Ten well-trained male triathletes (age 23.2 ± 4.5 years; height 180.8 ± 8.3 cm; weight 72.3 ± 6.6 kg) participated in the study. The participants had a training background in triathlon for 7 ± 2 years, and underwent intensive training during 14 ± 4 h/week for at least 3 years. The study was approved by the by the institutional ethics committee of Montpellier (France) and conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent before participation. They were familiarized with the incremental test procedure during training sessions performed prior to the testing, and were encouraged to give their best effort.

The subjects completed three different maximal incremental exercise tests to assess maximal exercise capacity and cardiorespiratory parameters during free swimming, arm cranking, and cycling. Peak oxygen uptake (V·O_2peak, ml·kg⁻¹·min⁻¹), ventilatory threshold (VT, % V·O_2peak), peak power output (PPO, W) for cycling and arm cranking, and mean swimming velocity (v = d/t, m·s⁻¹) were determined. Based on these parameters, workloads were calculated to eliciting equal relative intensity across the three exercise modes in a 6-min square-wave transition from rest to heavy-intensity for the assessment of on-transient V·O₂ kinetic parameters. All tests were performed in the same conditions, that is, at the same time of day (±2 h) and with identical pre-test warm up: 10 min at the intensity corresponding to 50% of V·O_2peak. The three different incremental tests were separated by a minimum of 72 h and completed in randomized order. The submaximal 6-min bouts mode were performed the day after the incremental test in the specific exercise mode. In the heavy-intensity bouts, parameters were estimated for the on-transient V·O₂ kinetics and compared across the three exercise modes. Figure 1 shows a schematic of the testing protocol used.

Gas exchanges during the three incremental tests were measured breath-by-breath (bxb) using a portable system (K4 b², Cosmed, Rome, Italy). During the swimming tests, the gas analyzer was attached to a swimming snorkel (Aquatrainer, Cosmed, Rome, Italy) that has been previously validated (Rodríguez et al., 2008) and used to measure gas exchange during swimming. In the incremental tests, bxb data were reduced to 30-s bin averages and V·O_2peak was determined as the highest 30-s V·O₂ average in either the swimming, arm cranking or cycling incremental tests. Using this protocol, the V·O_2peak was always attained at the final stage of the incremental test. VT was calculated using the v-slope method and coincided with V·O₂ that elicited the first departure from linearity in V· E/V·O₂. VT was expressed as % V·O_2peak in each exercise mode.

In the sub-maximal 6-min bouts, bxb data were first examined to exclude from the analysis the values greater than 3 SD from the local mean. Then data of the two square-wave transitions from rest to heavy-intensity were interpolated into 1-s values. In order to enhance the reliability in determining the parameters describing the V·O₂ kinetics, data were filtered through a 4th Butterworth low-pass digital filter with a cut-off frequency equal to 0.05 Hz (Robergs et al., 2010). To remove the influence of the cardiodynamic phase on the subsequent response we chose to exclude the first 20 s of data from the analysis. The on-transient V·O₂ kinetics was modeled according to the Equation (1): (1)V.O2(t)={V.O2basefort<tdpV.O2base+Ap(1−e−(t-tdp)/τp)for tdp≤t<tdsc     (primary component)V.O2base+Ap(1−e−(tdsc−tdp)/τp)+Asc(1−e−(t−tdsc)/τsc)for t≥tdsc    (slow component)} where V·O₂ (t) represents the weight-related V·O₂ at a given time t; V·O_2base is the baseline, resting V·O₂ (taken as the first 30-s average V·O₂ of the last minute before the start of exercise); td_p, τ_p, A_p represent the time delay, the time constant and the amplitude of the primary component, respectively; and td_sc, τ_sc, A_sc, represent the equivalent parameters for the slow component. A_p was used to determine the gain (G_p = A_p/P) of the primary component in cycling and arm cranking exercise.

