of four molecules with opposite FRET/FLIM efficiencies. On one side, we selected capsaicin and compound 5 , which mildly lower the AURKA/PHB2 interaction. On the other side, we selected compounds 12 and 13 , which show the highest FRET/FLIM efficiencies.

First, a reverse docking approach was used to discover possible interesting protein targets for all the selected compounds. Reverse docking results performed by GalaxySagittarius-AlphaFold ( Supplementary Data ) highlighted that the four compounds tested were predicted to bind AURKA, even if with different scores. Capsaicin and compound 5 could bind AURKA with a Galaxy best final score of 0.573 and 0.498, respectively. Conversely, compounds 12 and 13 could bind AURKA with higher Galaxy best final scores, which were of 0.765 and 0.846, respectively. Interestingly, the key capsaicin target Transient Receptor Potential cation channel subfamily V member 1 (TRPV1) 23 , 24 could be retrieved only in the presence of capsaicin and compound 5 for the “Binding compatibility prediction” search mode (Table 2 ). TRPV1 could be retrieved in the list of targets of compound 12 , but only with the “Similarity combined prediction” search mode. TRPV1 also showed a score significantly lower than the score of AURKA, and comparable to that of LC3, which occupies the lowest position on the list (Table 2 ). This may suggest a higher specificity of compound 12 and 13 , which could result in a superior affinity and/or selectivity for AURKA.

Second, we analyzed the affinity of capsaicin and of compound 5, 12 and 13 for AURKA and PHB2. To this end, we performed rigid and flexible docking analyses to detect possible relevant binding sites on the activated (pThr288) AURKA structure bound to ADP and Mg 2+ , on the PHB2 model, and on the tripartite AURKA/PHB2/LC3 complex model (Supplementary Table 2 ). The tripartite AURKA/PHB2/LC3 complex model was built by exploiting rigid protein-protein docking and homology modelling procedures. The best model obtained presents the Ser39 of PHB2 6 oriented toward the AURKA active site (Supplementary Fig. 2 ). This model also shows the region of LC3 involved in recognition of the LC3-Interacting Region (LIR) binding motif on PHB2 facing the LC3 binding motif YQRL (residues 121-124), and toward the putative LC3 binding motif FNRI (residues 52–55) (Supplementary Fig. 2 ). Binding energy is equal to −12.6 kCal/mol for the AURKA/PHB2 interaction, −10.8 kCal/mol for PHB2/ LC3 interaction, and −1.4 kCal/mol for the AURKA/LC3 interaction. This last value could be due to the very small interface area between AURKA and LC3, which is substantiated by low FRET/FLIM efficiencies in a previous report 6 . Of note, the present model lacks the first 125 residues of AURKA, which are a well-known disordered region. AURKA, LC3, and PHB2 present Z-scores equal to −7.1, −6.6, and −5.91, respectively, in line with the original structures having Z-score values of −7.06, and −6.54 for the crystallographic structure of AURKA and LC3, respectively, and −5.75 for the PHB2 model. The Ramachandran plot of the trimeric complex model obtained presents the following values: 97.8% of dihedral angles in permitted regions of the plot, 4.3% in regions allowed, and 0% in generously allowed and disallowed regions.

After generating the model of the tripartite complex (Supplementary Fig. 2 ), we then evaluated the effects of capsaicin and compound 5 on this model. Docking results for compound 5 and capsaicin highlighted a similar behavior of these two compounds (Table 3 ).

Blind and focused, rigid and flexible docking analyses highlighted that compound 5 and capsaicin bind a similar area in the active site of AURKA, both with two different possible conformations (Supplementary Table 2 ). Compound 5 shows interaction with ADP and Mg 2+ in its best conformation, and binding the AURKA active form. However, it should be noted that only few poses are present in the two clusters, potentially indicating a low selectivity for this area. Both conformations are comparable to the best energetic pose of capsaicin (Supplementary Fig. 3 ). While compound 5 binds exclusively to the cofactors, capsaicin appears to interact with the cofactor and with the active site residue D256. Interestingly, capsaicin binds the AURKA active site with a very low binding energy equal to -7.88 kCal/mol and a constant of inhibition (Ki) of 1.66 µM, and compound 5 does so with an energy value of −8.83 kCal/mol and a Ki of 336.01 nM. All the poses detected suggest a possible inhibitory role of these molecules on the AURKA active site. With this information, we sought to visualize possible steric clashes due to the presence of the ligands. Clashes are conformations occurring when two atoms, which are not covalently bonded to each other, are found to be impossibly close to each other. All the structures of the monomeric proteins bound to the ligands were superimposed on the model of the tripartite complex. The overlap highlighted a possible clash between the ligand—either capsaicin or compound 5 —and PHB2 (Fig. 5A ), suggesting that the binding of either ligand to AURKA could weaken the AURKA/PHB2 interaction.

To verify this hypothesis, additional docking simulations were performed on the tripartite complex model. Docking was focused at the interface areas of the three proteins. As expected, compound 5 was found to change its localization on AURKA, with a further 1 kCal/mol loss of its binding energy and reaching values of −7.78 or of −7.10 kCal/mol. Two binding sites were observed to reach the best binding energy with compound 5, and involve the AURKA K141-K143-Y219-K258-E260-W277-G291-T292 residues, the cofactors ADP-Mg2+ and PHB2 V32-G35-V36-S39-Q59-F70-I72-Q76 residues. They also involve AURKA F144-E168-K171-A172, PHB2 A31-Y34-E38-R71-W74-P73-Y77, and LC3 R14-T54 residues (Fig. 5B ). Similarly to compound 5 , capsaicin is able to interact with cofactors in the active site (Fig. 5C ). However, capsaicin is forced to move into a more dislocated binding site with the best binding energy of −7.62 kCal/mol. Alternatively, capsaicin is forced to adopt an alternative conformation similar to the second one already observed when binding to AURKA, and showing the lowest binding energy equal to −7.10 kCal/mol (Fig. 5C ).

