Introduction

Approximately two-thirds of patients with Alzheimer’s disease (AD) are women, and the loss of the ovarian steroid 17β-estradiol (E2) at menopause is strongly associated with increased vulnerability to neurodegeneration (Nebel et al., 2018). Although E2 replacement is neuroprotective in animal models, including non-human primates (Rapp et al., 2003; Morrison et al., 2006), hormone replacement therapy in post-menopausal women has produced adverse outcomes, including increased risk of coronary heart disease and vascular dementia (Rossouw et al., 2002; Shumaker et al., 2003). These limitations have driven the development of neuroprotective selective estrogen receptor modulators (SERMs), but most act through nuclear estrogen receptors and retain undesirable peripheral effects (Murphy et al., 2003).

In search of a selective neuroprotective SERM, we developed STX, a synthetic diphenyl acrylamide compound that selectively targets the central nervous system (CNS) and mimics the rapid, membrane-initiated estrogen signaling without activating classical estrogen receptors (Qiu et al., 2003; Qiu et al., 2006; Tobias et al., 2006; Roepke et al., 2010; Smith et al., 2014). STX activates G protein–coupled signaling pathways in hypothalamic neurons (Qiu et al., 2003; Qiu et al., 2006; Smith et al., 2013). In proopiomelanocortin (POMC) neurons, STX induces heterologous desensitization of inhibitory Gα_i/o-coupled receptors, including μ-opioid and GABA_B receptors, thereby increasing neuronal excitability (Qiu et al., 2003; Qiu et al., 2006). Importantly, STX crosses the blood–brain barrier, can be administered chronically, and reproduces key estrogenic effects on thermoregulation and metabolism without peripheral estrogenic actions (Roepke et al., 2010).

In addition to its rapid neuromodulatory effects, STX exhibits robust neuroprotective properties. It reduces ischemic neuronal damage in hippocampal neurons (Lebesgue et al., 2010), enhances synaptic plasticity, and protects against β-amyloid toxicity in cellular and animal models of AD (Gray et al., 2016; Quinn et al., 2022; Lee et al., 2025). These effects are accompanied by improvements in mitochondrial function, suggesting that regulation of cellular bioenergetics may be a central component of STX action. Mitochondrial dysfunction is a defining feature of neurodegenerative diseases (Ashleigh et al., 2023), and declines in brain metabolic capacity are strongly associated with menopause (Brinton et al., 2015; Mosconi et al., 2018). However, the molecular targets through which STX modulates mitochondrial function and neuronal activity remain unknown.

A key regulator of mitochondrial metabolism is the voltage-dependent anion channel (VDAC), located in the outer mitochondrial membrane. VDACs control the exchange of ATP, ADP, and other metabolites between mitochondria and the cytosol, thereby governing cellular energy homeostasis (Rostovtseva and Colombini, 1996; Colombini, 2004; Rostovtseva and Bezrukov, 2008). Modulation of VDAC gating and ion selectivity can directly influence mitochondrial respiration and metabolic flux. Although steroidal compounds have been reported to bind VDACs, their functional effects on channel activity are limited (Camara et al., 2017; Reina and De Pinto, 2017; Rostovtseva et al., 2020). In isolated liver mitochondria, the non-steroidal SERM tamoxifen has been shown to bind to VDACs and potentially regulate transport of anions across the mitochondrial membrane (Unten et al., 2022).

However, identifying the molecular targets of small molecules such as STX in intact cells remains a major challenge. Unlike protein–protein interactions, small molecule–protein interactions are often transient and difficult to capture biochemically. Diazirine-based photoaffinity labeling, combined with biorthogonal click chemistry, enables covalent capture and enrichment of ligand-bound proteins in living systems (Moellering and Cravatt, 2012; Schultz et al., 2022). This “flash-and-click” strategy has proven effective for mapping lipid and small-molecule interactomes in complex cellular environments (Haberkant et al., 2013; Hulce et al., 2013; Höglinger et al., 2017; Müller et al., 2020; Müller et al., 2021).

Therefore, we combined chemoproteomics, molecular biology, electrophysiology and mitochondrial physiology to identify the cellular targets of STX in hypothalamic POMC neurons. Using a photoactivatable STX probe, we identified all three VDAC isoforms as prominent intracellular binding partners. We show that STX directly modulates mitochondrial VDAC channels, enhancing voltage-dependent gating and altering ion selectivity, with corresponding increases in mitochondrial ATP production, respiratory capacity and glycolytic function.

Results

Chemoproteomic identification of STX targets reveals VDACs as prominent binding partners

In order to see if STX binds to VDACs similar to the structurally-related tamoxifen (Unten et al., 2022), we designed a novel bifunctional STX derivative (BF-STX) containing both a photo-cross-linkable diazirine group and an alkyne handle, which was synthesized in collaboration with SiChem GmbH (Germany) (Figure 1A). BF-STX retained full biological activity, reproducing the ability of STX to induce heterologous desensitization of Gi/o-coupled receptor signaling in POMC neurons, which inhibits the activity of these anorexigenic neurons (Qiu et al., 2003; Qiu et al., 2006) (Figure 1B). Also, with UV activation BF-STX enabled robust and cell-type–specific labeling in hypothalamic slices from female BL6 mice and mHypo43 cells (Figures 1C,D).

Figure 1.

To identify candidate STX-binding proteins, mHypo43 POMC cells were incubated with BF-STX (10 μM) for 30 min and then exposed to UV light (350 nm) to induce crosslinking (Figure 2A). Negative controls included samples without UV exposure and cells incubated with BF-STX together with a 3-fold higher concentration of STX. These controls ensured that the enriched hits were legitimate targets and not “off-targets” from photo-crosslinking (Kleiner et al., 2017; West et al., 2021). Protein–BF-STX conjugates were labeled via click chemistry using picolyl-azide agarose beads, digested on-bead, TMT-labeled, and analyzed by LC-MS/MS on an Orbitrap Fusion Lumos system at EMBL as described in the Materials and Methods. VDAC1, VDAC2, and VDAC3 emerged as prominent targets (Figure 2A, Suppl Table 1). Interestingly, even with a short exposure (5 min) to BF-STX, VDACs emerged as a significant targets (Figure 2B, Suppl Table 2), which is congruent with the rapid neuroprotective actions of STX to rescue CA1 hippocampal neurons following global ischemia in female rats (Lebesgue et al., 2010). Furthermore, fold-change correlation plots showed that VDAC proteins were strongly enriched in the UV crosslinking condition relative to the no-UV control (x-axis), while enrichment was reduced in the competition experiments, shifting the VDAC proteins toward the x-axis. This pattern is consistent with competition-sensitive BF-STX labeling (Figure S4A). Moreover, a heatmap plot of the binding (Figure S4B) revealed that when unlabeled STX (1 or 3 molar equivalents) was added, the signal for VDACs faded. This competition pattern is consistent with partial reduction of BF-STX enrichment in the presence of unlabeled STX.

