INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

SINCE THE EARLIEST MICRONEUROGRAPHIC recordings of sympathetic outflow in humans, the beat-by-beat modulation of muscle sympathetic nerve activity by the arterial baroreflex has been appreciated (13). This relationship has been used extensively over the last 20 years to probe the role of the arterial baroreflex in human health and disease (4). In practice, changes in muscle sympathetic nerve activity are assessed on a beat-by-beat basis during rapid changes in arterial pressure induced by intravenous boluses of vasoactive substances (e.g., nitroprusside and phenylephrine) (3, 5, 11). The data obtained during such protocols are typically analyzed "by hand" by an experienced microneurographer. This is disadvantageous for at least three reasons. First, manual analysis of muscle sympathetic nerve activity is subjective. In current practice, an observer visually scans the entire nerve recording and decides whether a burst has occurred after each heartbeat. The variability between observers has been reported to be 9% (8). Second, there is a threshold below which sympathetic bursts become lost in background noise and cannot be differentiated by an observer. Thus manual detection has an inherent bias against low levels of sympathetic activity. Last, the process of analyzing a microneurographic recording is time consuming and can take many hours.

Several attempts have been made to automate the analysis of muscle sympathetic nerve activity to overcome these limitations. In particular, two studies have validated the use of a triggered signal-average approach to analyze microneurographic recordings in humans (2, 10). Similar efforts have been made with animal nerve recordings (6). The signal average approach as developed in these prior studies was restricted to use during steady-state conditions in which muscle sympathetic nerve activity and arterial pressure are relatively stable. As such, the approach has not been used to assess the beat-by-beat changes of muscle sympathetic nerve activity that occur during dynamic baroreflex-mediated changes in sympathetic outflow. It was the goal of the present study to merge the methods currently used to assess beat-by-beat changes in muscle sympathetic nerve activity with a signal-averaging approach and overcome the problems inherent in manual analysis of baroreflex-mediated sympathetic responses. The hypothesis that this new approach would provide equivalent information in less time and with less subjectivity was tested.

The fundamental concept underlying this novel approach to signal analysis is the segregation of cardiac cycles by arterial pressure before application of signal averaging to the muscle sympathetic nerve activity recording. When this concept was applied to previously collected and analyzed data, the results showed a close agreement between the new method and the more tedious and subjective manual data analysis.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This is a retrospective study using data obtained during two prior studies (5, 9). The protocols used were approved by the human research committees of the Medical College of Virginia and the Hunter Holmes McGuire Department of Veterans Affairs Medical Center or the Institutional Review Board of the Mayo Clinic and Foundation. Each subject gave his or her informed consent before participation.

Protocol

Subject monitoring. Throughout the experiments, heart rate was determined from an electrocardiogram recording, beat-by-beat arterial pressure was measured in the finger by using a Finapres blood pressure monitor (model 2300, Ohmeda, Englewood, CO), and respiration was monitored continuously by using a bellows placed across the subject's abdomen and linked to a pressure transducer. Muscle sympathetic nerve activity was recorded via microneurography, as originally described by Sundlöf and Wallin (12). Briefly, multiunit, postganglionic muscle sympathetic nerve activity was recorded from the fibular (peroneal) nerve posterior to the fibular head with a tungsten microelectrode. The recorded signal was amplified 70,000- to 100,000-fold, band-pass filtered (700-2,000 Hz), rectified, and integrated (resistance-capacitance integrator circuit, time constant 0.1 s) by a nerve traffic analyzer (5).

Baroreflex control of sympathetic outflow. Sympathetic baroreflex responses were assessed by measuring muscle sympathetic nerve activity during arterial pressure changes induced by nitroprusside and phenylephrine as developed by Ebert and Cowley (3) and adapted by Halliwill et al. (5). After a 5-min baseline period, 100 µg sodium nitroprusside were given intravenously as a bolus, followed 1 min later by 150 µg phenylephrine HCl. This paradigm decreases arterial pressure ~15 mmHg below baseline levels and then increases it ~15 mmHg above baseline levels, over a short time course. This method of baroreflex testing has recently been reviewed and evaluated by Rudas et al. (11), who concluded it was a simple efficient method to quantify human sympathetic baroreflex responses.