We then analyzed the effects of capsaicin and compound 5 on PHB2. Docking analyses performed on the PHB2 model were focused on the most organized tertiary structure area of this protein. Results reveal a preference of capsaicin and compound 5 for the inhibitory site of the protein, originally detected when studying the PHB2 inhibitor YL-939 25 . Both capsaicin and compound 5 appear to bind to PHB2 by interacting with the same residues, P78-I79-I80-D82-L126-D127-Y128-E129. The binding energies of these compounds with PHB2 are 5.91 kCal/mol and −5.29 kCal/mol for capsaicin and compound 5 . They are higher than the binding energies measured for the two compounds in complex with AURKA, indicating that capsaicin and compound 5 could bind PHB2 with less affinity than for AURKA. However, it should also be noted that the number of poses in the cluster is higher for capsaicin in complex with PHB2 than with AURKA. Despite its lower binding affinity toward PHB2, the number of poses still indicates a good selectivity of capsaicin for the inhibitory site of PHB2.

Overall, capsaicin and compound 5 appear to move from the best energy binding pose to a second pose on AURKA alone. These compounds have a steric hindrance that induces a steric clash at the interface between AURKA and PHB2, potentially lowering or destabilizing the interaction between these two proteins. The molecular docking data thus substantiate the mild effect of capsaicin and of compound 5 in lowering the AURKA/PHB2 interaction observed with FRET/FLIM experiments.

Compounds 12 and 13 create a molecular glue between AURKA and PHB2, and they bind to the AURKA active site and the PHB2 inhibitory pocket

Given the previous results on capsaicin and compound 5 , we performed similar binding and docking analyses for compounds 12 and 13 .

Compounds 12 and 13 reach the best binding energy with the active conformation of AURKA, and they appear to specifically bind the substrate-binding site of the kinase (Table 4 ). Different from compound 5 , both molecules appear to interact with the D256 residue of the active site and with the cofactors. When observing the structure of the two compounds bound to the tripartite complex, no steric clash between AURKA and PHB2 is observed (Fig. 5D ). On the contrary, this superposition rather suggests a possible action of compounds 12 and 13 as a bridge between AURKA and PHB2. The docking performed on the complex also revealed that compounds 12 and 13 show a preference for the AURKA binding site (Fig. 5D ). The binding site is preserved, and no clashes occur to alter the binding. The two compounds appear to adapt their orientation in the pocket to preserve the binding and stabilize the complex. Compound 12 preserves its interaction with AURKA Q177, D274, W277 residues, ADP, and Mg 2+ . Compound 13 preserves its interaction with AURKA Q177, E181, D256, K258, N261, D274, W277 residues, and with ADP and the two Mg 2+ ions. Furthermore, the two compounds show an interaction with PHB2. Compound 12 shows interaction with V32, and compound 13 shows interaction with S39, F51, F70, R71, I72, and Q76 in one of the two alternative binding conformations. Binding energies for the compounds binding to the complex or AURKA alone are comparable, with −6.67 kCal/mol and −6.73 kCal/mol for compound 12 bound to the tripartite complex or AURKA alone, and with −6.74 kCal/mol and −6.77 kCal/mol for compound 13 bound to the complex or AURKA alone. This is additional evidence that the binding site does not undergo changes with these two compounds. In addition, the higher number of poses in the cluster suggests that the selectivity of these compounds for the AURKA active is higher than what measured for capsaicin and compound 5 .

As performed for capsaicin or compound 5 , we explored the effects of compounds 12 and 13 on PHB2 (Table 4 ). Both compounds appear to bind the D82-I80-D127 inhibitory site of PHB2, as another PHB2 ligand, YL-939 25 . The two compounds bind to this site with comparable mean binding energy (−4.81 kCal/mol and −4.68 kCal/mol for compounds 12 and 13 , respectively). The number of poses in each docking cluster is higher for compounds 12 and 13 than for compound 5 . This supports a higher selectivity of the former compounds for the inhibitory pocket than of the latter. The binding energies detectable are lower than those obtained for the AURKA active site. Despite this, the selectivity for the inhibitory pocket of PHB2 is higher, with almost 20 poses out of 100 detected in the same binding area.

Overall, the conformation of compounds 12 and 13 bridges AURKA and PHB2 and brings them close to each other. These compounds potentially act as a molecular glue 26 , stabilizing the interaction between AURKA and PHB2. In addition, docking analyses show that these two compounds not only bind to the AURKA active site but also to the PHB2 inhibitory pocket. Therefore, molecular docking data substantiate the effect of compounds 12 and 13 in strengthening the AURKA/PHB2 interaction observed with FRET/FLIM experiments.

Compound 13 is specific to the mitochondrial pool of AURKA, and it inhibits AURKA-dependent mitophagy

After modeling the binding of compound 13 to the kinetic pocket of AURKA, we asked whether this molecule was specific to the mitochondrial pool of the kinase. To this end, we first evaluated whether compound 13 could inhibit AURKA activation at alternative known locations of the kinase, such as centrosomes. AURKA activation was evaluated in MCF7 cells expressing the GFP-AURKA-mCherry FRET biosensor 27 . This probe detects the conformational changes of the kinase upon its activation by autophosphorylation on Thr288 28 – 30 . MLN8237 is widely known to inhibit AURKA activation regardless of its location, including at centrosomes and mitochondria 9 , 27 , 31 . As expected, MLN8237 lowered AURKA activation at centrosomes (Fig. 6A ). Conversely, compound 13 did not induce any significant FRET perturbation at centrosomes, demonstrating that this molecule is not effective in altering AURKA autophosphorylation at this location (Fig. 6A ).