Figure 2.

VDAC2 is the predominant isoform expressed in POMC neurons

At the molecular level, we first examined whether VDACs are expressed in hypothalamic anorexigenic POMC neurons, given their central role in mitochondrial metabolite exchange. Single-cell qPCR from pooled POMC neurons from female mice revealed a clear expression hierarchy, with Vdac2 ≥ Vdac3 >> Vdac1 mRNA (Figures 3A,B), indicating that VDAC2 is the dominant isoform in this neuronal population. This distribution contrasts with most tissues, where VDAC1 predominates (Zinghirino et al., 2021), suggesting a cell-type–specific specialization of mitochondrial outer membrane channels in POMC neurons.

Figure 3.

To assess whether STX influences mitochondrial function in these cells, we measured mitochondrial membrane potential, calcium and ATP levels in mHypo43 neurons. STX (100 nM) did not alter the mitochondrial membrane potential, but it significantly reduced mitochondrial calcium concentrations and increased mitochondrial ATP levels (Figures 3C-E). These findings indicate that STX enhances mitochondrial function without inducing depolarization, consistent with improved metabolic efficiency rather than mitochondrial stress.

STX directly modulates VDAC2 channel gating and ion selectivity

To determine whether STX directly targets VDAC channels, we performed electrophysiological recordings of recombinant human VDAC2 reconstituted in planar lipid membranes (Figure 4A) (Rostovtseva and Bezrukov, 2015). We focused on VDAC2, which had the highest mRNA expression in POMC neurons (Figure 3A,B). In a symmetrical 1 M KCl buffer solution, VDAC2 typically forms channels of ∼3.5 – 4.0 nS conductance (Figure 4A) (Rosencrans et al., 2025), which characteristically transition (i.e., gate) from the high conducting “open” state to a variety of lower-conducting or “closed” states in response to large applied voltages (Figure 4B, control trace). Under these experimental conditions, the application of 60 mV was required to observe consistent voltage gating. For the sake of illustrative clarity, in Figure 4B we show traces at negative potentials. The addition of 100 and 300 nM STX to the “cis” (grounded) side (Figure 4A) of the membrane caused channel closure at lower negative potentials, -20 and -10 mV, respectively (Figure 4B, lower traces). The transitions to the low conductance states were reversible such that the voltage jump to 0 mV fully reopened the channel, a characteristic behavior of reconstituted VDACs (Colombini, 1989). The single-channel recordings revealed that in the presence of STX, voltages of 40-50 mV lower than those in the control were sufficient for VDAC2 single-channel gating.

Figure 4.

A quantitative analysis of the reconstituted VDAC voltage gating requires recording from many channels, which is best achieved using a multichannel approach (Rostovtseva et al., 2006; Teijido et al., 2014; Rappaport et al., 2015). In this protocol, slow triangular voltage waves (5 mHz, ±60 mV) are applied to the membrane containing 20-100 channels. Figure 4C shows representative normalized conductance-voltage plots of G/G_max versus voltage at negative potentials, where G is the conductance at a given voltage, and G_max is the maximum conductance at voltages close to 0 mV, in control and after addition of 50, 100, and 1000 nM STX to the cis compartment. The STX-enhanced voltage gating was manifested as a decrease in the minimal conductance (G_min, indicated by dashed gray lines) calculated at |V| > between 45-55 mV (downward arrow in Figure 4C). The dose-dependence of STX-enhanced VDAC2 voltage gating was quantified as the normalized change in conductance (G_max-G_min)/G_min) plotted against STX concentrations (Figure 4D). The averaged normalized reduction in G_min with STX concentration was approximated by a first-order binding curve with a K_d of 15 ± 4 nM.

These results suggested a direct functional interaction between STX and VDAC2. To further test this hypothesis, we measured the ionic selectivity of VDAC2 in the presence of STX. The open state of VDAC2, similarly to VDAC1 and VDAC3 (Tan and Colombini, 2007; Queralt-Martín et al., 2020; Rosencrans et al., 2025), is anion-selective, and multiple voltage-induced closed states are characterized by a wide range of low conductances and predominantly cationic selectivity (Colombini, 1989). We measured VDAC2 selectivity in a 5x salt gradient, 1 M KCl (cis side) versus 0.2 M KCl (trans side). A representative trace of VDAC2 obtained in this salt gradient before and after the addition of 100 nM STX to the cis compartment is shown in Figure 4E. In the presence of STX, the fluctuations of channel conductance between different conductance levels were measured at ±10 mV. The corresponding I/V plots for the open and the three low-conducting states are shown in Figure 4F. The linear regressions through the data points determined the substate conductance as a slope, and the reversal potential Ψ_rev as the voltage at which the current is zero for each substate. While the selectivity of the open state (1.9 nS) remained unchanged (Ψ_rev = 8.3 mV), in the presence of STX two low-conducting states of 0.5 and 0.6 nS displayed anion selectivity with Ψ_rev equal to 9.4 and 3.4 mV, respectively, with one low-conducting state (0.9 nS) being non-selective (inset in Figure 4F). These results indicate that STX may alter the selectivity of VDAC2 closed states, without affecting either the conductance or selectivity of the open state.

