Ethical approval

Ethical clearance was obtained from the Victoria University Human Research Ethics Committee and the Alfred Hospital Ethics Committee and conformed to the standards set by the Declaration of Helsinki, except for registration in a database.

Participants

The study was undertaken in healthy young adults to avoid possible confounding pharmacological, disease or inactivity-related effects that might exist if studying digoxin effects in clinical populations. Ten healthy, untrained but recreationally active individuals, comprising nine males and one female, gave written informed consent and participated in the study [age 26.1 (5.9) years, mass 75.7 (11.3) kg, and height 178.4 (9.2) cm; $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$, 3.67 (0.42) l min⁻¹; mean (SD)]. All participants underwent an initial medical examination to screen for abnormal plasma electrolyte concentrations, kidney function, rest and exercise ECG, and history of adverse cardiovascular events. Participants consumed pre-packaged iso-energetic meals and beverages during 72 h before each experimental trial and refrained from vigorous activity and ingestion of caffeine and alcohol in the 24 h before each visit.

Study design and digoxin administration

After initial medical screening, participants attended the exercise physiology laboratory on six separate occasions. During their initial laboratory visit, anthropometric measurements of forearm length and circumferences were made for forearm volume estimation and participants were then familiarised with finger flexion contractions (Sostaric et al., 2006). Participants performed an initial incremental finger flexion exercise test and, after 30 min rest, were familiarised with the high-intensity, intermittent finger flexion protocol for use in subsequent trials. After a further 10 min rest, participants then performed an incremental two-legged cycling test to determine peak oxygen consumption. Participants repeated the finger flexion exercise test during the second and third laboratory visits, to determine intra-subject variability in power output and time to fatigue. Participants were also familiarised with the leg cycling exercise test protocol for use in subsequent trials, and also underwent two variability trials on separate days.

Participants were allocated to either a digoxin (DIG) or placebo (CON) initial treatment group, with trials conducted using a randomised, double-blind, crossover, counterbalanced design. Participants ingested an oral dose of digoxin 0.25 mg day⁻¹ (Lanoxin; GlaxoSmithKline, Victoria, Australia) or a placebo (sugar tablet) for 14 days. This dosage and delivery simulated typical clinical usage of digoxin and allowed sufficient time for [digoxin] to reach steady state concentrations. After a 4 week washout period, participants undertook the alternative treatment for a further 2 weeks and then repeated the experimental exercise tests. Thus, for those participants taking DIG first, the effective digoxin clearance time was 6 weeks, that is the standard 4 week washout plus the 2 weeks of placebo treatment. Similarly, for the placebo group, the time between biopsies was 6 weeks. This washout period should be more than sufficient, given that digoxin clearance half-time from serum after digoxin injection was ∼45 h (range 32−131 h) (Kramer et al., 1974) and from skeletal muscle after oral digoxin was ∼2.2 days (Jogestrand & Sundqvist, 1981).

Participants underwent quadriceps muscle function testing on day 13 of treatment, and on day 14 performed finger flexion (FF) followed by leg cycling (LC) exercise tests, with repeated arterial and venous blood sampling. Participants rested for 3 h between FF and LC trials, with arterial and venous [K⁺] analysed to ensure resting values had been re-established before commencement of the LC trial. An additional venous blood sample was taken at rest to measure serum [DIG] on days 7, 13 and 14 of the treatment period. Heart rate and rhythm were monitored during experimental exercise tests by a 12-lead ECG (Mortara, Boston, MA, USA). For ethical and safety reasons, the attending medical practitioner on day 14 trials was non-blinded; also, to minimise clinical risks, all digoxin measurements were performed by the Alfred Hospital for rapid reporting of [digoxin]. Study protocols are shown schematically in Fig. 1A.

Exercise protocols

Blood sampling procedures and analyses

Catheters (20 or 22G; Jelco) were inserted anterograde in the radial artery (a) of the non-contracting arm, under local anaesthesia (2% lignocaine injection) and retrograde into the deep antecubital vein (v) of the contracting forearm. Participants then rested for ∼30 min before the commencement of each DIG and CON trial. Intra-arterial and intravenous blood pressures were continually monitored (Marquette 710, Wisconsin, USA) using electronic pressure transducers (Abbott Critical care Systems, Chicago, IL, USA) connected to the saline-filled cannula via an extension line. Blood pressure signals were then interfaced with the finger flexion exercise computer system, enabling continual integration between power output, forearm circumference changes and blood pressure data. Arterial lines were kept patent by a slow, sterile, isotonic saline infusion.

During FF trials, arterial and venous blood samples (5 ml) were taken simultaneously at rest, in the final 10 s of each of the three initial 1 min exercise bouts (EB1, EB2, EB3), immediately before the start of each subsequent exercise bout (PreEB2, PreEB3 and PreEB4), during the final exercise bout at both 1 min (EB4+1) and during the final contractions at fatigue, as well as at 1, 2, 5, 10 and 30 min following exercise. Hand blood flow was occluded for 10 s before and during venous blood sampling by a high-pressure wrist cuff, to exclude collection of hand venous effluent blood (Fig. 1B). Forearm blood flow was determined using plethysmography and conducted immediately following blood sampling at rest, during exercise and in recovery, as previously described (Sostaric et al., 2006).

During LC trials, arterial and venous blood samples (5 ml) were taken simultaneously at rest, in the final 10 s of each of the two initial exercise bouts (EB1, EB2), immediately before the start of each subsequent exercise bout (PreEB2, PreEB3), during the final exercise bout at both 1 min (EB3+1) and at fatigue, and at 1, 2, 5, 10 and 30 min following exercise (Fig. 1C). Hand blood flow was occluded for 10 s before and during venous blood sampling by a high-pressure wrist cuff, to exclude collection of hand venous effluent blood. Forearm blood flow was determined using plethysmography and conducted immediately following blood sampling at rest, during exercise and in recovery, but the data were discarded as unreliable, being affected by movement artefacts.

