Digoxin Derivatives with Enhanced Selectivity for the α2 Isoform of Na,K-ATPase

Background: The α2 isoform of Na,K-ATPase may play a major role in aqueous humor production. Results: We have chemically modified digoxin and obtained derivatives with enhanced selectivity for α2. When applied topically, they effectively reduce intraocular pressure in rabbits. Conclusion: α2 is crucial for aqueous humor production. Significance: Potentially, the derivatives may be useful for control of intraocular pressure. In the ciliary epithelium of the eye, the pigmented cells express the α1β1 isoform of Na,K-ATPase, whereas the non-pigmented cells express mainly the α2β3 isoform of Na,K-ATPase. In principle, a Na,K-ATPase inhibitor with selectivity for α2 could effectively reduce intraocular pressure with only minimal local and systemic toxicity. Such an inhibitor could be applied topically provided it was sufficiently permeable via the cornea. Previous experiments with recombinant human α1β1, α2β1, and α3β1 isoforms showed that the classical cardiac glycoside, digoxin, is partially α2-selective and also that the trisdigitoxose moiety is responsible for isoform selectivity. This led to a prediction that modification of the third digitoxose might increase α2 selectivity. A series of perhydro-1,4-oxazepine derivatives of digoxin have been synthesized by periodate oxidation and reductive amination using a variety of R-NH2 substituents. Several derivatives show enhanced selectivity for α2 over α1, close to 8-fold in the best case. Effects of topically applied cardiac glycosides on intraocular pressure in rabbits have been assessed by their ability to either prevent or reverse acute intraocular pressure increases induced by 4-aminopyridine or a selective agonist of the A3 adenosine receptor. Two relatively α2-selective digoxin derivatives efficiently normalize the ocular hypertension, by comparison with digoxin, digoxigenin, or ouabain. This observation is consistent with a major role of α2 in aqueous humor production and suggests that, potentially, α2-selective digoxin derivatives could be of interest as novel drugs for control of intraocular pressure.

In the ciliary epithelium of the eye, the pigmented cells express the ␣1␤1 isoform of Na,K-ATPase, whereas the nonpigmented cells express mainly the ␣2␤3 isoform of Na,K-ATPase. In principle, a Na,K-ATPase inhibitor with selectivity for ␣2 could effectively reduce intraocular pressure with only minimal local and systemic toxicity. Such an inhibitor could be applied topically provided it was sufficiently permeable via the cornea. Previous experiments with recombinant human ␣1␤1, ␣2␤1, and ␣3␤1 isoforms showed that the classical cardiac glycoside, digoxin, is partially ␣2-selective and also that the trisdigitoxose moiety is responsible for isoform selectivity. This led to a prediction that modification of the third digitoxose might increase ␣2 selectivity. A series of perhydro-1,4-oxazepine derivatives of digoxin have been synthesized by periodate oxidation and reductive amination using a variety of R-NH 2 substituents. Several derivatives show enhanced selectivity for ␣2 over ␣1, close to 8-fold in the best case. Effects of topically applied cardiac glycosides on intraocular pressure in rabbits have been assessed by their ability to either prevent or reverse acute intraocular pressure increases induced by 4-aminopyridine or a selective agonist of the A3 adenosine receptor. Two relatively ␣2-selective digoxin derivatives efficiently normalize the ocular hypertension, by comparison with digoxin, digoxigenin, or ouabain. This observation is consistent with a major role of ␣2 in aqueous humor production and suggests that, potentially, ␣2-selective digoxin derivatives could be of interest as novel drugs for control of intraocular pressure.
In the ciliary epithelium, the Na,K-ATPase creates the cation gradients that power production of the aqueous humor (1). In principle, inhibition of the Na,K-ATPase should suppress the production of aqueous humor and control intraocular pressure (IOP). 2 Control of IOP is the mainstay of glaucoma therapy (2), but despite the selection of drugs available, fresh approaches to drug treatments are highly desirable. Previously, intravenous digoxin, a classical inhibitor of the Na,K-pump, used primarily to treat congestive heart failure, was considered for this role but was discarded due to systemic toxicity (3,4).
