This study was performed in compliance with the Declaration of Helsinki and approval of the local ethics committee at Rostock University was obtained. All participants gave their written informed consent to take part.

Twenty participants (10 females, 22.8 ± 2.2 years, 62.0 ± 5.3 kg, 169.2 ± 5.8 cm; 10 males, 23.9 ± 1.3 years, 76.1 ± 6.6 kg, 182.5 ± 5.9 cm) volunteered in this study. All participants were non-habitual smokers, free of medication and abstained from any exhaustive exercise and alcohol for >24 h prior to the experiment. Furthermore, the consumption of caffeine or nicotine was not allowed during the night and on the morning of the experiment. After a medical clearance participants performed an incremental cycling test (ER 900, Ergoline, Germany) until volitional exhaustion (initial load 30 W + 30 W 5 · min⁻¹) at a constant pedaling rate (60 revolutions · min⁻¹). Beat-to-beat HR (S810i, Polar, Finland) was continuously measured throughout the test. Individual R-R recordings were screened for five stationary 180 s-segments that covered a wide physiological activity range from low to maximal HR. HRV analysis of the resulting 100 R-R interval segments was performed using the free software Kubios HRV 2.2 (University of Kuopio, Finland). All analyzed R-R recordings exhibited low noise (rate of erroneous R-R intervals below 5%). Before the computation, R-R time series were detrended and corrected for artifacts using adaptive filtering. Frequency analysis was performed using a Fast Fourier Transform (Welch's periodogram: 256 s-window with 50% overlap). Internal validity was assessed for the parasympathetic HRV indices lnHFP and lnRMSSD as well as for lnLFP—mirroring both parasympathetic and sympathetic effects on HR (Task Force of the European Society of Cardiology the North American Society of Pacing and Electrophysiology, 1996; Goldberger et al., 2006; Smith et al., 2013). Agreement of the 180 s- with the 60 s-HRV indices was assessed using estimation equation, mean and standardized mean bias, typical error of estimate, and validity correlation. Statistical indices were calculated and interpreted according to Hopkins (2015). In a second step, lnRMSSD_60s was plotted against the corresponding HR to visualize the behavior of the vagally mediated HRV during incremental exercise.

Fifty healthy participants (13 females, 25.7 ± 5.8 years, 65.1 ± 10.6 kg, 169.7 ± 5.3 cm; 37 males, 26.4 ± 7.1 years, 77.9 ± 9.6 kg, 182.6 ± 6.3 cm) volunteered in this study. Inclusion criteria as well as load protocol are identical with and described under (i) validation study. Immediately after exhaustion, workload was reduced to 30 W and participants kept pedaling at 60 revolutions · min⁻¹ for additional 10 min (active recovery = POST). Beat-to-beat HR (S810i, Polar, Finland), oxygen uptake (VO₂), and carbon dioxide output (VCO₂) (EOS Sprint, Jaeger, Germany) were continuously recorded during the test. Blood lactate concentration was determined from capillary blood samples (LactateScout, SensLab, Germany), drawn from the left ear lobe at the end of each stage. Cardiorespiratory data and blood lactate concentration at exhaustion are described in Table 1. HR_LT was determined at the lowest value of the lactate-equivalent (ratio of blood lactate concentration and VO₂) during the incremental test (Dickhuth et al., 1999). For the statistical testing, exercise intensity was dichotomized into LOW [ = HR ≤ HR at the individual lactate threshold (HR_LT)] and HIGH (= HR > HR_LT) for both, PRE and POST, periods.

Due to measurement artifacts, data of one male subject had to be excluded from further analysis. For all other participants (N = 49), the first five adjacent 60 s-segments during the immediate active recovery (POST 1–5) were matched with corresponding 60 s-segments yielding the same average HR during the incremental cycling test. Thus, POST 1 was matched with PRE 5, POST 2 with PRE 4 and so on (Figure 1). All analyzed R-R interval recordings exhibited low noise (rate of erroneous R-R intervals below 5%). LnRMSSD_60s was used to quantify vagal HR modulation. Before the computation, R-R time series were detrended and corrected for artifacts using adaptive filtering (Kubios HRV 2.2, University of Kuopio, Finland). Bland-Altman Plots (Bland and Altman, 2007) and Student's t-test statistics were applied to statistically verify the agreement of the matched R-R segments. A repeated-measures analysis of variance (IBM SPSS Statistics 22.0, USA) was carried out to test the effect of measurement time, experimental period (PRE vs. POST) and their interaction on lnRMSSD. To test the hypothesis of exercise intensity being a significant contributor to distinct HRV patterns at similar HR levels, experimental period was used as repeated measures factor (PRE vs. POST) and intensity as between subject factor (HIGH vs. LOW). Additionally, high frequency peak values (HF_peak), a measure of respiration rate (Thayer et al., 2002), was entered as covariate to control for potential bias. If data violated the assumption of sphericity, Greenhouse-Geisser corrected p-values and respective degrees of freedom were reported. For post-hoc pair-wise comparisons significance levels were adjusted using Bonferroni's correction.

