The effect of passive lower limb training on heart rate asymmetry

Objective. Heart rate asymmetry (HRA) is an approach for quantitatively assessing the uneven distribution of heart rate accelerations and decelerations for sinus rhythm. We aimed to investigate whether automatic regulation led to HRA alternation during passive lower limb training. Approach. Thirty healthy participants were recruited in this study. The protocol included a baseline (Pre-E) and three passive lower limb training trials (E1, E2 and E3) with a randomized order. Several variance-based HRA variables were established. Heart rate variability (HRV) parameters, i.e. mean RR, SDNN, RMSSD, LF (n.u.), HF (n.u.) and VLF (ms2), and HRA variables, i.e. SD1a, SD1d, SD2a, SD2d, SDNNa and SDNNd, were calculated by using 5 min RR time series, as well as the normalized HRA variables, i.e. C1a, C1d, C2a, C2d, Ca and Cd. Main results. Our results showed that the performance of HRA was distinguished. The normalized HRA was observed with significant changes in E1, E2 and E3 compared to Pre -E. Moreover, parts of non-normalized HRA variables correlated with HRV parameters, which indicated that HRA might benefit in assessing cardiovascular modulation in passive lower limb training. Significance. In summary, this study suggested that passive training led to significant HRA alternation and the application of HRA gave us the possibility for autonomic assessment.


Introduction
Heart rate variability (HRV) has been extensively studied in clinical settings as it is modulated by the cardiovascular regulatory mechanisms, especially the interplay between sympathetic and parasympathetic activities (Mainardi 2009, Yan et al 2017. HRV is a non-invasive approach to studying heart rate by measuring the variation of RR-intervals, which beat at a non-constant frequency. There are different methods specifically being committed to analyzing the variance of RR time series, such as frequency-domain methods and nonlinear dynamic methods (Seely andMacklem 2004, Rajendra Acharya et al 2006).
Heart rate asymmetry (HRA), which is defined as the asymmetric distribution of heartbeat fluctuations, has been gaining momentum for the last few years. It takes into account the directionality of the RR-intervals, so that heart rate accelerations and decelerations can be distinguished (Guzik et al 2006, Piskorski and. It turns out that HRA is useful for quantifiable interpretations of chronic diseases (Guzik et al 2013, Karmakar et al 2014, Visnovcova et al 2014. Several measures have been developed to assess the asymmetry of heart rate, i.e. Guzik's index (GI) and Slope index (SI) uncovers the distance and phase angle information in the plot, respectively (Guzik et al 2006, while area index (AI) combines these two characteristics to perform the asymmetry (Yan et al 2017); Other measures to HRA includes the monotonic runs, Piskorski and Guzik proposed that RRintervals could be partitioned into acceleration and deceleration runs to count the numbers separately . All these measures are aiming at exploring the difference between accelerations and decelerations and interpreting the physiological meanings.
Despite literature regarding effects and associations between HRA metrics and pathological factors (Guzik et al 2013), few studies examined the HRA alterations response to passive limb training. Passive limb training is a kind of training in which the limbs are driven by external instruments or therapist. This training mode transmits instructions through proprioceptive information, so as to improve sports performance (Chiyohara et al 2020). Studies have shown that the increase of blood flow caused by individual passive leg movement can evaluate the peripheral blood vessels (Mortensen et al 2012, Trinity et al 2012. Several studies suggested that passive training elicited increased femoral blood flow and cardiac output, as well as a great change in HRV, which implied a great effect of passive training on cardiovascular and autonomic modulation (McDaniel et al 2010, Ives et al 2013, Shi et al 2016. Our previous studies had shown that there were significant differences in HRV indicators between different exercise levels (Shi et al 2016). However, the HRA mechanism in passive training remained unclear. The examination of changes in HRA measures with passive training can help better understanding the modulation of heartbeat, providing opportunities to comprehend the knowledge of how to standardize the training and make training effective for hemiplegia patients. In other words, HRA can be of great help to interpret how well the underlying control mechanism works and provide information in clinical settings.
In the present study, we focused on elucidating whether HRA differentiated between passive training trials and the baseline. Besides, we explored how HRA alteration is related to HRV indices and the physiological response to different passive limb training trials. This study can contribute to a more comprehensive understanding of the asymmetrical properties of heart rate and provide additional information about cardiovascular regulation during passive lower limb training.