The Na,K-ATPase consists of ␣ and ␤ subunits (␣␤) and accessory FXYD regulatory subunits (5,6). There are four isoforms of the ␣ subunit (␣1-␣4) and three isoforms of the ␤ subunit (␤1-␤3) expressed in a tissue-specific fashion (7). ␣1 is the common isoform that maintains sodium and potassium gradients in all tissues, whereas ␣2 is expressed mainly in heart and skeletal muscle and astrocytes, ␣3 in nerve, and ␣4 in testis. The ciliary epithelium in the eye is a functional syncytium consisting of pigmented cells (PE) facing the stroma and non-pigmented (NPE) cells facing the anterior chamber of the eye (1). The primary Na,K-ATPase isoform of the PE is ␣1␤1, whereas that of the NPE is ␣2␤3 (8). Thus, in principle, ␣2-selective cardiac glycosides (CGs) could effectively reduce IOP, and provided that they penetrate the intact eye and reach the ciliary epithelium, they could be applied topically. A potential advantage of topical application could be that systemic toxic effects typical of cardiac glycosides should be minimal (9).
␣2 isoform-selective cardiac glycosides are also, potentially, of great pharmacological interest as safer cardiotonic agents as well as in the present context of intraocular pressure and could provide important tools to study isoform function. For these reasons, previously we examined the isoform selectivity of known cardiac glycosides (10). The experiments utilized Pichia pastoris membranes expressing human Na,K-ATPase isoforms (␣1␤1, ␣2␤1, and ␣3␤1) and purified detergent-soluble isoform complexes of Na,K-ATPase (10 -13). The major findings were that digoxin and digitoxin showed 3-4-fold binding selectivity for ␣2 or ␣3 over ␣1, whereas aglycones, such as digoxigenin and digitoxigenin, showed no isoform selectivity. By contrast, ouabain showed some preference for ␣1 over ␣2 (see also Ref. 14). The conclusion from the work in Ref. 10 was that the isoform selectivity is determined by the sugar moiety of digoxin, especially the third digitoxose.
Much older and recent work has pointed to the importance of the sugars of cardiac glycosides for binding to Na,K-ATPase (10,(15)(16)(17)(18). Digoxin and digitoxin derivatives with zero, one, two, and three digitoxose moieties bind to purified lamb kidney or human Na,K-ATPase (␣1␤1) with increasing affinities (10,19). It was proposed many years ago that there are distinct sites and roles for the steroid-lactone and sugar moieties (15). The insights on binding and isoform selectivity are generally consistent with recent structures of Na,K-ATPase with bound ouabain (20 -22). The unsaturated lactone ring and steroid portion of ouabain are bound between trans-membrane segments M1, M4, and M5 of the ␣ subunit, in which there are no amino acid differences between isoforms. Assuming that the lactone-steroid moieties of all cardiac glycosides bind similarly, the implication is that isoforms cannot discriminate between any aglycones, as found experimentally (10). By contrast, the sugar is bound near extracellular loops, where there are a small number of amino acid differences between the isoforms. These residues might interact with the sugars of bound digoxin in an isoformselective way.
Based on a prediction (10) that selective chemical modification of the third digitoxose residue of digoxin might increase selectivity for ␣2 over ␣1, we have synthesized a series of perhydro-1,4-oxazepine derivatives of digoxin using a well worked out synthetic route (23) and tested selectivity. In parallel, we have tested whether topically applied cardiac glycosides, especially relatively ␣2-selective digoxin derivatives, effectively reduce intraocular pressure in rabbits and thus provide information on the functional role of ␣2 in ciliary epithelium.

MATERIALS AND METHODS
Ouabain (O3125) and digoxin (D6003), 4-aminopyridine (A78403), and IB-MECA (I146) were obtained from Sigma. HPLC grade methanol was from Baker. All of the organic solvents, reagents, and amines were of the highest purity analytical grade.
Assay of Na,K-ATPase Activity of Purified Isoform Complexes-Inhibition of Na,K-ATPase activity of the detergentsoluble ␣1␤1, ␣2␤1, and ␣3␤1 complexes by CGs was done exactly as described (10), using either the ␣␤ or ␣␤FXYD1 complexes. The presence or absence of FXYD1 does not affect inhibition of Na,K-ATPase activity by cardiac glycosides (10) but strongly stabilizes the complexes (13,24). The percentage inhibition of VCG/V0 was calculated, and K i values were obtained by fitting the data to the function, VCG/V0 ϭ K i /([CG] ϩ K i ) ϩ c. Inhibition was estimated in 3-8 separate experiments, and average K i values Ϯ S.E. were calculated. The significance of differences between K i ␣1 and K i ␣2 was calculated by the unpaired Student's t test (p values). The ratio of K i ␣1/␣2 Ϯ S.E. was calculated for each compound, and p values were calculated by comparison with digoxin. p values of Ͻ0.05 were considered significant.
