Stereospecific properties and intracellular transport of novel intrinsically fluorescent neurosteroids

Abstract

Allopregnanolone (AlloP) is a neuroactive steroid (NAS), which is a potent allosteric activator of the gamma-aminobutyric acid A (GABA_A) receptor. The mechanisms underlying the biological activity of AlloP and other NAS are only partially understood. Here, we present intrinsically fluorescent analogs of AlloP (MQ-323) and its 3β-epimer, E-AlloP (YX-11), and show, by a combination of spectroscopic and computational studies, that these analogs mimic the membrane properties of AlloP and E-AlloP very well. We found stereospecific differences in the orientation and dynamics of the NAS as well as in their impact on membrane permeability. However, all NAS are unable to condense the lipid bilayer, in stark contrast to cholesterol. Using Förster resonance energy transfer (FRET) and electrophysicological measurements, we show that MQ-323 but not YX-11 binds at the intersubunit site of the ELICα₁GABA_A receptor and potentiates GABA-induced receptor currents. In aqueous solvents, YX-11 forms aggregates at much lower concentrations than MQ-323, and loading both analogs onto cyclodextrin allows for their uptake by human astrocytes, where they become enriched in lipid droplets, as shown by quantitative fluorescence microscopy. Trafficking of the NAS analogs is stereospecific, as uptake and lipid droplet targeting is more pronounced for YX-11 compared to MQ-323. In summary, we present novel minimally modified analogs of AlloP and E-AlloP, which enable us to reveal stereospecific membrane properties, allosteric receptor activation, and intracellular transport of these neurosteroids. Our fluorescence design strategy will be very useful for the analysis of other NAS in the future.

Graphical Abstract

Introduction

Neuroactive steroids (NAS) are steroids that modulate brain function in various ways, thereby impacting neuronal activity, stress, memory, and mental health. NAS are produced from cholesterol, either in the periphery to cross the blood-brain barrier, or they are synthesized locally in the brain.¹ Upon cholesterol import into mitochondria, steroid-producing cells can convert it into pregnenolone by P450 side-chain cleaving enzyme.² Pregnenolone is the precursor for a variety of other sterols, including the sex hormones testosterone and estradiol, but also of progesterone and its metabolite, allopregnanolone (AlloP). AlloP is a NAS that rapidly alters neuronal activity by being a potent allosteric activator of γ-aminobutyric acid A (GABA_A) receptor, a chloride-conducting ligand-gated ion channel.³ By specific binding to distinct sites in the transmembrane domain (TMD) of GABA_A receptors, AlloP causes an allosteric transition in the extracellular domain (ECD) of the receptor, leading to increased ligand (i.e., GABA) binding and, thereby enhanced chloride influx into cells.^4–6 GABA_A receptor activation thus hyperpolarizes the plasma membrane of neurons, thereby reducing the ability to generate action potentials.^3,7 Mostly due to this non-genomic regulation of neuronal activity, AlloP and its formulations are used to treat patients with depression, and have been tested in clinical trials for treatment of insomnia, and anxiety disorders.³ This is because the balance between excitatory and inhibitory neuronal signaling is important for the regulation of mental health and sleep rhythm.⁸ AlloP and other NAS are also evaluated for clinical use to treat severe neurological and neurodegenerative diseases, such as epilepsy, major depressive disorder (MDD), multiple sclerosis (MS) as well as Parkinson’s disease (PD), and Alzheimer’s disease (AD).^9–11 AlloP and related NAS are also considered potent regenerative agents to reduce neuroinflammation after injury or in the context of the above-mentioned neurodegenerative diseases.¹² The potential of these NAS to reduce inflammation is at least partly a consequence of their action on microglia and astrocytes, either by binding to GABA_A receptors in these cells, thereby inhibiting NFκB signaling, by regulating the activity of Toll-like receptors or by exerting control on autophagy.^3,12,13 The ability of AlloP to stimulate repair could also become an important treatment option for MS.¹⁴ Mechanistically, AlloP can increase the activity of synaptic or extrasynaptic GABA_A receptors leading to phasic or tonic inhibition of neuronal activity, respectively.³ Interestingly, epi-allopregnanolone (E-AlloP), a diastereomer of AlloP, inhibits rather than activates GABA_A receptors and can reverse the stimulating effect of AlloP on GABA_A receptor mediated chloride influx by binding to distinct sites in the receptor.^4,6,15,16 To reach the different binding sites in the TMD of the various subunits of the GABA_A receptor, NAS like AlloP and E-AlloP must first be inserted into the lipid bilayer.¹⁷ Since membrane-active steroid hormones can modulate membrane properties,^18,19 it cannot be excluded that some of the effects of these two NAS are mediated by modulating bulk membrane properties, in addition to directly exerting control on receptor activity.^1,17 Accordingly, the question arises, whether differing membrane interactions affect the distinct biological activities of AlloP and E-AlloP. Despite being of central importance, so far, only few studies have directly compared the membrane properties of neurosteroids.^20,21 The impact of membrane interactions on function is a general feature of not only NAS but also a variety of other anesthetics and small molecules, whose absorption to the lipid bilayer can change the protein conformational energy landscape.^22–24 In fact, a recent hypothesis is that general anesthesia can impact ligand-gated ion channels, such as GABA_A receptors, via their impact on the lateral pressure profile in the lipid bilayer.^23,25 Theoretical studies suggest that this mechanism can even explain the complex electrophysiological behavior of the GABA_A receptor in the presence of ligands and agonists.^22–24

Additionally, the availability of both NAS for GABA_A receptor modulation might depend on their transport in cells, but almost nothing is known about the trafficking of AlloP and E-AlloP in neuronal cells. Since the effect of these NAS can last over hours and even days after single-dose application, intracellular sequestration and slow release from such depots could be possible. Studying intracellular transport of NAS depends on suitable analogs, which allow for tracking AlloP and E-AlloP in cells. One approach for that is to employ neurosteroid analogs tagged with an organic dye, such as nitrobenzoxadiazole (NBD) or a clickable Alexa488-azide, the latter linked via an extra alkyne group in the molecule.^21,26,27 Dye-conjugated neurosteroids, however, will have very different physicochemical properties compared to the natural NAS, as found in a large body of work for the related sterol cholesterol (reviewed in Ref. 28). Even though these dye-tagged neurosteroid analogs might activate the GABA_A receptor,²⁷ their membrane properties and interaction with other lipid species can differ from natural NAS. Thus, the polarity and charge of the attached dye will compromise the precise action of NAS in the lipid bilayer, such that potential differences between stereoisomers of the same neurosteroid cannot be uncovered.

