Photoaffinity labeling identifies an intersubunit steroid-binding site in heteromeric GABA type A (GABA A ) receptors

Allopregnanolone (3 a 5 a -P), pregnanolone, and their synthetic derivatives are potent positive allosteric modulators (PAMs) of GABA A receptors (GABA A Rs) with in vivo anesthetic, anxiolytic, and anti-convulsant effects. Mutational analysis, photoaffinity labeling, and structural studies have provided evidence for intersubunit and intrasubunit steroid-binding sites in the GABA A R transmembrane domain, but revealed only little definition of their binding properties. Here, we identified steroid-binding sites in purified human a 1 b 3 and a 1 b 3 g 2 GABA A Rs by photoaffinity labeling with [ 3 H]21-[4-(3-(trifluoromethyl)-3H-diazirine-3-yl) benzoxy]allopregnanolone ([ 3 H]21- p a potent GABA A R PAM. Protein microsequencing established 3 a 5 a -P inhibitable photolabeling of amino acids near the cytoplasmic end of the b subunit M4 ( b 3Pro-415, b 3Leu-417, and b 3Thr-418) and M3 ( to n H =1, which were favored by F -test comparison over fits with variable n H , with the exception of betaxalone ( a 1 b 3, n H = 0.5 6 0.1). Based upon an F -test comparison of fits of the data for a 1 b 3 and a 1 b 3 g 2 GABA A Rs to the same (null hypothesis) or separate IC 50 values, a common fit was favored for 3 a 5 a -P ( p = 0.7, F (DFn,DFd) = 0.22 (1,112)) and 3 a 5 b -THDOC ( p = 0.6, F (DFn,DFd) = 0.24(1,63)). Separate fits were favored for 21- p TFDBzox-AP ( p , 0.0001, F (DFn,DFd) = 35.9(1,94)), 3 a 5 b -P ( p = 0.002, F (DFn,DFd) = 10.4 (1,62)), andalphaxalone ( p = 0.01, F (DFn,DFd) = 6.7(1,62)).

Endogenous neurosteroids, including allopregnanolone (3a5a-P) and pregnanolone (3a5b-P), can produce anxiolytic, sedative, and anti-convulsive effects (1,2), and their synthetic analogs are in development as general anesthetics and for treatment of epilepsy, anxiety, depression, and other mood disorders (3,4). These neuroactive steroids act at submicromolar concentrations as potent positive allosteric modulators (PAMs) of g-aminobutyric acid type A receptors (GABA A R), and at higher concentrations as direct activators in the absence of GABA (5)(6)(7)(8). GABA A R potentiation by steroids demonstrates structural specificity in that the orientation of a hydroxyl group at the C-3 position ( Fig. 1) determines activity. Steroids with a 3a-OH, including 3a5a-P and the anesthetic alphaxalone, act as PAMs, whereas their 3b-OH epimers (3b5a-P and betaxalone) at higher concentrations inhibit GABA responses (9)(10)(11)(12). This structural specificity provided early evidence that steroids might interact with specific binding sites in GABA A Rs, identification and characterization of which would prove important for the development of novel steroid-based therapeutic agents.
Functional, structural, and photolabeling studies provide evidence for the existence of multiple steroid-binding sites in abg GABA A Rs. Steroids do not bind to the GABA and benzodiazepine-binding sites at subunit interfaces in the extracellular domain or to the homologous binding sites for intravenous general anesthetics such as propofol, etomidate, and barbiturates that are located at subunit interfaces in the extracellular third of the transmembrane domain (TMD) (Fig. 1) (13,14). Binding assays using channel blockers as well as electrophysiological assays identify multiple effects of steroids potentially mediated by distinct sites (15,16). Intersubunit and intrasubunit steroid-binding sites near the extracellular and cytoplasmic surfaces of the TMD are predicted based upon the recently determined a1b3g2 GABA A R structures (17,18) and the locations of amino acids identified by mutational analysis as determinants for GABA A R enhancement or direct activation. A site near the cytoplasmic end of the b 1 -a 2 subunit TMD interface was predicted based upon the identification of a1Gln-242 (human a1 numbering) as a position critical for enhancement by steroids (19,20). Consistent with this location, alphaxalone protected against the modification of cysteines substituted in the b3 M3 helix at positions contributing to this interface (21), and 3a5b-P, tetrahydrodeoxycorticosterone (3a5a-THDOC), and alphaxalone bind to a homologous pocket in crystallographic structures of homopentameric, chimeric receptors with GABA A R a subunit TMDs (22)(23)(24). In a1b3 GABA A Rs, there is 3a5a-P inhibitable steroid photolabeling of a residue at the cytoplasmic end of bM3 in proximity to this pocket, with additional residues identified near the extracellular end of the TMD within the a1 and b3 subunits (25).
Photoaffinity labeling with radiolabeled, photoreactive intravenous general anesthetics has allowed the identification of photolabeled amino acids for site identification and the determination of the pharmacological specificity of these sites by inhibition of photolabeling with nonradioactive anesthetics. Photolabeling with [ 3 H]azietomidate and a mephobarbital analog, [ 3 H]R-mTFD-MPAB, identified homologous binding sites in the a1b3g2 GABA A R TMD at the b 1 -a 2 and a 1 /g 1 -b 2 subunit interfaces, respectively (13,26). Etomidate and azietomidate bind with 100-fold selectivity to the b 1 sites, R-mTFD-MPAB with 50fold selectivity to the b 2 sites, and other barbiturates and propofol derivatives bind with variable selectivity to the two classes of sites.
Here we characterize a GABA A R steroid-binding site by use of 21-pTFDBzox-AP (21-[4-(3-(trifluoromethyl)-3Hdiazirin-3-yl)benzoxy]allopregnanolone), a photoreactive steroid that acts as a potent a1b3 and a1b3g2 GABA A R PAM (27). Previously, we reported that [ 3 H]21-pTFDBzox-AP primarily photoincorporated into the b3 subunit with ;80% of the subunit photolabeling inhibitable by 3a5a-P or by alphaxalone, but not by pregnenolone sulfate (PS), an inhibitory neurosteroid, or by etomidate or R-mTFD-MPAB (27). We now identify the amino acids photolabeled by [ 3 H]21-pTFDBzox-AP, which are located at the cytoplasmic ends of the bM3 and bM4 helices and form the base of a pocket at b 1 -a 2 intersubunit interface that extends up to the level of a1Gln-242 in aM1. By use of competition photolabeling with a panel of steroid GABA A R PAMs and inhibitors, we provide a first definition of the structural determinants important for high affinity binding to this site.