Because the asymptotic value of the second function is not necessarily reached at the end of the exercise, the amplitude of the V·O₂ slow component was defined as: (2)Asc′=Asc(1−e−(te−tdsc)/τsc) where te was the time at the end of the exercise bout.

To estimate V·O₂ kinetics parameters, equations were fitted to the exercise data (Microsoft Excel Solver) using an iterative procedure (Generalized Reduced Gradient), by minimizing the sum of the mean squares of the differences between modeled and measured V·O₂ values. The relative contribution of the slow component to the net increase in V·O₂ (%A_sc') was calculated as the ratio between the amplitude of A_sc' and the total increase of V·O₂ during exercise. Effort intensity perception was rated recorded using the Borg 6–20 RPE Scale during all maximal incremental tests.

All values are reported as mean ± standard deviation (SD). After analysis of normality (Kolmogorov-Smirnov test) and homogeneity of variance (Levene test) of the data, analysis of variance with repeated measures (RM-ANOVA) was used to compare the values from the three incremental tests (swimming, arm cranking, and cycling), as well as the V·O₂ kinetics parameters obtained from the three square-wave transition from rest to heavy-intensity exercise. Significant main effects were subsequently analyzed using the Student-Newman-Keuls post-hoc test. Before the analysis, significance level was set at 5%.

The bootstrap method was used in the present study to assess the coefficient of variation (CV%) of the model parameters. It consists in resampling the original data set with replacement to create a number of “bootstrap replicate” data sets of the same size as the original data set. A random number generator to determine which data of the original data set will be included in a replicate data set was used. This was repeated 1,000 times, and the parameters estimated were retained. The CV of these 1,000 replicates was calculated.

The maximal workload reached at exhaustion was 353 ± 42 W in cycling, 142 ± 22 W in arm cranking, and 1.214 ± 0.064 m·s⁻¹ in swimming. Both weight-related and absolute V·O_2peak were higher in cycling (65.6 ± 4.0 ml·kg⁻¹·min⁻¹ and 4.7 ± 0.6 l.min⁻¹) than in arm cranking (48.7 ± 8.0 ml·kg⁻¹·min⁻¹ and 3.4 ± 0.4 l.min⁻¹; P < 0.001, 0.007, respectively) and swimming (53.0 ± 6.7 ml·kg⁻¹·min⁻¹ and 3.8 ± 0.3 l.min⁻¹; P < 0.001, 0.024, respectively). With the exception of weight-related V·O_2peak arm cranking-swimming (r = 0.77, P = 0.026), no correlations were found for cycling-arm cranking and cycling-swimming (r = 0.40 and 0.36; P = 0.24 and 0.31). Absolute V·O_2peak values (r = −0.12, 0.25, and 0.23; P = 0.77, 0.48 and 0.23, respectively) for cycling-arm cranking, cycling-swimming, and arm cranking-swimming were not correlated. No differences were found in maximal HR values (188 ± 7, 179 ± 8 and 175 ± 9 bpm; P > 0.05) or RPE (18.3 ± 0.6, 18.6 ± 0.6 and 17.3 ± 0.9 a.u.; P > 0.05), for cycling, arm cranking and swimming, respectively.

The average workload during the submaximal 6-min bouts was 266 ± 37 W in cycling, 98 ± 18 W in arm cranking, and 1.044 ± 0.048 m·s⁻¹ in swimming, corresponding to 75 ± 2, 70 ± 6, and 86 ± 4 percent of maximal workload (but to 76 ± 2, 69 ± 6, and 71 ± 4 percent of V·O_2peak) for cycling, arm cranking and swimming, respectively. V·O_2peak attained in these submaximal tests was higher in cycling (55.2 ± 5.9 ml·kg⁻¹·min⁻¹) than in arm cranking, or swimming (31.9 ± 5.6 and 41.5 ± 5.5 ml·kg⁻¹·min⁻¹, respectively; P < 0.001 and 0.001 respectively), but values were not correlated (r = 0.40, 0.36, and 0.25 for cycling-arm cranking, cycling-swimming, and arm cranking-swimming, P = 0.40, 0.36, and 0.29, respectively).

