All animal experiments were conducted in accordance with the European Directive 2010/63/EU of the European Parliament (Council of 22 September 2010 on the protection of animals used for scientific purposes) and the Russian “GOST 33216-2014” Guidelines for the maintenance and care of laboratory animals and were approved by the local Bioethics Commission LLC of the Institute of Mitoengineering of MSU. Wildtype D. rerio (246 animals, shortfin phenotype, 6–8 months of age, male to female ratio 50:50) were kept in a ZebTEC (Tecniplast S.p.A, Buguggiate, Italy) recirculating system, and housed under a 14/10-h light/dark cycle (8 AM “on” and 10 PM “off”). The system parameters were maintained automatically with water set at 28°C, pH 6.8–7.5, 550–700 mOsm/L and constant aeration. Feeding was carried out twice a day with special food for fish (special diet services, scientific fish food, SDS 300–400). The mice (BALB/c, 6–8 week-old males) were obtained from the nursery for laboratory animals “Pushchino,” Russia. They were housed under a 12 h dark/light cycle in a temperature-controlled environment, with food and water ad libitum.

Fluvoxamine (flv, 5 and 10 mg/kg, in saline) and diazepam (diaz, 1.25 and 5 mg/kg, in 0.3% propylene glycol, 0.06% ethanol) were purchased from Sigma-Aldrich Inc., and administered in a volume of 10 μl. Control groups were treated in parallel with vehicle alone. For drug injection, the fish were anesthetized briefly by placing them in 10°C water. The vehicle controls were tested in parallel. LCGA peptides (supplied by Lactocore, Inc., 98% purity) were administered at a dose of 1 and 10 mg/kg (i.p., prepared in saline on the day of the experiment). The peptides had the following amino acid composition: LCGA-17 (FQSE = PheGlnSerGlu); LCGA-26 (DKTE = AspLysThrGlu); LCGA-59 (WDQV = TrpAspGlnVal) and LCGA-83 peptide (FLPY = PheLeuProTyr).

Behavioral testing was initiated 10 min after drug injection, starting with the novel tank test (NTT), followed immediately thereafter by the light–dark box (LDB) and the social preference test (SPT). The NTT was adapted from Maximino et al. (2013) with the behavior video-recorded and the data processed using EthoVision XT14 (Noldus). The distance traveled, speed, number of visits to the bottom, middle and top thirds of the aquarium, respectively, and the times spent in each of these zones during the initial 5 min in the tank were recorded. A decreased time spent at the surface of the NTT apparatus reflects a reduction in exploratory behavior or increased hiding motivation (Sackerman et al., 2010). The behavior in the LDB (Maximino et al., 2011) and SPT (Parker et al., 2014) was video recorded and processed using RealTimer (Open Science LCC, Russia). The fish were added to the center zone of a three-zone aquarium and allowed to adapt for 1–2 min before removal of the septa separating the center from the flanking zones. The time spent and the number of visits to the light and dark flanking zones and the latency to enter the lit zone were recorded during a 5-min session. Stress of fish is associated with increased time spent in the dark compartment (scototaxis) (Maximino et al., 2011). For the SPT, the residence time and the number of visits to the three equally sized zones (near the shoal, in the center zone of the aquarium, and near the wall opposite to the shoal) were assessed during a 5-min session. Increased shoal preference (shoaling reflex) is thought to represent a protective response to a predator (Nguyen et al., 2014). All behavioral testing was done in the light phase using 500 lux (NTT, LDB) or 200 lux (SPT) illumination.