Data Analysis

Baroreflex control of sympathetic outflow was determined from the relationship between muscle sympathetic nerve activity and diastolic pressure during vasoactive drug boluses (3, 5, 11). Diastolic pressure was used because muscle sympathetic nerve activity correlates closely with diastolic pressure but not with systolic pressure (11, 13). To perform a linear regression between nerve activity and pressure, values for nerve activity were first pooled over 3-mmHg pressure ranges. In other words, a single value for nerve activity was generated in 3-mmHg pressure increments from the lowest to the highest diastolic pressure seen during the baroreflex trial. This pooling procedure reduces the statistical impact of the inherent beat-by-beat variability in nerve activity due to nonbaroreflex influences (e.g., respiration) (3). The value for nerve activity in each 3-mmHg pressure range was determined by two independent methods: 1) manual detection and 2) segregated signal averaging.

Manual detection. The traditional method of manually detecting sympathetic bursts was performed. In brief, bursts are identified by their timing relative to the electrocardiogram QRS complex, height of the potential burst relative to the signal baseline, and overall shape of the potential burst. After bursts were identified by the investigator, the total integrated activity was calculated as the area under each manually detected burst. Any heartbeat not followed by a burst was assigned a total integrated activity of zero. Subsequently, heartbeats and their associated values for total integrated activity were pooled according to diastolic pressure and averaged as described previously (3, 5, 11).

Segregated signal averaging. Segregated signal averaging was performed by a computer program developed with LabVIEW (version 5.0, National Instruments, Austin, TX). The program first segregated heartbeats according to diastolic pressure and then averaged the muscle sympathetic nerve activity signals for heartbeats in each 3-mmHg pressure range (as outlined in Fig. 1). Specifically, a window of nerve activity that was 2.0 s in length and synchronized by the R wave of the electrocardiogram was signal averaged. The window was time shifted to account for the latency between R waves and sympathetic bursting. The duration of the shift was varied as needed from subject to subject but averaged 1.3 s. Subsequently, the extreme ends of the window, which represented sympathetic bursts occurring in the preceding and following heartbeats, were truncated, and the total integrated activity was determined as the area under the signal-averaged curve. This was done for each 3-mmHg pressure range. The points of truncation for all windows in a given trial were set at the obvious nadirs between the burst of interest and the preceding and following bursts, and the same timing was used for all windows in a given trial.

View larger version (31K): [in this window] [in a new window]   Fig. 1.   Example of segregated signal averaging in 1 representative subject. A: arterial pressure, electrocardiogram (ECG), and muscle sympathetic nerve activity (MSNA) tracings. As an example, 3-mmHg diastolic pressure range containing 74, 75, and 76 mmHg is highlighted by gray bar. R wave for each cardiac cycle within that pressure range is marked by arrows, and latency (1.4 s) and window (2.0 s) for first cardiac cycle are indicated with lines. B: MSNA tracings for 25 cardiac cycles taken from an extended segment of tracing in A that were within 3-mmHg diastolic pressure range containing 74, 75, and 76 mmHg are shown at left, and resulting signal average is shown at right. Each cardiac cycle within selected pressure range is included in the signal average. Area under solid section of the curve (right) was calculated to determine total integrated activity/beat.

Statistics

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Table 1 shows the characteristics of the subjects, including resting muscle sympathetic nerve activity. Figures 1 and 2 demonstrate the segregated signal-averaging method in one representative subject. Figure 1 demonstrates how cardiac cycles were first segregated according to diastolic pressure and then signal averaged. Figure 2 shows how the signal-average sympathetic nerve activity levels relate to arterial pressure, as anticipated because of baroreflex control of sympathetic outflow. Figure 3 compares the values for muscle sympathetic nerve activity, and Fig. 4 shows baroreflex sympathetic responses in one individual as assessed by the two methods. Manual detection and segregated signal averaging provided similar values for muscle sympathetic nerve activity that were highly correlated (r = 0.991, P < 0.05; Fig. 3). Across the 18 trials in 7 subjects, the correlation between muscle sympathetic nerve activity determined by the two methods was 0.914 ± 0.096 (SD). In the individual shown in Fig. 3, values for muscle sympathetic nerve activity across arterial pressures did not differ between the techniques. Consequently, the baroreflex relationships between muscle sympathetic nerve activity and diastolic pressure as determined by the two methods were nearly identical in this subject (see Fig. 4).

View larger version (19K): [in this window] [in a new window]   Fig. 2.   Example of segregated signal averaging in 1 representative subject. Shown in sequence are segregated signal averages for each 3-mmHg range (mean with range in parentheses). The well-known influence of baroreflex on sympathetic outflow is readily apparent by diminishing curve amplitude and area at higher pressures. In this figure, original 2-s windows have been truncated to 900 ms, which corresponds to region that was integrated to determine nerve activity.

