Selectivity of Digitalis Glycosides for Isoforms of Human Na,K-ATPase*

There are four isoforms of the α subunit (α1–4) and three isoforms of the β subunit (β1–3) of Na,K-ATPase, with distinct tissue-specific distribution and physiological functions. α2 is thought to play a key role in cardiac and smooth muscle contraction and be an important target of cardiac glycosides. An α2-selective cardiac glycoside could provide important insights into physiological and pharmacological properties of α2. The isoform selectivity of a large number of cardiac glycosides has been assessed utilizing α1β1, α2β1, and α3β1 isoforms of human Na,K-ATPase expressed in Pichia pastoris and the purified detergent-soluble isoform proteins. Binding affinities of the digitalis glycosides, digoxin, β-methyl digoxin, and digitoxin show moderate but highly significant selectivity (up to 4-fold) for α2/α3 over α1 (KD α1 > α2 = α3). By contrast, ouabain shows moderate selectivity (≈2.5-fold) for α1 over α2 (KD α1 ≤ α3 < α2). Binding affinities for the three isoforms of digoxigenin, digitoxigenin, and all other aglycones tested are indistinguishable (KD α1 = α3 = α2), showing that the sugar determines isoform selectivity. Selectivity patterns for inhibition of Na,K-ATPase activity of the purified isoform proteins are consistent with binding selectivities, modified somewhat by different affinities of K+ ions for antagonizing cardiac glycoside binding on the three isoforms. The mechanistic insight on the role of the sugars is strongly supported by a recent structure of Na,K-ATPase with bound ouabain, which implies that aglycones of cardiac glycosides cannot discriminate between isoforms. In conclusion, several digitalis glycosides, but not ouabain, are moderately α2-selective. This supports a major role of α2 in cardiac contraction and cardiotonic effects of digitalis glycosides.

There are four isoforms of the ␣ subunit (␣1-4) and three isoforms of the ␤ subunit (␤1-3) of Na,K-ATPase, with distinct tissue-specific distribution and physiological functions. ␣2 is thought to play a key role in cardiac and smooth muscle contraction and be an important target of cardiac glycosides. An ␣2-selective cardiac glycoside could provide important insights into physiological and pharmacological properties of ␣2. The isoform selectivity of a large number of cardiac glycosides has been assessed utilizing ␣1␤1, ␣2␤1, and ␣3␤1 isoforms of human Na,K-ATPase expressed in Pichia pastoris and the purified detergent-soluble isoform proteins. Binding affinities of the digitalis glycosides, digoxin, ␤-methyl digoxin, and digitoxin show moderate but highly significant selectivity (up to 4-fold) for ␣2/␣3 over ␣1 (K D ␣1 > ␣2 ‫؍‬ ␣3). By contrast, ouabain shows moderate selectivity (≈2.5-fold) for ␣1 over ␣2 (K D ␣1 < ␣3 < ␣2). Binding affinities for the three isoforms of digoxigenin, digitoxigenin, and all other aglycones tested are indistinguishable (K D ␣1 ‫؍‬ ␣3 ‫؍‬ ␣2), showing that the sugar determines isoform selectivity. Selectivity patterns for inhibition of Na,K-ATPase activity of the purified isoform proteins are consistent with binding selectivities, modified somewhat by different affinities of K ؉ ions for antagonizing cardiac glycoside binding on the three isoforms. The mechanistic insight on the role of the sugars is strongly supported by a recent structure of Na,K-ATPase with bound ouabain, which implies that aglycones of cardiac glycosides cannot discriminate between isoforms. In conclusion, several digitalis glycosides, but not ouabain, are moderately ␣2-selective. This supports a major role of ␣2 in cardiac contraction and cardiotonic effects of digitalis glycosides.
For more than 200 years congestive heart failure has been treated with the plant-derived digitalis cardiac glycosides such as digoxin, which increase force of contraction of failing cardiac muscle and reduce cardiac conduction rate. However, digoxin is now less used than in the past due to the narrow therapeutic window and drug toxicity. Cardiac glycosides are produced in mammals as well as plants. As reviewed recently (1) five different cardiac glycosides have been identified in mammalian tissues including the cardenolides ouabain and digoxin and the bufadienolides marinobufagenin, telocinobufagin, and 19-nor-bufalin. Cardiac glycosides inhibit Na,K-ATPase and control cellular Na ؉ and K ؉ gradients and also activate cellular signaling pathways that mediate control of gene expression and tissue growth (2,3). Thus, a multiplicity of physiological functions could be affected by the multiplicity of endogenous cardiac glycosides, but this is not well understood (1,4,5).