The mice were injected (i.p.) with LCGA-17 (in saline), diazepam (in 0.3% propylene glycol, 0.06% ethanol), bicuculline (in saline) or with vehicle alone, 30 min before testing and at the dosages indicated in the Figure legends. The open field test (OFT) was used to assess locomotion. The total distance traveled in a 5 min session was recorded using Ethovision XT14 (Prut and Belzung, 2003). The elevated plus maze (EPM) and the marble burying test (MBT) were used to assess anxiolytic-like drug activity. For the EPM (Pellow et al., 1985), the percentage of time spent in the open arms, the percentage of open arm entries, and the number of total arm entries during a 5-min session were recorded using Ethovision. “Anxiety Index” (AI), an integrative behavioral measure in the EPM was calculated as follows: 100 × {1 − [(time spent on open arms/total time on the maze)/2 + (number of entries to the open arms/total exploration on the maze)/2]} (Cohen et al., 2011). The MBT (Njung’e and Handley, 1991) was carried out in a 19 × 29 × 13 cm cage filled 5 cm deep with wood-chip bedding. Twenty marbles (5.2 g, 15 mm diameter, randomly colored) were evenly placed on the surface of the bedding in four rows of five pieces. At the end of a 30 min test, the animal was removed, and the marbles buried in bedding by at least two-thirds were recorded. The number of marbles buried in this test is reduced by anxiolytics and antidepressants such as BZDs and selective serotonin reuptake inhibitors (SSRIs), respectively (Takeuchi et al., 2002; Nicolas et al., 2006). The forced swim test (FST) was used for a preliminary assessment of antidepressant drug activity as described (Colelli et al., 2014). Briefly, on day one the mice were exposed to a 10 min pre-swim in a cylinder filled with 24 ± 1°C water. The next day they were re-exposed to the same apparatus for a 5 min test session with the behavior video recorded. The total time spent immobile was quantitated off-line visually. The interval between sequential behavioral tests was 2–3 days with the sequence OFT, EPM, MBT, and FST. All the testing and data analyses were performed by experimenters blind to treatment.

We used liquid chromatography-tandem mass spectrometry to separate Bos taurus milk hydrolysates (supplied by Lactocore, Inc.) into protein fractions. The major peptides of these fractions were identified and computationally converted into 288 unique tetrapeptides, using a sliding window with a step of one amino acid. We first performed blind docking of all 288 tetrapeptides to the α1β2γ2 GABA_AR, using the atomic receptor models 6 × 3S, 6 × 3T, 6 × 3U, 6 × 3V, 6 × 3W, 6 × 3X, 6 × 3Z, and 6 × 40 as templates (Kim et al., 2020) and QuickVina2.1 software (Alhossary et al., 2015) with modifications of Autodock Vina (Hassan et al., 2017). Prior to running the docking algorithm, all the receptor structures were stripped of co-crystallized modulators and antibodies. The box used for docking was set to include the entire receptor molecule. A transmembrane site of interest was identified visually (Supplementary Figure 1A). The same strategy was used for blind docking of allopregnanolone and pregnenolone sulfate (PREGS) to the GABA_AR and identification of a putative neurosteroid binding pocket (Supplementary Figures 1B,C). For subsequent peptide docking runs we used the proprietary Peptimize algorithm (Lactocore, Inc.), based on the Peptogrid algorithm (Zalevsky et al., 2019). The docking box for this run was designed as a bounding box for PREGS moieties and extended to account for the larger size of tetrapeptides compared to neurosteroids (Supplementary Figure 1C).

A α2δ-1-peptide interaction site was identified analogously. The 6JPA model of α2δ-1 was subjected to homology modeling with SwissModel to reconstruct the missing regions Thr831-Cys842 and Pro913-Met972 of the VGCC structure (Zhao et al., 2019). In particular, the second of these domains is large and the structure unknown. To account for all its possible structures we performed a molecular dynamic (MD) simulation of 1 μs of the whole α2δ-1 domain as described in Supplementary Data 1. We monitored the reconstructed loop’s root mean square deviation (RMSD) as a subjective measure of its stabilization. We observed that the system is mostly stabilized by 700 ns and picked this frame as a reference model for subsequent docking studies (Supplementary Figure 2). Gabapentin was blindly docked into the whole receptor, and the best pose was utilized to highlight the site for targeting. In search of additional support for this site, we extracted the frames from the trajectory every 100 ns and subjected them to blind docking. We obtained solid support for the identified site as most of the best gabapentin poses were spatially clustered within that site. Analogous to the GABA_A receptor study, the box for peptide docking to the α2δ-1 subunit encompassing the gabapentin site was defined (Supplementary Figure 3C). We note that the reconstructed Pro913-Met972 region is far removed from the optimal gabapentin binding site and therefore unlikely to influence the binding site selection process.

Next, the results of docking the peptides against the 6 × 3S GABA_AR model and against the 700 ns frame of the MD trajectory of VGCCs were rescored using Peptimize algorithm. To identify the most accurate interaction profiles of the peptides with the two receptors, we performed a redocking round with Autodock Vina (Trott and Olson, 2009) with the exhaustiveness parameter set to 512 (Rentzsch and Renard, 2015). The best peptide poses for each site were then subjected to 100 ns atomic MD simulations to equilibrate the peptide in its binding site and further refine the interaction profile while also taking water molecules into account (see MD protocol in Supplementary Data 1).