The absence of an effect of STX on VDAC2 open-channel conductance suggested that this hydrophobic compound most likely interacts with VDAC2 at the protein-lipid interface. We further postulate that by reversing the selectivity of VDAC2 closed states to anionic, STX modulates the transport properties of VDAC2 to favor ATP flux (Rostovtseva and Colombini, 1996). These results also explain the decreased mitochondrial calcium and increased mitochondrial ATP measured in mHypo43 (POMC) neurons (Figures 3D,E). Facilitation of ATP/ADP fluxes may contribute to the neuroprotective effects of STX in CNS neurons (Gray et al., 2016; Quinn et al., 2022; Lee et al., 2025).

STX enhances mitochondrial respiration and glycolytic function

Based on the findings that STX increased ATP concentrations in mHypo43 cells (Figure 3E), we decided to measure cellular metabolism in real-time using the Agilent Seahorse Assay system. We investigated the effects of STX on mitochondrial function, including mitochondrial respiration and glycolysis over time. Mitochondrial respiration (Figures 5, 6) was assessed by administrating agents that target different components of the electron transport chain (i.e., oligomycin, FCCP and rotenone/antimycin A) (Divakaruni et al., 2014) (see Materials and Methods). These measurements were conducted under control conditions (vehicle) and compared to treatment with 1 nM, 5 nM or 10 nM STX in mHypo43 cells. STX treatment led to an increase in several key parameters of mitochondrial respiration (Figure 5A). Quantitative analysis revealed that STX significantly increased basal respiration (Figure 5B), ATP production (Figure 5C), and maximal respiratory capacity (Figure 5D). Additionally, STX augmented spare respiratory capacity (Figure 5E), suggesting enhanced mitochondrial adaptability under stress conditions. Notably, proton leak (Figure 5G) but not non-mitochondrial respiration (Figure 5F) was significantly increased. Since neurons are highly dependent on glucose uptake and glycolysis as a source of energy (Li et al., 2023), we also analyzed glycolysis by exposing mHypo43 cells to oligomycin, FCCP and rotenone/antimycin A, while measuring the extracellular acidification rate (ECAR) over time. There were differences in ECAR with STX treatment versus control (Figure 5H). The lower concentrations of STX (1 and 5 nM) significantly increased glycolysis (Figure 5I), glycolytic capacity (Figure 5J), glycolytic reserve (Figure 5K) and non-glycolytic acidification (Figure 5L). However, since there was a non-monotonic dose-response to STX, i.e. 10 nM STX did not increase glycolytic function, we speculated that the concentration of estrogens in the serum were saturating the STX targets and therefore causing a desensitization. Therefore, we serum-starved the mHypo43 cells for 6 hours prior to the Seahorse Assay in order to remove any endogenous estrogens (Figure 6). The serum starvation did not alter the “bell-shaped” dose-response for glycolytic function (Figures 6H-L) and even showed a monotonic dose-response for mitochondrial respiratory parameters (Figures 6A-G). Since hepatocytes exhibit the highest mitochondrial density within the human body (Xu et al., 2023), we sought to see if STX was equally potent in immortalized human hepatocyte (IHH) cells to increase mitochondrial respiration. Indeed, 1 nM (and 10 nM) STX increased mitochondrial respiration in both serum (Figures S5A-G) and serum-starved conditions (Figures S6A-G). As one would predict, since IHH cells are not dependent on glucose uptake as an energy source, glycolytic function was essentially unaltered with STX treatment in serum (Figures 5H-L) and serum-starved conditions (Figures 6H-L). Collectively, these results indicate that STX acts independently of endogenous estrogenic compounds and are consistent with the fact that neurons require glucose uptake and glycolysis to maintain their electrophysiological activity (Li et al., 2023).

Figure 5.

Figure 6:

Discussion

A central strength of this study is the use of photoaffinity labeling coupled with click chemistry to capture STX–protein interactions in intact cellular systems. Because small molecule–protein interactions are often weak, reversible and difficult to preserve during biochemical isolation, the diazirine-containing BF-STX probe provided a means to covalently stabilize ligand-associated proteins after UV activation (Moellering and Cravatt, 2012; Schultz et al., 2022). The alkyne handle then enabled selective enrichment of BF-STX–labeled proteins through copper-catalyzed azide–alkyne cycloaddition, thereby reducing background and increasing confidence in target identification. Importantly, the use of no-UV controls and competition with excess unlabeled STX strengthened the specificity of the chemoproteomic hits. Therefore, the enrichment of VDAC1–3 under BF-STX labeling conditions, together with loss of signal after STX competition, supports the conclusion that VDACs are bona fide intracellular STX-interacting proteins rather than nonspecific contaminants. Thus, the flash-and-click approach provided a mechanistic bridge between the known cellular actions of STX (Lee et al., 2025) and the subsequent functional validation of VDAC2 gating and mitochondrial bioenergetics.

Given the abundant evidence that mitochondrial dysfunction plays a role in many neurodegenerative diseases (Ashleigh et al., 2023; DAlessandro et al., 2025), the current results are congruent with previous findings that STX is neuroprotective against β-amyloid toxicity both in neuronal cultures and transgenic mouse models of Alzheimer’s disease, in part by supporting mitochondrial function and synaptic integrity (Gray et al., 2016; Quinn et al., 2022; Lee et al., 2025). Conversely, the loss of mitochondrial function is clearly involved in neuronal dysfunction throughout the course of AD (Ashleigh et al., 2023; DAlessandro et al., 2025). Hence, our results suggest that the neuroprotective effects of STX likely include the regulation of VDAC function in both the healthy and diseased brain. Interestingly, Vdac2 was the most highly expressed isoform in POMC neurons, unlike the dominance of VDAC1 expression in most tissues with the exception of the testes and ovaries (Zinghirino et al., 2021). Therefore, we investigated the interaction of STX with VDAC2 using single-channel electrophysiological recordings to determine how nanomolar concentrations of STX affect the VDAC2 channel properties. In these experiments, we reconstituted VDAC2 channels in planar lipid membranes and analyzed their voltage-gating behavior, which is characterized by transitions between open and low-conducting, or closed states, in a voltage-dependent manner, a common feature of all VDAC isoforms (Queralt-Martín et al., 2020; Rosencrans et al., 2025). Nanomolar concentrations of STX increased the voltage sensitivity of VDAC2, reducing the characteristic gating voltage by about 40 mV, as observed upon the application of voltage ramps. Furthermore, in multi-channel recordings, STX enhanced voltage-dependent gating in a concentration-dependent manner with a K_d similar to its potent effects on POMC neurons ex vivo (Qiu et al., 2006). These results contrast with the neurosteroid allopregnanolone, which is a potent allosteric modulator of GABA_A channel gating (Majewska et al., 1986; Lambert et al., 1995). Although allopregnanolone can also bind to VDACs, based on photoaffinity labeling, it does not affect their gating properties (Cheng et al., 2019). The present results indicate that STX can function as a potent enhancer of VDAC gating, independent of other protein interactions. Moreover, the STX-induced shift of ion selectivity of VDAC2 closed states towards anions further confirms a direct STX-VDAC2 interaction. The absence of measurable effects on VDAC2 open state conductance or selectivity suggests that the hydrophobic STX interacts with VDAC2 at the protein-lipid interface (Rostovtseva et al., 2020; Rovini et al., 2020).