Figure 2 shows representative breath-by-breath and best-fit V·O₂ kinetics curves for the three exercise modes, expressed both in weight-related V·O₂ values (A), and in percentage of V·O₂ at the end of the exercise (B).

Table 1 shows the estimated V·O₂ kinetic parameters and the associated CV (%) during the 6-min square-wave transition from rest to heavy-intensity for all modes of exercise.

V·O_2base was not different between modes of exercise, but A_p was lower in both arm cranking and swimming compared with cycling (P < 0.001, for both), but higher in swimming compared with arm cranking (P < 0.005). Contrarily, gain was higher in arm cranking compared with cycling (P < 0.05). During swimming, subjects exhibited slower V·O₂ kinetics compared with arm cranking (P < 0.05) and cycling (P < 0.001), but faster during cycling compared with arm cranking (P < 0.05). The A_sc and A_sc' values were not different among exercise modes.

The mean V·O_2peak values observed are in accordance with data reported previously for swimmers (Rodríguez, 2000), however higher than healthy subjects familiarized with arm cranking exercise (Bernasconi et al., 2006). As would be expected, V·O_2peak in cycling was 26 and 19% higher than in arm cranking and swimming, respectively, in line with previous studies. Sawka (1986), in a review of 18 studies, reported an average V·O_2peak during arm cranking corresponding to ~22% of V·O_2peak during cycling. Two recent studies with well-trained triathletes also reported average V·O_2peak values during cycling that were 22% (Roels et al., 2005) and 36% (Barrero et al., 2014) higher compared with those in swimming. The smaller skeletal muscle mass involved during (mostly but not exclusively) upper body exercise is likely to be the main contributor to this ~20–35% lower V·O_2peak. Consequently, cardiorespiratory responses to swimming exercise will be influenced, as well, by other factors, such as HR_max and V·_E (Roels et al., 2005), which have been related with lower V·O_2peak measured in triathletes while swimming compared with cycling (Roels et al., 2005; Barrero et al., 2014) and running (Barrero et al., 2014). Although HR_max was not significantly different in our study, a trend for lower values in swimming (~175 bpm) and arm cranking (~179 bpm) compared with cycling (~188 bpm) was observed. Interestingly, in a study comparing trained swimmers and triathletes, the former exhibited 12% higher V·O_2peak in swimming than in cycling, whereas the opposite was found in the triathletes (22% higher V·O_2peak in cycling), highlighting the relevance of specific training (Roels et al., 2005). Therefore, both central and peripheral factors seem to interplay in determining the V·O_2peak attained during maximal exercise. Not withstanding the similar relative intensity performed during the heavy-intensity bouts, the subjects were also capable of performing well in the different exercise modes, and were similarly well trained in cycling and swimming;, therefore, the absence of differences in-between their V·O_2peak (~76, 69, and 71% in cycling, arm cranking and swimming, respectively) were observed.