Results of the internal validation study are provided in Table 2 and Figure 2. Mean biases between HRV_180s and HRV_60s as well as typical errors were very low and the correlations between the very short and the traditional short-term HRV very high.

LnHFP_60s, lnLFP_60s, and lnRMSSD_60s showed high agreement with their respective traditional short-term 180 s-HRV indices. Confidence limits of the validity correlation coefficients r were between 0.98 and 1.00.

Results of the external validity approach showed a three-phase behavior for lnRMSSD_60s. It almost linearly decreased to a minimum, plateauing at HR-values around 170 beats · min⁻¹, followed by a small rebound close to peak exercise (Figure 3).

Bland-Altman plot and t-test statistics confirmed the high correlation (r = 1.0, p < 0.001) and agreement (mean difference: 0.02 ± 1.29 ms, p = 0.792) between matched R-R segments, thus verifying the validity of the experimental approach (Figure 4). Despite identical HR for the matched R-R segments, lnRMSSD_60s was significantly affected by an interaction of measurement time × experimental period [F_{(4, 192)} = 30.733, p < 0.001, η_p² = 0.390]. Figure 5 illustrates the behavior of HR and lnRMSSD_60s during cycling at baseline, peak exercise, and during PRE and POST exhaustion cycling. HR and lnRMSSD_60s at peak exercise were significantly higher (p < 0.001) and lower (p = 0.017), respectively, if compared to the first minute of POST. While HR further decreased from POST 1 to POST 5, no additional rebound from POST 1 to POST 5 was evident for lnRMSSD_60s. In contrast, lnRMSSD_60s progressively decreased with increasing HR during PRE. Both HR and lnRMSSD_60s did not reach baseline values after 5 min of recovery (POST 5).

Individual workload responses during the different experimental periods may not be perfectly comparable between subjects due to differences in aerobic capacity, training state and individual recovery behavior. Therefore, metabolic state—estimated by individual lactate threshold—was used as a factor potentially influencing the behavior of lnRMSSD_60s. ANCOVA revealed that—despite no interaction effect on average R-R interval [F_{(1, 242)} = 1.888; p = 0.665; ņ_p² = 0.001]—lnRMSSD_60s was affected by the interaction of experimental period × intensity [F_{(1, 242)} = 30.233, p < 0.001, ņ_p² = 0.111; Figure 6].

Our data confirmed the internal validity of the 60 s HRV indices lnRMSSD_60s, lnLFP_60s, and lnHFP_60s compared to their respective 180 s indices. LnRMSSD_60s performed even better than the frequency domain indices. Based on the standardized mean bias as well as the standardized mean error, the difference between lnRMSSD_60s and lnRMSSD_180s was rated trivial. Thus, very short-term lnRMSSD_60s can be used interchangeably with traditional lnRMSSD_180s data segments. Furthermore, the behavior of lnRMSSD_60s during progressive exercise in our study was coherent with the current model of sympathetic-parasympathetic interaction during progressive exercise. According to this model and empirical data, vagal modulations of HR should be detectable until submaximal workloads around 150 · beats min⁻¹ (Ng et al., 2009; White and Raven, 2014). Thus, our results indirectly speak for the external validity of lnRMSSD_60s and its potential to reflect these vagal effects on HR control during submaximal exercise. Previous autonomic blocking experiments have also provided direct evidence for lnRMSSD_60s being a valid tool to describe parasympathetic reactivation following exercise (Goldberger et al., 2006).

The behavior of HR and lnRMMSD_60s during the final minute of cycling at maximum workload and during POST 1 indicated a strong parasympathetic reactivation immediately after peak exercise. In our experiment, RMSSD_60s plateaued at this increased level throughout the following POST periods. This is in contrast to the findings of Goldberger et al. (2006) who found RMSSD to progressively increase across the first 3 min of recovery after a shorter submaximal exercise protocol, but in line with other studies that have shown a slower recovery of vagally mediated HRV after maximal intensity exercise (see Michael et al. (2017) for review). Assuming that cardiac vagal activity plateaus on a high level as suggested by lnRMSSD_60s, the progressive HR decrease during POST can be mainly attributed to sympathetic withdrawal.