We also investigated the effects of STX on mitochondrial physiology using mHypo43 POMC cells, which were used for the photo-affinity labeling experiments, and IHH cells, which have been established as an effective assay for mitochondrial function (Lim et al., 2008). Hepatocytes exhibit the highest mitochondrial density within the human body (Xu et al., 2023). Based on the K_d values obtained from our biophysical measurements of channel gating, we evaluated the effects of nanomolar concentrations of STX on mitochondrial respiration and glycolytic function in both cell lines. Notably, we observed the types of changes expected following administration of agents targeting the electron transport chain (e.g., oligomycin); specifically, STX (1 nM) was able to potently increase not only basal respiration but spare respiratory capacity, ATP production and proton leak, which helps maintain Δψ_m, and non-mitochondrial respiration in both cell lines. Importantly, STX also increased glycolytic function in the mHypo43 POMC neural cell line, which is congruent with the fact that neurons require glucose uptake and glycolysis to support their high energy needs (Li et al., 2023). STX was equally effective in both serum-free conditions, which eliminates endogenous estrogens, and media that included serum. These results suggest that STX is more potent than E2 in its ability to enhance mitochondrial respiratory activity and affect VDAC gating through its direct interaction, as suggested by their prominent photo-affinity labeling with BF-STX.

In general, the effects of E2 on mitochondrial function are thought to be transduced via estrogen receptors α and β (ERα, ERβ), which regulate transcriptional responses to enhance the efficiency of respiratory chain complexes and improve ATP synthesis capacity (Simpkins et al., 2008; Huang et al., 2025). Immunoreactive ERα and ERβ have been identified within mitochondria (Simpkins et al., 2008; Klinge, 2020), which is consistent with the supposition that many if not all of the actions of E2 are transcriptionally mediated. E2 activates mitochondrial biogenesis, inhibits the accumulation of reactive oxygen species (ROS), and stabilizes the mitochondrial membrane potential (Klinge, 2020). STX, in contrast, does not bind to ERα or ERβ (Qiu et al., 2003), and its effects at both the plasma membrane and mitochondrial membrane are more rapid (<15 min) than a transcriptionally mediated response (Qiu et al., 2003; Smith et al., 2013; Conde et al., 2016). Most importantly, STX directly modulates VDAC gating in the absence of any transcription factors. Notably, STX potently (1 nM) increased both basal and stress-induced (spare capacity) oxidative phosphorylation in mHypo43 and IHH cells, whereas E2 has not been found to affect basal respiration in mammalian cells (Simpkins et al., 2008).

Ultimately, there are dramatic changes in brain oxidative metabolism during menopause that can promote AD (Brinton et al., 2015; Mosconi et al., 2018). In neuroblastoma cell models and primary hippocampal neuronal cultures, STX protects against Aβ-associated mitochondrial dysfunction and synaptic loss (Gray et al., 2016). Moreover, Quinn and colleagues have shown that STX treatment reduces Aβ burden in 5XFAD (AD) mice treated from 6 to 8 months of life (Quinn et al., 2022), during which time amyloid pathology becomes increasingly pronounced in 5XFAD mice (Oakley et al., 2006; Kimura and Ohno, 2009). STX does not alter the expression of Amyloid Precursor Protein (APP) or Aβ aggregation; however, it can indirectly modulate Aβ production by supporting mitochondrial function, which in turn would mitigate the accumulation of reactive oxygen species that promote amyloidogenic processing of APP to generate Aβ peptides (Tong et al., 2005; Jo et al., 2010). Indeed, our findings in mHypo43 cells and human hepatocytes support the role of STX to enhance mitochondrial respiration. By supporting mitochondrial function, STX might also augment the normal phagocytosis and clearance of pathogenic Aβ, which is similar to actions of insulin and other therapies that have Aβ-lowering effects in the absence of any direct effect on Aβ production (Logan et al., 2018; Matthews et al., 2019).

Material and methods

Animals

All the animal procedures described in this study were performed in accordance with institutional guidelines based on National Institutes of Health standards and approved by the Institutional Animal Care and Use Committee at Oregon Health and Science University (IACUC Protocol TR03-IP00000585).

Pomc^EGFP (RRID:IMSR_JAX:009593, strain of origin: C57BL/6 J) (Cowley et al., 2001) mice were used. All animals were maintained under controlled temperature and photoperiod (lights on at 0600 h and off at 1800 h) and given free access to food and water.

mHypo43 (mHypoE-N43/5) cells

The embryonic mouse hypothalamus cell line N43/5 (mHypoE-N43/5) was obtained from Cellutions Biosystems (Cedarlane, cat. no. CLU127) and cultured according to the supplier’s protocol. Cells were grown in 10 cm tissue culture dishes at 37 °C in 5% CO₂ using DMEM (Gibco, cat. no. 11960-044) containing 4.5 mg/mL D-glucose, supplemented with 2 mM L-glutamine (Sigma-Aldrich, cat. no. G7513), 1× non-essential amino acids (Gibco, cat. no. 11140-050), 1% penicillin/streptomycin (Sigma-Aldrich, cat. no. P4333), and 10% FBS. Freshly thawed cells were initially seeded and allowed to attach for 4–6 h before medium replacement. Cells were passaged at 70–90% confluence using trypsin and split at ratios between 1:5 and 1:10.