For measurement of serum [digoxin], blood samples were allowed to clot and then centrifuged at 4000 rpm for 10 min at 4°C (3K15 refrigerated centrifuge; Sigma, Laborzentrifugen, Germany). Serum was subsequently stored at −20°C until assayed for serum digoxin by standard clinical analysis using a Multigent (Abbot Laboratories, IL, USA) homogenous particle-enhanced turbidimetric immunoassay. The serum [digoxin] minimum detection level reported by the Alfred Hospital was 0.4 nmol l⁻¹ for the first seven participants, but changed during the study to 0.2 nmol l⁻¹ for the final three participants. However, this is not an important limitation given the clear elevation of [digoxin] to expected levels in the digoxin trials.

Arterial and venous blood samples from the experimental trials were immediately analysed in duplicate for plasma [K⁺] using an automated blood gas analyser (Sostaric et al., 2006). Whole blood [Hb], Hct and SO₂, and plasma $${P_{{O_2}}}$$ and $${P_{C{O_2}}}$$ were analysed (data not reported) using automated haematology and blood gas analysers (Sostaric et al., 2006). All measurements and analysers were calibrated immediately before, during and after measurements with precision standards in the range of the measurements (Sostaric et al., 2006).

Muscle sampling and analyses

Statistical analyses

Results are expressed as mean (SD). Data sets were tested for normality using the Shapiro–Wilk test and if criteria were not met, data were log transformed. Data sets were analysed using a linear mixed model, with time (rest, exercise, recovery) and treatment (DIG, CON) as fixed effects, and restricted maximum likelihood as the estimation method for any missing values. Post hoc analyses used the Least Significant Difference test. Where significant time effects were found, differences between times are indicated, but non-significant differences between times are not detailed. Significant treatment effects are reported (i.e. DIG vs. CON). To avoid repetition, treatment-by-time interactions are only stated when significant.

For the muscle data, a one-way ANOVA was used for the digoxin F_ab verification experiments. Paired data in resting muscle were analysed using a paired Student t test, for comparison of OB+F_ab vs. OB-F_ab. Sample size was balanced (n = 10) for all OB (±F_ab) and associated digoxin occupancy measures, and all 3-O-MFPase activity measures. The exact number of measurement points, defined as number of participants × number of trials × number of measurements minus missing data points, was 76, 76 and 77 for OB-F_ab, OB+F_ab and 3-O-MFPase activity, respectively.

Sample size for functional tests was n = 10 for Cybex tests and leg cycling exercise performance and n = 9 for finger flexion exercise performance. The exact number of data measurement points for Cybex tests was 138 for torque–velocity and 1000 for fatigue data and for time to fatigue during FF was 18 and during LC was 20. For blood analyses with exercise, balanced data sets between trials were used, with all data for a participant discarded if their alternative trial data were unavailable; sample sizes therefore differed between variables. During FF, sample sizes for arterial analyses were n = 9 for all variables. For FF venous and [ion]_{a-v differences} (uncorrected), sample sizes were n = 8 for [K⁺] and n = 7 for all other variables. For FF [ion]_{a-v difference} (corrected for fluid shifts), sample sizes were n = 7 for [K⁺]; whilst for plasma [ion] fluxes, n = 7 for [K⁺]. The exact number of measurement points during FF for [K⁺]_a, [K⁺]_v, [K⁺]_a-v was 242, 218 and 201, respectively. For calculated K⁺ flux data during FF, the exact number of data measurement points was 168. For LC, sample sizes for arterial analyses were n = 10 for all variables. For LC venous and [ion]_{a-v difference} (uncorrected), sample sizes were n = 8 for all variables. For LC [ion]_{a-v difference} (corrected for fluid shifts), sample sizes were n = 8 for [K⁺]. The exact number of measurement points during LC for [K⁺]_a, [K⁺]_v and [K⁺]_a-v, was 218, 216 and 209, respectively. Paired data were analysed using a paired Student t test (e.g. exercise time to fatigue). For paired comparisons, n = 10, except for FF where n = 9.

Individual variability was calculated for all participants within the exercise protocol (n = 10) and averaged to obtain an overall coefficient of variation (CV), with method variability also determined by CV. Statistical significance was accepted at P < 0.05. Statistical analyses were performed using IBM SPSS Statistics 27; since this package does not report exact P values lower than P < 0.001, for these data they are reported here as P = 0.001.

Serum digoxin concentration and compliance

All participants reported full compliance with digoxin, and no adverse events were reported. During DIG, serum [digoxin] on days 7, 13 and 14 were 0.7 (0.2), 0.7 (0.2) and 0.8 (0.2) nmol l⁻¹, respectively [mean (SD)]. During CON, serum [digoxin] was reported below the method detection limits in nine of 10 participants (n = 6, <0.4 nmol l⁻¹; n = 3, <0.2 nmol l⁻¹) and at the reported detection level (0.4 nmol l⁻¹) in one individual.

Muscle NKA results

Muscle [³H]-ouabain binding site content without Digibind (OB-F_ab)

Exercise

The [³H]-ouabain binding site content measured without incubation in Digibind (OB-F_ab) (pmol g wet weight⁻¹) was elevated above rest after exercise at 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (10%, P = 0.005) but not at fatigue (5%, P = 0.163); the OB-F_ab at 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ was also higher than at 3 h after exercise (12%, P = 0.005, Fig. 2A). The OB-F_ab (pmol g protein⁻¹) showed a similar pattern, being increased at both 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (21%, P = 0.001) and at fatigue (16%, P = 0.003); these were each respectively greater than at 3 h recovery (67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$: 20%, P = 0.001; and fatigue: 16%, P = 0.002, Fig. 2B).

Digoxin

Neither the OB-F_ab (pmol g wet weight⁻¹) (P = 0.253) nor the OB-F_ab (pmol g protein⁻¹) (P = 0.087, −5.6%) differed significantly between DIG and CON (Fig. 2).

Muscle [³H]-ouabain binding site content measured after clearance of bound digoxin by F_ab (OB+F_ab)

Exercise

No exercise effect was found for the OB+F_ab (pmol g wet weight⁻¹, P = 0.814, Fig. 3A), but the OB+F_ab expressed per gram of protein was higher at fatigue than at rest (11%, P = 0.022) and tended also to be higher at 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (9%, P = 0.087); OB+F_ab at 3 h after exercise was less than both 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (−13%, P = 0.041) and fatigue (−15%, P = 0.009, Fig. 3B).

