Increased heart rate fragmentation in those with Williams–Beuren syndrome suggests nonautonomic mechanistic contributors to sudden death risk

Abstract

Williams–Beuren syndrome (WBS) is a rare genetic condition caused by a chromosomal microdeletion at 7q11.23. It is a multisystem disorder characterized by distinct facies, intellectual disability, and supravalvar aortic stenosis (SVAS). Those with WBS are at increased risk of sudden death, but mechanisms underlying this remain poorly understood. We recently demonstrated autonomic abnormalities in those with WBS that are associated with increased susceptibility to arrhythmia and sudden cardiac death (SCD). A recently introduced method for heart rate variability (HRV) analysis called “heart rate fragmentation” (HRF) correlates with adverse cardiovascular events (CVEs) and death in studies where heart rate variability (HRV) failed to identify high-risk subjects. Some argue that HRF quantifies nonautonomic cardiovascular modulators. We, therefore, sought to apply HRF analysis to a WBS cohort to determine 1) if those with WBS show differences in HRF compared with healthy controls and 2) if HRF helps characterize HRV abnormalities in those with WBS. Similar to studies of those with coronary artery disease (CAD) and atherosclerosis, we found significantly higher HRF (4 out of 7 metrics) in those with WBS compared with healthy controls. Multivariable analyses showed a weak-to-moderate association between HRF and HRV, suggesting that HRF may reflect HRV characteristics not fully captured by traditional HRV metrics (autonomic markers). We also introduce a new metric inspired by HRF methodology, significant acute rate drop (SARD), which may detect vagal activity more directly. HRF and SARD may improve on traditional HRV measures to identify those at greatest risk for SCD both in those with WBS and in other populations.

NEW & NOTEWORTHY This work is the first to apply heart rate fragmentation analyses to individuals with Williams syndrome and posits that the heart rate fragmentation parameter W₃ may enable detection and investigation of phenomena underlying the proarrhythmic short-long-short RR interval sequences paradigm known to precede ventricular fibrillation and ventricular tachycardia. It also forwards a novel method for quantifying sinus arrhythmia and sinus pauses that likely correlate with parasympathetic activity.

INTRODUCTION

Williams–Beuren syndrome (WBS) is a rare genetic condition best known for supravalvar aortic stenosis (SVAS), distinct facies, and cognitive impairment (1). It is typically caused by a de novo microdeletion of 7q11.23 including the gene for elastin (ELN) (2). Elastin provides elastic recoil in blood vessels that increases the circulatory efficiency required for perfusion. Thus, elastin haploinsufficiency would be predicted to result in perfusion abnormalities in WBS. In fact, it is thought that elastin haploinsufficiency underlies most of the cardiovascular diseases seen in people with WBS (3). In addition to SVAS and branch pulmonary artery stenosis, it is well known that those with WBS have an increased risk of sudden death, particularly with the administration of anesthesia (4, 5). This risk is estimated to be 25–100 times higher than that seen in the general pediatric anesthesia population (6). Case reports and case series characterize sudden death in those with WBS in infants, children, and adults (6). Frequently, these episodes occur during or shortly after anesthesia administration. The mechanisms behind increased possible sudden cardiac death (SCD) risk in WBS, however, are poorly understood (6). Some cases are known to arise from coronary ostial stenosis (4). SCD episodes can be associated with hemodynamic collapse with and without ventricular fibrillation (VF). A unifying mechanism responsible for SCD in those with WBS has yet to be identified.

HRV Abnormalities in Those with WBS

We recently sought to investigate mechanisms underlying SCD in WBS using heart rate variability (HRV) (7). HRV has been used previously to investigate cardiovascular diseases, arrhythmia risk, and SCD (8–18). We demonstrated diminished HRV, decreased parasympathetic activity, and increased sympathetic activity in WBS using measures in the time, frequency, and nonlinear domains (7). These HRV abnormalities parallel those seen in other high-risk populations, including those following myocardial infarction, stroke, diabetes mellitus, heart failure, and left ventricular dysfunction (10, 19–23).