Compound 13 was renamed as HMBB after its chemical nomenclature N-(4-hydroxy-3-methoxybenzyl)butyramide, and we evaluated its capacity to inhibit AURKA-dependent mitophagy. We previously demonstrated that AURKA-dependent mitophagy is accompanied by a loss of mitochondrial mass, which is calculated by measuring the area covered by the mitochondrial matrix marker PMPCB and normalized over the total cell area 6 . We also showed that this process is PHB2-dependent, and that compounds strengthening the AURKA/PHB2 interaction, such as xanthohumol, can efficiently block AURKA-dependent mitochondrial clearance 6 .

MCF7 cells expressing AURKA-GFP and treated with DMSO showed a significant mitochondrial mass loss when compared to control cells (Fig. 6B ). On the contrary, treatment with HMBB led to an increase in the area covered by PMPCB in cells overexpressing AURKA (Fig. 6B ). This is indicative of an increase in mitochondrial mass levels, which are similar to those observed in control cells not overexpressing AURKA. Treatment with HMBB in control cells did not affect mitochondrial mass levels per se (Fig. 6B ). Therefore, HMBB significantly inhibited AURKA-dependent mitophagy in cells overexpressing the kinase. Accordingly, HMBB increased mitochondrial mass levels in triple-negative T47D breast cancer cells endogenously overexpressing AURKA 9 , 32 (Fig. 6C ). This is in agreement with the previously-described role of AURKA in regulating mitochondrial mass levels upon overexpression in both MCF7 and T47D cells 6 . Conversely, HMBB had no effect in triple-negative Hs578T cells that have low endogenous AURKA levels (Fig. 6D ) 9 , 32 . Silencing of AURKA or PHB2 in T47D cells revealed that HMBB cannot rescue mitochondrial mass loss when either component is missing (Supplementary Fig. 4 ). This further corroborates the specificity of HMBB in AURKA/PHB2-dependent mitophagy.

In conclusion, our results determined that capsaicin is a pharmacological component altering the AURKA/PHB2 protein-protein proximity. Newly-generated capsaicin derivatives showed differential effects on the capacity of AURKA to interact with PHB2, and they either lower or strengthen the interaction in FRET/FLIM experiments. Molecular docking analyses revealed that capsaicin derivatives stabilizing the AURKA/PHB2 proximity can bind to the AURKA/PHB2/LC3 tripartite complex, such as HMBB or compound 12 . These compounds target the AURKA active site and the PHB2 inhibitory pocket, resulting in enhanced binding features compared to molecules lowering the AURKA/PHB2 interaction. Last, we show that HMBB is specific to the mitochondrial pool of AURKA. While it cannot alter its activation at centrosomes, this molecule can efficiently revert AURKA-dependent mitophagy. Overall, we propose that capsaicin derivatives enhancing the AURKA/PHB2 interaction are promising inhibitors targeting the mitochondrial roles of AURKA.

Discussion

Aurora kinase A/AURKA is a hallmark of epithelial and hematological cancers, where it is frequently overexpressed at both the mRNA and protein levels 8 . It is known that AURKA has multiple subcellular locations, including mitochondria 9 , 10 .

Here, we identified capsaicin as a PHB2 ligand that reduces the AURKA/PHB2 interaction at mitochondria. This compound was previously shown to induce dissociation between PHB and the Adenine Nucleotide Translocator 2 (ANT2, SLC25A5). Proteomics studies suggest that the auto/mitophagy modulator LC3 may interact with ANT2, and in our previous study, we identified ANT2 as an interactor of AURKA using MS/MS-based approaches 9 . In addition, we previously observed that AURKA interacts with LC3 in a PHB2-dependent manner 6 . This raises the interesting possibility that PHB2, AURKA, ANT2, and LC3 may form a quaternary complex in mitochondria.

Capsaicin is known to be cytotoxic in a wide range of cancer cell lines. In addition to PHB2 19 , these effects have been attributed to the action of capsaicin on several targets. These include the neuronal vanilloid receptor TRPV1 23 , 24 , the Tumor-Associated NADH Oxidase class (tNOX/ENOX2) protein, which is ubiquitously expressed in cancer but not in non-transformed cells 33 , the nuclear lysine-specific demethylase 1 A (KDM1A/LSD1) 34 , and the nuclear transcription factor 2 (POU3F2) 35 . Finally, capsaicin was shown to bind to PHB2 and to induce its translocation from the inner mitochondrial membrane to the nucleus in HeLa cells 19 . This also contributes to the induction of apoptosis in human myeloid leukemia cells. Given these multiple targets of capsaicin, it is interesting to note that HMBB and compound 12 are unable to significantly bind to TRPV1 36 . We also observed that capsaicin and compound 5 are the only compounds capable of inducing a cytostatic effect in MCF7 cells, while none of the tested compounds exhibited cytotoxicity (Supplementary Table 3 ). The lack of overall cytotoxicity exerted by capsaicin and its ligands may suggest low expression levels of canonical capsaicin targets in MCF7 cells. The binding of HMBB and compound 12 is shown when the tripartite AURKA/PHB2/LC3 complex is formed, a condition observable only upon AURKA overexpression 6 . While this makes HMBB and compound 12 two more selective PHB ligands than capsaicin, it also suggests that the detection of cytotoxic effects with these compounds may require AURKA overexpression. Alternatively, the lack of cytotoxicity may also be explained by the reverse docking results obtained with GalaxySagittarius-AlphaFold predictions. Predictions uncovered that TRPV1 is a preferential target of capsaicin and compound 5 , whereas it was not a preferential target of HMBB or compound 12 (Table 2 , Supplementary Data ). This leads to the hypothesis that the cytostatic effect of capsaicin and compound 5 might be related to their activity on TRPV1. With HMBB and compound 12 instead, AURKA stood out as a preferential hit (Table 2 , Supplementary Data ). The results on TRPV1 highlight how HMBB and compound 12 , despite being derived from capsaicin, have developed their own selectivity and show a high specificity for other targets such as AURKA. The presence of TRPV1 in the capsaicin list with scores comparable to the best results further corroborates the predictions for the other compounds analyzed. It is important to emphasize that the absence of PHB2 from the GalaxySagittarius list is not related to a potential inability of the compounds to interact with it. Despite the use of GalaxySagittarius in AlphaFold mode, the PHB2 protein model is not present in the reference database. Therefore, evaluating the interaction score between PHB2 and the compounds studied was not possible.