View larger version (14K): [in this window] [in a new window]   Fig. 3.   Comparison of methods for determining MSNA in 1 representative subject. A: comparison of MSNA quantity in each pressure range between manual detection and segregated signal averaging. Values are similar, and correlation is high (r = 0.991, P < 0.05). B: Bland-Altman plot comparing manual detection and segregated signal averaging. Bias [mean difference () ± SD] in this individual was 3 ± 13 total integrated activity/beat (P > 0.05 vs. 0).

View larger version (16K): [in this window] [in a new window]   Fig. 4.   Comparison of methods for determining MSNA in 1 representative subject. Relationship between diastolic pressure and MSNA for both techniques is nearly identical in this subject. , Manual detection (r = 0.853, P < 0.05); , segregated signal averaging (r = 0.821, P < 0.05).

Across all tests, both methods provided values for muscle sympathetic nerve activity that correlated equally well with diastolic pressure [r for manual detection 0.868 ± 0.019 (SE) vs. r for segregated signal averaging 0.878 ± 0.026; P > 0.05]. Figure 5 compares the values for the slopes and intercepts of the baroreflex relationship between muscle sympathetic nerve activity and diastolic pressure as determined by the two methods. There was a high correlation between the baroreflex relationship slopes determined by manual detection and segregated signal averaging (r = 0.914, P < 0.05; see Fig. 5B), and there were no differences between the two methods [slope for manual detection 6.17 ± 0.91 (SE) vs. slope for segregated signal averaging, 5.98 ± 0.83 total integrated activity · beat¹ · mmHg¹; P = 0.60, see Fig. 5B].

View larger version (171717K): [in this window] [in a new window]   Fig. 5.   Comparison of methods for determining muscle MSNA in all subjects. A, left: comparison of MSNA quantity in each pressure range between manual detection and segregated signal averaging. Values are similar, and correlation is high (r = 0.902, P < 0.05). Right, Bland-Altman plot comparing manual detection and segregated signal averaging. Bias (mean difference ± SD) was 10 ± 24 total integrated activity/beat (P > 0.05 vs. 0). B, left: comparison of slopes of baroreflex relationships (MSNA vs. diastolic pressure) determined by manual detection and segregated signal averaging. Values are similar, and correlation is high (r = 0.914, P < 0.05). Right, Bland-Altman plot comparing slopes determined by manual detection and segregated signal averaging. Bias (mean difference ± SD) was 0.19 ± 1.52 total integrated activity · beat1 · mmHg1 (P > 0.05 vs. 0). C, left: comparison of intercepts of baroreflex relationships (MSNA vs. diastolic pressure) determined by manual detection and segregated signal averaging. Values are similar, and correlation is high (r = 0.926, P < 0.05). Right, Bland-Altman plot comparing intercepts determined by manual detection and segregated signal averaging. Bias (mean difference ± SD) was 1 ± 94 total integrated activity/beat (P > 0.05 vs. 0).

Similarly, there was a high correlation between the baroreflex relationship intercepts determined by manual detection and segregated signal averaging (r = 0.926, P < 0.05; see Fig. 5C), and there were no differences between the two methods [intercept for manual detection 427 ± 60 (SE) vs. intercept for segregated signal averaging 427 ± 57 total integrated activity/beat; P = 0.97, see Fig. 5C].

However, whereas baroreflex slopes did not differ (Fig. 6), there were some subtle differences between measurements of low levels of sympathetic activity between the two methods. Although nerve activity did not differ at the higher levels of nerve activity associated with the lowest pressure seen in each individual [manual detection 128 ± 18 (SE) vs. segregated signal averaging 138 ± 17 total integrated activity/beat; P = 0.17], nerve activity was lower with manual detection than with segregated signal averaging at the lower levels of nerve activity associated with higher pressures [manual detection 8 ± 2 (SE) vs. segregated signal averaging 18 ± 4 total integrated activity/beat; P < 0.05].

View larger version (15K): [in this window] [in a new window]   Fig. 6.   Group average relationship between diastolic pressure and MSNA for both techniques is nearly identical. Solid line, manual detection (MSNA = 427 ± 60  6.17 ± 0.91 · diastolic pressure); dashed line, segregated signal averaging (MSNA = 427 ± 57  5.98 ± 0.83 · diastolic pressure).