Synthesis and physiological validation of BF-STX probe

The novel bifunctional STX derivative (BF-STX) containing both a photo-cross-linkable diazirine group and an alkyne handle was designed by the Schultz Lab and synthesized in collaboration with SiChem GmbH (Germany) (Figure 1A and see Supplementary Files). To demonstrate that this construct was functional, we tested the efficacy of BF-STX to modulate the excitability of POMC neurons in ex vivo electrophysiological slice experiments. Indeed, BF-STX was as efficacious as STX in increasing POMC cell excitability (Qiu et al., 2003; Qiu et al., 2006) (Figure 1B).

Confocal imaging of BF-STX labelling in brain slices

Coronal hypothalamic arcuate slices (250 µm thick) containing POMC neurons were incubated with 10 µM BF-STX for 30 min, then subjected to UV photo-crosslinking (2.5 min, 1000 W Xe lamp with a 350 nm high-pass filter). Slices were fixed in cold methanol (–20 °C, 20 min), and BF-STX–protein conjugates were visualized by Cu(I)-catalyzed click chemistry using Alexa Fluor 488 picolyl azide. Nuclei were counterstained with DAPI. Imaging was performed on an Olympus FV1200 confocal microscope with a 60× oil-immersion objective, using the third ventricle (3V) as an anatomical landmark (Figure 1C).

Confocal Imaging of BF-STX labelling in mHypo43 cells

mHypo43 (POMC neurons) were treated with 10 µM BF-STX for 30 min, followed by UV crosslinking under the same conditions (2.5 min, 1000 W Xe lamp, 350 nm filter) (Figure 1D). After fixation in cold methanol (–20 °C, 20 min), cells were labeled with Alexa Fluor 488 picolyl azide and counterstained with DAPI. Confocal imaging (Olympus FV1200, 60× oil objective) showed robust intracellular accumulation of BF-STX in mHypo43 and MIN6 cells, whereas HeLa Kyoto cells (RRID:CVCL_1922) displayed little to no detectable signal.

Visualized whole-cell patch recording

Whole-cell current clamp and voltage clamp recordings were made from POMC neurons as previously described (Qiu et al., 2010; Qiu et al., 2014). Coronal arcuate slices (250 μm) were prepared from gonadectomized females, 10 weeks and older as previously described (Qiu et al., 2010; Qiu et al., 2014). The slices were then transferred to an auxiliary chamber in which they were kept at room temperature (25 °C) in artificial CSF (aCSF) consisting of the following (in mm): 124 NaCl, 5 KCl, 2.6 NaH₂PO₄, 2 MgSO₄, 2 CaCl₂, 26 NaHCO₃, 10 HEPES, 10 glucose, pH 7.4, until recording (recovery for 2 h). A single slice was transferred to the recording chamber at a time and was kept viable by continually perfusing with warm (35 °C), oxygenated aCSF at 1.25 ml/min. Whole-cell patch recordings were made from POMC neurons using an Olympus BX51 W1 fixed stage scope out-fitted with epifluorescence and infrared-differential interference contrast (IR-DIC) video microscopy. Patch pipettes (A-M Systems; 1.5 μm outer diameter borosilicate glass) were pulled on a Brown/Flaming puller (Sutter Instrument, model P-97) and filled with the following solution: 128 mM potassium gluconate, 10 mM NaCl, 1 mM MgCl₂, 11 mM EGTA, 10 mM HEPES, 3 mM ATP, and 0.25 mM GTP adjusted to pH 7.3 with KOH; 295 mOsm. Pipette resistances ranged from 3.5 to 4 MΩ. In whole-cell configuration, access resistance was less than 30 MΩ; the access resistance was 80% compensated. The input resistance was calculated by measuring the slope of the I–V relationship curve between −70 and −50 mV. Standard whole-cell patch recording procedures and pharmacological testing were performed as previously described (Qiu et al., 2010). Electrophysiological signals were digitized with a Digidata 1322 A (Axon Instruments) and the data were analyzed using p-Clamp software (Molecular Devices, Foster City, CA).

To study the effect of BF-STX on µ-opioid receptor-mediated responses in POMC neurons, a drug administration protocol was employed during whole-cell patch-clamp recordings in voltage-clamp mode (V_hold = –60 mV). After forming a gigaseal and achieving the whole-cell configuration, slices were perfused with tetrodotoxin (TTX, 1 μM) for 5 minutes in order to synaptically isolate the neurons. The µ-opioid receptor receptor-mediated response was evoked by perfusing [D-Ala2, N-MePhe4, Gly-ol]-enkephalin (DAMGO, 300 nM) until a steady-state outward current was reached. Following drug washout, the current returned to its pre-drug baseline. Cells were then treated with BF-STX for 15 minutes. DAMGO was perfused again, and the second response was recorded.

A standard artificial cerebrospinal fluid was used (Qiu et al., 2010). TTX (1 mM, Alomone Labs) and DAMGO (1 mM, d-Ala2, N-Me-Phe4, Gly-ol5-enkephalin; Peninsula Laboratories, Inc., Belmont, CA) were dissolved in H₂O. BF-STX (10 mM) was dissolved in 100% ethanol. Aliquots of the stock solution was stored as appropriate until needed.

Identification of candidate STX targets in POMC neurons using BF-STX

mHypo43 cells (3 x 10⁶ cells) were grown in culture medium (see above) and incubated with 10 μM BF-STX in DMEM for 5 min or 30 min. Photo-crosslinking was performed for 5 min. BF-STX without UV crosslinking and pre-incubation with STX before cross-linking served as negative controls. After washing and scraping, cells were lysed on ice by probe sonication (three 15 s bursts). Lysates were subjected to copper-catalyzed azide–alkyne cycloaddition (CuAAC) with picolyl azide agarose beads for selective enrichment of labeled proteins. Azide beads (200 µL) were washed in deionized water, added to lysate with CuSO₄ (1 mM), sodium ascorbate (1 mM), and TBTA (100 µM), and incubated at room temperature for 1 h with rotation. Beads were pelleted (1000 x g, 2 min), transferred to centrifuge columns, and washed with PBS (3×), bead wash buffer 1 (100 mM Tris-HCl, pH 8.0, 250 mM NaCl, 5 mM EDTA, 1% SDS; 5×), and bead wash buffer 2 (100 mM Tris-HCl, pH 8.0, 8 M urea; 10×). Beads were transferred to clean tubes and pelleted.