In this study, we present two new intrinsically fluorescent derivatives of AlloP (MQ-323) and E-AlloP (YX-11), which differ from their natural counterparts only by containing three conjugated double bonds in the ring system (Figure 1A and B). We and others have shown in previous studies that this approach of introducing extra double bonds into sterol molecules is very versatile and allows for studying sterol-protein interactions and intracellular trafficking for a variety of biologically relevant sterols without significantly compromising their activity.^28–30 Minimally modified sterols also allow for studying the impact of functional modifications of sterols on their membrane interaction, protein binding, or even intracellular transport, as we showed for analogs of cholesterol (i.e., cholestatrienol, CTL) versus oxysterol analogs, which differ from CTL only in having an extra hydroxy group in the side chain (i.e., 25- and 27-hydroxy-CTL).^31–34 We show here that the distinct properties of AlloP and E-AlloP for differential binding to and activation of the GABA_A receptor are preserved among the new fluorescent neurosteroids. Furthermore, we provide evidence for similar membrane properties of MQ-323, the analog of AlloP, and YX-11, the analog of E-AlloP, but we also observe differences, such as stereospecific self-aggregation. Finally, we assess cellular uptake and trafficking of both analogs by UV-sensitive fluorescence microscopy and find quantitative differences in their subcellular transport. Our results demonstrate the potential of intrinsically fluorescent neurosteroid analogs for elucidating the molecular function and biological activity of NAS.

Figure 1:

Allopregnanolone enantiomers and their fluorescent analogs can order phospholipid acyl chains. Chemical structures of allopregnanolone and epi-Allopregnalone (A) and their fluorescent analogs YX-11 and MQ-323 (B), respectively. The structure of allopregnanolone and epi-allopregnanolone differ from each other only by the orientation of the 3-hydroxyl group (highlighted in green and red, respectively). The fluorescent properties of the allopregnanolone and epi-allopregnanolone analogous, named YX-11 and MQ-323, are provided by the conjugated double bonds (highlighted in violet). Acyl chain order parameter calculated from MD trajectories for the sn-1 chain (palmitic acid, PA in panel C) and the sn-2 chain (oleic acid, OA in panel D) for membranes consisting of POPC and the indicated combinations of cholesterol (CHL) and neurosteroids.

Results

Membrane properties of AlloP, E-AlloP and their fluorescent analogs

AlloP and E-AlloP differ from each other only in the orientation of their 3-hydroxy group, and their analogs, MQ-323 and YX-11, show the same difference while containing three conjugated double bonds in the ring system, as shown in Figure 1A and B. Thus, the only difference between MQ-323 and YX-11 is the orientation of the 3-hydroxy group, exactly like the difference between AlloP and E-AlloP, allowing us to study the stereospecific effects of the NAS in a fluorescence setting. The parent molecule from which steroid hormones are synthesized, cholesterol, is well-known to condense fatty acyl chains in fluid lipid membranes, and it has been postulated that NAS can modulate membrane properties in a stereospecific manner.^25,35 To address this point and validate our novel fluorescent steroid analogs, we used molecular dynamics (MD) simulations of AlloP, E-AlloP, as well as MQ-323 and YX-11, and determined the acyl chain order parameter for the host phospholipid POPC in cholesterol-containing membranes. We found that none of the NAS caused additional acyl chain ordering in cholesterol-containing membranes (Figure 1C and D). Additional simulations of the natural NAS, AlloP and E-AlloP, using higher steroid mole fractions (i.e., between 10 and up to 60 mol%) in POPC membranes in the absence of cholesterol showed that these neurosteroids could order the phospholipid acyl chain of POPC, though to a much lower degree than cholesterol (Figure S1). Also, under those conditions, their membrane orientation became more confined and similar to that of 30 mol% cholesterol, while their dynamics were dramatically slowed (Figure S2 and S3). For all conditions tested, except the highest sterol/steroid mole fraction of 0.6, cholesterol has the highest membrane condensing effect, as reflected by the reduction in area per lipid (Figure S3). Together, NAS can affect the properties of lipid bilayers but only at very high membrane concentrations, which are not observed under physiological conditions.