Analysis of a single baroreflex trial required from 2 to 4 h when done manually. In contrast, the same trial required <15 min when done by segregated signal averaging.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The goal of this investigation was to merge the methods currently used to assess beat-by-beat changes in muscle sympathetic nerve activity with a signal-averaging approach to overcome the limitations inherent in manual analysis of baroreflex-mediated sympathetic responses. The resulting novel approach to signal analysis, segregation of cardiac cycles by arterial pressure before application of signal averaging to the muscle sympathetic nerve activity recording, shows a close agreement with the more tedious and subjective manual data analysis.

The traditional approach of analyzing nerve recordings by hand is disadvantageous for at least three reasons. First, manual analysis of muscle sympathetic nerve activity is subjective, and inter- and intraobserver differences may be considerable. In our laboratory, we see interobserver variability of 10%, consistent with published estimates of 9% (8). The use of segregated signal averaging eliminates this subjectivity. Second, with manual detection of sympathetic bursts there is likely to be a threshold effect such that, below a certain level of nerve activity, sympathetic bursts become lost in background noise and are undetectable to the observer. Thus manual detection has an inherent bias against low levels of sympathetic activity. Indeed, these results show that at lower levels of sympathetic activity associated with higher arterial pressures, manual detection produced values for sympathetic nerve activity that were lower than those for segregated signal averaging. Last, the process of analyzing a microneurographic recording by hand is time consuming. The program used to perform segregated signal averaging dramatically reduced the time spent analyzing data.

It should be clear that reliability and validity of measuring baroreflex sympathetic responses should not be forsaken to accelerate the pace of data analysis. However, the results of this investigation suggest that the new method of segregated signal averaging provides equivalent measurement of changes in sympathetic activity in response to arterial baroreflex-mediated changes in sympathetic outflow.

Although this method was designed with beat-by-beat analysis of baroreflex responses explicitly in mind, segregated signal averaging may have other uses. For example, it should be possible to adapt this approach to the analysis of respiratory variations in sympathetic outflow (i.e., segregate by respiratory phase instead of diastolic pressure) or to the analysis of signals other than sympathetic activity.

Limitations

Segregated signal averaging does not provide a means to determine the number of sympathetic bursts that occur. However, on the basis of recent insights gained from recordings of single sympathetic neurons, it would seem that the traditional measurement of bursts per minute or bursts per 100 heartbeats has clear limitations. Increases in sympathetic nerve activity are achieved in three ways: 1) recruiting inactive sympathetic neurons, 2) increasing the percentage of cardiac cycles in which already active neurons fire, and 3) increasing the discharge frequency of already active neurons within a cardiac cycle (i.e., multiple firing) (7). Burst count is only responsive to one of these mechanisms: increasing the percentage of cardiac cycles in which already active neurons fire. In contrast, segregated signal averaging and other methods that quantify the area of bursts are responsive to all three mechanisms. Thus, whereas early research was limited to counting bursts because of the difficulty of quantifying the area of the sympathetic bursts, newer techniques overcome these difficulties and provide additional insight into the total level of sympathetic activity because they are sensitive to the recruiting of additional neurons and multiple firing of neurons. In this context, quantifying the number of bursts becomes unnecessary.

Segregated signal averaging is dependent on the fixed time between the R wave of the electrocardiogram and the occurrence of sympathetic nerve activity. However, this interval is not truly fixed and can vary by as much as 50 ms from beat to beat. The potential impact that a change in latency might have on segregated signal averaging results should be considered. Along these lines, 50 ms was the typical shift in latency that occurred between the highest and the lowest blood pressure bins studied in this data set. In contrast, the narrowest signal-averaged window that was analyzed in any of the subjects was 524 ms long. If one can assume that the bell-shaped curve of a signal-averaged burst is similar to the normal distribution, then shifting the curve 50 ms relative to a 500-ms window would have predictable results. Assuming that 95% of sympathetic activity is centered within a 500-ms window, then shifting the window 50 ms in either direction reduces the measured area from 95 to 93.2%. If 99% of sympathetic activity is centered within a 500-ms window, then shifting the window 50 ms in either direction reduces the measured area from 99 to 97.9%. Thus the impact of shifts in latency between the R wave and nerve activity would be, at most, a <2% underestimation of sympathetic activity.

In conclusion, segregated signal averaging as developed in the present study provides a valid alternative to by-hand analysis of beat-by-beat changes in muscle sympathetic nerve activity that occur during dynamic baroreflex-mediated changes in sympathetic outflow. This approach provides an objective, rapid method to analyze nerve recordings. However, this approach is not a means to derive information from nerve recordings of poor quality.