Digoxin

The OB+F_ab did not differ between DIG and CON whether expressed per gram wet weight (P = 0.809) or per gram protein (P = 0.376, Fig. 3).

Maximal in vitro 3-O-MFPase activity

Exercise

The 3-O-MFPase activity (nmol min⁻¹ g wet weight⁻¹) did not differ from rest at 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ but declined at fatigue (−15%, P = 0.003), before then increasing (vs. fatigue: P = 0.016) to return to rest levels by 3 h after exercise (Fig. 4A). No changes were found when 3-O-MFPase activity was expressed relative to muscle protein content (Fig. 4B). The calculated ratio of 3-O-MFPase activity/ouabain binding content was reduced from rest at both 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (P = 0.008) and fatigue (P = 0.003) and recovered at 3 h after exercise.

Digoxin

The 3-O-MFPase activity did not differ between CON and DIG whether expressed relative to muscle wet weight (P = 0.287) or muscle protein (P = 0.180, Fig. 4). The ratio of 3-O-MFPase activity/ouabain binding content did not differ between DIG and CON (P = 0.770).

Quadriceps muscle strength and fatiguability

Muscle strength

Exercise

Peak torque at each velocity during Cybex isokinetic contractions was highly reproducible during repeat variability trials (CV 4.3−7.9%). During torque–velocity testing (0–360° s⁻¹) in experimental trials, peak torque declined with each increased velocity up to 300° s⁻¹ (P = 0.001) and further at 360° s⁻¹ (P = 0.026).

Digoxin

Peak torque across all velocities was lower in DIG than in CON (−4.3%, P = 0.010, Fig. 5A).

Muscle fatigue

Exercise

The calculated fatigue index was highly reproducible during variability trials (CV = 4.7%). During experimental trials, the contraction peak torque declined markedly during the 50 repeated contractions at 180° s⁻¹ (P = 0.001, Fig. 5B).

Digoxin

There were no differences between trials in contraction peak torque during the 50 repetitions (P = 0.221), or in the calculated fatigue index [DIG, 53.6 (9.0) vs. CON, 57.4 (10.0)%, P = 0.138] (Fig. 5B).

Finger flexion exercise results

Leg cycling test results

LC exercise performance, $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ and heart rate

LC incremental exercise test

The LC incremental exercise $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ and WR_peak were 3.67 (0.42) l min⁻¹ and 298 (23) W, respectively.

LC variability and experimental trials

During LC variability trials, good reproducibility was seen for VO₂ during each of 33% (CV, 2.2%), 67% (CV, 1.9%) and 90% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (CV, 4.4%), as well as for the time to fatigue at 90% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (CV, 5.9%). During LC experimental trials, no significant differences were found between DIG and CON for the time to fatigue at 90% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ (P = 0.775, Table 1), for VO₂ [33% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$, DIG, 1.21 (0.20) vs. CON, 1.19 (0.21); 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$, DIG, 2.63 (0.51) vs. CON, 2.68 (0.54); 90% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$, DIG, 3.56 (0.28) vs. CON, 3.55 (0.67) l min⁻¹, respectively], or heart rate [rest, DIG, 80 (4) vs. CON, 77 (5); fatigue, DIG, 193 (5) vs. CON, 191 (6); 30 min recovery DIG, 94 (7) vs. CON, 93 (5) bpm].

LC plasma [K⁺]

Exercise

Plasma [K⁺]_a increased above rest at EB2+1, EB2, EB3+1 and at fatigue where it peaked (P = 0.001), but did not differ from rest during the intervening rest periods; [K⁺]_a decreased rapidly after fatigue but remained elevated at 1 min (P = 0.001), fell below rest at 5 min (P = 0.012) and tended to be less also at 10 min recovery (P = 0.082; Fig. 7A). Plasma [K⁺]_v increased above rest from EB2 through fatigue until 2 min recovery (all P = 0.001, except PreEB3, P = 0.002, Fig. 7B). Plasma [K⁺]_a-v was mostly positive during the exercise period, indicating net K⁺ uptake into inactive forearm muscle, but was greater than rest only at fatigue (P = 0.001), with a tendency to also increase during EB3+1 (P = 0.059, Fig. 7C). After correction for the corresponding ΔPV_a-v, [K⁺]_a-v (corrected) displayed a similar pattern, tending to be greater than rest at EB2+1 (P = 0.058), EB2 (P = 0.053), EB3+1 (P = 0.053), increased at fatigue [1.71 (1.11), P = 0.001] and at 2 min recovery (P = 0.008) and tending to be less at 5 min recovery (P = 0.073).

Digoxin

There was no effect of DIG on plasma [K⁺]_a (P = 0.833, Fig. 7A), whereas [K⁺]_v was lower in DIG than in CON (0.15 mmol l⁻¹, P = 0.042, Fig. 7B), with a greater (more positive) [K⁺]_a-v in DIG compared to CON (0.08 mmol l⁻¹, P = 0.004, Fig. 7C); [K⁺]_a-v (corrected) did not differ between trials (P = 0.377). No differences were found between trials for the rise in [K⁺]_a from rest to fatigue [Δ[K⁺]_a, DIG, 2.59 (0.60) vs. CON, 2.65 (0.74), P = 0.675], or the decline from fatigue to 5 min recovery [−Δ[K⁺]_a, DIG, −2.88 (0.65) vs. CON, −2.95 (0.64), P = 0.667].

Clinically relevant serum digoxin concentration

Oral digoxin administration of 0.25 mg day⁻¹ for 14 days raised serum [digoxin] to 0.8 nmol l⁻¹ (∼0.62 ng ml⁻¹), with no difference between days 7 and 14, similar to a previous report using the same dose in healthy males (Schenck-Gustafsson et al., 1987) and only ∼20% less than was found after taking twice this digoxin dosage (0.5 mg day⁻¹) for 14 days in healthy individuals (Sundqvist et al., 1983). Importantly, the serum [digoxin] attained was within the optimal therapeutic range of ∼0.65–1.15 nmol l⁻¹ (∼0.5–0.9 ng ml⁻¹) (Bavendiek et al., 2017). Thus, our digoxin protocol achieved the desired outcome of a typical clinical [digoxin], allowing us to then investigate the consequent physiological and functional impacts on muscle NKA and plasma K⁺ regulation in healthy participants.