Captured proteins were reduced in digestion buffer (100 mM Tris-HCl, pH 8.0, 2 mM CaCl₂, 10% acetonitrile) with DTT (10 mM) at 42 °C for 30 min and alkylated with iodoacetamide (40 mM) at room temperature in the dark for 30 min. Bead-bound proteins were digested overnight at 37 °C with sequencing-grade trypsin. Peptides were desalted on C18 cartridges and stored at −80 °C prior to shipping to EMBL.

Identification of isolated proteins by LC-MS/MS (see Supplementary Files for complete protocol

Up to 10 µg of peptides were labeled using TMTpro™ 16plex reagent as previously described (Thompson et al., 2019). Briefly, 0.5 mg of TMT reagent was dissolved in 45 µL of 100% acetonitrile. Subsequently, 4 µL of this solution was added to each peptide sample, followed by incubation at room temperature for 1 hour. The labeling reaction was quenched by adding 4 µL of a 5% aqueous hydroxylamine solution and incubating for an additional 15 minutes at room temperature. Labeled samples were then combined for multiplexing, desalted using an Oasis® HLB µElution Plate (Waters) according to the manufacturer’s instructions, and dried by vacuum centrifugation.

Peptides were separated on an UltiMate 3000 RSLCnano system using a C18 trapping cartridge (µ-Precolumn PepMap™ 100, 300 µm × 5 mm, 5 µm, 100 Å) and an analytical column (nanoEase™ M/Z HSS T3, 75 µm × 250 mm, 1.8 µm, 100 Å). Samples were trapped at 30 µL/min in 0.05% TFA for 6 min and then gradient-eluted at 0.3 µL/min with solvent A (3% DMSO, 0.1% formic acid in water) and solvent B (3% DMSO, 0.1% formic acid in acetonitrile): 2–8% B (4 min), 8–26% B (104 min), 26–38% B (4 min), 40–80% B (0.1 min), 80% B (3.9 min), followed by re-equilibration to 2% B (4 min).

Peptides were analyzed on an Orbitrap Fusion™ Lumos™ Tribrid™ mass spectrometer in positive ion mode with a Pico-Tip nanospray emitter (2.2 kV spray voltage, capillary temperature 275 °C). Full MS scans (m/z 375–1500) were acquired in the Orbitrap at 120 000 resolution with a maximum injection time of 50 ms. Data-dependent MS/MS spectra were acquired at 30 000 resolution with a maximum injection time of 94 ms and AGC target of 200%. HCD fragmentation was used with normalized collision energy of 34%, a quadrupole isolation window of 0.7 m/z, and dynamic exclusion of 60 s; precursor charge states 2–7 were selected for fragmentation.

Data analysis

Raw files were converted to mzML format using MSConvert (ProteoWizard, RRID:SCR_012056) with peak picking and zlib compression. Spectra were searched with MSFragger in FragPipe (v19.0) against the UniProt (RRID:SCR_002380) Homo sapiens reference proteome (UP000005640) including common contaminants and reversed sequences. Carbamidomethylation (C, 57.0215) and TMTpro (K, 304.2072) were set as fixed modifications; oxidation (M, 15.9949), protein N-terminal acetylation (42.0106), and TMTpro (peptide N-terminus, 304.2072) were included as variable modifications. Trypsin specificity with up to two missed cleavages and a minimum peptide length of seven residues was used; precursor and fragment mass tolerances were set to 20 ppm; peptide and protein FDR were controlled at 1% using standard FragPipe settings.

FragPipe (protein.tsv files) were processed using the R programming environment (ISBN 3-900051-07-0). Initial data processing included filtering out contaminants and reverse proteins. Only proteins quantified with at least 2 unique peptides (Unique.Peptides >= 2) were considered for further analysis. 878 proteins passed the quality control filters. In order to correct for technical variability, batch effects were removed using the ‘removeBatchEffect’ function of the limma (RRID:SCR_010943) package (Ritchie et al., 2015) on the log2 transformed raw TMT reporter ion intensities (’channel’ columns). Subsequently, normalization was performed using the ‘normalizeVSN’ function of the limma package (VSN - variance stabilization normalization – (Huber et al., 2002)). For visualization, proteins abundances were normalized to the control condition by dividing by the median of control replicates (’ctrl’ column). These control ratios were used for clustering and heatmap display. Differential expression analysis was performed using the moderated t-test provided by the limma package (Ritchie et al., 2015). The model accounted for replicate information by including it as a factor in the design matrix passed to the ‘lmFit’ function. To obtain p-values and false discovery rates (FDRs), the ‘fdrtool’ function from the fdrtool package (Strimmer, 2008) was used to analyze the t-values produced by limma for certain comparisons. Proteins were annotated as hits if they had a false discovery rate (FDR) below 0.05 and an absolute fold change greater than 2. Proteins were considered candidates if they had an FDR below 0.2 and an absolute fold change greater than 1.5. Clustering based on the median protein abundances normalized by median of control condition was conducted to identify groups of proteins with similar patterns across conditions. The ‘kmeans’ method was employed, using Euclidean distance as the distance metric and ‘ward.D2’ linkage for hierarchical clustering. The optimal number of clusters (6) was determined using a hybrid Silhouette approach. The threshold was set as the maximum of either the 80th percentile of silhouette values or a fixed floor of 0.25 (threshold = 0.403), selecting the maximum number of clusters exceeding this threshold. This hybrid approach balances data-driven cluster selection while avoiding selection from noisy low-quality regions.