Increased force of contraction of cardiac muscle induced by cardiac glycosides is the result of inhibition of Na,K-ATPase. Raised intracellular Na ؉ concentration limits Ca 2ϩ extrusion via the 3Na ϩ /Ca 2ϩ exchanger, leading to enhanced Ca 2ϩ uptake into the SR by the Ca-ATPase and increased calciuminduced calcium release during excitation-contraction coupling. Cardiac glycoside-induced toxicity is associated with excessive inhibition of Na,K-ATPase, accumulation of Ca 2ϩ ions (i.e."calcium overload"), and "spontaneous" SR Ca 2ϩ release that can trigger delayed after-depolarizations and cardiac arrhythmias (11). Experiments with mice engineered to have ouabain-sensitive ␣1 and insensitive ␣2 isoforms (␣1 S/ S␣2 R/R ) or ouabain-insensitive ␣1 and ␣2 (␣1 R/R ␣2 R/R ) instead of the wild type with ouabain-insensitive ␣1 and ouabain-sensitive ␣2 (␣1 R/R ␣2 S/S ) have shown that ␣2 can play a predominant role in cardiac glycoside-induced positive inotropy (12) and also that ␣2 carries a large fraction of the Na,K-pump current in T-tubules of cardiac myocytes (13). Very recent work shows that ␣2 preferentially modulates Ca 2ϩ transients and SR Ca 2ϩ release in cardiac myocytes compared with ␣1 (14). Similarly, ␣2 plays an important role in contractility of vascular smooth muscle and control of blood pressure (15,16). Experiments with ␣1 S/S ␣2 R/R , ␣1 R/R ␣2 S/S , and ␣1 R/R ␣2 R/R mice also provide strong evidence for endogenous mammalian cardiac glycosides by showing differential responses to physiological stimuli and states that alter blood pressure (12,17). In view of the accumulated information on ␣1 and ␣2 isoforms, one could envisage that ␣2-selective cardiac glycosides could provide important insights into the physiological role of ␣2 and, in particular, complement the information obtained from genetically engineered mice. At the pharmacological level an ␣2-selective cardiac glycoside could induce cardiac contraction with minimal Ca 2ϩ overload and so act as a safer cardiotonic agent than conventional cardiac glycosides.
Structure-activity analyses of inhibition of Na,K-ATPase by cardiac glycosides or displacement of bound [ 3 H]ouabain have been conducted for many years; see for example Refs. 18 -22. As summarized in Ref. 23, essential structural features of active cardiac glycosides include rings A/B and C/D that are cis-fused and B/C rings that are trans-fused, a hydroxyl group at C14, and an unsaturated lactone attached at C17 of the steroid. The sugar at C3 of the steroid is not essential for inhibition but strongly affects binding affinity and rates; see Table 1B and Ref 23. Despite the great potential interest in isoform selectivity of cardiac glycosides, there is little clear-cut information. Variations in potency of digitoxigenin monoglycosides as inhibitors of kidney (␣1␤1) and brain (␣1, ␣2, ␣3) Na,K-ATPases were attributed to isoform selectivity, and a role of the sugar was proposed (24). However, most native tissues except kidney (␣1␤1) contain mixtures of isoforms, and it is difficult to differentiate interactions of cardiac glycosides with the individual isoforms and exclude possible complicating factors. Because, for example, human cardiac membranes express ␣1, ␣2, and ␣3 isoforms (8), healthy cardiac tissue would not be useful for this purpose even if it was readily available. Individual human isoforms have been expressed in Xenopus oocytes (␣1-3, ␤1-3) (25) and Saccharomyces cerevisiae (␣1-3, ␤1) (26,27) at low levels and used to characterize [ 3 H]ouabain binding. More recently we have expressed human isoforms (␣1, ␣2, and now also ␣3, with ␤1) at high levels in Pichia pastoris and purified the ␣1␤1, ␣2␤1, and ␣3␤1 protein complexes (28 -31). This system permits accurate analysis of cardiac glycoside binding and inhibition of Na,K-ATPase activity of the individual human isoforms, leading to the finding that several digitalis glycosides show moderate but highly significant selectivity for ␣2 (and ␣3) compared with ␣1.