Results
Positive and negative steroid GABA A R allosteric modulators enhance [ 3

H]muscimol binding
In equilibrium binding assays with the agonist [ 3 H]muscimol, GABA A R PAMs, including steroids and other general anes-thetics, enhance binding by increasing the fraction of GABA A Rs in a desensitized state that binds [ 3 H]muscimol with high affinity (28). 21-pTFDBzox-AP was shown previously to enhance [ 3 H] muscimol binding to expressed a1b3 and a1b3g2 GABA A Rs in membranes, and after purification in detergent/lipid micelles, with concentrations producing half-maximal enhancement (EC 50 , 0.2-0.5 mM) similar to those for 3a5a-P, 3a5b-P, and alphaxalone (27). We extended these studies by characterizing [ 3 H]muscimol binding to a1b3 GABA A Rs in the presence of steroids that act as GABA A R negative allosteric modulators, inhibiting GABA responses noncompetitively: the 3b-epimers of 3a5a-P, 3a5b-P, and alphaxalone, and two 3b-sulfated steroids (PS and dehydroepiandrosterone sulfate (DHEAS)) (10,12,29,30) (Fig. 2 and Table 1). The 3b-OH epimers of pregnanolone (3b5b-P) and alphaxalone (betaxalone) enhanced [ 3 H]muscimol binding with EC 50 values of 25 and 45 mM, respectively, whereas 3b5a-P at concentrations up to 100 mM did not. PS at concentrations up to 500 mM had no effect on [ 3 H]muscimol binding, whereas DHEAS reduced specific binding maximally by 50% (IC 50 = 10 mM). In addition, we found that (3a5a)-17-phenylandrost-16-en-3-ol (17-PA), which antagonizes steroid enhancement of GABA responses but not GABA responses (31) In initial photolabeling studies, we compared [ 3 H]21-pTFDBzox-AP photolabeling of a1b3 and a1b3g2 GABA A Rs. After photolabeling, GABA A R subunits were resolved by SDS-PAGE, and 3 H incorporation into the subunits was characterized by fluorography (Fig. 3A). As reported previously (27), for both receptor subtypes photolabeling was most prominent in Depicted are the four transmembrane helices in each subunit (M1-M4), the homologous binding sites for etomidate and R-mTFD-MPAB, an analog of mephobarbital, in the extracellular third of the b 1 2a 2 and a 1 /g 1 2b 2 subunit TMD interface(s), respectively, and a binding site for neuroactive steroids in the intracellular third of the b 1 2a 2 interface. The binding sites for GABA are located in the extracellular domain in the b 1 2a 2 subunit interfaces, and benzodiazepines bind at the homologous site in the a 1 -ginterface. B, steroid ring structure, with numbering of the carbons, and structures of representative neuroactive steroids that act as positive or negative GABA A R allosteric modulators.
the gel bands of 59 and 61 kDa that contain differentially glycosylated b3 subunits (13,32), and at a lower level in the 56-kDa gel band containing the a1 and g2 subunits. Photolabeling of the b3 subunit was inhibited by 30 mM 3a5a-P, but not by PS, etomidate, or R-mTFD-MPAB. To quantify the concentration dependence of inhibition of photolabeling by nonradioactive drugs, receptor aliquots were photolabeled with [ 3 H]21-pTFDBzox-AP in the presence of a range of drug concentrations, with receptor subunits excised from the stained gel after SDS-PAGE and 3 H incorporation into the b subunit determined by liquid scintillation counting. In a representative experiment (Fig. 3B), nonradioactive 21-pTFDBzox-AP maximally inhibited [ 3 H]21-pTFDBzox-AP photolabeling of a1b3 and a1b3g2 GABA A Rs to the same extent as 30 mM 3a5a-P, with IC 50 values of 0.7 and 0.9 mM, respectively. As described under "Experimental procedures," IC 50 values for drugs were determined by combining results from at least four independent experiments using two or more GABA A R purifications, with data from individual experiments combined after normalization to the total specific (i.e. 3a5a-P inhibitable) binding in the absence of competitor. The pooled data for inhibition by 21-pTFDBzox-AP are shown in Fig. 4.
In a1b3 and a1b3g2 GABA A Rs, a 3a-OH substituent is a major determinant of pregnanolone affinity for this site As a test of the pharmacological specificity of the sites identified in a1b3 and a1b3g2 GABA A Rs by [ 3 H]21-pTFDBzox-AP photolabeling, we compared inhibition by 3a5a-P with its antagonist 3b-OH isomer (3b5a-P) and with analogs modified at the 3-position by acetylation (3a-acetyl-5a-P), removal of the -OH (3-deoxy-5a-P), or oxidation into a ketone (3-oxo-5a-P) ( Fig. 4 and Table 1). 3a5a-P inhibited photolabeling of both receptor subtypes with an IC 50 of 0.4 mM, whereas 3b5a-P at 300 mM inhibited photolabeling of a1b3 and a1b3g2 GABA A Rs by ,10% and ;40%, respectively. At the highest concentration tested (100 mM), 3-deoxy-5a-P, which is a GABA A R PAM (20), as well as 3a-acetyl-5a-P and 3-oxo-5a-P each inhibited photolabeling by ,10%. We also determined that alphaxalone inhibited a1b3 and a1b3g2 GABA A R photolabeling with IC 50 values of 5 and 2 mM, respectively, whereas for betaxalone, 50% inhibition was seen at ;200 mM ( Fig. 4). Consistent with the importance of a 3a-OH for binding to this site, the sulfated 3b-OH antagonists PS and DHEAS at 100 mM each inhibited photolabeling by ,10% (Table 1 and Ref. 27). In contrast to the importance of the 3a-OH, the configuration at the 5-position was not important. 3a5b-P inhibited photolabeling with an IC 50 of 0.7 mM, similar to that for 3a5a-P, whereas 3a5a-THDOC and 3a5b-THDOC inhibited GABA A R photolabeling with IC 50 values of 2-3 mM (Table 1).
3a5a-P inhibits [ 3 H]21-pTFDBzox-AP photolabeling of amino acids located at the cytoplasmic ends of the bM3 and bM4 helices that contribute to a pocket at the b 1 -asubunit interface Based upon the similar pharmacological properties of the steroid-binding sites in a1b3 and a1b3g2 GABA A Rs defined by [ 3 H]21-pTFDBzox-AP photolabeling, we identified the photolabeled amino acids in a1b3 GABA A Rs, which can be expressed and purified at higher levels than a1b3g2 GABA A Rs. b3 subunits were isolated from a1b3 GABA A Rs photolabeled on a preparative scale with [ 3 H]21-pTFDBzox-AP (0.7 mM) in the presence of 300 mM GABA and in the absence or presence of 30 mM 3a5a-P. In five preparative photolabelings, the specific b subunit photolabeling (i.e. 3a5a-P inhibitable) was 320 6 70 3 H cpm/pmol, which indicated photolabeling of 1.2 6 0.2% of b subunits based upon the radiochemical specific activity of [ 3 H]21-pTFDBzox-AP (21.8 Ci/mmol) and the amount of GABA A R photolabeled. This efficiency of photolabeling was similar to that seen for GABA A R photolabeling by a photoreactive etomidate analog (32), but ;15% the efficiency seen for [ 3 H]R-mTFD-MPAB (33).