Materials and methods
Participants Thirty healthy, physically active participants (15 males, 15 females) without a history of cardiovascular or neurological disorder were included in the study. When the p value was less than 0.05, the statistical power was greater than 0.8. The average age was 23±2.3 years and body mass index (BMI) was between 19 and 24 kg m −2 . All participants were nonsmokers, normotensive, and asymptomatic for respiratory disease. None of them were taking any medication. Caffeine and alcohol were refrained approximately 48 h prior to the data collection, as well as intense exercise. None of the participants had received professional lower limb strength training. The study conformed to the Declaration of Helsinki and was approved by the ethics committee at University of Shanghai for Science and Technology, Shanghai, China. All participants gave written informed consent to participate in this study.

Experimental protocol
All the measurements were undertaken in a quiet and temperature-controlled (25±3°C) room free from external distractions. Throughout the experiment, participants were asked to maintain the body position with his/her back against the experimental chair (figure 1). Prior to the experimental measurement, participants were required to seat comfortably for at least 15 min as an adaptation period. Subsequently, the electrocardiograph (ECG) signal was recorded for 10 min as a baseline (Pre-E). Then, ECG signals were recorded for three 10 min trials with the passive cycling machine (Mode: ZP-K600A, Tianjin Zepu Technology Co., Ltd, China) adjusted to 5 cycs min −1 (E1), 10 cycs min −1 (E2) and 15 cycs min −1 (E3), respectively. These three trials were performed in random order. Each trial was separated by at least a 10 min period. Before the experiment, the participant conducted a sufficient number of practice sessions so that they could flex their leg to a comfortable range of motion and keep the body stable to avoid any motion artifacts.

Data acquisition and processing
In the experiment, the Power-Lab/16sp system (Castle Hill, AD Instruments, Australia, 2002) was used to record and amplify the ECG signal. Three electrodes for the lead II ECG signal were respectively placed on the right wrist, left wrist and right leg for each participant. The sampling frequency was set at 1 kHz and the ECG signal filtered by a 1 Hz high-pass filter and a 45 Hz low-pass filter (Karmakar et al 2014, Shi et al 2016. After removing abnormal R peaks on the QRS complexes, the values of normal-to-normal cardiac interval corresponding to sinus rhythm were automatically measured and were subsequently exported for further analyses. All ECG datasets used for subsequent analysis were free of any form of morphologically abnormal beats. To eliminate the effect of muscular compensation and ensure steady-state conditions, only the last 5 min ECG signal of each session was analyzed. HRV and HRA measures Short-term HRV analysis has been proved to overcome the high non-stationarities problem and is suitable for studying short-time autonomic response (Steeds et al 2004, Sandercock et al 2007. In this study, time-domain parameters including mean RR, SDNN and the square root of the mean squared differences of successive RRintervals (RMSSD) were calculated. Frequency-domain parameters were derived from the power spectral analysis of the last 5 min RR time series by using a fast Fourier transform (FFT) algorithm. The power spectrum is typically parsed into three frequency ranges (Heart rate variability 1996): very low frequency (VLF, 0.003-0.04 Hz); low frequency (LF, 0.04-0.15 Hz); high frequency (HF, 0.15-0.4 Hz). LF and HF components were measured in normalized units(n.u.) to minimize the inter-individual variation (Xu et al 2013). VLF was presented in absolute values of power (ms 2 ).
There were several effective measures established to assess HRA . Poincaré plot is a tool for depicting and quantifying the distribution of RR-intervals on 2D coordinates. HRA observed from the Poincaré plot represents the presence of complex dynamics in the physiological signal. There are two important lines: the first one is the identity line, which goes across all points representing no change in the duration of consecutive RR-intervals (RR n =RR n+1 ); the second line is perpendicular to the identity line, and it crosses the centroid of the whole plot. According to these two lines, the short-term and long-term HRV parameters, SD1 and SD2 can be partitioned into parts on accelerations and decelerations by the following way (Piskorski and Guzik 2011) SD1 a is calculated from the perpendicular distance of points above the identity line and SD1 d is calculated from the perpendicular distance of points below the identity line, representing the short-term variance of contributions of accelerations and decelerations, respectively. Consequently, SD2 is separated in the same way by referencing the perpendicular line where SD2 a and SD2 d represent the long-term variance of the contributions of accelerations and decelerations, respectively. The standard deviation of normal to normal RR-intervals (SDNN), which reflects the total variability, is extracted by calculating the total variance of all RR-intervals where n is the total number of RR-intervals and RR is the mean RR time series. As described by Piskorski and Guzik (2007), Piskorski and Guzik (2011), short-term variability SD1 is the variance of projection of points along the identity line, and perpendicular projection of points leads to long-term variability SD2. Combining the two variables, there is a known formula (Piskorski and Guzik 2011) Using the formulas above, SDNN can be partitioned into two parts, too To deal with the high inter-individual differences, each pair of the above asymmetric descriptors is normalized (Piskorski and Guzik 2012). For short-term variance This partition can detect a more specific asymmetry resulting from accelerations and decelerations of sinus heart rate.