EXPERIMENTAL PROCEDURES
Materials-Escherichia coli XL-1 blue strain was used for propagation and preparation of plasmid constructs. Yeast lytic enzyme from ICN Biomedicals Inc (catalog no. 152270) was used for transformation of P. pastoris protease-deficient strain SMD1165 (his4, prb1). n-Dodecyl-␤-D-maltopyranoside (catalog no. D310) and C 12 E 8 (25% w/w, catalog no. O330) were purchased from Anatrace. Synthetic SOPS (sodium salt)) was obtained from Avanti Polar Lipids and stored as a chloroform solution. BD Talon metal affinity resin (catalog no. 635503) was obtained from Clontech. Cholesterol and ouabain (O3125) were obtained from Sigma. [ 3 H]Ouabain and [ 3 H]digoxin were obtained from PerkinElmer Life Sciences. All other materials were of analytical grade.
Solubilization and Purification of Isoforms of Na,K-ATPase-Procedures for solubilization of membranes in n-dodecyl-␤-Dmaltopyranoside, binding to BD-Talon beads, and washing and elution of purified Na,K-ATPase have been described in detail (28 -30). Purification of ␣3␤1, which has not been described previously, follows exactly the same procedure. The three purified isoform complexes (0.3-0.5 mg/ml) were eluted from the BD-Talon beads in a solution containing 170 mM imidazole, 100 mM NaCl, 20 mM Tricine⅐HCl, pH 7.4, 0.1 mg/ml C 12 E 8 , 0.05 mg/ml SOPS, 0.01 mg/ml cholesterol, 25% glycerol. The proteins were stored at Ϫ80°C. Protein concentration was determined with BCA (B9643 Sigma).
Cardiac Glycoside Binding to P. pastoris Membranes-[ 3 H]Ouabain binding to yeast membranes (200 -300 g of protein) was assayed at 37°C for 1 h in medium containing 10 mM MOPS-Tris, pH 7.2, 3 mM MgCl 2 , 1 mM vanadate-Tris, 1 mM EGTA-Tris (32). Binding of ouabain or competitive displacement by other cardiac glycosides was assessed by varying the total concentrations of ouabain or other cardiac glycosides at constant [ 3 H]ouabain (between 1 and 2 nM; specific activity 30 -40 Ci/mmol). Full curves of bound versus free ouabain were generated, and K D values were estimated using a one-site model where [Ou f ] refers to the concentration of free ouabain, and B and Bmax have their usual meaning. Fitted dissociation constants, K D , were obtained using Kaleidagraph. For other inhibitors K 0.5 was calculated using a one-site inhibition model, B/B CGϭ0 ϭ K 0.5 /([CG] ϩ K 0.5 ). B refers to the [ 3 H]ouabain bound at a particular concentration of the cardiac glycoside (CG), and B CGϭ0 refers to the [ 3 H]ouabain bound at 1-2 nM [ 3 H]ouabain in the absence of other cardiac glycosides. The K D CG was calculated from K 0.5 by taking into account ouabain-CG competition as K D CG ϭ K 0.5 / (1ϩ[Ou f ]/ K D Ou ). With K D Ou values ␣1␤1 9.2 nM, ␣2␤1 21.5 nM, and ␣3␤1 11 nM, respectively (see Table 1), and 1 nM total ouabain, the values of 1 ϩ [Ou f ]/K D Ou were 1.06, 1.03, and 1.06, respectively. Binding of each cardiac glycoside was estimated in 3-5 separate experiments. Average K D values Ϯ S.E. for each isoform were calculated, and statistical significance was calculated by the unpaired Student's t test. p values Ͻ0.05 were considered significant. The selectivity ratios K D ␣1/K D ␣2 and K D ␣1/K D ␣3 were calculated as the quotient of the individual K D values. The error of the ratios of K D ␣1/ K D ␣2 and K D ␣1/ K D ␣3 are calculated from the S.E. K D values using the formula Digoxin binding was measured in the same conditions as ouabain ([ 3 H]digoxin ϳ7 nM stock, specific activity 40 Ci/mmol). Assay of Na,K-ATPase Activity of Purified Isoform Complexes-Inhibition of Na,K-ATPase activity of the detergent-soluble ␣1␤1, ␣2␤1, and ␣3␤1 complexes by cardiac glycosides was done as described (29,30). Before assay, the proteins were incubated at 25°C for 30 min. The inhibitors were added to the recombinant enzyme (Ϸ0.08 -0.2 g of protein) in 400 l of reaction medium containing 130 mM NaCl, 5 mM KCl, 3 mM MgCl 2 , 25 mM histidine, pH 7.4, 1 mM EGTA, 0.01 mg/ml SOPS, 0.001 mg/ml cholesterol, and 0.005 mg/ml C 12 E 8 in 48-well plates. The reaction (37°C for 1 h) was started by the addition of 1 mM ATP. P i release was measured with a malachite green dye to detect the phosphomolybdate (P i Color Lock, Innova Biosciences). The percent inhibition V CG /V 0 was calculated for each cardiac glycoside concentration, and K i values were obtained by fitting the data to the function V CG /V 0 ϭ K i /([CG] ϩ K i )ϩ c (using Kaleidagraph). V 0 and V CG represent the control rate and rate of Na,K-ATPase activity at particular concentrations of cardiac glycosides, [CG], respectively. The constant c represents a small fraction of uninhibited activity (range 0 -10%). Including c improves the fit. Inhibition was estimated in 3-5 separate experiments, average K i values Ϯ S.E. were calculated, and the significance of differences was calculated by the unpaired Student's t test. p values Ͻ0.05 were considered significant.