The photolabeled amino acids were identified by protein microsequencing of fragments beginning near the N termini of the b3M4, b3M3, and b3M1 helices that can be produced by digestion with endoproteinase Lys-C (Endo Lys-C) and resolved by reversed-phase HPLC (rpHPLC) (13,32,34). When Figure 2. Modulation of GABA A R agonist binding by steroid antagonists. The 3b-OH steroid antagonists 3b5b-P and betaxalone and the 3a-OH antagonist 17-PA enhance equilibrium binding of subsaturating concentrations of [ 3 H]muscimol (2 nM) with efficacies similar to that seen for the PAM 3a5a-P but with lower potencies, whereas no enhancement is seen for 3b5a-P. The 3b-sulfate antagonist PS did not enhance binding, whereas DHEAS reduced specific binding maximally by 50%. The data from n independent experiments were combined and fit to determine values of EC 50 (in mM), Hill coefficients (n H ), and maximal enhancements (B max , as % control), that were: 3b5b-P (26 6 7, 1. aliquots of the b subunit Endo Lys-C digests were sequenced, peaks of 3 H release were seen in cycles 3/4 and 6/7 that were inhibitable by 3a5a-P (Fig. 5A). When the digests were fractionated by rpHPLC (Fig. 5B), the peak of 3 H was recovered in a fraction that contained an unlabeled fragment beginning at b3Ala-280 near the N terminus of bM3, with the unlabeled fragment beginning at b3Ile-412 before the N terminus of bM4 eluting one fraction earlier. Additional 3 H-containing adducts eluted in the more hydrophobic fractions that contain the unlabeled fragment beginning at b3Arg-216 at the N terminus of bM1 that extends through bM2.
Protein sequencing protocols were designed to allow identification of photolabeled amino acids even if the incorporation of the hydrophobic steroid caused the 3 H-labeled fragment to elute in more hydrophobic HPLC gradient fractions than the unlabeled fragment directly identifiable by PTH-derivative analysis. When 50% of the fraction containing the peak of 3 H was sequenced, there were peaks of 3 H release in cycles 3-4 and 6-7 of Edman degradation that were reduced by 90% by 3a5a-P (Fig. 5C), as seen when the total digest was sequenced. There were no additional peaks of 3 H release above background in 30 cycles of Edman degradation (not shown). To determine whether the peaks of 3 H release originated from labeling in bM3 or bM4, we took advantage of the presence of b3Pro-415 in cycle 4 of Edman degradation of the bM4 fragment and the lack of a proline at that cycle in the bM3 or bM1 fragment. For the remaining 50% of the fraction, sequencing was interrupted at cycle 4 for treatment with o-pthalaldehyde (OPA) to prevent further sequencing of fragments not containing a proline at that cycle (35,36). After treatment with OPA in cycle 4, the 3 H releases in cycles 4, 6, and 7 were preserved, whereas sequencing of the M3 fragment was reduced by .95% (Fig.  5C). Thus, these 3 H releases did not originate from the bM3 fragment. Rather, the results were consistent with 3a5a-P inhibitable photolabeling of b3Pro-415 (cycle 4), b3Leu-417, and b3Thr-418 in the fragment beginning at b3Ile-412 before the N terminus of bM4. The 3 H release in cycle 3, although not tested by the use of OPA in cycle 4, indicated likely labeling of b3Ile-414. Based upon sequencing nine samples from five independent photolabeling experiments, b3Ile-414 and b3Leu-417 were photolabeled at 55 6 26 and 155 6 55 cpm/pmol, respectively, ;90% inhibitable by 3a5a-P ( Table 2). Because of uncertainties in calculating photolabeling efficiency for the second of two successive photolabeled amino acids, similar calculations were not made for b3Pro-415 and b3Thr-418.
Photolabeling of b3Arg-309 at the C terminus of bM3 was identified by sequencing the broad peak of 3 H that co-eluted with the unlabeled bM1 fragment (Fig. 6). A peak of 3a5a-P inhibitable 3 H release was seen in cycle 30, in addition to the peaks of 3 H release in cycles 3/4 and 6/7 attributable to labeling within the bM4 fragment and a peak in cycle 19 not reproduced in other experiments (Fig. 6A). The 3 H release in cycle 30 did not result from labeling in bM1, because for a sample sequenced with OPA treatment at cycle 13, the cycle containing b3Pro-228 in bM1, sequencing of the bM1 fragment persisted after treatment but no release of 3 H was seen in cycle 30 (not shown). This suggested that the labeled bM3 fragment, similar to the labeled bM4 fragment, eluted in more hydrophobic rpHPLC fractions than the unlabeled fragment, with the 3 H release in cycle 30 resulting from photolabeling of b3Arg-309. To test this, we generated a fragment beginning at b3Gly-287 in bM3 by use of cyanogen bromide to cleave at the C terminus of b3Met-287 (as well as other methionines in the sample on the sequencing filter). When this fragment was sequenced, there was a peak of 3a5a-P inhibitable 3 H release in cycle 23 consistent with photolabeling of b3Arg-309 (Fig. 6B). Based upon results from 4 independent photolabeling experiments, b3Arg-309 was photolabeled at ;25% the efficiency as compared with b3Leu-417 (Table 2).
Because b subunit photolabeling dominated over that in a and the gel band containing a subunit also contains b subunit at a low level (13), it was difficult to use our protocols to determine whether a subunit residues were photolabeled at low efficiency. Nonetheless, we searched in particular for 3a5a-P inhibitable photolabeling in aM4 at aAsn-408, which is an intrasubunit residue near the extracellular end of TMD that is a sensitivity determinant for steroid enhancement (19) and that was photolabeled by an allopreganolone derivative with a photoreactive group at C-21 (25). The latter also photolabeled b3Gly-308 or b3Arg-309 in the b 1 -a 2 steroid site. In parallel with the b subunit studies, we fractionated a subunit Endo Lys-  were photolabeled in the presence of GABA and varying concentrations of pregnane steroids containing a 3a-OH (3a5a-P, 3a5b-THDOC, alphaxalone, 21-pTFDBzox-AP), a 3b-OH (3b5a-P, betaxalone), or lacking a free 3-OH (3a-acetyl-5a-P, 3-deoxy-5a-P, 3-oxo-5a-P). After SDS-PAGE, covalent incorporation of 3 H in the b subunit was determined by liquid scintillation counting. For each independent experiment, nonspecific photolabeling (B ns ) was determined in the presence of 30 mM 3a5a-P, and specific binding was normalized to the 3 H cpm incorporated specifically in the control condition (B 0 2 B ns ). The plotted data are the averages (6 S.D.) from the independent experiments. As described under "Experimental procedures," the pooled data from the independent experiments were fit to Equation 2. Drug structures, parameters for the fits, and the number of independent experiments are tabulated in Table 1. The curves are plotted for fits to n H =1, which were favored by F-test comparison over fits with variable n H , with the exception of betaxalone (a1b3, n H = 0.5 6 0.1). Based upon an F-test comparison of fits of the data for a1b3 and a1b3g2 GABA A Rs to the same (null hypothesis) or separate IC 50 values, a common fit was favored for 3a5a-P (p = 0.7, F(DFn,DFd) = 0.22 (1,112)) and 3a5b-THDOC (p = 0.6, F(DFn,DFd) = 0.24 (1,63)). Separate fits were favored for 21-pTFDBzox-AP (p , 0.0001, F(DFn,DFd) = 35.9(1,94)), 3a5b-P (p = 0.002, F(DFn,DFd) = 10.4 (1,62)), and alphaxalone (p = 0.01, F(DFn,DFd) = 6.7(1,62)).