Within the heavy intensity exercise domain, τ_p reflects the rate at which the V·O₂ response achieves the steady state before the appearance of the “excess” V·O₂ of the slow component phase. Previous studies have reported a slower adjustment of V·O₂ in arm exercise compared with leg exercise occurs, either when expressed as the half-time for the initial adaptation of V·O₂ at exercise onset (Cerretelli et al., 1977), or as the mean response time for the overall response (Koga et al., 1996; Koppo et al., 2002), as long as the power output does equally elevate blood lactate (Casaburi et al., 1992). This difference in the kinetics was corroborated with the present results; i.e., arm cranking and swimming showing a substantially slower adjustment in V·O₂, i.e. higher τ_p, compared with cycling exercise (~19 and 32 vs. ~12 s, respectively). When comparing highly trained subjects of four sport modalities exercising in a time-to-exhaustion test at 100% of V·O_2peak, Sousa et al. (2015) also found slower V·O₂ kinetics for swimming (~21 s) compared with cycling (~16 s), rowing (~12 s), and running (~10 s). The faster rate of V·O₂ increase compared with the present study can be attributed to the dissimilar intensity and duration of the exercise (i.e., ~3.7 min at 100% V·O_2max vs. 6 min of heavy intensity exercise). Similarly, Rodríguez et al. (2015), in a recent study with swimmers comparing 100-m all-out arm stroke, leg kick and whole stroke swims, showed also slower V·O₂ kinetics in arm exercise (~12 s) compared with whole stroke (~9 s) and leg kicking exercise (~10 s). Testing physical education students who practiced a variety of sports, including those that predominantly utilize the upper body muscles, Koppo et al. (2002) also showed slower V·O₂ kinetics for arm cranking exercise, compared with 6-min high intensity cycling exercise. Therefore, this study extends previous findings by corroborating that the V·O₂ kinetics pattern in-between arm and leg exercises is different, even with trained triathletes performing well in these exercise modes—and not only with physically active or highly trained subjects not specifically and concomitantly trained in these modes—are tested.

This new feature of our study may help to explain the possible mechanisms responsible for the slower V·O₂ kinetics in arm exercise. It was previously suggested by Casaburi et al. (1992) and Koga et al. (1996) that such a slower response during arm exercise could be due to the lower training status of the upper-body muscles compared with the lower-body muscles, since the untrained subjects tested in both studies could have had a higher conditioning status in these than those of the upper limbs. Collectively, this and the above cited studies suggest that the “training status” of the subject's musculature seems not to be, or at least, not to a great extent, determinant in the slower V·O₂ kinetics of the upper vs. the lower body exercise at heavy intensity. However, it should be taking into account that although the present study was conducted with triathletes capable of performing well in different modes of exercise, arm cranking involves a distinct muscle recruitment pattern. This latter could, in some extend, reflect a different movement efficiency, and therefore, influence V·O₂ kinetics.

Another possible explanation for the modality dependence of V·O₂ slower kinetics could rely on the skeletal muscle fiber type found in the upper and lower body. On the one hand, this is supported by the fact that upper-arm muscles contain generally a higher proportion of type II fibers —e.g., in tennis players (Sanchis-Moysi et al., 2010) or cross-country skiers (Mygind, 1995)—, the proportions of type II fibers was shown to be higher in the triceps brachii compared to vastus lateralis muscles. On the other hand, it has also been reported that the percentage of type I fibers in the gastrocnemius, vastus lateralis, and deltoid was high in all, but not different among muscles, while the muscle respiratory capacity (Q·O₂) and mean citrate synthase activity of the deltoid were lower than the gastrocnemius (Flynn et al., 1987). Type II fibers are metabolically less efficient, and not only the time constant of V·O₂ primary component is significantly longer in upper-body muscles with a predominance of type II fibers (Kushmerick et al., 1992), but also is longer in subjects with a high proportion of type II fibers in the vastus lateralis (Pringle et al., 2002). Muraki et al. (2004), by studying triceps brachii muscle oxygen saturation using NIRS during arm cranking and cycling exercise in young women, noted a faster increase in the respiratory exchange ratio and a lower VT in arm compared with leg exercise, suggesting accelerated anaerobic glycolysis. They also concluded that the oxidative capacity of the arm muscles was limited due to early O₂ utilization rather than poor O₂ supply to the exercising musculature. The influence of fiber types composition and V·O₂ primary component was previously addressed by Doria et al. (2011) in an elegant study, where the authors showed that a prolonged and active sojourn in hypoxia promoted an increase in the expression of slow isoforms of both heavy and light myosin subunits of vastus lateralis muscle in mountaineers. Moreover, this shift in muscle fiber type was accompanied by a decrease in mean response time of V·O₂ kinetics at the onset of step submaximal cycling exercise. Therefore, in swimming exercise, the larger recruitment of type II muscle fibers could also have contributed to the slower V·O₂ kinetics observed.