Despite the multiple targets of capsaicin, it is important to note that they are not localized at mitochondria and that the effect of capsaicin at this subcellular compartment was previously unknown. We here provide the first evidence for the capacity of capsaicin and its derivatives to modulate the protein/protein proximity between PHB2 and a cancer-related kinase, such as AURKA, at this compartment. At mitochondria, AURKA was shown to play a variety of roles, including the modulation of mitochondrial morphology, ATP production, and mitochondrial clearance by mitophagy 6 , 9 . We previously showed that tagging the different proteins with fluorescent tags does not alter their intracellular distributions per se 6 , 9 , 27 , 37 . We also showed that the PHB2 ligand xanthohumol globally alters protein-protein interactions within the AURKA-PHB2-LC3 tripartite complex 6 . Indeed, xanthohumol lowers the AURKA/LC3 interaction while concomitantly increasing the AURKA/PHB2 interaction. Altering these tripartite interactions abolishes AURKA-dependent mitophagy and restores normal ATP levels in cancer cells with AURKA overexpression 6 . When looking at the newly-generated capsaicin derivatives, it is interesting to observe that compounds 11 , HMBB, and compound 12 significantly increase the AURKA/PHB2 interaction, as xanthohumol does 6 . Thanks to their low standard deviation compared to that of compound 11 in FRET/FLIM experiments, our data showcase the potential of these capsaicin derivatives for functional studies.

Molecular docking data further corroborate the notion that compounds reinforcing protein-protein interactions by FRET have binding mechanisms compatible with the inhibition of PHB2. All the compounds tested show a preference for a binding site on PHB2 already known to be the target of the YL-939 PHB2 ligand 25 . YL-939 appears to bind PHB2 with two H-bonds involving the D82 and D127 residues, while its distal phenyl group forms hydrophobic interactions with V43 and I80. Mutagenesis experiments performed on D82 and D127 highlighted that these residues are fundamental for inhibitor binding and to reduce the sensitivity to erastin in ferroptosis 25 . Similarly, the binding of capsaicin and compounds 5 , HMBB, and compound 12 to this pocket could exert an inhibitory effect. Moreover, the proximity of the PHB2 inhibition site to the LC3 binding site— YQRL motif, residues 121-124—may also indicate the involvement of these compounds in modulating the binding of LC3 to PHB2.

Interestingly, capsaicin and compounds 5, 12 , and HMBB all display an interaction with AURKA and show binding energy levels that are better than the ones obtained when interacting with PHB2. However, the different size of the molecules and their different hindrance may explain their differential effects on AURKA. First, they could compete for the binding to Ser39 of PHB2, a residue targeted by AURKA and required for AURKA-dependent mitophagy 6 . As a consequence, the functions of PHB2 within or outside the AURKA/PHB2/LC3 complex could also be compromised. In addition, the modeling of the tripartite complex shows that compound 5 has a steric hindrance for PHB2 accommodation, and this in the two conformations detected. Instead, HMBB and compound 12 accommodate at the interface of AURKA and PHB2. This may give rise to a stabilized AURKA/PHB2 interaction, which is corroborated by FRET/FLIM experiments in living cells.

Docking simulations performed directly on the tripartite complex also underline that HMBB and compound 12 can reach the AURKA active site. These analyses are supported by evidence in living cells, where HMBB can specifically block AURKA-dependent mitophagy. Its inability to target AURKA activation at centrosomes strongly suggests that it is a compound specific to mitochondria, and that it may require the simultaneous presence of AURKA and PHB2 to perform its inhibition. We also show that HMBB inhibits AURKA-dependent mitophagy with the same efficacy as xanthohumol or MLN8237 6 . In this light, the greater bioavailability of capsaicin in vivo compared to xanthohumol and the greater mitochondrial specificity of HMBB compared to standard catalytic AURKA inhibitors such as MLN8237 strongly suggest that HMBB and/or capsaicin derivatives with analogous properties may display great selectivity in preclinical or clinical setups.

Our results show that capsaicin derivatives are pertinent chemical tools to alter the formation of the tripartite AURKA/PHB2/LC3 complex both in live cells and at the molecular level. Future studies with a larger number of capsaicin derivatives will elucidate whether the AURKA/PHB2 interaction could be generally considered as a proxy to predict the efficacy of pharmacological compounds toward the mitochondrial functions of AURKA and, ultimately, the proliferation rates of cancer cells with AURKA overexpression. In conclusion, our study identifies a class of bioavailable compounds with the potential to become mitochondrial-specific inhibitors of AURKA.