[³H]-LCGA-17 (70 Ci/mmol) was synthesized by high-temperature solid-state catalytic isotope exchange (HSCIE) by A. Yu. Zolotarev at the Department of Physiologically Active Substances Chemistry, Institute of Molecular Genetics of RAS, Moscow, Russia (Zolotarev et al., 2003). Briefly, rat brain tissue was homogenized in 25 volumes of cold 50 mM Tris–HCl, pH 7.4 using a Teflon-glass homogenizer. The homogenate was centrifuged (20 min, 40,000 × g) and the precipitated membranes resuspended in the same buffer and centrifuged twice more. The pellet was resuspended in 15 mL of the same buffer and used for the radioligand analysis. Binding reactions (500 μL) contained 250 μL of the membrane suspension, 50 μL of [³H]-LCGA-17 (70 Ci/mmol, final concentration 16 nM) and 200 μL of 50 mM Tris–HCl buffer (pH = 7.0). The reaction was incubated at room temperature for 25 min, filtered through GF/B filters (Whatman) preconditioned with 0.3% polyethyleneimine, and the filters washed and air-dried, and the bound [³H] activity counted in a Tri-Carb 2900TR counter (Perkin Elmer). The IC₅₀ of putative competitor compounds was determined by supplementing binding reactions with corresponding test compounds at 10^–10 to 10^–4 M. The list of compounds tested is available in Supplementary Table 1.

Statistical analyses were performed using Prism 9.0 (GraphPad). Normality of distributions was assessed using Shapiro–Wilk’s W. Normally distributed data were analyzed by one-way ANOVA and Dunnett’s post hoc test. Non-normally distributed data were analyzed by Kruskal–Wallis followed by Dunn’s post hoc test. The data are presented as box and whisker plots. For pairwise comparisons (in vivo screening in D. rerio), we used Mann–Whitney (M–W) U-test. False discovery rate (FDR) correction using Benjamini and Hochberg (1995) was applied, with a significance threshold set of q = 0.05.

The objective of this study was to explore the suitability of milk-hydrolysate-derived peptides as a source of novel central nervous system (CNS)-active therapeutics, with a special focus on neuronal protein targets that are known to mediate the effects of clinically successful anxiolytics. To begin we performed in silico blind docking experiments with 288 milk hydrolysate-derived tetrapeptides to map a potentially druggable space on the entire α1β2γ2 GABA_AR structure (Supplementary Figure 1A). We then discarded all positional clusters situated on the exterior face of the transmembrane domain to avoid potential complications for peptides to enter and equilibrate within the membrane. We also excluded the BZD, ketamine, flurazepam, bromine, and barium binding sites, since therapeutic ligands for these sites have already been described (Puthenkalam et al., 2016). We then focused our attention on an internal transmembrane binding site, which we hypothesized may represent a binding site for PREGS and allopregnanolone. We found that allopregnanolone preferentially binds the external face of a transmembrane region in several distinct sites including ones recently identified for α1β3γ2 (Supplementary Figure 1B; Chen et al., 2019). Almost all of the best PREGS docking poses mapped to the transmembrane domain of the channel, which corresponded perfectly with the same site selected by modeling of peptide docking (Supplementary Figure 1C). Thus, a peptide modulator under development was provisioned to act as a functional analog of PREGS.

In addition to GABA_ARs, we scored all the hydrolysate-derived peptides for binding to the α2δ subunit of VGCC, exploring the possibility of candidates with high docking scores for both targets. In particular, we discovered a peptide docking site on α2δ that we had previously identified as a putative binding site for the α2δ ligand, gabapentin, using a blind docking experiment (Supplementary Figure 3). The two Peptimize scores from docking studies with the GABA_AR and α2δ were averaged to produce a unified binding score for each peptide. This score was then used to rank the hydrolysate-derived peptides by predicted affinity to both targets. The top three peptides identified in this way were HKEM, FFVA, and FQSE. Of these, the HKEM peptide was discarded due to a high probability of methionine oxidation that predicts peptide toxicity (Schöneich, 2005). The FFVA peptide was discarded due to low predicted solubility. The FQSE peptide (from here onwards named LCGA-17) and three lower-ranking control peptides (DKTE, WDQV, and FLPY, in the following referred to as LCGA-26, -59, and -83, respectively) were selected for further validation.