C digests by rpHPLC and found a 3 H distribution similar to that for b subunit digests shown in Fig. 5B. We sequenced fractions 24-27 that would contain the unlabeled and labeled fragments beginning at a1Ileu-392, with OPA treatment in cycle 10 of Edman degradation (a1Pro-401) to associate 3 H release beyond cycle 11 (a1Leu-402) with aM4. When the a1Ile-392 fragment (I 0 = 6 pmol) was sequenced for 25 cycles, no peaks of 3 H release were detected above background after cycle 11. Any photolabeling of a1Asn-408, or residues nearby in the primary structure, if it occurred, would be at less than 10% the efficiency of photolabeling of b3Leu-417.
Locations of photolabeled residues in a1b3g2 GABA A R structure In Fig. 7 we highlight the positions of four photolabeled residues (b3Pro-415, b3Leu-417, b3Thr-418, and b3Arg-309) in a structure recently solved by cryo-EM (18) of a1b3g2 GABA A Rs (Protein Data Base 6I53) purified from the same GABA A Rexpressing HEK 293T cell line used for our purifications. Most of the 116 amino acids comprising the b3 cytoplasmic domain between the M3 and M4 helices are not defined in this structure, which resolves amino acids beginning with b3Pro-415 and locates b3Val-420 at the cytoplasmic end of the bM4 helix. In this structure, b3Pro-415-b3Asp-419 form a turn at the cytoplasmic end of bM4, with the photolabeled residues (b3Pro-415, b3Leu-417, b3Thr-418, and b3-Arg-309 at the cytoplasmic end of bM3) contributing to the base of a pocket at the b 1 -a 2 subunit interface (Fig. 7, B and C) that extends between bM3 and aM1 up to the level of a1Gln-242, the amino acid in aM1 identified by mutational analysis as a major sensitivity determinant for many steroid PAMs (19), including 21-pTFDBzox-AP (27). This pocket is homologous to the intersubunit cleft identified as a binding site for 3a-OH steroid PAMs in crystal structures of homomeric, chimeric receptors containing GABA A R a subunit TMDs (22)(23)(24). Based upon computational docking using CDocker, 21-pTFDBzoxy-AP can be readily accommodated within this intersubunit pocket in the a1b3g2 GABA A R b 1 -a 2 subunit interface, with the lowest energy solutions adopting an orientation with the 3a-OH in proximity to a1Gln-242 and with the reactive diazirine in proximity to the photolabeled residues (Fig. 7C).
Anesthetic, anticonvulsant, and anxiolytic 3a-OH steroids bind to this site By use of competition photolabeling, we established that this site binds with high affinity a structurally diverse group of 3a-OH pregnane GABA A R PAMs that have a wide range of pharmacological activities in vivo ( Fig. 8A and Table 3). Org20599, an amino steroid anesthetic containing a 2b-morpholino-substituent to enhance water solubility (37), inhibited photolabeling with IC 50 = 0.2 mM. Substitutions at the 3b-and 17b-positions that improve bioavailability were well tolerated. Thus, GABA A R PAMs that act in vivo as an anticonvulsant (ganaxolone (38)), an anti-depressant (SAGE-217 (39)), or a sedative/ hypnotic (CCD-3693 (40)) each inhibited photolabeling with a1b3 GABA A Rs were photolabeled on a preparative scale in the presence of 300 mM GABA in the absence or presence of 30 mM 3a5a-P. GABA A R b subunits were isolated by SDS-PAGE and digested with Endo Lys-C. A, when digested, aliquots (10%) were sequenced without further purification, there were peaks of 3 H release in cycles 3/4 and 6/7 for the sample photolabeled in the absence (l) but not the presence (*) of 3a5a-P. Shown above are the sequences of the b3 subunit fragments produced by Endo Lys-C digestion that contain transmembrane helices (M1-M2, M3, and M4). B, 3 H elution profiles when the Endo Lys-C digests were fractionated by rpHPLC, determined by counting 10% of each fraction. Inset, Edman degradation determination of the masses (I 0 ) of b subunit fragments eluting in rpHPLC fractions 25 (bM4), 26/27 (bM3), and 28-29 (bM1). C, 3 H released during sequence analysis of the peak of 3 H (rpHPLC fraction 26) from receptors photolabeled in the absence (l, ♦) and presence (*,^) of 3a5a-P and released PTH-derivatives (h,~) in the absence of 3a5a-P. Equal aliquots were sequenced normally (l, *, h) or with sequencing interrupted at cycle 4 for treatment with OPA (♦,^,~) to prevent further sequencing of the bM3 fragment not containing a proline at that cycle. In the absence of OPA, the PTH-derivatives from the b3Ala-280 fragment (h, I 0 = 1.6 pmol) were detected for 20 cycles of Edman degradation. OPA treatment (~) prevented further sequencing of that fragment, but did not alter the pattern of 3 H release, with peaks in cycles 3/4 and 6/7 without (l) or with (♦) OPA. The persistence of 3 H release in cycles 4, 6, and 7 after OPA treatment was consistent with photolabeling of b3Pro-415, b3Leu-417, and b3Thr-418. This photolabeling was inhibitable by 3a5a-P, because the peaks of 3 H release were reduced by 90% (without (*) or with (^) OPA) when fraction 26 was sequenced from receptors photolabeled in the presence of 3a5a-P. The efficiency of photolabeling of a residue (in cpm/pmol of PTH-derivative) was calculated using Equation 4 ("Experimental procedures"). The data for the control condition are presented as mean (6 S.D.) from 4 (b3Arg-309) or 5 (b3Ile-414, b3Leu-417) independent photolabeling experiments with the number (n) of sequenced samples. Samples from GABA A Rs photolabeled in the absence (control) or presence of 30 mM 3a5a-P were sequenced in parallel. For each pair the percent inhibition at that residue was calculated from the ratio of calculated photolabeling efficiencies, with the mean ( IC 50 , 1 mM, and for UCI-50027, active orally as an anxiolytic (41), the IC 50 was 10 mM. 3b-CH 3 OCH 2 -3a,5a-THDOC (42) (IC 50 = 2 mM) was equipotent with 3a,5a-THDOC as an inhibitor. Each of these compounds inhibited photolabeling maximally to the same extent as 30 mM 3a5a-P, with the exception of ganaxolone, which inhibited maximally by only 71 6 2%.