The main finding of an interaction effect of experimental period (PRE vs. POST exhaustion cycling) × intensity (LOW vs. HIGH) confirmed the hypothesis of a different HRV pattern during these kinds of exercise, despite an identical net effect on average R-R interval length. We cautiously conclude that lnRMSSD reflects a different chronotropic pattern generated by the autonomic network in the medulla oblongata, which arises from differences in central command and afferent feedback during the different experimental periods. When HR is still above HR_LT during early POST, vagal HR modulation is stronger compared to cycling at corresponding HRs during PRE. Thus, during the very first minutes after exhaustive exercise, net HR might result from a stronger sympathetic-parasympathetic coactivation if compared to high intensity cycling at similar HR (Berntson et al., 1991, 1993). An alternative explanation for a decreased vagally modulated HRV during late PRE might be a high or even saturated sympathetic tone (White and Raven, 2014). However, because direct nerve recordings in the intact human heart are impracticable and HRV reflects only the end-organ response that integrates different (autonomic) facilitatory and inhibitory effects on HR, the underlying mechanisms remain speculative.

Several neural feedback and feedforward mechanisms and their interaction may play a role in generating the distinct HRV pattern during HR-matched PRE and POST periods. The reactivation of vagal activity during early POST might counteract or even overpower the potential sympathetically mediated HR elevation by the metaboreflex (Fisher, 2014). The vagal rebound may occur due to the removal of inhibitory inputs from central command and the muscle mechanoreceptors and/or baroreflexes (Fisher, 2014 and references cited herein). This model is consistent with our results and the assumption of a pronounced sympathetic-parasympathetic coactivation during early POST.

Principally, afferent signaling from several types of receptors, including peripheral and central chemoreceptors, arterial baroreceptors, metabolite-, mechano-, and nocisensitive afferents within and around the working muscle (Fisher, 2014; Amann et al., 2015; Michelini et al., 2015) should profoundly differ during PRE and POST. For example, baroreceptor inputs change by the rapid decline of arterial blood pressure upon cessation of heavy exercise, inter alia by a decreased venous return due to a reduced activity of the skeletal muscle pump during POST (Takahashi et al., 2005). Along with decreases in central command at the beginning of POST, the cardiac baroreflex sensitivity increases via an augmented vagal activity (Ogoh et al., 2002, 2005; Gallagher et al., 2006). An alteration of the baroreflex sensitivity might in turn affect HRV (Di Rienzo et al., 1991). Additionally, altered feedback from muscle mechano- and metabolite-sensitive afferents during POST compared to PRE might contribute to different ANS outflow to the heart (Gladwell et al., 2005; Amann et al., 2010; Fisher, 2014). Mechanoreceptor inputs to the cardiovascular control centers in the brainstem decrease due to lesser muscular engagement during active recovery and would thereby alter cardiac autonomic drive (Crisafulli et al., 2003; Trinity et al., 2010). A stronger phasic compromised muscle blood flow in the exercising muscles (Holtz, 1996) during PRE might have also contributed to a distinct activity pattern of metabolite-sensitive afferents, resulting in a different autonomic HR modulation during PRE and POST (Alam and Smirk, 1938; O'Leary, 1993; Spranger et al., 2013; Fisher, 2014; Amann et al., 2015). Altered feedback from central and peripheral chemoreceptors during PRE and POST periods might have also modulated cardiac control. During mild to moderate dynamic exercise, arterial CO₂ pressure slightly increases toward hypercapnic tension, whereas it returns to resting values or even hypocapnic tension during heavy to exhaustive exercise due to the respiratory compensation of acidosis (Dempsey, 1988; Forster and Pan, 1988; Raven et al., 2013). The compensatory hyperventilation during heavy exercise can result in a respiratory alkalosis in the cerebral fluid that is expected to reduce central chemoreceptor discharge and to stimulate neural projections that inhibit the carotid afferent chemosensitive activity (Whipp and Ward, 1998). Taken together, it can be assumed that a different chemical milieu of the arterial blood and/or cerebral fluid during the matched PRE and POST leads to an altered signaling from peripheral and/or central chemosensitive afferents to the cardiovascular control centers in the medulla oblongata. Despite chemoreflexes being powerful modulators of the neural ventilatory and circulatory control, they are itself subject to important negative feedback interactions with baro- and pulmonary stretch receptors as well as to modulatory effects of circulating catecholamines (Kara et al., 2003; Mansukhani et al., 2015). The inhibitory effect of the baro- and pulmonary stretch reflexes in turn depends on arterial CO₂-pressure (Somers et al., 1989, 1991), which—in turn—depends on exercise intensity and the ventilatory response.

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.