Preservation of NKA content in skeletal muscle despite elevated digoxin

These are the first measures of digoxin effects on muscle [³H]-ouabain binding site content (OB-F_ab) in healthy individuals, and the first serial measures of OB-F_ab taken both before and following digoxin treatment, with a major finding being that OB-F_ab did not decline after 14 days of digoxin treatment in healthy adults. The typical OB-F_ab assay measures [³H]-ouabain binding to all functional αβ complexes, which, in human skeletal muscle, normally enables complete quantification of NKA and is therefore the gold standard measure of NKA content (McKenna et al., 1993; Nørgaard et al., 1983). We hypothesised that digoxin binding to, and inhibition of, NKA in skeletal muscle (Jogestrand & Sundqvist, 1981; Joreteg, 1986) would reduce the OB-F_ab in resting muscle, based on reports of substantially reduced OB-F_ab in muscle in digitalised heart failure patients (Schmidt et al., 1991, 1995; Schmidt, Allen et al., 1993). Also, another study on heart failure patients receiving digoxin reported similar muscle OB to controls, but these analyses were conducted after first removing bound digoxin with Digibind, thus indicating that pre-Digibind OB (i.e. OB-F_ab) must have been substantially reduced with digitalisation in their cohort (Green et al., 2001). Our findings of unchanged muscle OB-F_ab in healthy participants with digoxin therefore clearly contrast earlier studies reporting reduced muscle OB-F_ab in heart failure patients with digitalisation.

In muscle exposed to digoxin, a fraction of NKA already bound by digoxin is unavailable for [³H]-ouabain binding, thereby underestimating true NKA content (Schmidt & Kjeldsen, 1991). Hence, we also determined the digoxin–NKA occupancy using digoxin antibody fragments (F_ab) to first clear any digoxin bound in muscle, followed by the standard ouabain binding measures (OB+F_ab) (Schmidt & Kjeldsen, 1991). We confirmed that this F_ab wash methodology enabled almost complete NKA recovery (96%) in muscle biopsy samples from healthy humans incubated in digoxin, almost identical to that reported in skeletal muscle obtained post-mortem (Schmidt & Kjeldsen, 1991) and in heart failure patients (Green et al., 2001). A key finding here was that when measured after Digibind incubation, the ouabain binding sites in resting muscle samples had in fact increased by 8.2% after 14 days of digoxin (i.e. OB+F_ab was 8.2% greater than OB-F_ab in DIG). This indicated a greater overall total NKA (i.e. OB+F_ab) existed in muscle after digitalisation in these healthy participants, which was not the case in cardiac patients. The muscle NKA–digoxin occupancy in these healthy participants was 7.6% of the total ouabain binding sites (i.e. 7.6% of OB+F_ab). This is consistent with the 9% occupancy reported in muscle from heart failure patients digitalised for 3 days, and although lower, also with the 13% occupancy in muscle obtained post-mortem from previously digitalised patients (Schmidt et al., 1995; Schmidt, Holm-Nielsen et al., 1993). The higher occupancy in these two clinical studies is probably due to the patient's higher serum [digoxin] of 1.2−2.3 nm (Schmidt et al., 1995; Schmidt, Holm-Nielsen et al., 1993), already lower muscle NKA content with heart failure (Norgaard et al., 1990), and probably reduced muscle mass due to sarcopenia and/or cachexia. Importantly, none of these studies contrasted digoxin occupancy before vs. after digitalisation within the same individuals. Our findings may therefore also give some insights into understanding muscle adaptation in heart failure patients. It is possible that NKA synthesis evident in healthy muscle is diminished in advanced heart failure, eventually contributing to reduced muscle NKA content (Norgaard et al., 1990). This would also be consistent with their lack of muscle NKA upregulation in response to physical training, as NKA content actually declined after 3–4 months of training (Green et al., 2001), in contrast to the consistent NKA upregulation with training in healthy adults (Wyckelsma et al., 2019). Our findings of preservation of NKA content and maximal 3-O-MFPase activity with digoxin in resting muscle are also internally consistent. Since 3-O-MFPase activity was not also measured with Digibind, it is not possible to determine whether this might also have been greater after digoxin removal. Despite the uncertainties with maximal 3-O-MFPase activity as a measure of NKA activity after exercise (Juel et al., 2013), we have utilised this measure here as additional evidence verifying the lack of depression in NKA after digoxin treatment. For future studies, alternative measures of NKA activity would be beneficial to further explore the functional effects of NKA inhibition. In rats, high-dose digoxin initially depressed 3-O-MFPase activity in muscle by 13% after 3–7 days, but this then recovered after 3 months of further digoxin treatment, suggesting NKA adaptability (Li et al., 1993). Interestingly, the 8.2% gain in OB+F_ab in human muscle after 14 days of digoxin is substantially less than the 30% elevation in NKA activity found in guinea pig myocardium after digitoxin for 10–15 days (Bonn & Greeff, 1978). Further research on the acute and chronic effects of digoxin on NKA in muscle is of interest not only because of digoxin's ongoing clinical usage, but also given the growing focus on cardiotonic steroid signal transducing, protein–protein interactions and intracellular signalling functions via Src-dependent and Src-independent pathways (Aperia et al., 2016; Cui & Xie, 2017).