The binding of C-11 substituted pregnanolones
As the carbonyl at C-11 in alphaxalone reduced its IC 50 value by 20-fold compared with 3a5a-P (Table 1), we used competition photolabeling to determine the effects of other C-11 substituents on binding to this site ( Fig. 9 and Table 5). As seen for the C-11 carbonyl in alphaxalone, the affinity for the 5b analog of alphaxalone (renanolone, IC 50 =10 mM) was 20-fold weaker than that for 3a5b-P. Substitution of an 11b-OH further reduced potency by 20-fold (3a5b-P-11b-ol, IC 50 ; 200 mM). In contrast, high affinity binding was retained in the presence at C-11 of either the small azi-group (11-Azi-AP, IC 50 = 0.4 mM) or the bulky azidotetrafluorophenyl-group (11-F4N3Bzoxy-AP, IC 50 = 0.1 mM) in two recently introduced photoreactive 3a5a-P derivatives that act as potent GABA A R PAMs and general anesthetics (45).

Simultaneous binding with nonsteroidal GABA A R PAMs at the etomidate site
We used competition photolabeling to determine whether PAMs of large size that bind to the etomidate site near the extracellular end of the TMD b 1 -a 2 subunit interface would inhibit [ 3 H]21-pTFDBzox-AP photolabeling of a1b3 GABA A Rs, whether by steric overlap or by negative allosteric interaction. Less than 10% inhibition was seen at the highest concentrations tested for propofol (300 mM, molecular volume, 191 Å 3 ), etomidate (300 mM, volume 208 Å 3 Fig. 5B) from receptors photolabeled in the absence (l,^) and presence (*) of 3a5a-P. Samples were sequenced in duplicate, and the 3 H release is plotted as mean cpm (6½ range). The primary sequence began at b3Arg-216 at the N terminus of bM1 (^, I 0 = 6 pmol) with the b3Ala-280 fragment present at 10% of that level (not shown). Also plotted are the PTH-derivatives released (h) for the total amount of the b3Ala-280 fragment sequenced in fractions 26-30 (I 0 = 9 pmol). The peak of 3a5aP-inhibitable 3 H release in cycles 30/31 of Edman degradation was not seen when fractions 26 or 27 were sequenced (not shown). If the increased hydrophobicity of the photolabeled b3Ala-280 fragment shifted its elution to more hydrophobic fractions than the unlabeled fragment, which eluted in fractions 26-27, the peak of 3 H release in cycle 30 would result from 3a5a-P inhibitable photolabeling of b3Arg-309 near the C terminus of bM3. The following experiment tested this hypothesis. B, 3 H (l, *) and PTH-derivatives (h) released during sequence analysis of a b3 subunit fragment beginning at b3Gly-287 confirms 3a5a-P inhibitable photolabeling of b3Arg-309. From an independent preparative photolabeling of GABA A Rs in the absence (l, h) and presence (*) of 3a5a-P, rpHPLC fractions 25-29 were pooled for sequencing from Endo Lys-C digests of b subunits. Samples were first sequenced for 20 cycles with OPA treatment at cycle 4 (not shown), then sequencing was interrupted for treatment with cyanogen bromide to cleave at methionines (see "Experimental procedures"). After treatment, the fragment beginning at b3Gly-287 (h, I 0 = 1.8 pmol) was sequenced along with fragments beginning at b3Pro-228 in bM1 and b3Thr-262 in bM2. The peak of 3 H release in cycle 23 seen for photolabeling in the absence (l) but not in the presence (*) of 3a5a-P was consistent with photolabeling of b3Arg-309 at 30 cpm/pmol. This efficiency was the same as that calculated for b3Arg-309 photolabeling based upon the peak of 3  We also tested ivermectin (volume 880 Å 3 ), a nonanesthetic activator and PAM of GABA A Rs and other pentameric ligandgated ion channels (47) that binds in a homomeric, invertebrate glutamate-gated chloride channel (Glu-Cl) to an intersubunit site (48) homologous to the etomidate and R-mTFD-MPAB sites in a1b3 or a1b3g2 GABA A Rs. In a1b2g2 GABA A Rs, however, mutational analysis indicate that ivermectin interacts nonequivalently with these sites, with the g 1 -bsite most important for activation (49). We found that 100 mM ivermectin inhibited [ 3 H]21-pTFDBzox-AP photolabeling of a1b3 or a1b3g2 GABA A Rs by ,10% (Fig. 10). In contrast, ivermectin potently inhibited photolabeling by [ 3 H]R-mTFD-MPAB (IC 50 = 0.02 6 0.005 mM, n H = 0.6 6 0.1) and inhibited [ 3 H]azietomidate photolabeling at higher concentrations (IC 50 = 6.4 6 1.0 mM) (Fig. 10). Thus, ivermectin at 100 mM fully occupies those sites without inhibiting [ 3 H]21-pTFDBzox-AP photolabeling, and consistent with the functional studies, ivermectin binds with higher affinity to the a 1 /g 1 -bsites than to the b 1 -asites. Furthermore, the concentration dependence of inhibition of [ 3 H] R-mTFD-MPAB photolabeling (n H = 0.6) established that ivermectin binds nonequivalently to the a 1 -band g 1 -bsites in the presence of GABA. This is not the case for R-mTFD-MPAB, but it is for other GABA A R PAMs including the anesthetic p-benzoyl-propofol and the sedative/anticonvulsant loreclezole (26). Based upon a fit of the inhibition data to a two-site model, ivermectin binds to the b 2 sites with IC 50 values of 3.1 6 1.1 and  ). A, a partial alignment of the human a1, b3, and g2L GABA A R subunits' M3 and M4 helices (denoted by heavy lines) with cytoplasmic extensions, with amino acid numbering of the subunits after signal sequence cleavage. Asterisks (*) denote conserved residues in the alignments, and dashed lines designate residues resolved in the PDB 6I53 GABA A R structure (18). Residues photolabeled by [ 3 H]21p-TFDBzox-AP are color-coded: b3Arg-309 (crimson); b3Ile-414 (red); b3Pro-415 (orange); b3Leu-417 (lime green); and b3Thr-418 (magenta). B and C, images of the PDB 6I53 structure with horizontal lines approximating membrane-aqueous interfaces. In B, a-helices are cylinders and b-sheets are ribbons. Binding sites for GABA (green, overlaid from PDB 6HUJ), etomidate (red, docked), and aTHDOC (blue, docked) are included. C, an expanded view of the TMD at a b 1 -ainterface with the [ 3 H]21p-TFDBzox-AP labeled residues (b3Arg-309, b3Pro-415, b3Leu-417, and b3Thr-418) highlighted. These residues are shown as Connolly surfaces, as are the others that contribute to an intersubunit pocket extending to aGln-242 (purple), a residue identified by mutational analysis as a steroid sensitivity determinant (19). 21p-TFDBzox-AP is in stick format, docked in this pocket in its lowest energy orientation with a transparent Connolly surface of the 9 lowest energy solutions. In this orientation, the photoreactive diazirine is within 5 Å of b3Arg-309, b3Leu-417, and b3Thr-418, and the 3a-OH is within 5 Å of aGln-242 and 3 Å of aTrp-246, a residue also identified by mutational analysis as a steroid sensitivity determinant (20). Also shown is the etomidate-binding site in the b 1 -ainterface near the extracellular end of the TMD, defined by the residues photolabeled by etomidate analogs and by mutational analysis (b3Met-286, b3Val-290, a1Leu-232, and a1Met-236 (32, 69)) and a docked etomidate (in stick figure). 150 6 65 nM. Further studies would be required to determine whether it is the a 1 -bor g 1 -bsite that binds with highest affinity.