The fiber type controlling mechanism was also previously investigated within animal models. In a study with rats, muscles comprised of a high proportion of type II fibers have been shown to optimally increase their fractional O₂ extraction during intensive exercise compared with predominantly slow-twitch muscles, which may be important for ensuring high blood-myocyte O₂ flux and, therefore, a greater oxidative contribution to energetic requirements (McDonough et al., 2005). Moreover, for the same power output, “steady-state” V·O₂ was reported to be larger in arm than in leg exercise, implying the mechanical efficiency to be lower during arm exercise (Vokac et al., 1975). Of interest are the reports that in vitro V·O₂ on-kinetics of single muscle fibers at high intensity differ between high and intermediate oxidative muscle fiber with high oxidative muscle fibers yielding shorter time constant values than intermediate muscle fibers (30 vs. 50 s, respectively) during stimulated isometric contractions (Wüst et al., 2013). This difference seems to be in the same order of magnitude as that shown between arm cranking and cycling in the present study. Although one cannot infer full-body exercise responses from results on isolated fiber, this indirectly supports the speculation that fiber type differences underlie the differences in fast component between the different exercise modes. Collectively, these studies point out that one likely explanation for the slower V·O₂ kinetics in arm cranking and swimming heavy intensity exercise, compared with cycling, is the greater recruitment of type II muscle fibers in the upper-body musculature.

Another potential explanation for the slower rate of V·O₂ increase during swimming is related to the horizontal body position adopted during exercise. In the supine position, the blood flow to the working leg muscles is reduced, presumably as a consequence of lower arterial pressure in the legs when the effect of gravity (i.e., hydrostatic gradient effect) is removed (Koga et al., 1999). The same authors, Koga et al. (1999) reported that heavy exercise in the supine position was associated with a reduced amplitude of the fast component, which may be due, at least in part, to an attenuated early rise in HR in the supine position. The resultant fall in perfusion pressure to the active muscles leads to reduced exercise tolerance and slower V·O₂ kinetics compared to upright exercise (Hughson et al., 1991). Results from MacDonald et al. (1998), who found that slower V·O₂ kinetics at the onset of knee extension and flexion exercise in the supine compared with the upright position was accompanied by a slower increase of lower femoral artery blood flow, seem to support this hypothesis. In addition, the inability to produce maximal muscle contractions due to environmental constraints —i.e., body posture and propulsion is achieved by applying forces in a fluid medium— could also have limited the rate of increase in V·O₂ during swimming (Sousa et al., 2015). Other possible contributing mechanisms could be the involvement of static muscle actions for trunk and leg stabilization and differences in the muscular action regime (i.e., concentric vs. eccentric exercise), whether by themselves or related to fiber type recruitment or decreased muscle perfusion to the working muscles (see previous discussion).

Other differences were found in the fast component response: (i) the amplitude in cycling was ~37% larger than in swimming and ~50% greater than in arm cranking, which is proportional to the greater V·O_2peak in cycling compared with swimming (~20%) and arm cranking (~25%); and (ii) the gain was higher in arm cranking (~15 ml·min⁻¹·W⁻¹) compared with cycling (~11 ml·min⁻¹·W⁻¹). This latter trend corroborates previous findings for high intensity exercise, although the values reported in our study are slightly higher (Koppo et al., 2002). These findings are consistent with the hypothesis of a greater recruitment of type II fibers during arm cranking and swimming than during cycling exercise. In fact, Bernasconi et al. (2006) documented an increase in the electromyography signal of biceps brachii, triceps brachii, anterior deltoid, and infraspinatus muscles during heavy arm cranking exercise. In addition, Pringle et al. (2002) reported that G_p during heavy cycle exercise was positively correlated with the percentage of type I fibers in the vastus lateralis. Therefore, differences in fiber type composition between the muscles of the upper and lower body play an important role in the fast component gain.