There are several interpretations for why muscle OB-F_ab was maintained and not depressed with 14 days of digoxin. The first is a possible compensatory upregulation of NKA in muscle after digoxin, which acted to preserve muscle functional NKA content and would be consistent with both the unchanged OB-F_ab and the 8.2% greater OB+F_ab after digoxin. This could be due to increased synthesis and/or reduced catabolism of NKA subunits, but several of our findings are inconsistent with this. These include first that OB+F_ab in DIG was not greater than in CON as would be expected (Fig. 3; the relevant interaction effect was non-significant). Second, increases in protein abundances of the dominant NKA isoforms would be expected after digoxin, namely α₁, α₂, β₁ or β₂ isoforms, whereas in fact, no significant changes were detected (our unpublished data). Another finding inconsistent with possible upregulation of muscle NKA with digoxin is the lack of effect on muscle NKA α_1-3 or β_1-3 isoform mRNA expression, with no effects evident after 14 days (our unpublished data). This was despite digoxin occupancy of 7.6%, and contrasts with other reports that cardiac glycosides differentially affect NKA isoform mRNA and transcription rates (Kometiani et al., 2000; Murphy et al., 2006; Wang et al., 2000; Yamamoto et al., 1993). However, because those studies used different species, tissues and cell preparations, with different glycoside concentrations and also with conflicting findings, it is difficult to compare their findings to ours in skeletal muscle in healthy humans. It is possible that an initial effect of NKA inhibition due to digoxin binding did affect individual NKA isoform gene transcripts, but this cannot be answered without a time-course study. Therefore, at this time we cannot satisfactorily explain how muscle OB was sustained in the face of elevated serum and presumably also tissue levels of digoxin. Further experiments are required to ascertain whether any NKA isoforms are upregulated at the protein level to correspond with the greater muscle OB after digoxin (OB+F_ab). It should also be noted, however, that no NKA upregulation is also inconsistent with the calculated digoxin binding in muscle of ∼29 pmol g⁻¹ wet weight. This digoxin binding was greater than muscle digoxin content measured by radioimmunoassay in earlier studies, of ∼8–10 pmol g⁻¹ wet weight (∼32 −39 nmol kg⁻¹ dry weight) in digitalised healthy males with a similar [digoxin] (Ericsson et al., 1981; Jogestrand & Sundqvist, 1981; Schenck-Gustafsson et al., 1987). In future studies, an initial muscle biopsy taken after an acute dose of digoxin would clarify the extent of initial digoxin binding and allow measures of OB+F_ab without any possible compensatory upregulation.

It seems unlikely that post-translational modifications of NKA due to glutathionylation contributed to our findings of greater OB+F_ab than OB-F_ab in muscle. Studies in numerous cell types have shown that a fraction of NKA subunits are glutathionylated under basal conditions and that glutathionylation can be rapidly and reversibly increased with oxidative stress (Petrushanko et al., 2012, 2017), which inactivates NKA and therefore can regulate NKA activity (Dergousova et al., 2018). Glutathionylation of NKA α subunits and of β₁ and β₂ isoforms also occurs in human skeletal muscle, and increased glutathionylation via GSSG incubation was associated with decreased NKA activity (Juel et al., 2015). Hence a fraction of NKA in muscle may be inhibited by glutathionylation, with implications for functional NKA content and activity, although the regulation is complex (Pirkmajer & Chibalin, 2016), as glutathionylation of FXYD-1 may exert protective effects against glutathionylation of NKA subunits (Bibert et al., 2011). Ouabain fixes NKA in the E2P conformation, thereby reducing cysteine availability for glutathionylation and protecting NKA from inactivation (Poluektov et al., 2019). High-dose ouabain (10 μm) reduced both glutathionylation and NKA activity in heart homogenates (Petrushanko et al., 2012), but the effects of nanomolar ouabain, or of digoxin, on NKA glutathionylation are unclear. However, it would seem likely that, if anything, digoxin binding in muscle may in fact reduce NKA glutathionylation, thereby removing NKA inhibition and increasing NKA availability. Hence, application of Digibind to remove digoxin would not then allow detection of previously inactive digoxin-bound NKA due to glutathionylation, but perhaps the opposite.

An important finding was that a significant elevation in OB+F_ab was found only after digoxin, not in control muscle, where a non-significant difference with F_ab of 4.5% was found, similar to earlier findings of 5–7% in muscle incubated in F_ab but without digoxin exposure (Schmidt & Kjeldsen, 1991; Schmidt, Holm-Nielsen et al., 1993; Green et al., 2001). We suggest that this low ‘occupancy’ value of 4.3% in controls probably mainly reflects variability in the OB method, but that it might also include some actual NKA occupancy, via binding to endogenous ouabain or ouabain-like compounds in muscle. In humans, plasma [ouabain] of between 0.05 and 1 nm and plasma [marinobufagenin] of between 0.2 and 0.6 nm have been reported (Orlov et al., 2020), with much higher values also reported for ouabain-like compounds of 2.5 and of 7.3 nm for an ‘NKA-inhibitor’ in plasma for humans at rest, both of which were markedly elevated with exercise (Bauer et al., 2005). Although Digibind does bind both endogenous ouabain and ouabain-like compounds (Pullen et al., 2004), we cannot determine whether this also occurred here, as we did not measure these compounds in plasma or muscle. However, if present, they would contribute to the low apparent occupancy in control muscle in these healthy participants, but also to the calculated digoxin occupancy in the DIG muscle. However, this possibility remains to be determined.

Acute exercise effects on NKA content and 3-O-MFPase activity

A novel finding was the transient and rapid increase in muscle OB-F_ab after only 20–30 min of exercise. The OB-F_ab was elevated by 10% after 20 min of exercise comprising 10 min each at 33% and 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$, when expressed per gram wet weight (NS, 5% at fatigue) and by 21% when expressed per gram protein, as well as by 16% at fatigue when expressed per gram protein. These intriguing findings indicate that muscle NKA content was acutely increased with exercise. It is unlikely that this can be explained by any of increased NKA synthesis and/or reduced degradation, increased recruitment of NKA subunits to the sarcolemma and t-tubules from intracellular stores, increased rate of binding of ouabain due to elevated NKA activity, methodological artefacts and/or digoxin effects, as detailed below. We suggest alternatively that this might reflect increased formation of functional NKA complexes from already existing, but non-bound, α and β subunits in muscle.

The elevated muscle OB-F_ab with exercise differs from our previous reports that OB-F_ab was unchanged with acute exercise (see references in McKenna et al., 2008), but appears consistent with the 13% increase reported in muscle OB-F_ab after prolonged running for ∼10 h and attributed to NKA synthesis (Overgaard et al., 2002). However, increased NKA synthesis or decreased degradation seems unlikely. The increase here occurred after only 20–30 min of exercise, where it is unlikely that muscle could synthesise ∼35–70 pmol NKA (per gram muscle).