Materials and methods

Cell culture procedures

MCF7 (HTB-22), T47D (HTB-133), and Hs578T (HTB-126) cells were purchased from the American Type Culture Collection and were kept free from mycoplasma throughout experiments. They were grown in Dulbecco’s Modified Eagle’s Medium (DMEM) already containing 1% L-glutamine (Thermo Fisher Scientific), and supplemented with 10% FBS (Thermo Fisher Scientific) and 1% penicillin-streptomycin (Thermo Fisher Scientific). For all live microscopy experiments, cells were grown at 37 °C either in Nunc Lab-Tek Chamber slides (Thermo Fisher Scientific) or in polymer-coated µ-Slide 8 Well high slides (Ibidi). Before proceeding with FLIM imaging, standard growth media were replaced with phenol red-free Leibovitz’s L-15 medium (Thermo Fisher Scientific) supplemented with 20% FBS and 1% penicillin–streptomycin. Plasmid transfections were performed with Lipofectamine 2000 (Thermo Fisher Scientific), according to the manufacturer’s instructions, and incubated for 48 h before FLIM analyses. pcDNA3 AURKA-GFP, pcDNA 3.1 empty vector, and pmCherry N1-PHB2 were previously described 6 . PHB2-Dark mCherry (mCherry Y72C 22 ) was created with the Quikchange site-directed mutagenesis strategy (Agilent), directly on the pmCherry N1-PHB2 vector using the 5′- CTGTCCCCTCAGTTCATGTGCGGCTCCAAG-3′ and the 5′- CTTGGAGCCGCACATGAACTGAGGGGACAG-3′ forward and reverse primers, respectively. The AURKA conformational biosensor and its donor-only counterpart (GFP-AURKA), originally published in ref. 27 were subcloned into a pcDNA 3.1 vector expressing GFP-AURKA-mCherry or GFP-AURKA under the control of the CMV promoter. The siRNA against AURKA was synthesized and purchased from Eurogenetec (sequence: 5′-AUGCCCUGUCUUACUGUCA-3′), as previously described 27 . Allstars negative control (SI03650318) siRNAs and a functionally-validated siRNA against PHB2 (SI02780918) were purchased from Qiagen. Fluorizoline, FL3, capsaicin, and all capsaicin derivatives were resuspended in DMSO and stored at −20 °C for long-term storage. Each compound was used at a final concentration of 50 µM, and incubated for 6 h for FRET experiments, or 24 h for mitochondrial mass experiments. In dose-response experiments, HMBB (compound 13) was used at 50 µM, 10 µM, and 5 µM, respectively. MLN8237 (Alisertib) was purchased from Euromedex, resuspended in DMSO, and stored at −80 °C for long-term storage. It was used at a final concentration of 100 nM and incubated for 6 h before imaging.

FRET/FLIM microscopy

Unless stated otherwise, FLIM analyses were performed as previously described 27 , with a time-gated custom-built system attached to a Leica DMI6000 microscope (Leica) with a CSU-X1 spinning disk module (Yokogawa) and a 63 oil immersion objective (NA 1.4), a picosecond pulsed supercontinuum white laser at 40 MHz frequency (Fianium), and a High-Rate Intensifier (LaVision) coupled to a CoolSNAP HQ2 camera (Roper Scientific). GFP was used as a FRET donor in all experiments, and it was excited at 480 ± 10 nm. Emission was selected using a bandpass filter at 483/35 nm (Semrock). To calculate fluorescence lifetime, five temporal gates with a step of 2 ns each allowed the sequential acquisition of five images covering a total delay time spanning from 0 to 10 ns. The five images were used to calculate the pixel-by-pixel mean fluorescence lifetime according to the equation: τ = ΣΔ τ i × I i /ΣI i , where Δτ i corresponds to the delay time after a laser pulse of the i th image acquired. I indicates the pixel-by-pixel fluorescence intensity in each image. The FLIM setup was controlled by the Inscoper Suite solution (Inscoper, France), and GFP lifetime was measured in real-time during acquisition. In all experiments, AURKA-GFP lifetime was calculated by the Inscoper software only when the pixel-by-pixel fluorescence intensity in the first gate was above 2000 gray levels. Raw lifetime values were then extracted by analyzing images post-acquisition with the Fiji software (NIH). To calculate ΔLifetime values, the mean lifetime of the cells in the donor-only condition (AURKA-GFP + Empty vector or AURKA-GFP + PHB2-Dark mCherry for experiments reporting on AURKA/PHB2 interaction; GFP-AURKA for experiments with the GFP-AURKA-mCherry biosensor) was calculated and then used to normalize data in all the analyzed conditions and for each independent experiment as previously reported 38 . FRET efficiency was calculated as previously described 17 , 27 , and the mean FRET efficiency of the cells in the donor-only condition (AURKA-GFP + PHB2-Dark mCherry) was calculated and then used to normalize data in all the analyzed conditions and for each independent experiment. FRET/FLIM was also calculated in TCSPC mode (Fig. 4D, E and Supplementary Fig. 1 ), and GFP lifetime was measured with an inverted SP8 Leica confocal microscope (Manheim, Germany) equipped with a single-molecule detection (SMD) module based on a Picoquant hardware solution (Berlin, Germany), a 470 nm pulsed laser with a 40 MHz repetition rate, a 500/50 nm band pass filter, a single-photon avalanche diode (SPAD) detector, and a 63X oil immersion objective (N.A. = 1.4) as in ref. 32 . A Picoharp 300 was used for time-correlation and for image reconstruction using scanning signals to recover FLIM images (time-tagged time-resolved (TTTR) method). Lifetime values in each cell were determined by fitting the fluorescence decay with the built-in Symphotime software. This was performed by integrating the signal from the pixels in a region of interest with a single exponential model.

Immunocytochemistry, confocal microscopy, and mitochondrial mass analyses

Cells were fixed in 4% paraformaldehyde (Euromedex), stained using standard immunocytochemical procedures, and mounted in ProLong Gold Antifade reagent (ThermoFisher Scientific). Mitochondria were stained with a polyclonal rabbit anti-PMPCB (16064–1-AP; Proteintech) used at a 1:500 dilution and a secondary anti-rabbit antibody coupled to Alexa 647 at a 1:5000 dilution (ThermoFisher Scientific). Two-color images of AURKA-GFP and PMPCB-Alexa 647 were acquired with a Leica SP8 inverted confocal microscope (Leica) driven by the Leica Acquisition Suite (LAS) software and a 63x oil-immersion objective (NA 1.4). The excitation/emission wavelengths for GFP were 488 and 525/50 nm and 633 and 650/20 nm for Alexa 647. Mitochondrial mass was determined by calculating the relative mitochondrial area ( A PMPCB ) for each image using the PMPCB staining. A PMPCB was obtained with the Fiji software (NIH) on maximal projections of confocal images acquired as above. As previously reported 6 , A PMPCB was determined as the ratio between the area covered by PMPCB and selected with an automatic threshold mask and the total cell area.