Discussion
In this report we show that a novel photoreactive steroid, [ 3 H]21-pTFDBzoxy-AP, binds with high affinity to a site in the TMD of heteromeric GABA A Rs at the cytoplasmic end of the b 1 -asubunit interface, and we use a photolabeling inhibition assay to provide a first definition of the structure-affinity relationships for a GABA A R steroid-binding site. In a1b3 and a1b3g2 GABA A Rs, pharmacologically specific [ 3 H]21-pTFD-Bzoxy-AP photolabeling was primarily within the b subunit, with the photolabeled amino acids located in the GABA A R structure near the cytoplasmic ends of the bM4 (b3Pro-415, b3Leu-417, and b3Thr-418) and bM3 (b3Arg-309) helices that contribute to the base of a pocket at the b 1 -asubunit interface. This binding site extends upward to the level of a1Gln-242, a position identified by mutational analysis as a major determinant for steroid enhancement of GABA responses. Many 3a-OH pregnane and androstane GABA A R PAMs bind to this site at concentrations similar to those necessary for GABA A R enhancement, but we also identified potent steroid GABA A R PAMs that do not bind to this site. High affinity binding depends on the presence of a free 3a-OH and is highly sensitive to the nature of the substitution at the C-17 position. 3-Deoxy-5a-P and steroids with an -OH in place of the carbonyl at C-20 of 3a5a-P enhance GABA responses with potencies similar to 3a5a-P (20,30,43), but their binding affinities for the b 1 -a 2 steroid site are reduced by more than 1000-fold. These potent steroid PAMs that do not bind to the b 1 -asteroid site should serve as useful lead compounds for the development of novel reagents to identify additional GABA A R steroid-binding sites.

Structural determinants for binding to the b 1 -asubunit interface steroid site
Because the b subunit amino acids photolabeled by [ 3 H]21-pTFDBzoxy-AP were located within a common binding pocket at the b 1 -asubunit interface, characterization of the effects of nonradioactive steroids on GABA A R photolabeling at the level of the b subunit could be used to determine the affinities (IC 50 values) of nonradioactive drugs for that site. We did not identify any nonsteroidal GABA A R PAMs that inhibited photolabeling, including drugs varying in size from propofol to ivermectin that bind to the adjacent etomidate site at the b 1 -asubunit interface. Many steroid 3a-OH GABA A R PAMs were potent inhibitors of [ 3 H]21-pTFDBzoxy-AP photolabeling, reducing b subunit photolabeling maximally to the same extent as 30 mM 3a5a-P with a concentration dependence characterized by a Hill coefficient of 1. The simplest interpretation of this pattern of inhibition is that it results from competitive interactions at a common binding site. For a1b3 and a1b3g2 GABA A Rs, the IC 50 values for 5b-isomers differed by less than a factor of 2 from those for 3a5a-P, 3a5a-THDOC, and alphaxalone (Tables 1 and 4), and the IC 50 values for inhibition of photolabeling were within a factor of 2 of EC 50 values reported for enhancement of GABA responses (12,23,45). A similar good correlation between photolabeling inhibition IC 50 and GABA A R enhancement EC 50 was seen for many substituted pregnanolones, including those acting in vivo as an anesthetic (Org20599), anti-convulsant (ganaxolone), or antidepressant (SAGE-217) ( Tables 3 and 5).
Our results establish that high affinity binding to the GABA A R b 1 -a --binding site depends critically on the presence of a free 3a-OH, consistent with the inactivity of 3b-OH steroids, 3-oxo-5a-P, and 3a-acetyl-5a-P as GABA A R PAMs (9)(10)(11)28). We found that they bound at least 1000-fold more weakly than 3a5a-P. 3-Deoxy-5a-P also did not bind to this site, despite the fact that it acts as a GABA A R PAM with a  Tables 3 and  4. A, substitutions at the 2b-(Org-20599) and 3b-(ganaxolone, SAGE-217, CCD-3693, UCI-50027) positions are well tolerated, as is the presence at C-19 of an -H (SAGE-217, CCD-3693) rather than -CH 3 . Pregnanes with a carbonyl at C-20 bind with high affinity, but those with an -OH do not. With the exception of ganaxalone (B ns = 28.6 6 1.6%), curves were calculated from fits with B ns = 0 and n H = 1. B, substituents at steroid carbon 17 (C-17) are a major determinant of binding affinity. 5a-Androstan-3a-ol (3a5a-A), with hydrogens at C-17, and 3a5b-P-20-deoxy, with a 17b-ethyl substituent, bind to this site, but 3a5a-A17a-ol does not. Inhibition curves were calculated from fits with B ns = 0 and variable n H for 5aA3a-ol,17-one (IC 50  potency similar to 3a5a-P (20). Thus, 3-deoxy-5a-P enhances GABA responses without binding to this site.
We found substitutions at the steroid C-17 position that were unexpectedly important determinants of binding to the b 1 -a 2 site. Early studies of 3a-OH steroids as GABA A R PAMs established that the C-20 carbonyl of 3a5a-P was not essential, because androsterone (3a5aA-17-one) and pregnan-3a,20-diols were potent PAMs (11,30,43). The C-20 carbonyl is also not essential for binding to the b 1 -asite, because 3a5a-A and 3a5b-P-20-deoxy, PAMs with an -H or b-ethyl at C-17, bound with high affinity (Table 4). However, two potent PAMs, 5a-pregnan-3a,20a-diol and 5b-pregnan-3a,20b-diol, did not bind to the b 1 -asite at concentrations 100-fold higher than necessary for GABA A R enhancement. 3a5a-A-17a-ol, with an a-OH at C-17, also did not bind to the b 1 -asite. However, this may not be simply a consequence of the -OH, because 3a5a-A-17b-ol is a PAM (50), as are other steroids with a C-17 side chain in a b-configuration (11,51).
Although many 3a-OH steroid PAMs potently inhibited GABA A R photolabeling to the same extent as 3a5a-P and with a concentration dependence characterized by a Hill coefficient of 1, ganaxalone and 3a5aA-17-one were exceptions. Ganaxolone was a potent inhibitor (IC 50 = 0.3 mM), but at high concentrations, maximal inhibition (B ns = 29 6 2%) was less than that seen in the presence of 3a5a-P (B ns = 0%), whereas other PAMs with 3b-substituents inhibited fully. For 3a5aA-17-one, which enhances a1b2g GABA A R responses with an EC 50 of ;3 mM (52), inhibition was fit equally well either to n H = 1 and variable B ns (IC 50 = 5 6 2 mM, B ns = 63 6 3%) or with B ns equal to 0 and a variable Hill coefficient (IC 50 = 700 6 390 mM, n H = 0.32 6 0.05). There was no evidence that the partial inhibition resulted from limited solubility of these two steroids in the detergent/lipid environment used for GABA A R purification, and further studies are required to clarify the mechanism of inhibition.