The differing orientation of the 3-hydroxy group in AlloP versus E-AlloP could affect the interaction of either NAS with membrane phospholipids. To test this hypothesis, we quantified the extent of tilting of the steroids from MD simulations of the bilayers at physiologically relevant concentrations, i.e., in membranes consisting of POPC with 25 mol% cholesterol and 5 mol% neurosteroid. We found that all four NAS studied have a rather broad tilt angle distribution in both membrane leaflets (Figure 2A–D). This is in contrast to cholesterol, whose membrane orientation is much more confined around 15 degree tilt relative to the bilayer normal (Figure 2E). All sterols show excursions in their membrane orientation, which vary for cholesterol between 0 and max. 40 degrees, while the NAS can lie flat in the bilayer and even flip to the opposite membrane leaflet. We found a slightly more tilted orientation of AlloP and its fluorescent analog MQ-323 compared to E-AlloP and its analog YX-11. In contrast, both fluorescent neurosteroids resembled their natural counterparts closely (Figure 2A–D). Thus, the orientation of the 3-hydroxy group has a slight effect on membrane anchoring, and a β-orientation, like in cholesterol, gives stronger interactions compared to an α-orientation, as in AlloP. To sustain this conclusion, we analyzed the hydrogen bonding pattern and reorientation dynamics of the four NAS from the MD simulations. We found that E-AlloP and YX-11 form slightly more hydrogen bonds to water and neighboring POPC molecules compared to AlloP and MQ-323, respectively (Figure 3A–D). Furthermore, the larger tilting caused a much slower reorientation dynamics of AlloP and MQ-323 in the membrane compared to E-AlloP and YX-11 (Figure 3E). The fastest reorientation was found for cholesterol, which is a result of its upright orientation, allowing for unhindered rotation around its long molecular axis. The much larger tilting and slowed rotational dynamics of the four studied NAS compared to cholesterol is in line with our earlier observations with side chain oxysterols.^32,34

Figure 2:

Membrane localization and orientation of allopregnanolone enantiomers (A-B) and their fluorescent analogs (C-D) and cholesterol (E). Density heatmaps (2D-histograms) over the tilt angle and distance from the membrane center (z) are shown.

Figure 3:

Hydrogen bonding capacity and membrane dynamics of allopregnanolone enantiomers and their fluorescent analogs. The number of hydrogen bonds formed between the allopregnanolone enantiomers and water (blue lines), the POPC carbonyl groups (orange lines), and cholesterol’s hydroxyl group (green lines) are shown for all neurosteroids (A-D). The black lines indicate the total number of hydrogen bonds formed by membrane-embedded neurosteroids. Membrane dynamics of neurosteroids were assessed in relation to that of cholesterol from the decay of the autocorrelation function as a function of time, ACF(t), as shown in (E). Panels (F-D) show snapshots from the MD simulations of MQ-323 and YX-11. The neurosteroid molecules are highlighted as spheres.

To follow up on these observations, we carried out experiments with the NAS in model membranes using ²H-NMR spectroscopy. Assessment of the acyl chain order parameter of POPC membranes containing either AlloP, E-AlloP, MQ-323, or YX-11 by NMR spectroscopy showed that none of these NAS can order phospholipid acyl chains to a significant extent (Figure 4A and B). This is, again, in stark contrast to cholesterol, while substituting some cholesterol with the NAS, also did not affect membrane ordering (Figure 4C). Based on these observations, we conclude that - at least at physiologically relevant concentrations (i.e., here 10 mol%) - neither AlloP nor E-AlloP or their fluorescent derivatives can cause structural ordering of lipid bilayers on the time scale accessible by NMR spectroscopy. There was a slight tendency of acyl chain ordering by 10 mol% of the NAS in POPC membranes as assessed by MD simulations (Figure S1). This minor discrepancy between NMR and MD results could be due to the different time scales accessible by both methods. Again, at un-physiologically high concentrations, we do observe membrane ordering by the NAS paralleled by less tilting and faster rotation around the long molecular axis (Figure S2A and B).

Figure 4:

Order parameter of neurosteroids and their effect on membrane permeability to dithionite. The order parameter of the sn-1 chain of deuterated POPC in the indicated membranes was measured by ²H-NMR spectroscopy (A-C). Permeability of large unilamellar vesicles consisting of the indicated lipids and containing 0.5 mol% NBD-PC to sodium dithionite was measured by fluorescence spectroscopy (D). A bi-exponential fit to the data gives the rate constant for access of dithionite to the inner membrane leaflet.

While we did not see the opposite effect, i.e., a decreased ordering of acyl chains in the presence of NAS, the largely upright orientation of the neurosteroids could imply that they can reduce the membrane permeability barrier. To test this idea, we measured the permeation of dithionite across lipid membranes labeled with the fluorescence phospholipid NBD-PC. Reduction of the fluorescence of NBD-PC by dithionite is biphasic with a fast phase resembling the chemical reduction step and a slow phase resembling dithionite permeation and/or flip-flop of the NBD-PC across the membrane.³⁶ This assay has been widely used to assess membrane perturbation by drug molecules,^37,38 but also to determine the impact of structural modifications of cholesterol on membrane permeability.^39–42 Using this assay, we found that YX-11, and to a lower extent E-AlloP, reduced membrane permeability slightly (Figure 4D). This effect was limited to POPC membranes, while in POPC/cholesterol membranes, the permeability for dithionite was reduced irrespective of the added NAS (Figure 4D). The latter is a consequence of the presence of cholesterol, which is known to reduce the permeability of liposomes for dithionite.

Neurosteroid analogs bind to the ELICα₁GABA_A receptor and regulate its activity in a stereo-specific manner