The rapid increase in OB-F_ab with exercise is also unlikely to simply reflect an increased rate of ouabain binding in muscle due to increased NKA activity, due to neuro-humoral responses to exercise or elevated intracellular [Na⁺] with muscle contractions. First, the assay utilises 2 h of incubation to ensure saturation of [³H]-ouabain binding occurs (Nørgaard et al., 1984). Second, in rat muscle, saturation of [³H]-ouabain binding and also unchanged OB-F_ab were evident under conditions of elevated NKA activity due to electrical stimulation or insulin, where an increase in OB-F_ab would be expected if NKA activity or availability of NKA had increased (McKenna et al., 2003). In that study, ouabain binding was greater in the early period of incubation, but this effect had disappeared with a plateau after 2 h of incubation. This suggests that a functional elevation in pump activity cannot account for our results of increased OB-F_ab with exercise.

In non-digitalised human skeletal muscle biopsy pieces, the typical ouabain binding assay fully quantifies NKA (Nørgaard et al., 1984) and in rodent muscle, binding does not differ between intact muscles and cut muscle pieces (Clausen, 2013). Thus, OB-F_ab should measure all functional NKA in pieces of human muscle (Clausen, 2013). The OB-F_ab method was recently criticised as being too slow to detect increased NKA recruitment due to translocation, as these pumps would then already be included in the final OB-F_ab measure (Pirkmajer & Chibalin, 2016). This possibility cannot explain our finding because OB-F_ab was increased with exercise. Therefore, it seems unlikely that the increase in OB-F_ab with exercise could be due to increased recruitment to the sarcolemma and t-tubules of otherwise unavailable NKA, or of NKA subunit isoforms (Benziane & Chibalin, 2008; Pirkmajer & Chibalin, 2016) from undefined intracellular stores, as reported in membrane fractions following exercise (Tsakiridis et al., 1996; Juel, Nielsen et al., 2000; Juel et al., 2001), or possibly from caveolae (Kristensen et al., 2008). The increase in OB-F_ab with exercise also cannot simply be due to fluid shifts, since dilution might be expected to reduce OB-F_ab per wet weight but not change OB-F_ab per protein. We suggest that this also cannot simply be excluded as a methodological artefact, since an increase in ouabain binding with exercise was similarly detected for both OB-F_ab and OB+F_ab when expressed per gram protein. Our finding is also unexpected with digoxin, as we anticipated that OB-F_ab might even decline with exercise in DIG, since digoxin binding to muscle was previously reported to increase by up to 20% during exercise (Joreteg & Jogestrand, 1983). In that study, however, the digoxin dosage taken was double that of the present study (i.e. 0.5 vs. 0.25 mg day⁻¹, for 14 days) and the serum [digoxin] achieved was 75–88% higher than in our study (i.e. 1.4–1.5 vs. 0.8 nm). Whether this explains the discrepancy between the two studies on acute exercise effects on muscle NKA content is unclear.

One possibility is that structural alterations to NKA with exercise, caused by glutathionylation or phosphorylation of NKA subunits or of phospholemman might enable these increases in OB-F_ab with exercise. Glutathionylation of NKA β₁, but not α subunits or β₂ isoforms, increased in muscle with intense exercise and was also associated with reduced NKA activity measured by both direct Pi release and the 3-O-MFPase method (Juel et al., 2015). However, whilst fractional NKA inactivation could reduce muscle OB-F_ab, this is opposite to the effect seen here with exercise. It therefore seems unlikely that NKA glutathionylation or de-glutathionylation is responsible for the increase found in OB-F_ab with exercise. Whilst phosphorylation of phospholemman can increase NKA activity and also increases with muscle contraction (Pirkmajer & Chibalin, 2016), this also seems unlikely to account for the increased OB-F_ab with exercise. Glycosylation of the β-subunit is essential for NKA functional expression in cultured chick muscle cells, as measured via ouabain binding and activity (Alboim et al., 1992), but it is unknown whether β-glycosylation is modified by muscle contraction and therefore whether this also affects NKA activity or ouabain binding.

We therefore consider an alternative possible mechanism for the elevated OB-F_ab with acute exercise, in the absence of increases in α subunit proteins. We suggest that an increased formation of functional NKA complexes occurred from already existing, but non-bound, α and β subunits in muscle, that were either inserted into, or being trafficked to, sarcolemmal/t-tubular membranes (DeTomaso et al., 1994) that could then be detected by ouabain binding. Further research using different techniques is required to test this possibility.

Our findings of acute exercise effects on muscle NKA were somewhat paradoxical between the approximate increases seen in OB-F_ab vs. approximate depressive effects on maximal in vitro 3-O-MFPase activity. The elevated OB-F_ab (per wet weight) after 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ but not after fatiguing exercise are first at odds with the reduction in the 3-O-MFPase activity (per wet weight) after fatiguing exercise. This divergence is consistent with the previous finding of a decline in muscle 3-O-MFPase activity with intense exercise but not in OB (Petersen et al., 2005) and with the well-described reduction in 3-O-MFPase activity with intense exercise (McKenna et al., 2008). Suggested factors contributing to this reduction in 3-O-MFPase activity independent of any effects on OB include elevations in reactive oxygen species, myoplasmic calcium (McKenna et al., 2008) and NKA subunit glutathionylation (Juel et al., 2015). Our findings are inconsistent when compared between results expressed per muscle wet weight vs. per protein. Whilst an increase (per protein) in OB-F_ab was found after both 67% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ and after fatiguing exercise, there was a lack of effect (per protein) on 3-O-MFPase activity. A reduction in 3-O-MFPase activity (per protein) with exercise has been reported previously (Aughey et al., 2007; Fraser et al., 2002), suggesting the reductions in activity per wet weight were not simply due to fluid shifts; it is unclear why no reduction was similarly found here. Others have reported that a decline in 3-O-MFPase activity with exercise was not replicated by a reduction in NKA activity when this was measured by Pi liberation (Juel et al., 2013). However, their subsequent paper did in fact report a decline in the latter measurement with exercise (Hostrup et al., 2014). Further detailed investigations into acute exercise effects on muscle NKA activity utilising different techniques are warranted. Nonetheless, it seems likely that acute exercise exerts diverging effects on NKA activity and content.