Possible atomic interactions of LCGA-17 with its α2δ binding domain were investigated in greater detail using a redocking round with enhanced accuracy followed by a 100 ns molecular dynamics study, to account for potential solvation effects and assess the overall short-term stability of particular binding poses (Figures 1A–E and Supplementary Videos 1–3). The width of the transmembrane channel pore varies among different GABA_AR crystal structures (Kim et al., 2020), leading us to compare the binding poses of LCGA-17 in GABA_AR structures with both a narrow (6 × 3S) and wider channel pore (6 × 3W). The initial binding pose of LCGA-17 in the structure 6 × 3W resembled the binding pose of PREGS in α2δ. However, when simulating receptor-ligand complex movement by MD, we found that the conformation of LCGA-17 in this structure was unconstrained and highly dynamic. The peptide had rapidly adopted a distinctive “enclosed” structure characterized almost exclusively by intra- rather than intermolecular interactions (Figures 1A,B and Supplementary Video 1). Indeed, attractive interactions between LCGA-17 and the receptor were largely absent, suggesting shape complementarity as the only evidence in support of interaction in the “wider-pore” case. Autodock Vina in this case predicted a binding energy of −7.0 kcal/mol. The same enclosed peptide structure was observed as an initial binding pose for the narrower channel conformation of the 6 × 3S receptor structure, although this pose was located more toward the extracellular portion of the channel. In this pose, the peptide conformation was remarkably stable, involving numerous interactions with the receptor (Figures 1C,D and Supplementary Video 2), which was also reflected in a greater binding energy (−7.8 kcal/mol). Comparing the interaction profiles, we propose that this second binding mode represents a possible functional form of LCGA-17 when bound to GABA_ARs. The binding mode of LCGA-17 in the α2δ site was dynamically stable (Supplementary Video 3) and supported by an extensive network of stable, attractive interactions within the α2δ binding pocket depicted in Figure 1F and is reflected in a high predicted binding energy of −8.5 kcal/mol. Notably and in contrast to the conformation of LCGA-17 in the GABA_AR PREGS binding pocket, LCGA-17 did not fold into an enclosed conformation when bound to α2δ (Figures 1E,F). Collectively, these studies identify LCGA-17 as a promising ligand for α2δ, with additional potential for binding to the GABA_AR PREGS binding pocket.

To explore the potential of LCGA-17 and three other LCGA peptides as behaviorally active anxiolytics we first employed high throughput behavioral tests in D. rerio, using diazepam and fluvoxamine as active control compounds with established anxiolytic efficacy in patients. The three tests employed have in common that they examined the fishes’ resolve to overcome different naturally aversive conditions such as novelty (NTT), bright light (LDB), or moving away from the shoal (SPT). In previous characterizations of diazepam and fluvoxamine using these tests, both compounds affected mainly the time spent in the upper half of the tank (NTT), the time in the light (LDB), and the time away from the “shoal” (SPT), respectively (data not shown), which led us to focus on these three parameters as the most informative. Each of the six compounds was tested at two dosages with the summary statistics of the most informative dosages depicted in Figure 2. For a summary of all the dosages tested see Supplementary Table 2.

In the NTT, LCGA-17 increased the time spent in the upper half of the tank, thereby mimicking the anxiolytic-like effect of fluvoxamine (Figure 2A) (for statistics see Figure legend). None of the other LCGA peptides had a measurable effect in this test. However, diazepam (5 mg/kg) reduced the time spent in the upper half of the tank. This drug effect did not represent sedation as the total distance traveled remained unaffected (Figure 2B). In the LDB, LCGA-17 increased the time spent in the lit compartment, thereby reproducing the anxiolytic-like effect of diazepam (Figure 2C). Neither fluvoxamine nor the other LCGA peptides affected the behavior in this test. In the SPT, both LCGA-17 and LCGA-26 increased the time the fish spent out of the shoaling zone, thereby reproducing the anxiolytic-like effect of diazepam (Figure 2D). However, fluvoxamine, LCGA-59, and LCGA-83 failed to affect behavior in this test. Collectively, these experiments predict that LCGA-17 has anxiolytic activity, with an efficacy comparable to diazepam and fluvoxamine.