For HRA, the well-established variables including SD1 a/d , SD2 a/d , SDNN a/d , C1 a/d , C2 a/d and C a/d were calculated in this study. For these variance-based variables, SD1 a/d and C1 a/d were the contributions of acceleration and deceleration to short-term variance, respectively; SD2 a/d and C2 a/d were to long-term variance; SDNN a/d and C a/d were to total variance (Piskorski and Guzik 2012). It was worth noting that this interpretation referred to the construction of variables rather than the length of analyzed time series (Brennan et al 2001, Piskorski andGuzik 2012). Taken together, the combined HRV and HRA analysis of time series offered in-depth insight into the dynamics of the heart.

Statistical analysis
All quantitative variables were presented with mean±standard deviation (SD). Shapiro-Wilk test was used to investigate if the normality assumption was satisfied. Furthermore, one-way repeated measures ANOVA was performed for the normally distributed HRV indices among trials, i.e. Pre-E, E1, E2 and E3, followed by a Student-Newman-Keuls post hoc test. The difference between accelerations and decelerations of normalized paired HRA variables was examined using paired sample T-test. Additionally, the interclass correlation coefficient (ICC) was used for evaluating the reproducibility of non-normalized HRA variables in different trials. A nonparametric Spearman correlation test was performed to analyze the relation between non-normalized HRA variables and HRV parameters. All test results yielding p-value < 0.05 were considered statistically significant, and analyses were conducted by using SPSS statistical software (Version 24).

Results
As shown in figure 2, HRV parameters were calculated in a baseline (Pre-E) and three passive movement trials (E1, E2 and E3). Compared to Pre-E, the mean RR and HF (n.u.) in E1, E2 and E3 increased, and the SDNN, LF (n.u.) and VLF (ms 2 ) decreased. For mean RR, LF (n.u.) and VLF (ms 2 ), the statistical significance was found in E2 and E3 compared to Pre-E (p<0.05). For SDNN and HF (n.u.), the statistical significance was found in E1 and E2 (p < 0.05). For RMSSD, no differences were observed in three passive training trials (p>0.1).
The normalized HRA variables including C1 a , C1 d , C2 a , C2 d , C a and C d were examined and compared in figure 3. In Pre-E, no asymmetry was observed for these HRA variables, suggested that the behavior of heart rate accelerations and decelerations had no difference in the baseline. While in E1, E2 and E3, the HRA changes were observed with higher decelerations than accelerations. A significant difference was found between C1 a and C1 d in all three passive training trials, while the only significant difference between C2 a and C2 d was found in E1. For the difference between C a and C d , a significant difference existed in E1 and E3.
Additionally, 24 participants (80%) with SD1 a <SD1 d and 19 cases (63%) with SD2 a <SD2 d were observed and 21 participants (70%) exhibited SDNN a <SDNN d in E1, E2 and E3. The non-normalized HRA variables were summarized in table 1. Furthermore, ICC values in table 1 showed that the non-normalized HRA variables produced high reproducibility with ICC>0.8 (p<0.001). Figure 4 shows the comparison of Poincar´e plot under the four trials. The fitted ellipses in the figure can see different changes in acceleration and deceleration patterns under different trials. In all cases, SD2 was reduced under E1, E2, E3 as compared to the Pre-E. E1, E2, E3 were more fit compared to Pre-E. Figure 5 shows the frequency histograms of the time series. The negative numbers represent the deceleration part and positive numbers represent the acceleration part. The sample asymmetry was used to quantify the observed asymmetry with weighting parameters α=β=2 (Kovatchev et al 2003). As can be seen in the figure, the asymmetry was most obvious under Pre-E. The asymmetry changed as the motion changes. As the number of circles increased, the asymmetry decreased with an increasing contribution of deceleration part.