Elevated serum digoxin was not associated with impaired K⁺ homeostasis with exercise

Contrary to our hypothesis, elevated serum digoxin did not exacerbate plasma [K⁺] with exercise in these healthy adults. Instead, [K⁺] variables were remarkably consistent between DIG and CON, for both FF and LC exercise tests. Plasma [K⁺]_a during and following each of FF and LC exercise was unchanged by digoxin, whilst during FF, plasma [K⁺]_v, [K⁺]_a-v difference and K⁺ efflux into plasma across the contracting forearm muscles were also all unchanged by digoxin. In contrast, during LC, a lower [K⁺]_v and a greater [K⁺]_a-v were found with digoxin, indicating a greater net K⁺ uptake by inactive forearm muscles. These small differences with digoxin are opposite to expected changes with NKA inhibition in inactive muscle and cannot readily be explained. We anticipated that digoxin would exacerbate the rise in plasma [K⁺] with exercise, based on studies in patients with atrial fibrillation and heart failure, where 1.2–2.3 nmol l⁻¹ [digoxin] further increased [K⁺] by ∼0.2–0.3 mmol l⁻¹ during cycling exercise (Norgaard et al., 1991; Schmidt et al., 1995), increased the K⁺ loss from the exercising leg by ∼138% (Schmidt et al., 1995), and also in healthy adults, where digoxin elevated resting venous [K⁺] (Edner et al., 1993). The non-impairment by digoxin of K⁺ homeostasis with exhaustive dynamic exercise in these healthy participants is therefore clearly at odds with previous clinical studies (Norgaard et al., 1991; Schmidt et al., 1995). This was not due to inadequate systemic digitalisation, as [digoxin] was at expected clinical levels, at 0.8 nm. One possibility is that this [digoxin] was below a threshold required to perturb K⁺ homeostasis during exercise; that is, 1.1–1.2 nm was previously shown to exert an effect (Norgaard et al., 1991; Schmidt et al., 1995). The lack of digoxin impairment of K⁺ homeostasis is highly unlikely to be due to insufficient activation of NKA in contracting muscles, given the large disturbances to systemic [K⁺] during both LC and FF exercise, the expected large increases in NKA activity with intense muscular contractions and the high in vivo muscle NKA activity in both LC and FF trials, evidenced by the rapid post-exercise declines in [K⁺]_a and/or in [K⁺]_v after early and fatiguing exercise bouts. These rapid post-exercise reductions in systemic [K⁺] with intense exercise are probably consequent to decreased muscle interstitial [K⁺] induced by muscle NKA activation in the previously contracting musculature (Sejersted & Sjogaard, 2000; Lindinger & Cairns, 2021). The net K⁺ uptake into forearm muscles during LC was also consistent with substantial in vivo NKA activity also in non-contracting muscles. Why this net K⁺ uptake was elevated with DIG during LC is unclear. Catecholamines were not measured here to determine whether these might be elevated above placebo trials, which might have independently affected NKA activation and produced a counterbalancing effect to digoxin on K⁺ homeostasis. However, this seems unlikely, given noradrenaline was reduced with digoxin in heart failure patients (van Veldhuisen et al., 1993). Furthermore, altered fluid shifts cannot explain the unchanged [K⁺] with digoxin, since arterio-venous ΔPV or ΔBV were unchanged by digoxin, in both FF or LC trials. Early studies in healthy males found that intravenous ouabain infusion of 0.05 mg min⁻¹ for 10 min did not affect forearm or hand blood flow (Glover et al., 1967). We found slightly lower forearm blood flow during FF with digoxin, but this seems unlikely to explain the lack of effect of digoxin on the [K⁺]_a-v and K⁺ fluxes during FF. We were unable to obtain reliable forearm blood flow during leg exercise so cannot conclude whether any changes in perfusion might be related to the lower [K⁺]_v and elevated [K⁺]_a-v across the inactive forearm. Rather, these seem more likely to reflect a paradoxical greater NKA activation with digoxin during leg exercise. One possibility for the largely unchanged [K⁺] perturbations with exercise after digoxin is that intracellular [Na⁺] could have been slightly elevated due to digoxin inhibition of NKA, which would then increase activity of the remaining functional NKA, thereby counterbalancing inhibitory effects. Importantly, this study was conducted with healthy, young adults, whereas studies that reported exacerbated increases with digoxin in [K⁺]_a or [K⁺]_v in venous blood draining contracting muscles during exercise utilised clinical populations (Norgaard et al., 1991; Schmidt et al., 1995). Hence the impacts of digoxin therapy on K⁺ regulation may be quite different in these populations. One previous finding in healthy individuals also reported no effect of acute digoxin on venous [K⁺] during handgrip exercise (Janssen et al., 2009). However, their findings are difficult to apply here, since their exercise intensity was low, induced only small increases in venous [K⁺] and [Lac⁻], and as it is unclear whether their blood sampling occurred during or after exercise, which markedly affects [K⁺]; and also, since none of serum [digoxin], digoxin binding to muscle or forearm blood flow were measured.

The preservation of NKA in skeletal muscle after 14 days of oral digoxin in healthy humans is the most likely explanation for the remarkable consistency in [K⁺] observed between DIG and CON trials across several exercise modalities. Our findings are therefore consistent with K⁺ being a tightly regulated physiological variable during and following exercise (McKenna et al., 2008) and in non-exercising conditions (McDonough & Youn, 2005). Although digoxin occupancy of NKA was not analysed for the forearm muscle, it is likely to be similar to quadriceps muscle, given these muscles all have a similar mixed fibre composition (Johnson et al., 1973). Whilst we cannot exclude the possibility that a higher serum [digoxin] and greater digoxin binding fraction in muscle are required to impair K⁺ homeostasis in healthy individuals, such a study might be difficult to conduct due to the increased risks involved.