Inhibitors-Most cardiac glycosides were dissolved in 70-100% ethanol. Bufalin, marinobufagenin, and gitoxin were dissolved in DMSO. Stock solutions were diluted in 50% ethanol or 50% DMSO. Inhibitors were added directly to the reaction medium containing membranes ([ 3 H]ouabain binding) or pure enzymes (Na,K-ATPase activity), so that the final ethanol or DMSO concentration was 1%. Didigitoxosyl digoxigenin was prepared by selective degradation of digoxin as described (33). The product was analyzed by high performance liquid chromatography and shown to be close to 80% pure.
Cardiac Glycoside Binding to P. pastoris Membranes Expressing ␣1␤1, ␣2␤1, and ␣3␤1-The high level isoform expression permits accurate determination of [ 3 H]ouabain binding and competitive displacement of [ 3 H]ouabain by many cardiac glycosides (Fig. 2, Table 1). 3 The ligand conditions (3 mM with MgCl 2 and 1 mM vanadate-Tris) are optimal for high affinity ouabain binding to an E 2 -vanadate conformation (32), vanadate serving as a transition-state analogue of inorganic phosphate. Full binding curves of [ 3 H]ouabain yielded dissociation constants K D (see Table 1, line 1) with values in the tens of nM range and order ␣1 Ͻ ␣3 Ͻ ␣2. These values are consistent with previous observations on [ 3 H]ouabain binding to human isoforms expressed in S. cerevisiae (26,27) or Xenopus oocytes (25). Fig. 2 depicts typical displacement curves for three digitalis glycosides, digoxin, digitoxin, and a semisynthetic derivative, ␤-methyl digoxin as well as three aglycones, digoxigenin, digitoxigenin, and bufalin, a bufadienolide (see Table 1 for structures). For digoxin, digitoxin, and ␤-methyl digoxin, the curves for ␣2 and ␣3 are shifted to the left of the curves for ␣1, with little or no difference between ␣2 and ␣3. By contrast, for all three aglycones the curves are superimposable. A single-site 3 The accuracy of these measurements crucially depends on two factors. The first is a high signal to background ratio.   Table 1. Digoxin, ␤-methyl digoxin, and digitoxin indeed show a moderate but highly significant selectivity for both ␣2 and ␣3 over ␣1, up to 4-fold (see also  Table 1 were carried out on four different preparations of membranes expressing the different isoforms, with essentially similar results. The observation that the aglycones do not discriminate between the isoforms implies that ␣2/␣3:␣1 selectivity depends on the sugar residues. To test this hypothesis more systematically as well as the effect of the number and nature of the sugar substituents and the possible influence of the steroid and lactone moieties, we have performed similar experiments on a large number of glycosides and aglycones. These results are collected together in Table 1 into five main groups, B, C, D, E, and F. Group B consists of the digoxigenin and digitoxigenin glycosides with one, two, three, or four digitoxose substituents (see also Fig. 3). The addition of a single digitoxose to digoxigenin strongly increases the affinity for all isoforms but even more for ␣2 and ␣3 compared with ␣1, leading to a significant ␣2/␣3:␣1 selectivity (ϳ2fold). The addition of the two and then three sugars to digoxigenin increases ␣2/␣3:␣1 selectivity further, to more than 3-fold in the case of digoxin, whereas the