Having ruled out major differences in membrane properties of the neurosteroid epimers, we assessed next their interaction with GABA_A receptors. While AlloP is known as a positive allosteric modulator of GABA_A receptors, Epi-AlloP does not bind to the same site as AlloP, and it is an important property for fluorescent analogs of both steroids to show the same stereospecific binding pattern.¹ A particular advantage of intrinsically fluorescent sterols and steroids is that their excitation spectra significantly overlap with the emission spectra of aromatic amino acids, allowing for direct measurement of binding by Förster resonance energy transfer (FRET). This approach has been widely used to study protein-sterol interactions, e.g.,^33,43,44 and we recently also applied it to two fluorescent neurosteroid analogs.⁴⁵ To determine the affinity of the new fluorescent neurosteroids, purified ELICα₁GABA_A receptor was first incubated with increasing concentrations of the AlloP analog MQ-323. FRET was measured by exciting the donor fluorescence (i.e., that of the protein at 280 nm) and recording the emission of MQ-323 (Figure 5A). Due to energy transfer, the sensitized emission of MQ-323 increases upon binding to the receptor in a saturable manner (Figure 5B). Since FRET efficiency is a function of proximity and mutual orientation of donor and acceptor dipoles, not only specific binding but also non-specific association of MQ-323 or its insertion into the detergent micelle of the reconstituted receptor could result in a FRET signal. To account for those contributions, the binding experiment was repeated in the presence of 30 μM AlloP, and the resulting FRET signal was subtracted from the determined binding curve (Figure 5B). As a further test for binding specificity, the experiments were repeated with a mutant form of ELICα₁GABA_A receptor, in which Trp246, in transmembrane helix 1 of the receptor, was mutated to leucine (W246L). The Trp246 residue is essential for the binding of AlloP (analogs) in the canonical intersubunit site on GABA_A receptors and for the potentiation of GABA-elicited currents by AlloP.^4,46–48 In the W246L mutant receptor, the FRET intensity of MQ-323 was indistinguishable from the non-specific FRET signal observed in the wild-type ELICα₁GABA_A receptor, supporting that MQ-323 specifically binds at the previously identified site involving transmembrane helix 1 of the receptor.^4,45 In contrast to MQ-323, adding YX-11 to the wild-type ELICα₁GABA_A receptor did not result in any FRET signal, clearly showing that the fluorescent analog of E-AlloP (like E-AlloP) does not bind to the canonical neurosteroid site on the receptor (Figure 5C).⁴ Thus, the new intrinsically fluorescent analogs of AlloP and its epimer are powerful tools for detecting stereospecific interactions with the GABA_A receptors.

Figure 5:

Förster resonance energy transfer (FRET) of neurosteroid binding to GABAa receptors. Purified ELICα₁GABA_A receptor was incubated with either MQ-323 or YX-11 dissolved in a buffer containing 0.015% DDM, and FRET from the protein to the respective NAS was measured upon excitation at 280 nm. Raw emission spectra (A) and normalized FRET intensities (B, n = 3 for each plot) for increasing concentrations of MQ-323. Nonspecific binding in (B) is the FRET signal obtained in the presence of 30 μM allopregnanolone, and W246L shows that the FRET is reduced to non-specific binding when W246 in the receptor is replaced by leucine. Comparison of total and non-specific FRET obtained with MQ-323 and YX-11 (C), showing that YX-11 does not bind to the ELICα₁GABA_A receptor.

Next, we asked whether the stereospecific binding of the novel neurosteroid analogs also contributes to allosteric regulation of the GABA_A receptor. We found that MQ-323 potentiates the human α1β3γ2L GABA_A receptor activated by low (<EC₁₀) GABA strongly, while YX-11 had a very weak potentiating effect on receptor function (Figure 6A and B). When the receptors were maximally activated in the presence of saturating GABA and 50 μM propofol, YX-11 weakly inhibited receptor function (Figure 6C). Together, these results demonstrate that the new intrinsically fluorescent analogs of AlloP and its epimer, E-AlloP, maintain the ability of endogenous NAS to modulate GABA_A receptor activity. This makes these probes highly suitable for functional studies of neurosteroid function in the future.

Figure 6:

Electrophysiological measurements of GABA_A receptor currents. The α1β3γ2L receptors were activated by 2 μM GABA (A, B) or 1 mM GABA + 50 μM propofol (C), and modulated by 10 μM MQ-323 or YX-11. Drug applications are shown by horizontal lines above current traces. The dot plots show steroid effects in % of control response (100% = no effect) calculated as ratios of current amplitudes after and before application of steroid.

Neurosteroid analogs show stereo-specific solubility in aqueous solution

Based on their intrinsic fluorescence, the analogs of AlloP and E-AlloP, should be suitable for live-cell imaging studies, so we aimed for labeling human astrocytes with either MQ-323 or YX-11. We found that diluting both probes into the culture medium resulted in poor labeling but gave bright extracellular aggregates, particularly in the case of YX-11 (Figure S4). Aggregates show delayed photobleaching, which, besides their bright fluorescence, can be used to identify them in the microscope. To follow up on these observations, we studied the emission spectra of the fluorescent neurosteroids in water and found that YX-11 but not MQ-323 shows a pronounced red-shift and change in spectral shape when the steroid concentration exceeds ca. 9 μM (Figure 7A and B). By plotting the maximal emission at 378 nm for MQ-323 and 416 nm for YX-11 as a function of steroid concentration and using a bi-linear regression, a breakpoint at 8.99 μM was found for YX-11, which likely resembles the critical aggregation concentration (CAC) for this steroid in water (Figure 7C). As for MQ-323, no breakpoint was observed, and the emission spectra had the same shape over the entire studied concentration range (Figure 7B). The integrated emission of YX-11 and MQ-323 is similar in most solvents, and the red-shifted emission of YX-11 was only observed in water and PBS (Figure S5). These results suggest that the self-aggregation of the fluorescent steroids is stereospecific. In an attempt to uncover the molecular mechanisms underlying this stereospecific aggregation, we carried out MD simulations of YX-11 and MQ-323 monomers in water/ethanol mixtures at such high concentrations (0.1 M), that aggregate formation can be followed in a few μsec, achievable by atomistic molecular simulations. Under those conditions, the formation of aggregates was readily observed for both steroid probes once the ethanol content dropped below ca. 50% (v/v) (Figure S6). Thus, both neurosteroids can form aggregates in aqueous solution. To test whether self-aggregation might also take place in membranes, both NAS analogs were incorporated in large unilamellar vesicles (LUVs), and fluorescence spectra were recorded. The characteristic red-shift in emission was found for YX-11 but not for MQ-323, even at 20 mol% steroid probe in the membrane, but it was only observed for YX-11 at a high total lipid concentration of 200 μM (Figure S7).