Dissociation Rates of Cardiac Glycosides-Purified ␣2␤1FXYD1 complexes (0.3-0.5 mg/ml) were incubated for 30 min at 37°C in a medium containing 1 mM ATP, 100 mM NaCl, 4 mM MgCl 2 , 25 mM histidine⅐HCl, pH 7.4, without (control) or with 1 M of different cardiac glycosides. The enzyme solutions were then diluted 100-fold into a medium containing 100 mM NaCl, 5 mM KCl, 1 mM EDTA (Tris), 0.005 mg/ml C12E8, 0.01 mg/ml (1-stearoyl-2-oleoyl-sn-glycero-3-[phosphor-L-serine], 0.001 mg/ml cholesterol and incubated at 37°C for different lengths of time. Aliquots were removed at different times, and Na,K-ATPase activity was measured in triplicate over 0.5 min (digoxigenin) or 2 min (other cardiac glycosides) in the standard activity medium containing ATP (200 M). The activity of test samples was divided by the activity of the control samples, and the time course for reversal of inhibition was analyzed by fitting the data to the function v t ϭ v ∞ e Ϫkt ϩ c. Normalized curves for comparison of different experiments (e.g. as in Fig. 7) were obtained by subtracting the constant value c from each value of the activity and refitting the ratio v t /v ∞ ϭ 1 Ϫ e Ϫkt .
Synthesis of Perhydro-1,4-oxazepine Derivatives of Digoxin-Syntheses of digoxin perhydro-1,4-oxazepine derivatives were performed in two steps: 1) oxidation of digoxin with sodium periodate to give an open ring dialdehyde in the third sugar moiety (25) and 2) reductive amination with a primary amine, in the presence of NaCNBH 3 (23), closing a seven-membered ring to give the digoxin perhydro-1,4-oxazepine derivative. Details are given here for the case of the glycinamide derivative, DGlyN.
Oxidation of Digoxin with NaIO 4 (25)-A solution of NaIO 4 (400 mg, 1840 mol) in H 2 O (4 ml) was added under stirring at room temperature to a suspension of digoxin (400 mg, 512 mol) in 95% EtOH (36 ml, not fully soluble). The mixture that dissolved immediately was allowed to stand at room temperature for 1 h. A precipitate of NaIO 3 was formed and was removed by centrifugation and filtration through a syringe filter (polytetrafluorethane, 0.2 m). The solution was concentrated and extracted with CHCl 3 . The organic layer was washed with water, dried over Na 2 SO 4 , filtered, and evaporated in an evaporator and high vacuum to give the dialdehyde, which is dissolved in 48 ml of methanol to give a 10 mM solution of dialdehyde.
Reductive Amination with Glycinamide Hydrochloride (23)-Glycinamide hydrochloride (28.2 mg, 256 mol, Aldrich) and digoxin dialdehyde (240 mol) were dissolved in absolute methanol at 12 and 10 mM, respectively. The apparent pH was corrected to 5-6 with concentrated acetic acid in methanol, and the mixture was kept at room temperature for 5 min. The Schiff base that forms was reduced with an excess of NaCNBH 3 with stirring. Progress of the reaction was monitored by TLC. The mixture was left for 1.5 h, and the methanol was evaporated.
The DGlyN reaction mixture was dissolved in a minimal amount of 50% methanol, filtered, and used for purification by HPLC. HPLC purification was done on a Purospher STAR RP-18e semipreparative column, eluted with a gradient of 50 -80% methanol in water in 15 column volumes at a flow rate of 4 ml/min. Other derivatives were purified using optimal gradients of methanol established in analytical HPLC runs (Chromolith RP-18e) prior to application to the semipreparative column. The methanol was JT Baker HPLC gradient grade.
Animals-New Zealand White rabbits (3-3.5 kg) about 1 year old, of either sex, were housed individually in separate cages in animal room conditions on a reversed, 12-h dark/light cycle. For the experiments the animals were transferred to rabbit restrainers in a quiet and calm atmosphere. No ocular abnormalities were detected prior or during the experiments. Animal care and treatment were subject to the approval of the institutional committee for animal experiments (Weizmann Institute IACUC Permission 04270911-2).