Cell harvesting of dispersed and POMC neurons and quantitative real-time PCR (qPCR)

Cell harvesting and qPCR was conducted as previously described (Bosch et al., 2013). The ARH was microdissected from basal hypothalamic coronal slices obtained from female Pomc^EGFP mice in which POMC neurons were labeled with YFP (n = 3 animals/group). The dispersed cells were visualized, patched, and then harvested (10 cells/tube) as described previously (Bosch et al., 2013). Briefly, the tissue was incubated in papain (7 mg/ml in oxygenated aCSF) for 50 min at 37°C and washed 4 times in low Ca²⁺ aCSF and two times in aCSF. Gentle trituration with Pasteur pipettes were used to disperse the neurons onto a glass bottom dish. Oxygenated aCSF circulated into the plate keeping the cells clear of debris. Only healthy cells with processes and a smooth cell membrane were harvested. The cells were harvested using the XenoWorks Microinjector System (Sutter Instruments, Navato, CA), which provided negative pressure in the pipette and fine control to draw the cell up into the pipette. Cells were harvested as pools of 10 individual cells/tube.

Primers for the genes that encode for Vdac1, Vdac2, Vdac3 and β-actin were designed using Clone Manager software (Sci Ed Software) (RRID:SCR_014521) to cross at least one intron-exon boundary and optimized as previously described using Power SYBR Green method (Bosch et al., 2013; Nestor et al., 2016). We have published previously the primers for β-actin (Qiu et al., 2018), which have similar efficiency as the Vdac primers (see Table1). Controls included neuronal pools reacted without reverse transcriptase (RT), hypothalamic RNA reacted with RT and without RT, as well as water blanks. Primers (Table 1) for qPCR were further tested for efficiency (E = 10^(−1/m) – 1) (Livak and Schmittgen, 2001; Pfaffl, 2001; Bosch et al., 2013).

Table 1.

qPCR was performed on a Quantstudio 7 Flex Real-Time PCR System (Life Technologies) using Power SYBR Green (Life Technologies) Master Mix according to established protocols (Bosch et al., 2013). The comparative ΔΔC_T method (Livak and Schmittgen, 2001; Pfaffl, 2001) was used to determine values from duplicate samples of 4 μl for the target genes and 2 μl for the reference gene β-actin in a 20 μl reaction volume containing 1× Power SYBR Green PCR Master Mix (Applied Biosystems) and 0.5 μM forward and reverse primers. Three 10 cell neuronal pools/animal (n=3 animals) were used to determine the relative linear quantity using the 2^-ΔΔCT equation (Bosch et al., 2013). In order to compare the relative expression levels of target genes Vdac1, Vdac2 and Vdac3 in POMC^EGFP neurons the mean Δ CT for target gene Vdac2 was used as the calibrator.

Measurements of VDAC activity in planar lipid membranes

Planar lipid bilayers were made by the apposition of two monolayers comprised of the 1:1 (w:w) mixture of dioleoyl-phosphatidylcholine (DOPC) and dioleoyl-phosphatidylethanolamine (DOPE) lipids (1:1 w/w) or diphytanoyl-phosphatidylcholine (DPhPC) (Avanti Polar Lipids, Alabaster, AL) dissolved in pentane. These bilayers were formed across a ∼125 μm (in multichannel experiments) and 75 μm (in single-channel experiments) diameter aperture in the 15-μm-thick Teflon partition separating two ∼1.5-mL compartments, as previously described (Rostovtseva et al., 2006). The recombinant human VDAC2 was a generous gift of Tsyr-Yan Dharma Yu (Institute of Atomic and Molecular Sciences, Academia Sinica, Taipei, Taiwan). VDAC2 channel insertion was achieved by the addition of 0.2-0.5 µl of VDAC2 diluted in 100 mM Tris, 50 mM KCl, 1 mM EDTA, 15% (v/v) DMSO, 2.5% (v/v) Triton X-100, pH 7.35 to the cis (grounded) compartment of the experimental chamber while stirring, as described previously (Rostovtseva et al., 2006; Rostovtseva and Bezrukov, 2015).). Ion current recordings were acquired using an Axopatch 200B amplifier (Axon Instruments) in the voltage clamp mode, following previously reported protocols (Rostovtseva and Bezrukov, 2015). Current signals were filtered at 10 kHz with a low-pass Bessel (8 pole) filter, digitized at a sampling frequency of 50 kHz, and analyzed using pClamp10.7 software (Axon Instruments For single-channel analyses in Clampfit 10.7, recordings were further digitally filtered with a 5 kHz 8-pole Bessel low-pass filter. Potential is defined as positive when it is greater on the cis side, the side of compound STX and VDAC2 addition. STX was added from its stock solution in DMSO.

Voltage dependence of VDAC2 gating was analyzed using a previously described protocol (Rostovtseva et al., 2006; Teijido et al., 2014) under the application of a slow, symmetric triangular voltage wave (±60 mV, 5 mHz) generated by an Arbitrary Waveform Generator 33220A (Agilent Technologies). Data were sampled at 2 Hz frequency and analyzed as detailed earlier (Teijido et al., 2014) using a custom in-house algorithm (Rappaport et al., 2015) and pClamp 10.7 software (RRID:SCR_011323). In each experiment, current records were collected from membranes containing more than 20 channels in response to 5–10 periods of voltage waves to ensure data collection from more than 100 channels per experiment. Only the part of the wave, during which the channels were reopening, was used for the subsequent analysis for the reasons described elsewhere (Colombini, 1989; Teijido et al., 2014). G_min, the minimal multichannel conductance in multichannel experiments, was measured at applied potentials in the range of 50 mV < |V| < 60 mV.

VDAC selectivity was measured in 1M (cis) / 200mM (trans) KCl gradient buffered with 5 mM HEPES at pH 7.4, as described previously (Teijido et al., 2014). In all experiments, VDAC2 was added to the cis compartment. VDAC2 ion selectivity was calculated from the potential corresponding to the zero current (Ψ_rev) using a linear regression fit to the data.