Heart Rate Fragmentation

In addition to autonomic modulators of arrhythmia, nonautonomic modulators contribute to both physiological and pathological sinus node function as well (24–26). A new noninvasive tool that attempts to quantify these modulators, called heart rate fragmentation (HRF), was motivated by a paradoxical increase in short-term HRV that may confound traditional HRV analyses in patients with diminished vagal activity (24, 27). This short-term variability is evidenced by beat-to-beat changes in the sign of heart rate (HR) acceleration. Some argue that sustained, frequent reversals in the direction of HR acceleration are inconsistent with parasympathetic control and instead reflective of an “erratic sinus rhythm” that creates more random variation (24, 27, 28). The term “fragmentation” is used to describe the frequency of such reversals, where more reversals signify increased fragmentation of an NN interval time series, where NN refers to the equivalent of an “RR interval” between normal (i.e., sinus) beats. Those who introduced HRF suggest that increased fragmentation signifies a breakdown of regulatory networks that may modulate the sinoatrial node (SAN) (24). Recent studies show that the HRF approach outperforms some traditional HRV analyses in predicting adverse cardiovascular events (CVEs) and cardiac mortality, as well as in distinguishing both clinical and subclinical coronary artery disease (CAD) from healthy individuals (24, 29, 30). Furthermore, others have reported limitations of traditional HRV measures, which have yet to fully reveal the biological correlates of each HRV measure (22, 23, 27, 31–34). Based on these arguments, we sought to apply HRF analysis to the WBS cohort to elucidate potential mechanisms underlying SCD in WBS that are not fully captured by conventional HRV measures. In doing so, we aim to identify additional prognostic indicators that may reflect SCD mechanisms seen in WBS and aid in SCD risk stratification. This study may also serve to demonstrate the broad applicability of HRF in predicting adverse cardiac outcomes in cardiovascular diseases. We hypothesized that those with WBS would exhibit increased fragmentation and that the HRF indices would not strongly correlate with traditional HRV metrics, including measures of autonomic activity.

Novel Metric “SARD” for Detection of Acute Drops in HR

In addition to studying HRF, we developed a novel metric called significant acute rate drop (SARD). Inspired by HRF methodology, SARD attempts to quantify acute vagally mediated compensatory events that are commonly identified clinically as sinus arrhythmia, sinus pauses, or sinus bradycardia events (22).

Study Design and Major Findings

We performed a cross-sectional analysis of HRF indices collected in a cohort of those with WBS versus age- and sex-matched healthy controls collected as part of an ongoing natural history study. We further conducted within-diagnosis correlation and multivariable regression analyses to evaluate the relationship, if any, between HRF and conventional HRV metrics [heart rate (HR), percentage of successive NN intervals that differ by >50 ms (pNN50), high-frequency power (HFP), short axis of Poincaré plot (SD1), standard deviation of NN intervals (SDNN), and sample entropy (SampEn)] that were previously shown to exhibit significant differences in the WBS cohort. We applied univariate linear regressions to evaluate the relationship between SARD and conventional HRV. The major findings of this study are 1) those with WBS exhibit significantly increased HRF compared with healthy controls (in 4 out of 7 parameters), which have been shown by others to be predictive of increased cardiovascular risk; 2) HRF may reflect HR modulators not fully captured by conventional HRV measures; and 3) those with WBS exhibit a significantly lower prevalence of SARDs compared with healthy controls, suggesting that SARD may serve as a novel, clinically informed quantitative measure.

MATERIALS AND METHODS

Human Subjects

Study approval was obtained from the National Heart, Lung, and Blood Institute Institutional Review Board. Written informed consent was provided by subjects or legal guardians. Assent was obtained when appropriate. As in our prior HRV study, data for all study participants were obtained under the Impact of Elastin Mediated Vascular Stiffness on End Organs protocol (NCT02840448) that, at the time of this analysis, included people with WBS (n = 65), people with WBS-like conditions (familial ELN-associated SVAS and smaller than typical WBS deletions sparing ELN n = 8), and control subjects (n = 41) (numbers reflect enrollment at the time) (7). All participants underwent molecular testing and those included in the WBS group all exhibited copy number 1 over ELN (the equivalent of ELN fluorescent in situ hybridization positive testing). All participants underwent a standard clinical history and physical exam.