Figure 7:

Fluorescence emission spectra of YX-11 and MQ-323 titration in water. (A+B) Normalized fluorescence emission spectra (Ex. 328nm) of YX-11 and MQ-323 with various concentrations (0 to 50 μM) in water. The spectra represent the average from three independent measurements and are normalized to the maximum peak intensity. (C) The mean of the maximum emission at 416 nm and 378 nm from three individual experiments of YX-11 (orange) and MQ-323 (blue), respectively, plotted as a function of neurosteroid concentration and fitted with a straight-line model with breakpoints using a piece-wise linear regression routine.

Loading of AlloP onto cyclodextrin has been shown to improve its bioavailability in drug formulations¹⁰ and we, therefore, mixed MQ-323 and YX11 with low concentrations of MβCD to prevent self-aggregation of the neurosteroids prior to cell labeling. We found that the emission spectrum of YX-11 above its CAC in PBS/cyclodextrin solutions became identical to that below its CAC in PBS (Figure S8) and was indistinguishable from that of MQ-323 (Figure S8B). In line with these observations, we observed by UV-sensitive fluorescence microscopy more intense and homogeneous labeling of human astrocytes by both fluorescent NAS probes upon loading them onto cyclodextrin (Figure 8) but not when loading them onto albumin (not shown). Strikingly, the uptake of YX-11 was higher than that of the bioactive AlloP analog, MQ-323, when comparing cell-associated intensities of both probes with autofluorescence (Figure 8). We chose human astrocytes for these experiments, as these glia cells are very important in GABAergic signaling and use neurosteroids in their communication with neurons.⁴⁹

Figure 8:

Uptake of neurosteroid analogs loaded on methyl-β-cyclodextrin (MCβD) by astrocytes. Human astrocytes were loaded with either YX-11 and MQ-323 on MCβD (final concentration of 20 μM) and incubated for 1 hr before imaging on a UV-sensitive wide-field microscope (A-B). Scale bar, 20 μm. Cells were segmented using Cellpose (see Image analysis for fluorescence microscopy data) in order to find the mean UV fluorescence intensity for cells loaded with YX-11 (n = 580) or MQ-323 (n = 488) (C). Control in panel C are cells without loading of neurosteriods (n = 290). By uing a Mann-Whitney U test on the distributions for YX-11 and MQ-323 in panel C, a p-value of 9.510 × 10⁻³² was found.

Neurosteroid analogs accumulate in lipid-droplets in astrocytes

To determine the trafficking itineraries of our novel fluorescent NAS analogs, we carried out co-staining experiments with known organelle markers. NAS were administered in MβCD and allowed to equilibrate for 30 minutes before loading to cells. Both YX-11 and MQ-323 showed little - if any - co-localization with NBD-Ceramide, a marker for the Golgi apparatus and trans-Golgi network (TGN, Figure S9). Co-staining with the Golgi marker was slightly higher for YX-11 compared to MQ-323, but this could also be due to the increased uptake and, therefore, higher signal for the E-AlloP analog. Both NAS probes became enriched in lipid droplets co-stained with the droplet marker BODIPY 493/503 (Figure 9). Importantly, the extent of this droplet targeting was much more pronounced for YX-11 compared to MQ-323, as validated by image quantification (Figure 9). No colocalization was found with a marker for late endosomes and lysosomes (LE/LYSs), ruling out that any of the punctuate stainings of the fluorescent neurosteroids are lysosomal (Figure S10). Similarly, the staining of the PM was rather low compared to the intracellular accumulation of both NAS probes. We conclude that fluorescent analogs of AlloP and E-AlloP are transported to lipid droplets and that the extent of this intracellular accumulation is stereospecific. We repeated the uptake experiment with an ACAT inhibitor (Avasimibe) and found a strongly reduced formation of lipid droplets and consequently no or very little droplet targeting of YX-11 and MQ-323 to the few remaining droplets, despite a slightly higher uptake of YX-11 by the cells upon ACAT inhibition (Figure S11). Thus, droplet formation requires ACAT activity. A hypothesis for the preferred droplet targeting of YX-11 is that this E-AlloP analog is preferentially esterified by ACAT and therefore enriched in lipid droplets. Supporting this hypothesis are studies carried out for cholesterol and its 3’-hydroxy α-epimer, epi-cholesterol, which showed that the β-orientation of the 3’-hydroxy group, as in cholesterol, is necessary for sterol esterification.^51,52 Thus, it is possible that YX-11 with its 3’-hydroxy group in β-orientation is recognized by ACAT and therefore esterified and stored in droplets, while MQ-323 is no ACAT substrate. Unfortunately, the fluorescence of the neurosteroid probes in the very few remaining droplets after ACAT inhibition is too low to test this model directly. Together, our results show that transport of the fluorescent neurosteroids to droplets requires ACAT activity, and they suggest that the stereospecific accumulation of allopregnanolone epimers is at least partly the result of different esterification e iciency of fluorescent analogs of AlloP and E-AlloP.

Figure 9:

Neurosteriods are localized in lipid droplets (LDs). Human astrocytes were loaded with either YX-11 or MQ-323 on MCβD (final concentration of 20 μM), incubated for 1 hr and afterward stained with bodipy for 5 min before imaging on a UV-sensitive wide-field microscope (A-B). Scale bar, 20 μm. Cells and LDs were segmented as described in Image analysis for fluorescence microscopy datas, and the fraction of YX-11 (n = 290) or MQ-323 (n = 223) found inside LDs per cell was quantified (C). A fraction above 1 indicates the upregulation of neurosteroids in LDs. By using a Mann-Whitney U test on the distributions in panel C, a p-value of 7.66 × 10⁻³⁰ was found.