Cell proliferation and viability

MCF7 cells were cultured in Eagle’s Minimal Essential Medium (EMEM, Sigma-Aldrich) supplemented with 10% FBS (Thermo Fisher Scientific), 2 mM L-G lutamine (Sigma-Aldrich), 1 mM sodium pyruvate (Sigma), 100 U/ml penicillin (Sigma-Aldrich), and 100 µg/ml streptomycin (Sigma-Aldrich). Cells were grown at 37 °C in NUNC Edge96 Nunclon Delta Surface 96-well plates. All compounds were incubated for 48 h prior to cell proliferation and viability experiments, at the following final concentrations: 0.1 µM; 0.5 µM; 1 µM; 5 µM; 10 µM, 50 µM with three independent points measured for each concentration. Only data for 50 µM were reported in Supplementary Table 3 . Cellular proliferation was calculated with an IncuCyte S3, driven by the IncuCyte 2023A software, and by measuring cellular confluence. Cellular viability was measured with the WST-1 test (Takara). 450 nm absorbance was recorded using an Envision (Perkin Elmer) plate reader.

Molecular modelling of capsaicin and capsaicin derivatives, structural models of the AURKA-PHB2-LC3 tripartite complex, and molecular docking simulations

The molecular structures of compounds 5, 12, and 13 were built by Discovery Studio 4.5 (Dassault Systèmes BIOVIA) and minimized by Chimera 1.14 39 . The molecular structure of capsaicin was instead downloaded from PubChem 40 (CID: 1548943). The putative protein targets of capsaicin and compounds 5-12-13 were estimated with GalaxySagittarius-AF 41 , exploiting both similarity and compatibility mode, and re-ranked using docking. The structures of Aurora kinase A (AURKA) in its active conformation, and of Microtubule-associated protein 1A/1B light chain 3B (LC3) were retrieved from PDB 42 (PDB code: 5G1X, and 7ELG, respectively). For the 3D-structure of Prohibitin-2 (PHB2), the AlphaFold complete model (identifier: AF- Q99623 -F1) was used 43 . The H-dock software 44 was used to create a model of the AURKA-PHB2-LC3 tripartite complex. The complexes obtained were analyzed in terms of compatibility of the interface orientation with the information on the binding site from the literature. Once the most suitable complex was selected, it was used as a template for Modeller9.22 45 to refine the interactions among the chains of the complex. PDBePISA 46 was used to calculate the binding energy and the interactions established at the interfaces. ProSA-web 47 was used to analyze the quality of the model obtained.

Blind and focused docking simulations were performed by AutoDock 4.2.5.1 and AutoDockTools 1.5.6 48 . For blind docking with AURKA active conformation, a 126 × 126 × 126 grid box with grid center coordinates equal to (37.908; −15.701; 37.348) was set; spacing was set at 0.397 Å. Focused flexible docking was performed on the AURKA active site to refine the binding energy and the conformation of the compounds in the pocket. A 60 × 60 × 60 grid box with center coordinates (39.621; −19.626; 37.357) and a spacing of 0.375 Å was set as default. Four residues—D256-N261-D274-W277—were set as flexible. This is because D256 is the active site, and the other three residues occupy positions with a relevant role for the access to the active site.

For PHB2, docking was focused on the portion of the protein having a more organized tertiary structure. The greed box was equal to 126x126x126 with center coordinates equal to (−15.757; −2.72; 16.28) and a spacing value of 0.386 Å. Docking simulations performed on the complex with capsaicin and compounds 5, 12, or 13 were performed, focusing the box at the interface of the three molecules with a greed box equal to 126 × 126 × 126, a greed center with coordinates (−31.084; −6.4; 22.525), and spacing value of 0.386 Å.

All the docking simulations allowed the flexibility of the ligand. The results obtained were analyzed by AutoDockTools and LigPlot v2.2 49 .

Statistical analyses

Two-way ANOVA was used to compare the effect of the transfection conditions and of the pharmacological treatments on ΔLifetime (Figs. 2 , 4D and Supplementary Fig. 1D, E ), and on mitochondrial mass loss (Fig. 6B and Supplementary Fig. 4 ). Statistics and summary statistics for biological replicates comparing the effect of the transfection conditions and the pharmacological treatments were performed with SuperPlotsofData 50 (Table 1 , Supplementary Data ). Ordinary one-way ANOVA was used to compare the effect of the pharmacological treatments on the AURKA/PHB2 interaction (Fig. 4E ) and on the FRET efficiency within the GFP-AURKA-mCherry biosensor (Fig. 6A ). T -tests were used to compare the effect of transfection conditions on raw and ΔLifetime values (Fig. 4C and Supplementary Fig. 1B ), and of the pharmacological treatment with compound 13 (HMBB) on mitochondrial mass loss (Fig. 6C, D ). Statistics for biological replicates were performed using SuperPlotsofData under GraphPad Prism v. 10 51 .

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary information

Description of Additional Supplementary Files

Supplementary Data

Reporting summary

Acknowledgements

We thank S. Dutertre, X. Pinson, and G. Le Marchand at the Microscopy Rennes Imaging Center (MRic, BIOSIT, Biogenouest) for assistance with FLIM experiments. MRic is a member of the national infrastructure France-BioImaging, supported by the French National Research Agency ( ANR-24-INBS-0005 FBI BIOGEN ). We thank the Plateforme de chimie biologique integrative (PCBIS, University of Strasbourg) for assistance with cell proliferation and viability experiments. We are grateful to N. Jolivet (IGDR, Rennes) for critical reading of the manuscript, to C. Bertin (IGDR, Rennes) for technical assistance, and to all lab members for helpful comments. This work was supported by the Centre National de la Recherche Scientifique (CNRS), the University of Rennes, the Ligue Contre le Cancer, Comités d’Ille et Vilaine et du Finistère to G.B. C.C was supported by a PhD fellowship from the Ligue Nationale Contre le Cancer (grant n. IP/SC – 17653) and by the Fondation ARC pour la Recherche sur le Cancer .