The Spearman correlation coefficients between non-normalized HRA variables and HRV parameters were summarized in table 2. A considerable part of the coefficients was statistically significant (p<0.05). Mean RR and RMSSD were found to have strong positive correlations with SD1 a/d , while the correlation between SD1 a/d and others was moderate. For SD2 a/d and SDNN a/d , the positive correlation with SDNN and VLF (ms 2 ) was strong, whereas the correlation with others was weak.

Discussion
In this study, we elucidated the response of HRA to passive lower limb training and explored the correlation between non-normalized HRA variables and HRV parameters. Three passive training trials were performed in a (b) corresponds to long-term asymmetry (C2a versus C2d); (c) corresponds to total asymmetry (Ca versus Cd). Paired t-test was used to detect the significance of differences between groups with a significance level of p<0.05. randomized order to examine the HRA alternation compared to the baseline. The well-established HRA variables, i.e. SD1 a/d , SD2 a/d , SDNN a/d , C1 a/d , C2 a/d and C a/d were examined. Additionally, the response of the autonomic mechanism to passive lower limb training was assessed by HRV parameters, i.e. mean RR, SDNN, RMSSD, LF (n.u.), HF (n.u.) and VLF (ms 2 ). Marked reduction in SDNN, LF (n.u.) and VLF (ms 2 ) of HRV parameters was found during passive lower limb exercise, while mean RR and HF (n.u.) significantly increased compared with the baseline. These findings were in agreement with previous studies (Weippert et al 2013, Shi et al 2016, Fouladi et al 2019 that the passive limb movement led to a shift of reduced sympathetic tone and increased parasympathetic activity. In this study, the relationships between HRA variables and HRV parameters were investigated, which indicated that HRA correlated with changes in sympathetic and parasympathetic activities (see table 2). Accordingly, short-term HRA variables SD1 a/d correlated highly with mean RR and RMSSD, which was a quantitative marker of parasympathetic predominance. The long-term and total HRA variables SD2 a/d and SDNN a/d had a strong positive correlation with SDNN and VLF (ms 2 ). SDNN marked both sympathetic and parasympathetic modulation (Malik et al 1996) and VLF (ms 2 ) was suggested as a reflection of sympathetic tone in some recent studies (D'Ascenzi et al 2014, Brar et al 2015, Thomas et al 2019. Other significant coefficients were moderate as implied that parts of the same regulation mechanisms were reflected on both HRA and HRV. We assumed that the reduction in SD1 a/d , SD2 a/d and SDNN a/d resulted from suppression of sympathetic activity and enhancement of parasympathetic regulation after passive training. However, this conclusion must be drawn cautiously in future work due to a considerable part of the weak coefficients, which indicted that HRA and HRV were correlated but had distinct metrics in assessing intricate and nonlinear autonomic system. Another reason for the weak correlation might due to the hemodynamic responses to passive training. It was worth noting that increased mean RR represented decreased heart rate, which showed that this protocol led to a non-enhancement in cardiac output (Venturelli et al 2014, Harvey 2016, Fouladi et al 2019. Passive lower limb training increased peripheral vasomotor and induced vascular congestion (Gifford and Richardson 2017). Acute inhibition of sensory afferent leads to decreased congestion induced by passive training of lower limbs through central response (Trinity et al 2010). The central command was believed to play an important role in cardiovascular response and muscle activation during exercise (Decety et al 1993, Calbet et al 2015. Moderate-to-vigorous intensity physical activity activates stronger sympathetic nervous system activity and higher vagal activity (Radtke et al 2013). Nevertheless, in the present study, the completely attempted muscle contraction made the  absence of central command activation (Shi et al 2016). This was thought to be responsible for the imbalance of heart rate accelerations and decelerations during passive movement.
Physiologically, HRA was related to the sinus node innervated by the autonomic nervous system, which modulated heart rhythm through neurotransmitter release (D'Souza et al 2019). This was the reason that the disturbed HRA could provide diagnostic evidence for many pathologies (Stein et al 2005, Porta et al 2008, Guzik et al 2013. The effect of sympathetic and vagal stimulation on cardiac periods can be reflected by the change of acceleration and deceleration mode. The reduction of deceleration capacity may be a sign of hidden cardiovascular disease. Acceleration capacity has been shown to predict dilated cardiomyopathy (Yan et al 2020).