No significant differences were found in the V·O_2sc amplitude (absolute and relative) or the time constant in-between modes of exercise, which does not corroborate previous reports conducted in cycling and arm cranking exercise (Koga et al., 1999). Multiple putative mechanisms of the V·O_2sc phenomenon have been postulated (Poole and Jones, 2012), the discussion of which exceeds the scope of this paper. However, in a recent work where a computer model of the skeletal muscle bioenergetic system was used, Korzeniewski and Zoladz (2015) postulated that the generation of V·O_2sc during heavy exercise could be related with: (i) the progressive inhibition of anaerobic glycolysis by accumulating protons (together with a slow decrease of the net creatine kinase reaction rate), (ii) the gradual increase of ATP usage during exercise, and (iii) perhaps, a decrease in the P/O ratio (ATP molecules/O₂ molecules). It follows that a greater recruitment of type II muscle fibers would theoretically have an effect on V·O_2sc. Koga et al. (1999) reported that heavy exercise in the supine position was associated with a greater amplitude of V·O_2sc but also, concomitantly, with a reduced amplitude of the fast component; this latter effect may be due, at least in part, to an attenuated early rise in HR in the supine position. These and other data seem to indicate that the amplitudes of the primary and slow components are sensitive to changes in muscle blood flow, which may influence, in turn, motor unit recruitment patterns. However, assuming that our subjects were equally well trained in all modes of exercise, we interpreted our results to indicate that either an equivalent anaerobic energy contribution to muscle metabolism, or that a similar progressive loss of muscle contractile efficiency—associated with the development of fatigue—occurred in the three exercise modalities, not measurably affecting the amplitude of the fast or slow components.

A limitation of the present study should be acknowledged, as the athletes performed only one transition to heavy exercise in each of the three modalities. Despite the number of transitions necessary for an adequate confidence in model parameter estimates is not clearly defined, it should be recognized that three (Spencer et al., 2011) or more exercise transitions might have improved such confidence. However, in endurance trained subjects, the number of repetitions may be reduced due to higher V·O_2max and lactate threshold, as suggested by Koga et al. (2005). Moreover, since we did not perform muscle biopsies on the subjects, it is impossible to determine the exact fiber type composition or phenotype of the arm vs. legs muscles. We agree that knowing the fiber type distribution and the oxidative capacity (e.g., mitochondrial content/efficiency, CS activity) would help clarifying some above-speculated mechanistic differences between the three activities modes used in this study.

In conclusion, this study provides further evidence of meaningful differences in the V·O₂ kinetics across exercise modalities, noted in the same group of athletes similarly trained in-between them. Specifically, (i) V·O_2peak in cycling was ~26 and ~19% higher than in arm cranking and swimming, respectively, mainly attributable to the smaller skeletal muscle mass involved during upper body exercise; (ii) albeit the same trained athletes performed heavy intensity cycling, arm cranking and swimming exercises at a similar relative intensity, substantially slower V·O₂ kinetics in swimming (~31 s) and arm cranking (~19 s) compared with cycling exercise (~12 s) was observed; (iii) also the amplitude of the primary component in cycling was ~37% larger than in swimming and ~50% greater than in arm cranking, in proportion to the greater V·O_2peak in cycling compared with swimming, although muscular efficiency (i.e., G_p) was higher in arm cranking compared with cycling; (iv) although the slow component was evident in all exercise modes, unlike in previous studies, no significant differences were noted in the amplitude of the slow component (absolute and relative); and (v) overall, it is suggested that the muscle mass involved, the greater and/or faster recruitment of type II muscle fibers during upper body exercise, the dynamics of fatigue onset, the horizontal position adopted during swimming, differences in muscle perfusion, and the involvement of trunk and lower-body stabilizing muscles are plausible explanations for the differences found in the V·O₂ kinetics patterns during heavy exercise, independently of the “training status” of the subjects.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.