Author contributions

CRediT taxonomy: conceptualization (L.D. and G.B.) data curation (A.D., C.C., D.G., V.P., K.F., A.F., L.D., and G.B.), Formal analysis (A.D., C.C., D.G, V.P., K.F., A.F., L.D., and G.B.), Funding acquisition (L.D. and G.B.), Investigation (A.D., C.C., D.G., V.P., K.F., A.F., L.D., and G.B.), Methodology (A.F., L.D., and G.B.), Project administration (L.D. and G.B.), Resources (L.D. and G.B.), Software (C.C., D.G., V.P., A.F., and G.B.), Supervision (A.F, L.D., and G.B.), Validation (L.D. and G.B.), Visualization (C.C., D.G., V.P., A.F., L.D., and G.B.), Writing–original, review and editing (C.C., D.G, A.F., L.D., and G.B.).

Peer review

Peer review information

Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: Laura Rodríguez Pérez.

Data availability

Source microscopy data are available on Zenodo 52 (10.5281/zenodo.17610406). All other data are available from the corresponding authors (L.D. and G.B.) upon request.

Competing interests

The authors declare no competing or financial interests. G. Bertolin is an Editorial Board Member for Communications Biology , but was not involved in the editorial review of, nor the decision to publish this article.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

These authors contributed equally: Amel Djehal, Claire Caron.

Contributor Information

Laurent Désaubry, Email: desaubry@unistra.fr.

Giulia Bertolin, Email: giulia.bertolin@univ-rennes.fr.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-026-09573-3.

References

1. Wang, T., Ma, F. & Qian, H. Defueling the cancer: ATP synthase as an emerging target in cancer therapy. Mol. Ther. Oncol. 23 , 82–95 (2021).

2. Pickrell, A. M. & Youle, R. J. The roles of PINK1, parkin and mitochondrial fidelity in Parkinson’s disease. Neuron 85 , 257–273 (2015).

3. Wang, S. et al. The mitophagy pathway and its implications in human diseases. Signal Transduct. Target Ther. 8 , 1–28 (2023).

4. Wei, Y., Chiang, W.-C., Sumpter, R., Mishra, P. & Levine, B. Prohibitin 2 is an inner mitochondrial membrane mitophagy receptor. Cell 168 , 224–238.e10 (2017).

5. Wang, D. et al. Prohibitin ligands: a growing armamentarium to tackle cancers, osteoporosis, inflammatory, cardiac and neurological diseases. Cell. Mol. Life Sci. 77 , 3525–3546 (2020).

6. Bertolin, G. et al. Mitochondrial Aurora kinase A induces mitophagy by interacting with MAP1LC3 and prohibitin 2. Life Sci. Alliance 10.26508/lsa.202000806 (2021).

7. Kashatus, D. F. et al. RALA and RALBP1 regulate mitochondrial fission at mitosis. Nat. Cell Biol. 13 , 1108–1115 (2011).

8. Nikonova, A. S., Astsaturov, I., Serebriiskii, I. G., Dunbrack, R. L. & Golemis, E. A. Aurora A kinase (AURKA) in normal and pathological cell division. Cell. Mol. Life Sci. 70 , 661–687 (2013).

9. Bertolin, G., et al. Aurora kinase A localises to mitochondria to control organelle dynamics and energy production. eLife 7 , e38111 (2018).

10. Grant, R. et al. Constitutive regulation of mitochondrial morphology by Aurora A kinase depends on a predicted cryptic targeting sequence at the N-terminus. Open Biol. 8 , 170272 (2018).

11. Bertolin, G. & Tramier, M. Insights into the non-mitotic functions of Aurora kinase A: more than just cell division. Cell. Mol. Life Sci. 77 , 1031–1047 (2020).

12. Chouha, N. et al. Development of fluorizoline analogues as prohibitin ligands that modulate C-RAF signaling, p21 expression and melanogenesis. Eur. J. Med. Chem. 242 , 114635 (2022).

13. Tabti, R. et al. Development of prohibitin ligands against osteoporosis. Eur. J. Med. Chem. 210 , 112961 (2021).

14. Yurugi, H. et al. A subset of flavaglines inhibits KRAS nanoclustering and activation. J. Cell Sci. 133 , jcs244111 (2020).

15. Elderwish, S., Audebrand, A., Nebigil, C. G. & Désaubry, L. Discovery of 3,3′-pyrrolidinyl-spirooxindoles as cardioprotectant prohibitin ligands. Eur. J. Med. Chem. 186 , 111859 (2020).

16. Djehal, A. et al. Targeting prohibitin with small molecules to promote melanogenesis and apoptosis in melanoma cells. Eur. J. Med. Chem. 155 , 880–888 (2018).

17. Padilla-Parra, S., Audugé, N., Coppey-Moisan, M. & Tramier, M. Quantitative FRET analysis by fast acquisition time domain FLIM at high spatial resolution in living cells. Biophys. J. 95 , 2976–2988 (2008).

18. Leray, A., Padilla-Parra, S., Roul, J., Héliot, L. & Tramier, M. Spatio-temporal quantification of FRET in living cells by fast time-domain FLIM: a comparative study of non-fitting methods [corrected]. PLoS ONE 8 , e69335 (2013).

19. Kuramori, C. et al. Capsaicin binds to prohibitin 2 and displaces it from the mitochondria to the nucleus. Biochem. Biophys. Res. Commun. 379 , 519–525 (2009).

20. Janusz, J. M. et al. Vanilloids. 1. Analogs of capsaicin with antinociceptive and antiinflammatory activity. J. Med. Chem. 36 , 2595–2604 (1993).

21. Padilla-Parra, S. et al. Quantitative comparison of different fluorescent protein couples for fast FRET-FLIM acquisition. Biophys. J. 97 , 2368–2376 (2009).

22. Nguyen, T. A., Puhl, H. L., Hines, K., Liput, D. J. & Vogel, S. S. Binary-FRET reveals transient excited-state structure associated with activity-dependent CaMKII - NR2B binding and adaptation. Nat. Commun. 13 , 6335 (2022).