Discussion

NAS, like AlloP, are potent regulators of mood, stress handling, and sleep rhythm due to their ability to potentiate the effect of GABA on the GABA_A-mediated chloride currents. The effects of these NAS are site-specific and stereo-specific, largely due to specific receptor-ligand interactions. Dissecting such interactions at a molecular level requires suitable NAS probes, which mimic endogenous neurosteroids closely but preserve the key pharmacological characteristics of each steroid analog. Here, we present two novel analogs of AlloP and its 3β-epimer, E-AlloP, which can be used to study not only neurosteroid binding to specific sites in the GABA_A receptor by an on-line fluorescence spectroscopy assay, but also to directly observe intracellular trafficking of NAS by UV-sensitive fluorescence microscopy. We show that the membrane properties of MQ-323, the fluorescent derivative of AlloP, and of YX-11, the analog of E-AlloP, are very similar at low steroid concentrations, which is in line with previous studies.^1,20 MD simulations and permeation studies with dithionite indicate that E-AlloP and its analog YX-11 are slightly better anchored in lipid membranes compared to AlloP and its analog MQ-323, suggesting that the β-orientation of the 3-hydroxy group, as also found for cholesterol, is important for hydrogen bonding of sterols and their interaction with phospholipids. However, the difference is minor, and the capacity to order fatty acyl chains is negligible for all studied NAS, in strong contrast to cholesterol. Since neurosteroids have a much shorter side chain compared to cholesterol, the alkyl chain of cholesterol is essential for its strong membrane condensing effect, in line with previous studies.⁴⁰ Unexpectedly, we found that YX-11 self-aggregates at much lower concentrations than MQ-323 in aqueous solutions and likely also in membranes. Whether aggregates are inserted into the lipid bilayer or directly formed once the E-AlloP analog accumulates in the membrane remains to be shown. Similarly, the molecular mechanisms underlying stereospecific aggregation of AlloP analogs need to be determined.

Our new intrinsically fluorescent NAS not only allow for studying properties of neurosteroids in solutions and in membranes, but also enabled us to infer stereospecific interactions with GABA_A receptors. We demonstrate that the stereospecific interaction and regulation of the ELICα₁GABA_A receptor by AlloP is preserved for its fluorescent analog MQ-323, while the analog of E-AlloP cannot bind or activate the receptor to a comparable extent. Thus, due to their minimal chemical modification compared to the natural NAS epimers, MQ-323 and YX-11 are sutiable probes for studying stereospecific allosteric regulation of GABA_A receptors by combined FRET spectroscopy and electrophysicology.

When loading the fluorescent NAS onto cyclodextrin, their aqueous solubility was comparable, and efficient labeling of human astrocytes could be achieved. This underlines the importance of proper formulations for neurosteroid targeting to cells.¹⁰ It also supports and explains previous findings showing that the slow action of NAS on GABA_A receptors can be accelerated with cyclodextrin:⁵³ based on our imaging results, we conclude that increased retention of NAS in cells, when loaded onto cyclodextrin, results in a larger cellular pool being available for receptor activation. Thus, our findings support a model of membrane-mediated access of AlloP to the GABA_A receptor, which is in full support of the lipophilic nature of the NAS.²⁶ While we could load the NAS probes onto albumin, labeling of cells with steroid-albumin complexes was inefficient (not shown). Future studies could attempt cellular delivery by other proteins, such as steroid carrier proteins, which have been proposed to be key regulators of trafficking and activity of NAS.² Using multi-color live-cell microscopy, we could show that human astrocytes internalize more of the E-AlloP analog, YX-11, compared to the AlloP analog, MQ-323. This cellular enrichment with the E-AlloP analog is largely due to its preferred transport to lipid droplets.

Our results are in line with previous findings, showing that (analogs of) NAS accumulate at intracellular sites,^26,53,54 and extend these observations by showing that the intracellular pool is primarily consisting of lipid droplets and to a smaller extent the Golgi/TGN and likely the endoplasmic reticulum (ER). Accumulation of a clickable analog of AlloP in the Golgi has been shown previously in neurons.²⁷ Based on these findings, a model can be envisioned, according to which AlloP and other NAS are continuously supplied to the plasma membrane (PM) from lipid droplets, either directly by non-vesicular trafficking or by secretory membrane traffic via the Golgi/TGN. Future studies using MQ-323 and YX-11 in combination with genetic or chemical inhibition of selected trafficking pathways will provide further insight into the transport pathways of NAS in cells.

Materials and Methods

Computational details

Molecular dynamics simulations

Membrane bilayers containing in each leaflet 70 POPC lipids, 25 cholesterol molecules, and 5 neurosteroid molecules were assembled using packmol. Membranes were assembled for allopregnanolone, epiallopregnanolone, and their fluorescent analogs. Additionally, a membrane with a 70:30 ratio of POPC:cholesterol was generated for comparison purposes. The systems also included 10,000 water molecules, 22 K⁺, and 22 Cl⁻ ions, corresponding to a concentration of about 120 mM KCl. The water molecules were described by the TIP3P force field,⁵⁵ while the POPC lipids and cholesterol were described by the amber lipid14 force field.⁵⁶ Force field parameters for the neurosteroids were derived using the QForce⁵⁷ program package, based on r2scan-3c⁵⁸ Hessian calculations carried out using the Orca program,⁵⁹ version 5.0.3. Charges were derived using electrostatic potential fitting. Lennard-Jones parameters were assigned from GAFF.⁶⁰ Flexible dihedral parameters were fitted to relaxed dihedral scans, also using r2scan-3c.

All molecular dynamics simulations were carried out using the Gromacs program,⁶¹ version 2023.1. The membranes were equilibrated in a three-stage procedure. First, the structures were minimized for 5000 steps with a steepest descent minimizer to remove any bad contacts. Next, we carried out NVT equilibration (200 ps), and NPT equilibration (2 ns) was carried out. A time-step of 2 fs was used. Long-range electrostatics were treated with the particle mesh Ewald method⁶² with a cutoff distance of 12 Å. Bonds involving hydrogens were constrained using LINCS.⁶³ The temperature was controlled by a Nose-Hoover thermostat^64,65 (towards 298.15K), while the pressure (in the NPT simulations only) was controlled by a semi-isotopic Berendsen barostat⁶⁶ (towards 1 bar). After equilibration, production molecular dynamics runs (1000 ns) were carried out. In the production simulations, we switched to a Parrinello-Rahman barostat.⁶⁷ The resulting trajectories were analyzed using Gromacs (for deuterium order parameters), and in-house Python scripts relying on the MDAnalysis library⁶⁸ for tilt angle distributions, hydrogen bonding, and autocorrelations of the tilting angle.