Mode of steroid binding at the b 1 -asteroid site
Our photolabeling results establish that 21-pTFDBzoxy-AP binds in heteromeric GABA A Rs at the b 1 -asubunit interface. Based upon computational docking, in its most energetically favorable binding mode, 21-pTFDBzoxy-AP's photoreactive diazirine is in proximity to the photolabeled amino acids at the cytoplasmic surface of the TMD and the 3a-OH is in proximity to a1Gln-242 and a1Trp-246. Thus, 21-pTFDBzoxy-AP binds at the b 1 -asubunit interface site in an orientation similar to that of 3a5a-THDOC, 3a5b-P, or alphaxalone at a 1 -asubunit interfaces in the crystal structures of homopentameric, chimeric receptors with GABA A R a subunit TMDs (22)(23)(24). Consistent with this mode of binding, positive modulation of GABA responses by 21-pTFDBzoxy-AP is lost in the a1Q242W mutant receptor (27). Although direct interactions between the free 3a-OH and aGln-242 were predicted based upon the loss of 3a5a-P PAM activity seen for substitutions at aGln-242 (19), substitutions at aGln-242 also caused loss of PAM activity for 3-deoxy-5a-P (20), a PAM that did not inhibit [ 3 H]21-pTFD-Bzoxy-AP photolabeling. This discrepancy indicates that substitutions at aGln-242 can interfere with PAM activity even for steroids that do not bind to the b 1 -asite, and the GABA A R amino acids interacting directly with the 3a-OH remain to be determined. Based upon competition photolabeling, the presence of a free -OH at C-20 is as deleterious for binding to the b 1 -asite as is its absence at the C-3 position, even though a free hydroxyl group at C-21 (3a5a-THDOC) or certain bulky substitutions (21-pTFDBzoxy-AP, SAGE-217) are well-tolerated. Just as the high affinity binding associated with the 3a-OH must result from specific interactions with GABA A R amino acids, the 1000-fold reduction of binding affinity (IC 50 ) for the pregnane diols compared with 3a5a/b-P is likely to result from energetically unfavorable interactions between a C-20 hydroxyl and GABA A R amino acids in the b 1 -ainterface steroid-binding site. If a pregnan-3a,20a/b-diol binds to the b 1 -asteroid site in the same orientation as alphaxalone or pregananolone in the crystal structures of chimeric receptors with a subunit TMDs (23, 24), the C-20 hydroxyl would be in proximity to b3Phe-301 in bM3, the position equivalent to a1Thr-206/a5Thr-309 that is in proximity to the C-20 carbonyl of alphaxalone and pregnanolone. Simple solubility considerations cannot account for the loss of binding, because incorporation of a hydroxyl residue instead of a carbonyl group increases the octanol-water partition coefficient calculated by the ALOGPS 2.1 program (RRID: SCR_018786) for any of the steroids tested by less than a factor of two, whereas 3a5b-P binds 7-fold more tightly and 5b-pregnan-3a,20b-diol binds .100-fold more weakly than 3a5b-P-20-deoxy (Table 4). Thus, the carbonyl at C-20 allows for favorable energetic interactions with GABA A R amino acids not possible for a hydroxyl function.
In contrast to the differential effect on binding affinity seen for incorporation of a carbonyl or hydroxyl at C-20, direct incorporation of a carbonyl (androsterone) or a hydroxyl (3a5aA-17a-ol) on the steroid ring system at C-17 weakens binding by more than 50-fold compared with 3a5a-A. Likewise, the presence of a carbonyl (renanolone) or a hydroxyl (3a5b-P-11b-OH) at C-11 weakens binding by 20-and 300fold compared with 3a5b-P. Because incorporation of a carbonyl or a hydroxyl will decrease the steroid partition coefficient by 10-30-fold, the decreased affinities (increased IC 50 values, measured as the total steroid concentration) for these steroids may result in large measure from their decreased hydrophobicity and lipid partitioning rather than from energetically unfavorable specific interactions of the carbonyl or hydroxyl with the GABA A R.

Additional binding sites for steroid PAMs
Our sequencing results established that the b subunit amino acids photolabeled most efficiently by [ 3 H]21-pTFDBzoxy-AP all contribute to the steroid-binding site at the b 1 -asubunit interface. Although we did not identify photolabeled amino acids that would contribute to other steroid-binding sites, our competition photolabeling results identified three potent PAMs (EC 50 ,10 mM, 3-deoxy-5a-P, 5a-pregnan-3a,20a-diol, and 5b-pregnan-3a,20b-diol) that did not bind to this site. The absence of an a hydroxyl at C-3 at one end of the steroid ring system or the presence of a hydroxyl at C-20, 10 Å away at the other end of the ring system, prevents binding to the b 1 -asteroid site. The presence of the 3a-OH is insufficient to overcome unfavorable interactions at the other end of the molecule. It remains to be determined whether one or more of these  "orphan" PAMs bind to the intrasubunit sites near the extracellular end of the TMD recently identified by photolabeling in aM4 by steroids containing photoreactive groups at C-21 or C-6 and in bM3 by a steroid containing a C-3a photoreactive group (25).

Antagonist steroids
Although early studies suggested that 3b-OH steroids competitively antagonize steroid enhancement of GABA A R function (53)(54)(55), subsequent studies indicate that they noncompetitively inhibit GABA responses in the absence of enhancing steroids, acting in a manner more similar to the sulfated 3b-OH neurosteroids PS and DHEAS (12,56). Although steroid PAMs generally enhance [ 3 H]muscimol or [ 3 H]flunitrazepam equilibrium binding (51,57,58), our results indicate that inhibitory 3b-OH steroids can modulate [ 3 H]muscimol binding in 3 different ways. (i) The enhancement of binding seen for betaxalone and 3b5b-P indicates that those steroids stabilize the GABA A R in a desensitized state with high affinity for [ 3 H]muscimol. 17-PA, a 3a-OH androstene that does not inhibit GABA responses in the absence of a steroid PAM (31,59), also enhanced [ 3 H]muscimol binding. (ii) The lack of modulation seen for 3b5a-P and PS at concentrations as high as 100 mM suggests that they do not perturb the receptor conformational state. (iii) That DHEAS partially inhibits binding indicates that it stabilizes the receptor in a state with low affinity for [ 3 H]muscimol, potentially a resting, closed channel state. The effects we observed for 3b5a-P, PS, and DHEAS are consistent with previous studies of [ 3 H]muscimol binding to rat brain membranes (29,60).
Our competition photolabeling results are consistent with functional studies indicating that free and sulfated 3b-OH steroids inhibit GABA responses without binding to the same site as steroid PAMs. Thus, PS, 3b5a-P, and 3b5b-P inhibit GABA responses with IC 50 values of less than 5 mM (12, 61), but any inhibition of [ 3 H]21-pTFDBzoxy-AP, photolabeling, if it occurred, would be characterized by IC 50 values greater than 300 mM. Betaxalone was a possible exception, because inhibition (IC 50 = 175 mM) occurred at similar concentrations as enhancement of [ 3 H]muscimol binding (EC 50 = 50 mM). However, the concentration dependence of inhibition of photolabeling (n H = 0.5) was inconsistent with a simple model of direct completion for the [ 3 H]21-pTFDBzoxy-AP-binding site.