Molecular modeling of LCGA-17 binding to GABA_AR and α2δ suggested the presence of distinctive binding sites. In support of this notion, radioligand binding of [³H]-LCGA-17 to rat brain membranes and competition with unlabeled LCGA-17 revealed an IC₅₀ of ∼2 μM. Gabapentin showed similar affinity (IC₅₀ ∼ 11 μM) for the same site, while diazepam and PREGS were ineffective as competitors (IC₅₀ > 100 μM) (Figure 3). Moreover, we detected no cross-reactivity with other GABA_AR binding sites (GABA, muscimol, bicuculline, gabazine, CGS-9895), nor with ligand binding sites of other major neurotransmitter receptors (GABA_B: baclofen; dopamine receptors: haloperidol, sulpiride, spiperone, 7-OH-DPAT; serotonin receptors: ketanserin; acetylcholine receptors: nicotine; and glutamate receptors: glutamate, glycine, Ro-256981, LY-354740, MK-801, spermine, arkain). Notably, the LCGA-17 binding site was found to be distinct from those for other GABA_AR ligands including isoguvacine, salicylidene salicylhydrazide, bretazenil, SL651498, MK0343, THDOC, TB21007, gaboxadol, FGIN-1-27, and allopregnanolone (Supplementary Table 1).

To explore the potential of LCGA-17 as a therapeutic in mammals we conducted dose-finding studies in mice, using a series of behavioral tests with predictive validity for anxiolytic and antidepressant drug activity in patients. In the EPM, LCGA-17 (20 mg/kg) and diazepam (0.75 mg/kg) similarly increased the percent time spent on open arms. Moreover, LCGA (unlike diazepam) had no effects on total arm entries, indicating anxiolytic-like activity without locomotor effects in this test (Figures 4A,D). An anxiolytic effect of LCGA-17 was also evident based on the increased percent time spent on open arms, which was significant at both the 1 and 20 mg/kg dosages (Figure 4B). This effect was evident across the entire dose range of the peptide (1–20 mg/kg) when the data were plotted as AI, consisting of the mean of these EPM parameters (Figure 4C). Moreover, LCGA had no effect on total arm entries, again indicating absence of locomotor effects (Figure 4D). In the OFT, LCGA-17 (5, 20 mg/ml) and diazepam modestly increased the distance traveled (Figure 4E), as previously reported for diazepam and expected as part of an anxiolytic response (Crawley, 1981). More importantly, a more overt anxiolytic response of LCGA-17 (but not diazepam) was evident based on the increased time spent in the center with all doses combined (Figure 4F and Supplementary Figure 4F) and especially significant at the lowest dose (1 mg/kg, Figure 4F). In the MBT, LCGA-17 reduced the number of marbles buried at both 5 and 20 mg/kg, thereby mimicking the anxiolytic-like effect of diazepam (Figure 4G). Collectively, these data strongly suggest that LCGA-17 has anxiolytic properties, with a rapid onset of action similar to or more potent than diazepam, yet without any sign of sedative effects even at the highest doses.

To begin to explore the potential of LCGA-17 as an antidepressant, we further subjected the mice to the FST. Although the biological underpinnings of this test are controversial, the time spent immobile in this test is known to be reduced across multiple classes of antidepressants including SSRIs, tricyclics (Detke et al., 1995), and ketamine (Ren et al., 2016), and insensitive to anxiolytic concentrations of diazepam (Detke et al., 1995). Interestingly, LCGA-17 (5 mg/kg) significantly decreased the behavioral immobility of mice (Figure 4H) and also evident when all the doses (1–20 mg/kg) were combined (Supplementary Figure 4H). Moreover, the behavioral effects of different dosages of LCGA-17 were indistinguishable from each other, for all test paradigms (for statistics see legend of Supplementary Figure 4). As predicted, diazepam failed to affect the behavior in this test. While these findings are encouraging, much more extensive experimentation will be necessary to assess the antidepressant and stress-protective potential of LCGA-17, which is beyond the scope of this initial study.

The rapid anxiolytic actions of gabapentinoids are thought to involve increased GABA_AR activity, reminiscent of diazepam but mediated by increased cell surface expression instead of direct activation of GABA_ARs (Yu et al., 2019). Therefore, to begin to test whether the behavioral effects of LCGA-17 in the EPM involved enhanced GABAergic transmission, we co-treated the mice with LCGA-17 and the GABA_AR antagonist bicuculline. Interestingly, the LCGA-17-induced increase in the percent time on open arms was completely abolished by bicuculline, without effects of bicuculline alone (Figure 5A). Similarly, co-treatment of mice with LCGA-17 and bicuculline reduced the % Open Arm Entries compared to the effect of LCGA-17 alone (Figure 5B) (for ANOVA statistics see Figure legend). Lastly, bicuculline blocked the effect of LCGA-17 on the AI, again without effects of bicuculline alone (Figure 5C) and without effects on locomotion (Figure 5D). In summary, LCGA-17 has been identified as a novel rapid-acting anxiolytic and potential antidepressant that acts via α2δ and possibly GABA_ARs to enhance ionic GABAergic neurotransmission.