Flow cytometry assay

mHypo43 cells were seeded in 24-well plates at 1x10⁵ cells/well and incubated overnight in serum-free DMEM without phenol red in the presence of 50 nM or 100 nM STX. 1 µM LIVE/DEAD^TM Fixable Near-IR Dead Cell stain (Thermo Fisher Scientific, L34976) was added to each sample and the fluorescence was measured by exciting cells with 637 nm laser and detecting emission with 780/60 nm bandpass filter. Labeled cells were acquired on a flow cytometer (BD FACSymphony A5) with FACSDiva software v9.3.1(RRID:SCR_001456). Analysis was performed in FlowJo v10.10 (RRID:SCR_008520) with gating on singlet live cells, and the geometric mean for label fluorescence was determined for each condition.

Mitochondrial membrane potential

Cells were trypsinized and incubated with 5nM TMRM (Tetramethylrhodamine methyl ester, Thermo Fisher Scientific, M20036) and LIVE/DEAD^TM Fixable Near-IR Dead Cell stain in PBS for 30 mins at 37 °C in a 5% CO₂ incubator. Cells were washed and resuspended in 250 µL PBS and acquired on low speed. The fluorescence of TMRM was measured by exciting the cells with a 561 nm laser, and the emission was detected with bandpass filter of 586/15 nm.

Mitochondrial calcium

Cells were trypsinized and incubated with 2 µM Rhod2-AM (Thermo Fisher Scientific, R1245MP) and 0.02% Pluronic^® F-127 (Sigma-Aldrich, P2443) in PBS for 45 mins at room temperature. Cells were washed and incubated with LIVE/DEAD^TM Fixable Near-IR Dead Cell stain in 250 µL PBS for 30 mins at room temperature and acquired on low speed. The fluorescence of Rhod2 was measured by exciting the cells with a 561 nm laser, and the emission was detected with bandpass filter of 586/15 nm.

Mitochondrial ATP

Cells were trypsinized and incubated with 5 µM Biotracker ATP-Red (Sigma-Aldrich, SCT045) and LIVE/DEAD^TM Fixable Near-IR Dead Cell stain in PBS for 15 mins at 37 °C and 5% CO₂. Cells were washed and resuspended in 250 µL PBS and acquired on low speed. The fluorescence of ATP-Red was measured by exciting cells with a 561 nm laser, and the emission was detected with a bandpass filter of 586/15 nm.

Seahorse Assays

Cellular bioenergetics were assessed using the Seahorse XFe96 Analyzer (Agilent Technologies) with either the XF Cell Mito Stress Test Kit (Agilent Technologies, 103015-100) or the XF Glycolysis Stress Test Kit (Agilent Technologies, 103020-100), following the manufacturer’s protocols. Seahorse XFe 96-well plates (Agilent Technologies, 103794-100) were coated with rat tail collagen type I (Gibco, A10483-01) for 1 h at room temperature, air-dried under sterile conditions, and seeded with immortalized human hepatocytes (IHH) or mHypo43 cells at 20,000 cells per well, followed by overnight incubation at 37°C with 5% CO₂. The Seahorse XFe96 sensor cartridge (Agilent Technologies, W11425) was hydrated overnight in Seahorse XF Calibrant (Agilent Technologies, 100840-000) at 37°C in a non-CO₂ incubator. On the day of each assay, cells were washed twice with Dulbecco’s phosphate-buffered saline (Gibco, 14190-144), serum-starved for 6 h, and then washed twice with Seahorse XF DMEM medium, pH 7.4 (Agilent Technologies, 103575-100). Cells were incubated for 1 h at 37°C in a non-CO₂ incubator with either 1 nM STX, 5nM STX (for mHypo43 cells), 10 nM STX, or Seahorse XF assay medium alone (control) immediately prior to the assay. To assess the impact of serum starvation, additional assays were performed in the same manner but without the 6 h serum starvation step.

For the Mito Stress Test, the Seahorse assay medium was supplemented with 10 mM glucose, 1 mM pyruvate, and 2 mM L-glutamine, and injection ports were loaded with 1.5 µM oligomycin, 1 µM FCCP, and 0.5 µM rotenone/antimycin A. Oxygen consumption rate (OCR) was measured in real time using cycles of mixing (3 min), waiting (2 min), and measuring (3 min), and basal respiration, ATP-linked respiration, maximal respiration, spare respiratory capacity, and non-mitochondrial respiration were calculated.

For the Glycolysis Stress Test, the Seahorse assay medium was supplemented with 2 mM L-glutamine only, and injection ports were loaded with 10 mM glucose, 1 µM oligomycin, and 50 mM 2-deoxy-D-glucose (2-DG).

Extracellular acidification rate (ECAR) was measured using the same cycle parameters described above, and glycolysis, glycolytic capacity, glycolytic reserve, and non-glycolytic acidification were calculated. All data were analyzed using Wave software (Agilent Technologies). Each treatment group was assayed in at least eight technical replicates per plate, and all experiments were independently repeated a minimum of three times. Wells with poor adherence or OCR/ECAR values exceeding two standard deviations from the mean were excluded from analysis.

Data are presented as mean ± standard error (SE). Statistical significance was determined using one-way ANOVA followed by Dunnett’s multiple comparison test.

Data availability

All data generated or analyzed during this study are included in the manuscript and supporting files. Source data files have been provided for all figures. Additional data and analysis code are available from the corresponding author upon reasonable request.

Acknowledgements

The authors would like to thank Jin Hui-Deng for his assistance in breeding and genotyping the transgenic mouse lines at OHSU. The click chemistry, molecular biological and electrophysiological studies were supported by the National Institutes of Health grants: NIA grant R21 AG080057 (MJK & CS) and NIDDK grant R01 DK068098 (MJK & OKR). Research at the Intramural Research Program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health was funded by NIH grant ZIA HD000072-18 (SMB). The contributions of the NIH authors (SMB, TKR, MR, MW, WF) are considered Works of the United States Government. The findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the National Institutes of Health or the U.S. Department of Health and Human Services.

Additional files

Supplementary file 1. Additional experimental details, materials, methods and supporting figures, including NMR spectra.

Additional information

Funding

National Institute on Aging (R21 AG080057)

• Carsten Schultz

National Institute of Diabetes and Digestive and Kidney Diseases (R01 DK068098)

• Oline K Rønnekleiv

National Institutes of Health (ZIA HD000072-18)

• Sergey Bezrukov