Conclusions
We have shown that [ 3 H]21-pTFDBzoxy-AP, a photoreactive steroid and GABA A R PAM, binds with high affinity in the b 1 -asubunit interface of heteromeric, human, full-length a1b3 and a1b3g2L GABA A Rs in a site homologous to that revealed in crystal structures of chimeric homomeric pentameric ligand-gated ion channels of the same superfamily. We used competition photolabeling to establish that the steroid structure-activity relationships of this site parallel that observed in many functional pharmacological studies. These studies also reveal that some potent PAMs, such as 3a-deoxy-5a-P and pregnan-3a,20-diols, bind to a different site or sites, Thus, [ 3 H] 21-pTFD-Bzoxy-AP is a useful tool for the development of steroids that selectively target specific sites on GABA A Rs including those with other subunit compositions.
Human a1b3 and a1b3g2 GABA A Rs with the a1 subunits containing a FLAG epitope at the N terminus of the mature subunit were expressed in tetracycline-inducible, stably transfected HEK293-TetR cell lines, and purified from detergent  Table  5. Covalent incorporation of 3 H was determined by liquid scintillation counting of b3 subunits isolated by SDS-PAGE, and data from independent experiments were normalized and combined as described under "Experimental procedures" and in Fig. 4. The plotted data are the mean 6 S.D. from the independent experiments.
After photolabeling, GABA A R subunits were separated by SDS-PAGE as described (13), and gel bands containing a/g (56 kDa) and b (59/61 kDa) subunits were identified by Gel Code Blue Safe Protein Stain (ThermoFisher Scientific) for analytical labelings (26). For analytical scale experiments, [ 3 H]21-pTFDBzox-AP incorporation was measured by scintillation counting of excised gel bands (in 3 H cpm) or by fluorography (13). For preparative scale experiments, gel bands of interest were excised and eluted passively in elution buffer (100 mM NH 4 HCO 3 , 2.5 mM DTT, 0.1% SDS, pH 8.4) for 3 days at room temperature. The eluates were filtered and concentrated, and the proteins in the eluate were precipitated (75% acetone, overnight at 220°C) and then resuspended in digestion buffer (15 mM Tris, 0.5 mM EDTA, 0.1% SDS, pH 8.4).

Quantitation of concentration dependence of inhibition of photolabeling
The concentration dependence of inhibition of 3 H incorporation into GABA A R subunits was fit by nonlinear least squares to Equation 2, where, B(x) is the 3 H cpm incorporated into a subunit gel band at a total inhibitor concentration of x. B 0 is incorporation in the absence of inhibitor, B ns is the nonspecific incorporation, IC 50  Enzymatic and chemical fragmentation a1 and b3 subunits isolated by SDS-PAGE from a1b3 GABA A Rs photolabeled on a preparative scale were digested with Endo Lys-C (3-5 mg, 3 days, 20°C), which produces fragments beginning at the N termini of each subunit's M1, M3 and M4 helices that can be separated and purified by rpHPLC (13). To cleave at the C-terminal side of methionines, samples already loaded onto sequencing supports were treated with cyanogen bromide as described (65,66).

HPLC purification and protein microsequencing
Subunit digests were fractionated by rpHPLC on an Agilent 1100 binary pump system using a Brownlee C4 Aquapore column (100 3 2.1 mm, 7-mm particle size) at 40°C with an upstream guard column (Newguard RP-2). The aqueous solvent contained 0.08% TFA and the organic solvent contained 60% acetonitrile, 40% isopropyl alcohol, 0.05% TFA. Elution was achieved using a nonlinear gradient increasing from 5 to 100% organic solvent over 80 min at a flow rate of 0.2 ml/min. Fractions of 0.5 ml were collected, and 10% aliquots were assayed for determination of 3 H. Fractions of interest were pooled for sequencing and droploaded at 45°C onto Micro TFA glass fiber sequencing filters (Applied Biosytems) that were treated after loading with Biobrene Plus (Applied Biosystems).
Samples were sequenced on an Applied Biosystems Procise 492 Protein sequencer programed to use 2/3 (;80 of 120 ml) of the material from each cycle of Edman degradation for PTHderivative identification and quantitation and to collect 1/3 for 3 H determination by liquid scintillation counting. For some samples, sequencing was interrupted at a designated cycle for treatment of the sequencing filter with o-phthalaldehyde (35,36) to prevent further sequencing of any peptide not containing a proline at that cycle. The amount of PTH-derivative released (in picomoles) for a given residue was quantified using their peak height in the chromatogram, background-subtracted, compared with a standard peak, and the PTH-derivatives released for the detected peptide were fit to the equation, where F(x) is the pmol of the amino acid in cycle x, I 0 is the calculated initial amount of the peptide, and R is the repetitive yield. The 1st residue in the peptide as well as Cys, Trp, His, and Ser were not used in the calculation due to known problems with their quantitation. The efficiency of photolabeling (E in cpm/pmol) at a labeled amino acid in cycle x was calculated by the equation, where cpm x is the 3 H released in cycle x.

Molecular modeling and computational docking
For computational docking studies, we used the recently solved cryo-EM structure of a desensitized state of a1b3g2L GABA A R (PDB 6I53) (18) in a lipid-nanodisc with a bound positive allosteric modulator megabody in the extracellular domain. This structure was determined from GABA A Rs purified from the same cell line as that used in our photolabeling studies, a cell line expressing full-length receptor subunits with intact cytoplasmic domains. Although most of the ;120 amino acids comprising each subunit cytoplasmic domain were not resolved in this structure, the locations of 4 of the 5 amino acids specifically photolabeled by [ 3 H]21-pTFDBzox-AP were resolved. In contrast, only the photolabeled b3Arg-309 was resolved in 5 other structures using the same source of GABA A Rs that were determined in the presence of GABA, picrotoxinin, or bicuculline (17).
Docking of 21-pTFDBzox-AP and other steroids to the PDB 6I53 model was performed using the Discovery Studio CDOCKER module. Potential binding sites at each subunit interface of the PDB 6I53 structure were defined by 14-Å radius binding site spheres centered by the position of 3a5a-THDOC molecules overlaid (Discovery Studio: Tools: Superimpose Proteins: Sequence Alignment) from the PDB 5OSB structure (after removal of the extracellular domain and cytoplasmic linker). For docking, a structure of 21-pTFDBzox-AP was created by appropriate additions at the 21 position of 3a5a-THDOC (PubChem structure CID No. 101,771). Four copies of 21-pTFDBzox-AP, differing by rotations of ;180°, were seeded into the binding site spheres. The 50 lowest energy solutions (simulated annealing with full potential minimization) were collected for each molecule from 50 random conformations (high temperature molecular dynamics) and 50 randomized orientations within the sphere (i.e. 2,500 attempted dockings per molecule). In two independent docking runs, we found that 21-pTFDBzox-AP was predicted to bind most favorably at the gb 1 -asubunit interface with the lowest CDocker interaction energy (249.17 kcal/mol) at that site 4.5 kcal/mol more favorable than at the b 1 -ag-binding site and more than 10 kcal/ mol more favorable than at the homologous a 1 -g -, a 1 -b -, or g 1 -bintersubunit sites. At the gb 1 -asite, for the energetically most favored solution and 56% of all collected solutions, 21-pTFDBzox-AP adopted a common orientation with the 3a-OH directed toward a1Gln-242 and the aromatic diazirine extending linearly from the steroid backbone into a groove between b3Arg-309 and b3Leu-417, residues photolabeled by [ 3 H]21-pTFDBzox-AP (see "Results"). 3a5a-THDOC and 3a5a-P were also predicted to bind in a similar orientation at the gb 1 -asite, with most favorable CDOCKER interaction energies of 240.6 and 235.0 kcal/mol. Although both molecules were predicted to bind in an orientation with the 3a-OH in proximity to a1Gln-242, no consistent prediction was made concerning the energetic importance of a 3a-OH. For THDOC, the CDOCKER interaction energy for the 3a-OH isomer was 1.8 kcal/mol more favorable than for the 3b-epimer, whereas the interaction energy was 1.4 kcal/mol more favorable for 3b5a-P than for 3a5a-P.

Data availability
All data are contained within the article.