addition of a fourth sugar slightly reduces the affinity for all isoforms but does not affect the ␣2/␣3:␣1 selectivity. Fig. 3 illustrates these features and confirms that ␣2 and ␣3 show the same profile compared with ␣1. In the case of digitoxigenin glycosides the addition of either two or three digitoxoses to digitoxigenin has a similar effect on affinity and ␣2/a3:␣1-selectivity. Group C consists of nine synthetic monoglycosyl derivatives of digitoxigenin (20). Several of these derivatives show a higher affinity for all three isoforms compared with the parent digitoxigenin, but only the ␤-L-rhamnosyl derivative, evomonoside, seems to show significant selectivity, in this case for ␣3:␣1. Group D consists of natural or semisynthetic 16-OH steroid and formylated or acetylated derivatives of gitoxin (19) and oleandrin. The aglycone gitoxigenin has a low affinity, whereas the tridigitoxosyl gitoxigenin (gitoxin) binds much better to all isoforms, but in neither case is there significant isoform selectivity. The 16-formylated gitoxigenin has a much higher affinity compared with the 16-OH gitoxigenin, but no isoform selectivity. By contrast, the 3␤-digitoxosyl derivatives of 16-formyl and 16-acetyl gitoxigenin both show significant ␣2/␣3:␣1 selectivity. Similarly, oleandrin, with a 16-acetyl modification and a single oleandrose sugar, shows very high affinity and significant ␣2/␣3:␣1-selectivity. Group E consists of uzarin derivatives, which are rare cardenolides with planar steroid moieties. The aglycone shows low affinity and no selectivity, whereas the uzarigenin diglucoside shows even lower affinity and significant selectivity, at least for ␣2:␣1. The rhamnoside shows a much a higher affinity for all three isoforms but no selectivity. Group F consists of two bufadienolide

Cardiac Glycosides and Isoforms of Na/K-ATPase
aglycones, marinobufagenin and bufalin. Bufalin was discussed above. Marinobufagenin, which is reported to have ␣1 selectivity in some systems (for review, see Ref. 34), shows low affinity in this assay and, like bufalin, shows no isoform selectivity. These data confirm the importance of the sugars in ␣2/␣3:␣1 selectivity and provide several other conclusions on structural requirements of the sugar and steroid moieties ( Fig. 3 and "Discussion").
We have also looked directly at [ 3 H]digoxin binding to the three isoforms. Estimates of the K D values based on full binding curves up to M concentrations are unsatisfactory due to a high background count (about 10-fold higher for [ 3 H]digoxin com-  (Fig. 4). Fig. 4A depicts the well known [ 3 H]digoxin-K ϩ antagonism at increasing concentrations of K ϩ . The striking point is that K 0.5 K ϩ for displacing [ 3 H]digoxin is 5-6-fold higher for ␣2 compared with ␣1, whereas that for ␣3 is closer to that for ␣1 (see Fig. 4A). Fig. 4B shows the rate of dissociation of [ 3 H]digoxin. The rate is about 5-fold faster for ␣2 compared with ␣1, and that for ␣3 is closer to ␣2. The order of the isoform-dependent differences in K 0.5 K ϩ ions and dissociation rates is similar to properties of [ 3 H]ouabain binding to human isoforms expressed in Xenopus oocytes (25). Thus, these features are intrinsic properties of the isoforms, and all other cardiac glycosides should show the same features.