In this study, more participants (80%) exhibited short-term asymmetry with SD1 d >SD1 a , which showed consistency with some researches. For example, Piskorski et al Guzik 2012, Piskorski et al 2019) suggested that most subjects showed the HRA phenomenon with SD1 d >SD1 a , SD2 d <SD2 a and SDNN d <SDNN a in 420 young healthy participants. However, more participants with SD2 d >SD2 a and SDNN d >SDNN a were found for long-term and total asymmetry, which was different from Piskorski et al 's research. This discrepancy may partially result from different lengths of the data used in the present study. Another reason may be due to a compensation mechanism in HRA, which was that a larger contribution of decelerations to short-term asymmetry was compensated by a larger contribution of accelerations to long-term asymmetry Guzik 2012). It could be physiologically interpreted that the prolongations of AH and HV intervals contributed larger than the shortenings to short-term variability of the intervals . On the other hand, the shift toward heart rate deceleration could stem from respiratory change. It was reasonable to believe that the respiratory maneuvers had an effect on HRA property, i.e. Klintworth et al found an increased HRA with inspiration/expiration ratio=1:1 or 2:1 by recording 5 min ECGs (Klintworth et al 2012). Their results suggested that there was a time difference for vagally and sympathetically mediated effects of the baroreceptor reflex on the heart following the related inspiration/ expiration ratio. Therefore, the asymmetry in accelerations and decelerations was mainly caused by the unsynchronized influence from the two factors. In the present study, the passive movement could promote the activation of vagal tone and the suppression of sympathetic activity (Fouladi et al 2019). On the other hand, it would increase oxygen consumption and metabolites production, which had profound effects on respiratory rate (Eckberg 2003). This consequence could lead to the imbalance of sympathovagal interaction and the enhancement of heart rate deceleration.
Note that our results suggested symmetry in the baseline (see figure 3) which was not in line with the previous study (Guzik et al 2006, Yan et al 2017. We assumed that it was the different uses of reference points that led to this consequence. It was reported that the performance of HRA, i.e. SI and AI, varied a lot when using different reference points (Yan et al 2019). Yan et al's study demonstrated that using the minimum RR-interval time series as reference points could achieve optimization of the result (Yan et al 2019). In the present study, the origin of the global coordinate was used. Therefore, there was no evidence indicating that SDNN a/d was a lack of sensitivity to sympathetic reduction during passive movement and this result remained more investigations. Furthermore, we also used AI and SI to assess the asymmetry phenomenon in response to passive training in this study as Yan et al did (Yan et al 2019), however, no asymmetric phenomenon was observed in four trials and there was no statistical significance (p>0.1). The outcomes proved that these variance-based HRA variables performed better than SI and AI in assessing autonomic regulation for passive lower limb training. Additionally, the ICC results underlined the suitability of HRA for application.
Despite these interesting results in our study, some limitations needed to be acknowledged. It is worth mentioning that human information processing is a multilayered process involving a variety of physiological reactions (Blechert et al 2016). A multi-measure approach was encouraged to explore the comprehensive and detailed assessment of cardiovascular modulation (Blechert et al 2016, Kiselev et al 2016. Besides, it was well known that nonlinear analysis of HRV were significant complementary explanations of cardiovascular response, which provided additional insights into physiological behavior (Zamuner et al 2015, He 2020. It was important to highlight that future studies should quantify some nonlinear parameters, such as Shannon entropy (Cavalcante Neto et al 2018) and symbolic analysis parameters (Porta et al 2009). Previous studies revealed that men and women had different automatic responses to passive training (Dutra et al 2013, Shi et al 2016. In addition, it was verified that HRA could be perturbed by pathologies (Porta et al 2008, Yan et al 2017. Hence, the gender influence on HRA should be taken into account in further studies, and also a larger population of healthy participants and patients from different age groups are required to confirm and extend these conclusions.
To sum up, this study investigated the influence of passive movement on HRA and demonstrated that HRA could provide information in assessing autonomic response to passive training. The asymmetry of accelerations and decelerations was a universal phenomenon during three passive training trials in this study. The variancebased HRA variables performed well in detecting asymmetric phenomenon and suggested suppression of sympathetic activity and enhancement of vagal tone responding to passive movement. The correlations between HRA and HRV parameters reinforced the point.