23. Amantini, C. et al. Capsaicin-induced apoptosis of glioma cells is mediated by TRPV1 vanilloid receptor and requires p38 MAPK activation. J. Neurochem. 102 , 977–990 (2007).

24. Xu, S. et al. Capsaicin induces mitochondrial dysfunction and apoptosis in anaplastic thyroid carcinoma cells via TRPV1-mediated mitochondrial calcium overload. Cell. Signal. 75 , 109733 (2020).

25. Yang, W. et al. Non-classical ferroptosis inhibition by a small molecule targeting PHB2. Nat. Commun. 13 , 7473 (2022).

26. Schreiber, S. L. The rise of molecular glues. Cell 184 , 3–9 (2021).

27. Bertolin, G. et al. A FRET biosensor reveals spatiotemporal activation and functions of aurora kinase A in living cells. Nat. Commun. 7 , 12674 (2016).

28. Cheetham, G. M. T. Crystal structure of aurora-2, an oncogenic serine/threonine kinase. J. Biol. Chem. 277 , 42419–42422 (2002).

29. Bayliss, R., Sardon, T., Vernos, I. & Conti, E. Structural basis of Aurora-A activation by TPX2 at the mitotic spindle. Mol. cell 12 , 851–862 (2003).

30. Zhang, Y. et al. Identification of the auto-inhibitory domains of Aurora-A kinase. Biochem. Biophys. Res. Commun. 357 , 347–352 (2007).

31. Damodaran, A. P., Vaufrey, L., Gavard, O. & Prigent, C. Aurora A kinase is a priority pharmaceutical target for the treatment of cancers. Trends Pharmacol. Sci. 38 , 687–700 (2017).

32. Sharma, R. K., Chafik, A. & Bertolin, G. Aurora kinase A/AURKA functionally interacts with the mitochondrial ATP synthase to regulate energy metabolism and cell death. Cell Death Discov. 9 , 1–12 (2023).

33. Islam, A., Su, A. J., Zeng, Z.-M., Chueh, P. J. & Lin, M.-H. Capsaicin targets tNOX (ENOX2) to inhibit G1 cyclin/CDK complex, as assessed by the Cellular Thermal Shift Assay (CETSA). Cells 8 , 1275 (2019).

34. Jia, G., Cang, S., Ma, P. & Song, Z. Capsaicin: a “hot” KDM1A/LSD1 inhibitor from peppers. Bioorg. Chem. 103 , 104161 (2020).

35. Chen, H. Y. et al. Capsaicin inhibited aggressive phenotypes through downregulation of tumor-associated NADH oxidase (tNOX) by POU domain transcription factor POU3F2. Molecules 21 , 733 (2016).

36. Yang, F. et al. Structural mechanism underlying capsaicin binding and activation of the TRPV1 ion channel. Nat. Chem. Biol. 11 , 518–524 (2015).

37. Durel, B., Kervrann, C. & Bertolin, G. Quantitative dSTORM super-resolution microscopy localizes Aurora kinase A/AURKA in the mitochondrial matrix. Biol. Cell 113 , 458–473 (2021).

38. Bertolin, G., Marchand, G. L., & Tramier, M. Real-time monitoring of aurora kinase a activation using conformational fret biosensors in live cells. J. Vis. Exp. 161 , e61611 (2020).

39. Pettersen, E. F. et al. UCSF Chimera-a visualization system for exploratory research and analysis. J. Comput. Chem. 25 , 1605–1612 (2004).

40. Kim, S. et al. PubChem 2023 update. Nucleic Acids Res. 51 , D1373–D1380 (2023).

41. Kwon, S., Jung, N., Yang, J., and Seok, C. GalaxySagittarius-AF: predicting targets for drug-like compounds in the extended human 3D proteome. J. Mol. Biol . 436 , 168617 (2024).

42. Berman, H. M. et al. The protein data bank. Nucleic Acids Res. 28 , 235–242 (2000).

43. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596 , 583–589 (2021).

44. Yan, Y., Tao, H., He, J. & Huang, S.-Y. The HDOCK server for integrated protein–protein docking. Nat. Protoc. 15 , 1829–1852 (2020).

45. Eswar, N. et al. Comparative protein structure modeling using Modeller. Curr. Protoc. Bioinformatics 10.1002/0471250953.bi0506s15 (2006).

46. Krissinel, E. & Henrick, K. Inference of macromolecular assemblies from crystalline state. J. Mol. Biol. 372 , 774–797 (2007).

47. Wiederstein, M. & Sippl, M. J. ProSA-web: interactive web service for the recognition of errors in three-dimensional structures of proteins. Nucleic Acids Res. 35 , W407–W410 (2007).

48. Morris, G. M. et al. AutoDock4 and AutoDockTools4: automated docking with selective receptor flexibility. J. Comput. Chem. 30 , 2785–2791 (2009).

49. Wallace, A. C., Laskowski, R. A. & Thornton, J. M. LIGPLOT: a program to generate schematic diagrams of protein-ligand interactions. Protein Eng. 8 , 127–134 (1995).

50. Goedhart, J. SuperPlotsOfData – a web app for the transparent display and quantitative comparison of continuous data from different conditions. 10.1091/mbc.E20-09-0583 (2021).

51. Lord, S. J., Velle, K. B., Mullins, R. D. & Fritz-Laylin, L. K. SuperPlots: Communicating reproducibility and variability in cell biology. J. Cell Biol. 219 , e202001064 (2020).

52. Caron, C., Bertolin, G. & Djehal, C. et al. Raw images (Zenodo) 10.5281/ZENODO.17610405 (2025).

Associated Data

Supplementary Materials

Supplementary information

Description of Additional Supplementary Files

Supplementary Data

Reporting summary

Data Availability Statement

Source microscopy data are available on Zenodo 52 (10.5281/zenodo.17610406). All other data are available from the corresponding authors (L.D. and G.B.) upon request.