Intraocular Pressure and Corneal Thickness Measurements-IOP (mm Hg) of rabbits was measured using a calibrated Pneumatonometer (model 30, Reichert Technologies). A local anesthetic, oxybuprocaine HCl (0.4%, 25 l), was applied to each cornea about 1 min before IOP measurements. Two baseline IOP readings were taken before topical administration of the CG (or PBS as control) and after a half-hour (zero time). The readings of the two measurements were almost identical, suggesting that the CGs had little or no effect on the basal IOP. At zero time, one drop of 4-amino pyridine (4AP) (40 mg/ml, 30 l) or IB-MECA (1 M, 30 l) was administered to both eyes of each rabbit, and IOP measurements were made at different times, as indicated in each experiment. For experiments in which the IOP was elevated for several hours, 4AP was added every 1.5 h, or IB-MECA was added every 2 h. The pneumatonometer readings were accepted when the S.D. of the value was between 0.1 and 0.4 mm Hg, representing a possible error of 6 -13% compared with the minimal 3-mm Hg increase and 1.6 -6.7% compared with the maximal 6-mm Hg increase in IOP induced by 4AP or IB-MECA. Each experiment was repeated two or three times with similar results. In all cases, the figures depict the average effect on IOP (i.e. for four or six eyes) compared with control Ϯ S.E. Where error bars are not seen in the figures, the errors are smaller than the symbols used. Significance of differences from the control was calculated by the unpaired Student's t test (p values). p values of Ͻ0.05 were considered significant.
Corneal thickness (m) was measured using an ultrasonic pachometer (Sonogage pachometer, Cleveland, OH), before and during the experiment with CG and 4AP or IB-MECA treatments. The values represent averages of three independent measurements for each eye.
Drug Preparation and Administration-Stock solutions of cardiac glycosides were dissolved in ethanol and freshly diluted in PBS for each experiment such that the final ethanol concentration did not exceed 1%.
Modeling-Digoxin (Protein Data Bank entry 3B0W) was introduced manually into the structure of pig kidney Na,K-ATPase bound with ouabain (4HYT) so that the steroid and lactone moieties of ouabain and digoxin superimposed as closely as possible (see Ref. 17). The structure file with bound digoxin was then submitted to the YASARA Energy Minimization Server (26). The structural figure was prepared with PyMOL. Fig. 1 shows the synthetic route involving 1) selective periodate oxidation of the third digitoxose moiety (25) and 2) reductive amination of the dialdehyde using the free amine (R-NH 2 ) plus NaCNBH3 (23) for the case of glycinamide. Details of the synthesis and purification of compounds are given under "Materials and Methods." For verification of structures, masses of the pure compounds were then determined (supplemental Table S1), and 1 H and 13 C NMR spectra were obtained (full assignment of several derivatives will be published elsewhere). Supplemental Table S1 shows the structure and names of the different amine substituents and the theoretical and experimentally found masses of 15 digoxin derivatives and the glycine derivative of bis-digitoxose digoxigenin. Fig. 2 shows curves for inhibition of Na,K-ATPase activity of purified human isoforms (␣1␤1, ␣2␤1, and ␣3␤1) of DGlyN and DMe, the glycinamide and methylamine derivatives (see Table S1) with improved selectivity for ␣2 compared with digoxin (see Fig. 5 of Ref. 10) for curves of inhibition by digoxin). Table 1 provides information on the inhibitory effects of 14 digoxin perhydro-1,4-oxazepine derivatives (with identifying numbers) produced for this study. The isoform selectivity ratios (K i ␣1/␣2) of several derivatives (DGlyN (7.45 Ϯ 0.46), DMe (6.47 Ϯ 0.71), DGly (5.1 Ϯ 0.54), DPrN, (5.28 Ϯ 0.75), and DScar (4.98 Ϯ 1.2)) are significantly greater than that of digoxin (3.44 Ϯ 0.34). In most cases, K i values for both ␣1 and ␣2 were lower than for digoxin, but the effect was greater for ␣2 compared with ␣1, so the calculated ratio K i ␣1/␣2 was higher compared with digoxin. In all cases, the K i for ␣3 was closer to that for ␣1 than to ␣2 (as seen also in Ref. 10). 3 For clarity, the detailed data are not shown, but this feature is seen clearly for DMe and DGlyN in Fig. 2. Thus, it is primarily the K i ␣1/␣2 that is affected by the modification of the third digitoxose. The K i values of several other derivatives in Table 1 (e.g. compounds 10 and 12) were also significantly lower than for digoxin itself, but a differential effect between the isoforms was not observed. In other cases (e.g. compounds 1, 3, and 6) the K i values were higher than for digoxin and, again, the selectivity for ␣2 was not improved. The result in Table 1 that the glycine derivative of bis-digitoxose digoxigenin (compound 14) showed lower selectivity for ␣2 over ␣1 compared with the glycine derivative of the tridigitoxose (compound 2) indicates that modification of the third digitoxose residue is optimal for this effect, consistent with a similar conclusion