Aggregation simulations of MQ-323 and YX-11 were carried out in ethanol/water solvent mixtures of varying solvent composition (0%, 20%, 40%, 60%, 80%, and 100% ethanol concentration by volume). In each simulation, 100 neurosteroid molecules were solvated in a volume of about 1570 nm³, corresponding to a neurosteroid concentration of about 0.1M. Force-field parameters were assigned as described above, with parameters for ethanol being derived using Q-Force based on r2scan-3c QM calculations. Solvent/neurosteroid mixtures were assembled using the Packmol program, with the neurosteroids were randomly distributed in the simulation cell. The assembled systems were equilibrated using the same protocol as described above. After this, production molecular dynamics runs (200 ns) were carried out, and the resulting trajectories were analyzed for aggregation.

Experimental procedures

Reagents

1-palmitoyl-2-oleoyl-phosphatidylcholine (POPC #850457P) and cholesterol (#700000P) were purchased from Avanti Polar Lipids, Inc. (Alabaster, Al). Bodipy 493/503(#D3922), NBD-C6-Ceramide (#N1154, NBD-Cer) and Dextran, Tetramethylrhodamine (#D1819) was purchased from ThermoFisher. Methyl-β-cyclodextrin.(MCβD)(#332615) was purchased from Sigma-Aldrich. Immortalized human astrocytes (IM-NHA) (#P10251) and the acquired growth media (astrocyte medium kit (#P60101)) were purchased from Innoprot.

Preparation of large unilamellar vesicles (LUVs)

LUVs were made by extrusion using a 400nm pore diameter filter and a micro extruder from Avanti Polar Lipids (Alabaster, AL), as described in.⁶⁹ LUVs were made with different lipid compositions: POPC: neurosteroids (80:20mol%) and POPC: cholesterol: neurosteroids (70:20:10mol%), and made with a final lipid concentration of either 50 μM or 200 μM corresponding to 100mol%).

Assessment of membrane permeability

Access of dithionite to the phospholipid analog NBD-PC was measured in LUVs exactly as described.³⁸ Briefly, the fluorescence of LUVs of the respective lipid composition with 1mol% NBD-PC was continuously monitored for an excitation wavelength of 470 nm and an emission wavelength of 540 nm. Sodium dithionite was added from a freshly prepared 1 M stock solution in 100 mM Tris (pH 10) giving a final dithionite concentration of 50 mM. After monitoring the reduction kinetics, Triton X-100 (final concentration of 0.5% (w/v) was added, resulting in disruption of the liposomes and full access of the quencher to the lipid analog. After normalization to the intital intensity, the decay curves were fit with a bi-exponential model, and from the rate constant of the second phase, the kinetics of access of dithionite to NBD-PC in the inner membrane leaflet was inferred.

Fluorescence spectroscopy

Emission spectra were obtained using an ISS Chronos BH spectrofluorometer (Urbana-Champaign, IL). The excitation wavelength was set to 328 nm, and recorded from 350–500 nm. For all measurements, a slit width of 0.5 mm (excitation and emission path) with no polarization was used.

Photo-physical properties of YX-11 and MQ-323

For each experiment, a fresh stock of neurosteroids with a final concentration of 2.5 mM in absolute ethanol was prepared. Emission spectra were obtained in various solvents with different polarities with a final concentration of 20 μM. Each spectrum represents the average from three independent measurements. The averaged spectra were smoothed using the Savitzky-Golay method in Python. The integrated intensity (area under the curve) was calculated for the raw spectra and displayed as the mean with its correlation standard error (sem). The titration experiments were conducted in water with a final concentration ranging from 0–50 μM. The emission maximum of the raw spectra for YX-11 was found at 416 nm and for MQ-323 at 378 nm. The maximum emission from three individual experiments was plotted as a function of neurosteroid concentration and fitted with a straight-line model with breakpoints using piece-wise regression in Python.

FRET measurements of MQ-323 and YX-11 binding to ELICα₁GABA_A receptor

Binding of MQ-323 and YX-11 to ELIC- 1GABAAR was measured as previously described.⁴⁵ Briefly, 100 μl of ELICα₁GABA_A receptor (0.3 μM) was transferred into a 50 μl window quartz cuvette, which was maintained at 4°C. NaI (final concentration of 100 mM) was added to the protein sample to reduce the emission of hydrophilic tryptophan residues. 1 μL aliquots of 100X stock solutions of MQ-323 or YX-11 (prepared in Buffer A + 0.015% DDM) were sequentially added to the protein sample and incubated for 30 min after each addition. The emission spectra (background spectrum) of MQ-323 or YX-11 in the absence of protein were obtained and subtracted from the protein-MQ-323 (or YX-11) emission spectra to obtain the background-corrected spectra. Finally, the contribution of ELICα₁GABA_A receptor tryptophan emission to the FRET spectra was calculated as previously described⁴⁵ and subtracted from the background-corrected spectrum to obtain the emission spectrum of the signal resulting from energy transfer (FRET spectrum) from tryptophan to MQ-323 or YX-11. The 370 nm intensity values of the FRET spectra were plotted against MQ-323 concentration to yield the MQ-323 binding curve. Non-specific MQ-323 binding was measured by determining MQ-323 binding (as described above) in the presence of 30 μM allopregnanolone and specific binding was obtained by subtracting non-specific binding from total binding. Data is presented as mean ± S.D. from three independent experiments.