Inhibition of Na,K-ATPase Activity of Purified ␣1␤1, ␣2␤1, and ␣3␤1 Complexes-As an independent test of isoform selectivity, we have compared inhibition by the different cardiac glycosides of Na,K-ATPase activities of the three purified detergent-soluble isoform complexes. 4 Representative inhibition experiments are shown in Fig. 5 for digoxin, digoxigenin, and ␤-methyl digoxin, and fitted K i values for a selection of glycosides and aglycones are presented in Table 2. An obvious feature in Fig. 5 is that the curves for inhibition of ␣2 by digoxin and ␤-methyl digoxin lies to the left of that for ␣1, and those for ␣3 lie between those for ␣1 and ␣2. The curves for inhibition by digoxigenin are similar for all three isoforms. The fitted K i values with statistical tests for significance in Table 2 confirm these observations and also illustrate an important additional property. The ratio of K i for ␣1/␣2 is 3-5-fold in the case of digoxin, digitoxin, ␤-methyl digoxin, oleandrin, and 16-formate 3-␤-digitoxose gitoxigenin. By contrast, the ratio K i ␣1/␣3 for all these glycosides is lower than K i ␣1/␣2, and in most cases it is nearer to 1. Again, ouabain is different in that it is the only glycoside for which K i ␣1 Ϸ ␣2 Ϸ ␣3. For all the aglycones, ouabagenin, digoxigenin, and digitoxigenin, there are no or only minor differences between the isoforms. By comparison with K D values for cardiac glycoside binding in Table 1, the K i values in Table 2 are all higher, which is expected on account of the presence of K ϩ and the K ϩ -cardiac glycoside antagonism. Because, however, K 0.5 K ϩ values for antagonism of digoxin binding lie on the order ␣2 Ͼ ␣3 Ͼ ␣1 (Fig. 4A), CG binding could be antagonized by the K ϩ to different extents in the order ␣1 Ͼ ␣3 Ͼ ␣2. Differential K ϩ -cardiac glycoside antagonism on ␣1, ␣3, and ␣2, respectively, could explain the observed order of inhibitory potency of ␤-methyl digoxin or digoxin (K i ␣2 Ͻ ␣3 Ͻ ␣1) and also the finding that for ouabain K i ␣1 Ϸ ␣2 Ϸ ␣3 ( Table 2). To test this notion we have looked at the inhibition by ␤-methyl digoxin at 20 mM K ϩ rather than the usual 5 mM K ϩ . Under this condition, the K i for ␣1 and ␣3 was substantially raised, whereas the K i for ␣2 was raised only slightly (see Fig. 5 and Table 2). As a result the apparent selectivity of ␣2:␣1 increased further to 7.3-fold, compared with 4.5-fold at 5 mM K ϩ , and the order is clear-cut, K i ␣2 Ͻ ␣3 Յ ␣1. Experiments like those in Fig. 5 were carried on three different preparations of the purified isoform complexes and also complexes of ␣1␤1 purified from membranes prepared from yeast grown at 20°C as well as 25°C, with essentially similar results. Furthermore, reconstitution of ␣1␤1 and ␣2␤1 with FXYD1 (see Ref. 30) does not significantly affect the K i values for inhibition by digoxin or the ␣1/␣2 selectivity ratio (unpublished result, to be presented in detail in a separate paper).

Isoform Selectivity and Structures of Cardiac Glycosides-
The central finding in this work is that digoxin, digitoxin, ␤-methyl digoxin, and some other glycosides show a partial but highly significant selectivity for binding to ␣2 and ␣3 compared with ␣1. By contrast, ␣2/␣3:␣1 selectivity was not detected for any aglycone tested, indicating the essential role of the sugar. Ouabain behaves differently to the digitalis glycosides, showing some selectivity for ␣1:␣2 and little difference between ␣1 and ␣3, confirming the findings in Refs. [25][26][27]. The pattern of inhibition of Na,K-ATPase activity by the digitalis glycosides is consistent with the binding data but is modified by a large difference in apparent affinity of K ϩ for antagonizing cardiac glycoside binding. As a result, the order of potency of the digitalis glycosides as inhibitors of Na,K-ATPase is K i ␣2 Ͻ ␣3 Յ ␣1. Again, ouabain stands out in that K i ␣2 Ϸ ␣3 Ϸ ␣1. It is of course significant that the selectivities observed for cardiac glycoside binding and inhibition of Na,K-ATPase activities are mutually consistent. Furthermore, similar results were obtained on several independent membrane preparations and purified ␣1␤1, ␣2␤1, and ␣3␤1 complexes and for the purified ␣1␤1 prepared from cells grown at 20°C as well as the usual 25°C and also for the ␣1␤1FXYD1 and ␣2␤1FXYD1 complexes. These findings all suggest that the moderate ␣2 selectivity of the digitalis glycosides is a robust result and cannot be attributed to the type and conditions of the assays, the state of the protein (intact membranes versus 4 An important methodological consideration is that the reaction time must be long enough to ensure full equilibration of all cardiac glycosides with all three isoforms throughout the assay. It is known, for example, that binding of ouabain to ␣1 is much slower than to ␣2 and ␣3 (25). In practice, for assay times Ն40 min at 37°C, the K i values for inhibition of all three isoforms were constant. For reaction times that are too short (e.g. 5-10 min), K i values for ␣1, but not for ␣2, were significantly raised, strongly distorting the selectivity patterns. detergent-soluble proteins), type of the purified complex, or mere chance. As mentioned in the Introduction, extensive structure-activity relationships for binding of many cardiac glycosides and inhibition of Na,K-ATPase activities, utilizing native enzyme preparations, have been published previously (see for example Refs. 18 -22 and 35). These have established the basic structural features necessary for cardiac glycoside action. Overly detailed comparisons between the different studies or with the current study using recombinant human proteins are not worthwhile because of wide differences in experimental conditions, species differences, and enzyme sources with mixed isoform content. Results for the recombinant human ␣1␤1 and native sheep and pig kidney Na,K-ATPase (␣1␤1) (19,20,22), the only native source of the enzyme with the defined isoform content, do appear in general to be consistent despite the species difference.