reported previously (10). In summary, the strategy of modifying the third digitoxose moiety produced compounds with enhanced selectivity measured as K i ␣1/␣2. Reduction of Intraocular Pressure by Topically Applied Digoxin and Perhydro-1,4-oxazepine Derivatives-Intraocular pressure in rabbits was measured accurately after anesthetizing the cornea with local anesthetic. The basal IOP values for different rabbits were in the range of 15-21 mm Hg. Nevertheless,  in initial experiments, we observed that topically applied digoxin and DMe (1 mM, 1 drop in each eye) produced only modest and variable effects on basal IOP. Possible explanations are that the drugs are not sufficiently permeable through the cornea or, alternatively, that they are sufficiently permeable and reach the ciliary epithelium, but partial inhibition of Na,K-ATPase does not affect aqueous humor production or IOP because active sodium pumping is not rate-limiting in the basal condition (see Ref. 27 and "Discussion"). It seemed that a simple well controlled procedure for acute elevation of IOP was required in order to examine whether topically applied cardiac glycosides are able to counter such an effect. Two different pharmacological agents have been used for this purpose. First, topical 4AP acutely and transiently raises the IOP in rabbits' eyes (28). The mechanism involves blocking of a voltage-dependent potassium channel by the 4AP in sympathetic nerves of the iris-ciliary body and release of norepinephrine, leading to an increased aqueous humor inflow. Subsequently, we have used topical IB-MECA, which is a selective agonist of the A3-adenosine receptor and raises aqueous humor inflow and IOP by activating chloride channels of the NPE cells (29,30).

Synthesis and Testing of Perhydro-1,4-oxazepine Derivatives of Digoxin-
A standard experimental design involved topical application of the cardiac glycosides (1 drop in each eye) 30 min prior to application of 4AP and measurement of IOP every 30 min over 5 h. Fig. 3 shows the effects of 4AP alone and effects of digoxin, DGlyN, DMe, and digoxigenin on IOP using this protocol. As expected, one drop of 4AP in each eye (control) raised the IOP by 20 -30% (3-6 mm Hg), and the effect was dissipated after 5 h. Digoxin (Fig. 3A) at a high concentration (1 mM) prevented the 4AP-induced rise in IOP, whereas 0.25 mM digoxin was poorly effective. By comparison, both DGlyN and DMe (Fig. 3,  B and C), the most ␣2-selective of the new perhydro-1,4-oxaz-epine derivatives, were effective at much lower concentrations (0.05-0.1 mM) than digoxin. Similarly, the aglycone of digoxin, digoxigenin, effectively reduced IOP at lower concentrations than digoxin (Fig. 3D). Ouabain, a widely used water-soluble cardiac glycoside, somewhat reduced IOP only at 1 mM, whereas lower concentrations were poorly effective (not shown). Finally, Fig. 4 compares the relative effects of DGlyN, DMe, digoxigenin, digoxin, and ouabain, on IOP all at 0.1 mM after 1.5 h and confirm the order as DGlyN Ϸ DMe Ϸ digoxigenin Ͼ digoxin Ͼ ouabain.
Over time, the cardiac glycosides that penetrate to the ciliary epithelium will be washed out of the eye into the general circulation, and the effect on IOP will dissipate. Although Figs. 3 and 4 demonstrate that the cardiac glycosides reduce IOP with greater or less efficacy, by this protocol, the duration of the effect cannot be evaluated due to the transient nature of the 4AP effect itself. Thus, we have done additional measurements in which the effect of a single drop of digoxin, digoxigenin, DGlyN, and DMe was compared when 4AP was then added every 2 h so as to maintain the IOP at the elevated level for 7-8 h without added cardiac glycoside (Fig. 5). By this protocol, the reduction of IOP was indeed seen to be transient (Fig. 5). Fig. 5A shows an experiment with DMe that demonstrates a clear dependence of the washout time on concentration. IOP was held at the low level for 5.8, 4.5, and 2.5 h for 1, 0.5, and 0.2 mM, respectively, before the IOP rose back to the elevated level with 4AP. Striking differences in washout times were detected between DMe, DGlyN, digoxigenin, and digoxin applied at equal concentrations (1 mM). As seen in Fig. 5B, DMe maintained IOP at the low level for about 5.5 h, by comparison with 3.5 h for DGlyN, about 2 h for digoxigenin, and only 1 h for digoxin. Ouabain was washed out at a rate between that of digoxin and digoxigenin. In short, the relatively ␣2-selective derivatives DMe and DGlyN produce the longest acting effect, compared with either digoxin, a less ␣2-selective CG, or digoxigenin, a non-selective CG. Finally, we have used this protocol to show that DGlyN and DMe rapidly normalize pre-established ocular hypertension (see Fig. 5C). DGlyN or DMe (1 mM) were applied 1 h after the first application of 4AP, which was added every 2 h. DGlyN and DMe reversed the initial rise in IOP within 30 min, indicating that the compounds permeated the cornea and bound to the pump in the NPE cells. The normalized IOP was maintained for about 4 h for both compounds.