Electrophysiology

The electrophysiological experiments were conducted on human α1β3γ2L GABA_A receptors. Receptor expression in Xenopus laevis oocytes and details of two-electrode voltage clamp have been described in detail previously.^4,70 The modulatory effects of MQ-323 and YX-11 were tested on receptors activated by 2 μM (<EC₁₀) GABA. In addition, YX-11 was tested on receptors activated by the combination of 1 mM GABA + 50 μM propofol. This drug combination fully activates the receptors.⁷⁰ The drug application protocol consisted of activating the receptors with GABA (or GABA + propofol), followed by co-application of steroid, and wash in GABA (or GABA + propofol). The effect of steroid was calculated from the ratio of current amplitudes in the presence and absence of steroid.

NMR spectroscopy

The desired molar ratio of respective molecules was first mixed in ethanol. For overnight lyophilization the solvent was evaporated and the samples were redissolved in cyclohexane. The obtained fluffy powder was were hydrated with 50 wt% H2O-buffer (10 mM Hepes, 100 mM NaCl, pH 7.4) and equilibrated by ten freeze-thaw cycles. The static 2H NMR spectra were acquired on a Bruker (Bruker Biospin, Rheinstetten, Germany) DRX300 NMR spectrometer using the quadrupole echo pulse sequence⁷¹ on a high-power probe with a 5-mm solenoid sample coil. The relaxation delay was 1 s, the delays between the 90° pulses (of ~ 3.2 μs) were 50 μs. For the depaked spectra⁷² smoothed order parameter profiles were calculated.⁷³ The measurements were carried out at a temperature of 25°C.

Cell culture and labeling

Immortalized human astrocytes (IM-NHA) were cultured in T25 Nunc^™ EasyFlask^™ (#156367, Thermo Scientific) and grown in astrocyte medium (AM) supplemented with 10% Fetal Bovine Serum (FBS) and 1% pencilin/streptomycin solution (P/S solution) and incubated at 37°C, 5% CO₂, and 95% humidity. Media was changed every second day and subcultured at 90% confluence using 1X trypsin-EDTA (#T47174, Sigma-Aldrich). For microscopy cells were plated on 35mm microscope dishes (#TKO-P351184–408, MatTek) and ready for imaging when they reached 70–80% confluence.

To prevent aggregation of the neurosteroid analogs they were loaded on methyl-β-cyclodextrin (MβCD) in PBS (final concentration of YX-11 and MQ-323 (20 μM) and MβCD (40 μM)). The MβCD-sterol complex was thoroughly vortexed and settled for minimum 30 minutes at RT before loading to cells in Medium 1 (M1) (containing 150 mM NaCl, 5 mM KCl, 1 mM CaCl₂, 1 mM MgCl₂, 5 mM glucose and 20 mM HEPES at pH 7.4) for 1 hr at 37°C. For labelling of lipid droplets bodipy (final concentration of 0.2 μg/mL) was added to cells for 5 minutes at 37°C. For Golgi and TGN labelling cells were stained with NBD-Cer (final concentration of 5 μM) for 10 minutes at 37°C. For Labelling late endosomes and lysosomes (LE/LYSs) cells were stanied with 0.25 mg/mL Rh-dextran overnight. Before imaging, cells were washed with M1 media.

Fluorescence microscopy

Widefield epifluorescence microscopy was carried out on a Leica DMIRBE microscope with a 63 × 1.4 NA oil immersion objective (Leica Lasertechnik GmbH) with a Lambda SC smart shutter (Sutter Instrument Company) as illumination control. Images were acquired with an Andor Ixon blue EMCCD camera operated at −75°C and driven by a Solis software. The microscope contains an 10× extra magnification lens in the emission light path, resulting in a final pixel size of 193 nm for the 63× objective. YX-11 and MQ-323 were imaged in the UV range of the spectra using a specially designed filter cube obtained from Chroma Technology (Corp., Brattleboro, VA, USA) with a 335-nm (20-nm bandpass) excitation filter, 365-nm dichromatic mirror, and 405-nm (40-nm bandpass) emission filter. For the neurosteriods a bleach stack with 100 frames were recorded with 400 msec acquisition time. Bodipy was imaged using a fluorescein filter cube with 480 nm (40 nm bandpass) excitations filter, 505 nm longpass dichromatic mirror, and 527 nm (30 nm bandpass) emission filter.

Image analysis for fluorescence microscopy data

Image correction and deconvolution

Autofluorescence correction is done for all UV images for the neurosteriods by subtracting the last image of each bleach stack from the first one in an automated manner using batch processing in Python. The resulting image only contain the bleaching fluorescent probe signal. For visualization purposes images obtained for the lipid droplets, Golgi and TGN, and Rd-dextran (green and red emission) were deconvolved in an automated manner using batch processing in Python. The Richard-Lucy algorithm was used with 20 iterations and a theoretical PSF for each specific channel was used for deconvolution and generated using the Diffraction PSF 3D plugin in ImageJ (https://imagej.net/plugins/diffraction-psf-3d) Analysis of photobleaching kinetics was carried out using the ImageJ plugin PixBleach, as described.⁷⁴

Segmentation of cells using Cellpose

Cells were segmented using the deep learning-based segmentation method Cellpose, as described in.⁷⁵ The images were segmented using the built-in Cytoplasm 2.0 model (‘cyto2’) in Cellpose with a diameter of 100 in an automated manner in Python. The mean fluorescence intensity of the neurosteriods per cell was found using the masked area. To quantify the fluorescence intensity of the neurosteriods in lipid droplets (LDs), the LDs were segmented using a custom training pipeline. A ratio of how much of the neurosteroids found in LDs per cell was displayed in boxplots. A ratio above 1 indicates that more of neurosteriods is found in LDs than outside.