An attempt to measure competition of various cardiac glycosides with [ 3 H]ouabain for binding to human ␣1, ␣2, and ␣3 expressed in S. cerevisiae was reported recently (36). With the exception of [ 3 H]ouabain binding data, the results for other cardiac glycosides were quite different from the present findings. In particular, in comparable conditions of high affinity binding, no difference was detected for digoxin or digitoxin binding to ␣1, ␣2, and ␣3, and ␤-methyl digoxin was stated to be partially selective for ␣1 over ␣2 and ␣3. We believe that the most likely explanation of the inconsistency is that very low levels of expression of the isoforms in S. cerevisiae (␣1, 4.95; ␣2, 1.32; ␣3, 4.51 pmol/mg) compared with P. pastoris (␣1, 25-30; ␣2, 20 -25; ␣3, 20 pmol/mg) compromised the accuracy of the measurements and the ability to detect moderate isoform-dependent differences. 5 The data in Table 1 provide several additional conclusions on the role of the number and nature of the sugars. First, compared with digoxigenin and digitoxigenin, the addition of only one digitoxose residue raises affinity for all isoforms and produces some ␣2/␣3:␣1 selectivity, whereas two or more digitoxoses added to digoxigenin or digitoxigenin induce the maximal affinity and ␣2/␣3:␣1 selectivity ratio (see also Fig. 3). Second, only cardiac glycosides with sugars in ␤-glycosidic links display ␣2/␣3:␣1 selectivity. Third, only sugars having 5-methyl groups (digitoxose, oleandrose and to some extent L-rhamnose) display ␣2/␣3:␣1 selectivity. Fourth, the nature of the steroid is crucial as well as the sugar (as seen for groups D (gitoxin) and E (uzarin) derivatives in Table 1). The latter observations make it less surprising that ouabain behaves differently to digoxin and digitoxin. Presumably the hydrophilic steroid and ␣-L-rhamnoside of ouabain interact better with ␣1 than ␣2 compared with the hydrophobic steroid and ␤-digitoxose in digoxin and digitoxin, which bind better to ␣2 and ␣3 than to ␣1.
Compatibility with the Structure of Na,K-ATPase with Bound Ouabain-The structures of shark rectal gland Na,K-ATPase (37), especially a structure with bound ouabain (38), provide an interesting insight into the proposed role of the sugars in iso-form selectivity. Fig. 6 shows a detail of the structure of Na,K-ATPase with two occluded K ϩ ions and ouabain bound in a deep cavity facing the external surface. Because the digitalis glycoside binding affinities are the same and higher for both human ␣2 and ␣3 compared with ␣1, one could assume that the higher affinity is due to interactions with residues that are the same in ␣2 and ␣3 but differ in ␣1. All such residues at the extracellular surface and trans-membrane segments have been mapped onto the structure (␣1 residues). The interesting feature is that all these residues are located in the extracellular loops in proximity to the rhamnose moiety of ouabain, but within the steroid-lactone binding cavity there are no such isoform-specific differences. This structure is not an ideal model because ouabain is bound with low affinity due to the presence of occluded Rb ϩ , but as discussed in Ref. 38, it explains most effects of mutations on high affinity ouabain binding. In addition, the high affinity state was modeled by movement of transmembrane segments TM1/TM2 to close off the binding cavity, with ouabain bound in an almost identical position (38). With this caveat, the structure predicts quite naturally that no aglycones can show isoform selectivity, because the differences between the isoforms are found only in proximity to the sugar. This prediction is exactly consistent with all our observations. Conversely, our observations imply that the steroid-lactone moieties of all cardiac glycosides should be bound similarly to that of ouabain. By comparison with this structure, prior models of cardiac glycoside binding have assumed that the cardiac glycoside molecule lies more or less horizontally across the extracellular surface and do not explain our results. Of course, the structure cannot explain why ouabain somewhat favors ␣1 or why the digitalis glycosides somewhat favor ␣2/␣3, but it confirms that the key lies in interactions of the sugars near the extracellular loops. This structural insight fits very well with a two-step binding model proposed by Yoda (39) to explain the fact that the sugar is not essential for inhibition but strongly affects the binding affinity and rates (23). On the Yoda model (39), the steroid-lactone and sugar moieties occupy specific subsites. An independent kinetic contribution of the sugar to binding is consistent with the proposed role in isoform selectivity, assuming that only the sugar subsite differs in ␣1 compared with ␣2/␣3 isoforms.