Despite the evidence on the mechanism of action of 4AP in Ref. 28, one might raise the objection that digoxin derivatives do not inhibit aqueous humor inflow directly but act indirectly by, for example, interfering with the 4AP itself. In order to test this hypothesis, we have used IB-MECA, which induces acute ocular hypertension by a different and well defined mechanism (29,30). A single drop of IB-MECA (1 M) induced a significant but transient increase in IOP, whereas repeated application each 1.5 h maintained increased IOP over 4 -5 h (see Fig. 6, Control). Fig. 6A depicts the effects of DMe (0.1-1 mM) applied prior to the IB-MECA, and a similar result was obtained for DGlyN (not shown). Fig. 6B depicts the effects of digoxigenin, DGlyN and DMe (1 mM) applied after the IB-MECA. Clearly, the effects of digoxigenin, DGlyN, and DMe at 1 mM were almost the same as seen with 4AP in Fig. 5, obviously exclud- Amines used were as follows. DOH, hydroxylamine; DGly, glycine; DGlMe, glycine methyl ester; DGlyN, glycinamide; DAlaN, alaninamide; DSer, serine; DSerN, serinamide; DSCar, semicarbazide; DPrN, propionamide; DEtDA, ethylene diamine; DMe, methylamine; DEt, ethylamine; DMeCF3, trifluorethylamine; DbisGly, glycine; Dbis, digoxigenin bis-digitoxoside. p values were calculated by Student's t test and denoted as follows: *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. n, number of independent experiments. p (␣2v␣1), significance of differences between K i ␣2␤1 and K i ␣1␤1. p (v digoxin), significance of the difference of the selectivity ratio ing the notion that CGs interfere with the action of 4AP itself. Note again that the duration of the effect is significantly greater for DMe and DGlyN than for digoxigenin. Furthermore, when the concentration of DGlyN and digoxigenin was raised to 3 mM, the difference in duration of the effect was greatly amplified compared with the experiment at 1 mM (Fig. 6, compare B and C). Corneal thickness was also measured after application of digoxin (1 mM), DGlyN (0.5 mM), DMe (0.5 mM), and ouabain (1 mM) after 4AP. Over 4 h, the corneal thickness, measured in m, was not significantly affected (supplemental Table S2). Similar results were obtained with CGs applied after IB-MECA. Dissociation of Cardiac Glycosides from ␣2␤1-Because the principal isoform in NPE cells is ␣2 and dissociation from the pump is expected to affect the duration of the effects on IOP, we have compared the dissociation rates of different cardiac glycosides from the purified ␣2␤1 isoform. The experiment used a protocol similar to that described previously (16) (see "Materials and Methods") to compare dissociation rates of digoxin, digoxigenin, DGlyN, and DMe. Fig. 7 depicts representative experiments for each of the four compounds, and Table 2 shows the average rate constants and half-times from three or four experiments. Evidently, the aglycone digoxigenin dissociates much faster than digoxin or any other glycone, and DMe and DGlyN also dissociate significantly slower than digoxin itself. See "Discussion" for the significance of these findings in relation to the IOP measurements.

DISCUSSION
Digoxin Derivatives with Enhanced ␣2 Selectivity-The present findings confirm the prediction that selective chemical modification of the third digitoxose residue of digoxin can produce derivatives with enhanced selectivity for ␣2 over ␣1. Compared with digoxin (K i ␣1/␣2 ϭ 3.4), the selectivity ratio was significantly increased in the order DGlyN Ͼ DMe Ͼ DGly Ϸ DPrN Ϸ DScar, reaching a maximal value of K i ␣1/␣2 ϭ 7.45 for DGlyN (Table 1). Although we have now prepared 14 different perhydro-1,4-oxazepine digoxin derivatives, it is, of course, possible that additional substitutions will produce even more ␣2-selective compounds and lead to further mechanistic insights.