Digoxin impaired skeletal muscle strength, but not exercise performance or fatiguability

An interesting finding was the reduced maximal voluntary strength of the quadriceps muscles with digoxin by 4%, measured across a range of limb velocities. Our research design in these 10 healthy participants featured careful control of test procedures, familiarisation and determination of within-subject variability for quadriceps strength and fatiguability, which were highly reproducible. This might explain why we detected a reduction whereas previous studies reported no change in muscle strength in only six, well-trained participants after digoxin (Sundqvist et al., 1983) or only tendency to reduced strength in only four participants (Bruce et al., 1968; Sundqvist et al., 1983). Others have reported that intra-arterial injection of high-dose ouabain enhanced electrically evoked muscle strength, whereas low-dose ouabain did not (Smulyan & Eich, 1976). The digoxin dose here was sufficient to induce small reductions in muscle strength, whereas a higher digoxin dosage might be required to induce more marked reductions in force, as demonstrated in isolated muscle preparations with inhibition by ouabain (Clausen, 2003). In genetically altered mice, reduced or absent NKA α₂ isoform in muscle did not reduce evoked tetanic force in diaphragm (Radzyukevich et al., 2004), but did lower twitch and tetanic force in EDL muscle (Radzyukevich et al., 2013). This latter finding is unlikely to be applicable to our participants, however, due to the relatively small fraction of NKA bound by digoxin at this clinical serum concentration. Mice exhibiting lowered NKA α₁ isoforms also had reduced mass and cross-sectional area in soleus muscle (Kutz et al., 2018). This suggests potential linkages between NKA, muscle mass and strength might also be present in human muscles. It is therefore interesting that a reduction in strength persisted with DIG even despite preservation of functional muscle NKA. Another possibility is that these effects might also be via neural effects, linked to inhibition of neuronal NKA α₃ isoforms, in the motor pathways and/or motor nerves, but this remains to be tested.

We hypothesised that fatiguability would be exacerbated with digoxin, due to NKA inhibition at rest and during exercise (Joreteg & Jogestrand, 1983), consistent with marked inhibitory effects of ouabain in isolated muscle preparations (Clausen, 2003) and with impaired treadmill running in mice with skeletal muscle-selective NKA α₂ knockout (Radzyukevich et al., 2013). However, an important and consistent finding was the lack of effect of digoxin on fatiguability during three different exercise protocols, namely 50 repeated dynamic contractions of the quadriceps muscles and each of intense FF and LC exercise that concluded with a final exercise bout continued to fatigue. The quadriceps fatiguability and LC performance tests were all highly reproducible during variability trials, although a higher variability was evident for time to fatigue during FF. The lack of a digoxin effect on all these repeat contraction performance measures was internally consistent, suggesting variability was not the underlying cause for failure to detect an effect. The lack of effect of digoxin on torque during the fatigue test at 180° s⁻¹ probably reflects the typical variability, especially in initial efforts, in peak torque during these 50 repeated contractions (Fig. 5). Our findings were also consistent with a lack of effect of digoxin on isometric endurance during sustained handgrip contractions at 30% MVC (Bruce et al., 1968).

Elevated extracellular [K⁺] at high concentrations can depress maximal force in rodent muscle preparations, but can also potentiate muscle force during twitch and submaximal contractions, thereby exerting dual effects (Pedersen et al., 2019). Thus, small increases in [K⁺] with digoxin might actually benefit muscle performance during submaximal exercise, yet also contribute to fatiguability during fatiguing contractions when muscle activation increases. An important question is whether the peak [K⁺] reached during exhaustive exercise (FF, ∼4.2 and ∼5.2 mm; LC, ∼6.5 and ∼5.0 mm, in arterial and antecubital venous plasma, respectively) were sufficiently high to produce muscular fatigue. Muscle interstitial [K⁺] typically reaches ∼11–13 mm with exhaustive exercise and substantially exceeds arterial plasma [K⁺], or plasma [K⁺] in the venous effluent from contracting muscles (McKenna et al., 2008) with reported gradients of ∼4.3 and ∼3.8 mm (Nielsen et al., 2004) and of 6.5 and 5.8 mm (Green et al., 2000). It therefore seems likely that muscle interstitial [K⁺] was similarly elevated 4–6 mm above our measured plasma values during these exhaustive contractions. If so, the interstitial [K⁺] might have reached 9–11 mm during FF and 10.5–12.5 mm during LC and therefore at critical levels that may contribute to impaired membrane excitability and muscle force reduction during intense contractions (McKenna et al., 2008). However, given the lack of changes in [K⁺] after 14 days of digoxin, presumably linked to preservation of muscle functional NKA, we cannot determine here whether there was any effect of elevated [K⁺] on fatigue.

The unchanged performance time and $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ during two-legged cycling to exhaustion at 90% $${\rm{V}}_{{{\rm{O}}}_{\rm{2}}{\rm{peak}}}$$ with DIG contrasts with findings of worsened incremental treadmill running time with digoxin in four healthy adults (Bruce et al., 1968), but is consistent with unchanged VO_2max reported with a serum [digoxin] of ∼1.0 nmol l⁻¹ in well-trained (Sundqvist et al., 1983) and in untrained men (Russell & Reeves, 1963). Two of these studies were performed after short-term, higher dose digitalisation (Bruce et al., 1968; Russell & Reeves, 1963), also suggesting that the extent of digitalisation cannot explain the different findings. In contrast, digitalisation in heart failure patients, with serum [digoxin] of 1.1 nmol l⁻¹, increased peak exercise VO₂ by 16%, consistent with its myocardial inotropic effects (Sullivan et al., 1989). No effects of digoxin were found here during FF on O₂ uptake by the contracting forearm muscles. Together, these all point to no effect of digoxin on performance during intense exercise. However, any possible effect might not have been detected due to NKA being preserved in skeletal muscle in these participants.

Competing interests

None.

Author contributions

M.M., H.K., D.C.S. and R.S. conceived and designed the research. S.S., A.P., C.G., X.G., M.B., J.A., A.G., C.S., K.M., K.C., S.F., J.L., D.C.S., R.S. and M.M. conducted the experiments. S.S., A.P., X.G., K.M., K.C. and M.M. analysed the data. M.M., S.S., A.P. and X.G. performed the statistical analyses. M.M., S.S., A.P. and X.G. wrote the initial manuscript draft. All authors read and approved the final manuscript, except H.K. (deceased) and X.G. (departed university and lost contact) who each read earlier drafts.

Funding

The study was funded by the National Health and Medical Research Council of Australia (No. 256603).