Pharmacological and Physiological Implications-Of all the cardiac glycosides we have tested, those that show the greatest, even if partial selectivity for ␣2:␣1, digoxin, and ␤-methyl digoxin are the ones in most extensive clinical use. This sup-ports the evidence mentioned in the Introduction that ␣2 is the major player in cardiac contractility and primary target for cardiac glycosides such as digoxin. By contrast, ouabain, which shows no selectivity for ␣2:␣1, is not used clinically. Although in a comprehensive clinical trial (DIG trial), digoxin did not reduce overall mortality (40), a subsequent post hoc analysis found that a lower serum concentration (0.5-0.9 compared with Ͼ1 ng⅐ml) has overall benefit and reduces mortality (41). This could fit a hypothesis that therapeutic (␣2?) and toxic (␣1?) effects are separable, and in principle, a more ␣2-selective cardiac glycoside may increase the therapeutic window and reduce toxicity. Expression of Na,K-ATPase in human cardiac muscle is lowered up to 40% in congestive heart failure, especially ␣1, whereas ␣3 or ␣2 levels may be unchanged or lower depending on the region of the heart (42). Therefore, a more ␣2-selective cardiac glycoside than digoxin could be considered to be tailored to the failing heart.
The different selectivity of digoxin and ouabain for human ␣1 and ␣2 at the protein level was unexpected and may be relevant to differences between ouabain and digoxin observed in vivo. For example, injection of rats with low doses of ouabain over several days raise BP, but paradoxically, digoxin does not have this effect and even protects against the ouabain (43). The effect of ouabain derivatives, modified in the lactone ring, on BP does not correlate directly with inhibition of dog kidney Na,K-ATPase (␣1␤1) (44). Thus, the mechanism, especially the difference between ouabain and digoxin, is not well understood. The inability of digoxin to raise BP suggests that inhibition of ␣2 may not be involved. Furthermore, digoxin is not known to raise BP in man. Both ouabain and digoxin have been detected in mammalian tissues and fluids (1). Taken together, the selectivity differences of ouabain and digoxin and their different effects in vivo support the notion that these two cardiac glycosides have specific but distinct physiological roles (for reviews, see also Refs. 1 and 5). For example, digoxin might affect primarily ␣2-dependent and ouabain primarily ␣1-dependent functions. Furthermore, ouabain induces Ca 2ϩ oscillations (45), and long-term effects of ouabain on cell Ca 2ϩ and BP might be mediated via ␣1 and the cellular signaling pathways, as suggested in Ref. 2. The first step in ouabain-mediated signaling in vivo is tyrosine phosphorylation of the soluble tyrosine kinase, Src (3). In this respect, it is of interest that, unlike FIGURE 6. Structure of Na,K-ATPase with bound ouabain. The figure shows the Protein Data Bank codes 3A3Y structure (38). Residues identical in human ␣2 and ␣3 and different in ␣1 are shown in cyan. Residues and numbering are those for human ␣1. Dark green, ouabain, steroid-lactone; light green, rhamnose. The figure was drawn by PYMOL.

TABLE 2
Inhibition of Na,K-ATPase activity of ␣1␤1, ␣2␤1, and ␣3␤1 complexes ouabain, digoxin does not activate tyrosine phosphorylation of Src in an in vitro assay. 6 Perspective-The observations that several digital glycosides are partially selective for ␣2:␣1 and the mechanistic insight that aglycones show no isoform selectivity suggest that improved ␣2 selectivity might be achieved by modifying the digitoxose. A more ␣2-selective cardiac glycoside could have important applications as a tool to analyze toxicity of cardiac glycosides, physiological functions of ␣2, and different endogenous cardiac glycosides and possibly as a safer cardiotonic agent compared with digoxin. Isoform-selective cardiac glycosides could also be important in recently discussed applications of cardiac glycosides such as treatment of cancer (46).