Although the structures of the ouabain-bound conformations of renal Na,K-ATPase (20 -22) are consistent, in general, with the observed lack of isoform selectivity of aglycones, these structures cannot explain in detail the moderate selectivity of digoxin for ␣2 or increased selectivity for ␣2 of perhydro-1,4oxazepine derivatives. The enhanced ␣2-selectivity of the perhydro-1,4-oxazepine derivatives, such as DGlyN and DMe, detected as lower K i values for ␣2 compared with ␣1, indicates a differential interaction of the third digitoxose with isoformspecific residues in the exterior loops of ␣2 and ␣1. The very large difference of dissociation rates between aglycones and glycones is quite similar to the findings in Ref. 16 using brain enzyme (␣1, with some ␣2 and ␣3) and emphasize the role of the sugars in binding to ␣2. Specific interactions with ␣2 of the modified digitoxose derivatives of DGlyN and DMe moieties are also indicated directly by the slower dissociation rates compared with digoxin ( Fig. 7 and Table 2). Obviously, new structures with bound digoxin or digitoxin are required in order to obtain more detailed information on binding of the sugar moiety. With structural information it may also be possible to design derivatives with improved selectivity for ␣2 compared with those found by the semiempirical methods used here (and also explain other properties attributed to the sugars (17)).  An interesting molecular insight into interactions of the third digitoxose residue and isoform selectivity is illustrated by Fig. 8A. The figure presents a molecular model in which the digoxin molecule (Protein Data Bank entry 3B0W) was introduced onto the high affinity ouabain-bound molecule (Protein Data Bank entry 4HYT) (22) so that the lactone and steroid portions of ouabain and digoxin overlap closely, and then a minimal energy structure was obtained (26) (see Ref. 17 for a similar approach). The three digitoxose residues point outward toward both ␣ and ␤ subunits. The detail in Fig. 8B shows the digitoxose moiety in proximity (Ͻ3.5 Å) to residues Asp-Asp-Arg-Trp 887 in L7/8 of ␣, the third digitoxose being close to both ␣Trp 887 and also ␤Gln 84 . Support for this orientation toward the ␤ subunit comes from an old observation that photoaffinity probes located in the third digitoxose of digitoxin label both ␣ and ␤ subunits, whereas photoaffinity probes located in other regions of cardiac glycoside molecules label only the ␣ subunit (31,32). In the present context, the interesting insight is that ␣Trp 887 is one of only four residues in extracellular loops that are different in ␣2 (and ␣3) from ␣1 (Gln 119 , Glu 307 , Val 881 , and Trp 887 in pig ␣1) and were inferred previously to be candidates for determining isoform selectivity (10). Obviously, close proximity to one of these four residues fits very well with the notion    JULY 25, 2014 • VOLUME 289 • NUMBER 30 that interactions of the third digitoxose are important for isoform selectivity and the present findings that derivatives of the third digitoxose can enhance isoform selectivity. In ␣2, Trp 887 is replaced by a threonine. Fig. 8 is also relevant to a recent proposal that ouabain can bind to the Na,K-ATPase with the rhamnose moiety pointing inward (33) (i.e. rotated 180°with respect to the crystal structure). As seen from this model, a similar inversion for bound digoxin with three digitoxose moieties pointing inward is hard to conceive. Our findings that perhydro-1,4-oxazepine derivatives of the third digitoxose raise binding affinities also seem to make the concept untenable at least for equilibrium-bound states.

Effects of Digoxin Derivatives on Intraocular Pressure
Reduction of Intraocular Pressure by Digoxin Derivatives-Overall efficacy of topically applied cardiac glycosides to reduce IOP is measured both by the lowest dose required to produce a maximal effect and the duration of the effect. The required dose reflects the concentration of drug in the aqueous humor, determined largely by drug permeability via the cornea, and the affinity for Na,K-ATPase in NPE cells. The duration reflects the rate of fluid wash-out via the trabecular network, the effective free drug concentration in the aqueous humor, and the rate of drug dissociation from the Na,K-ATPase. Although it is difficult to distinguish between these factors on the basis of IOP measurements alone, when combined with the prior data on enzyme inhibition and dissociation rates 4 and an assumption that lipidsoluble compounds permeate more easily than water-soluble compounds, likely explanations can be given. In particular, the findings that DGlyN and DMe are active at low concentrations with extended durations of the effects, compared with either digoxin or digoxigenin (Figs. 5B and 6, B and C), show that both DGlyN and DMe are effectively permeable via the cornea and are correlated with their higher affinity for ␣2 (Table 1) and slow dissociation rates ( Fig. 7 and Table 2). The most significant difference between DGlyN or DMe and digoxigenin was the duration of the effect (Figs. 5B and 6B), especially when the concentration was raised from 1 to 3 mM (Fig. 6C). A simple explanation is that at the higher topical concentration (3 mM), differences in concentration of the drugs within the eye, due to different permeabilities, are evened out, and the large difference in duration of the effect more accurately reflects the much slower dissociation rate of DGlyN from ␣2 ( Fig. 7 and Table 2). In other words, in this condition, the confounding "pharmacokinetic" effects are minimized, and a better estimate of the dif-ference between the ␣2-selective DGlyN and the non-selective digoxigenin is obtained.
As another example of this reasoning, compared with digoxin, the rather lipid-soluble aglycone, digoxigenin, is effective at lower concentrations (Fig. 3), and the duration of the effect is even longer (Fig. 5B). Because digoxigenin does not have a higher affinity for ␣2 compared with digoxin (Table 1) and also dissociates much more quickly (Fig. 7 and Table 2), in this case, the explanation for the differences must be that there is a higher concentration of digoxigenin than digoxin in the aqueous humor due to better permeation via the cornea.
In summary, when evaluated by the dose and especially duration of effects, the most ␣2-selective compounds, DMe and DGlyN, are significantly more effective than either the moderately ␣2-selective digoxin or non-selective digoxigenin. One important conclusion is that ␣2 indeed plays a major role in production of the aqueous humor, as could be predicted from its prominent expression in NPE cells (8).
The well defined mechanism of action of IB-MECA to activate an NPE cell chloride channel and aqueous humor inflow (29,30), as well as the similar effects of DMe and DGlyN when IOP is raised by either 4AP or IB-MECA, lead to several additional mechanistic insights. First, the effects of the CGs are, almost certainly, due to inhibition of the aqueous humor inflow stimulated by either 4AP or IB-MECA. Second, the modest effect of the CGs on the basal IOP fits well with much previous evidence that chloride flux across the NPE cells is rate-limiting for basal aqueous humor production (27,30). Stimulation of fluid inflow by IB-MECA itself provides one strong piece of evidence for this conclusion. In this situation, the active sodium and potassium fluxes are not themselves rate-limiting, and partial inhibition of the Na,K-pump by cardiac glycosides cannot greatly affect fluid transport. By contrast, when the fluid inflow is stimulated by IB-MECA or 4AP, the active sodium and potassium fluxes become rate-limiting for net salt flux and fluid transport, so inhibition by DMe and DGlyN can normalize IOP. Third, it has been reported that incubation of anterior segments of the eye with ouabain for extended periods increases outflow capacity of aqueous humor (34), and the mechanism may involve restructuring of trabecular meshwork cells to decrease cell volume and resistance to outflow (35). Although, in theory, such a mechanism might explain reduction of IOP by DMe and DGlyN, if this was the case, one could expect basal IOP to be reduced. In addition, the superior efficacy and rapid effects of the ␣2-selective digoxin derivatives fit much better with inhibition of ␣2 in NPE cells.
Potential Pharmacological Applications?-Could relatively ␣2-selective cardiac glycosides, such as DMe and DGlyN, or even more optimized derivatives lead to development of novel drugs for control of IOP and glaucoma? Obviously, the demonstrated efficacy of topical DMe and DGlyN in reducing IOP and the extended duration of their effects compared with the nonselective digoxigenin are interesting and could also be significant. In addition, local toxicity of ␣2-selective cardiac glycosides, namely swelling of the cornea and lens, should be minimal because corneal endothelium expresses ␣1 and a minor amount of ␣3 but no ␣2 (36), and lens epithelium expresses only ␣1 (37). Also, systemic cardiotoxic effects should be minimal. Nevertheless, and despite the fact that the acute ocular hypertension models provide a convenient tool to assess effects of compounds on IOP, an obvious limitation is that they do not represent realistic models of glaucoma. Evaluation of the drug potential of ␣2-selective cardiac glycosides will require the use of more realistic animal models of chronic ocular hypertension, similar to glaucoma itself (see, for example, Ref. 38 or 39), in order to assess their efficacy in reducing IOP as well as local and systemic toxicity over extended periods.