Our current study includes those with WBS and control subjects, comprising the same cohort used in our prior HRV study of WBS (7). Children < 12 yr of age were excluded. Four controls were excluded because of divergent HRV results, including one with a documented anxiety disorder and one with spontaneous ventricular tachycardia. Some patients enrolled in the longitudinal study exhibited SVT and were therefore on chronic β-blocker therapy. This resulted in exclusion from the present study.

In addition to chronic β-blockade, exclusions were due to bundle branch morphology preventing accurate rhythm assessment, or failure of the recording to meet quality control metrics. When the remaining 23 subjects with WBS and 24 control subjects were matched for age (<2 yr in those <21 yr of age and <4 yr in those >21 yr of age) and sex, five subjects with WBS and six control subjects did not have a suitable match and were thus excluded. The remaining 18 subjects and 18 controls were included in this study. See Table 1 for patient demographics.

Table 1.

Patient demographics

	Control	WBS Subject	P Value
n	18	18
Female, %	72.2	72.2
Age, yr	[17.35–26.7]	[17.40–27.85]
BMI, kg/m2	[20.05–26.2]	[18.33–26.80]
EF >60%	100	100
%CT surgery	0	16.7	<0.01b

Values are percentages and medians [interquartile ranges]. P values were derived from a Mann–Whitney U test^b. BMI, body mass index; EF, ejection fraction; CT, cardiothoracic; WBS, Williams–Beuren syndrome.

Ambulatory ECG Recordings

Holter monitor recordings (24 h) were obtained from each participant. The recordings were reviewed by an on-site telemetry nurse (K.S.) and a pediatric cardiologist with advanced electrophysiology training (M.D.L.). Initial analysis was performed using Spacelabs Impresario (v.3.07.0158, Spacelabs Healthcare). Non-normal beats, as well as tracings of electrical and mechanical noise artifact, were excluded. Recordings comprising >5% artifact or with <22 h of nonartifactual data were excluded from the study. RStudio (v.1.1.463) was used to standardize the start time of each recording (.txt format) to 8:00 am. Supplemental Fig. S1 (all Supplemental figures are available at https://doi.org/10.6084/m9.figshare.25719708.v1) illustrates the data processing steps.

Heart Rate Variability and Heart Rate Fragmentation Analysis

HRV findings were previously published (7). They were performed as recommended by the 1996 European Society of Cardiology and North American Society of Pacing and Electrophysiology HRV task force (8) and the European Heart Rhythm Association (35). RMSSD and SD1 are numerically identical, so only SD1 has been included here (36). HRF analysis of the NN interval time series (in seconds) was performed using MATLAB 2021a. The methodology of HRF analysis, described in detail by Costa et al. (24, 29, 30), is conceptually illustrated in Fig. 1 using an example. In the NN interval time series, HR increases, decreases, or stays the same from beat to beat. These beat-to-beat changes in the sign of HR acceleration are represented by a series of +1 s, −1 s, and 0 s that correspond to an increase, decrease, or no change in HR, respectively (Fig. 1A). A change in HR was defined as a difference of >1/s from the previous interval, where “s” is the sampling frequency of the monitor (2,048 Hz in this case). When the sign of the acceleration changes from one beat to the next, this signifies an “inflection point” [measured by the percentage of inflection points (PIP) in the NN interval time series]. A change from +1 to −1 (or vice versa) is considered a “hard” inflection, whereas a change to or from 0 is coded as a “soft” inflection (measured by PIP^H and PIP^S, respectively) (Fig. 1B). An increase in PIP, PIP^H, or PIP^S signifies an increase in fragmentation.

Figure 1.

Heart rate (HR) fragmentation (HRF) methods can be graphically explained. A: direction of HR acceleration from beat to beat is represented by a “+1” (increase), “−1” (decrease), or “0” (no change) (green). B: inflection points (blue) occur where the direction of HR accelerations change. A “soft” inflection point (blue “S”) occurs when the acceleration direction changes to or from “0,” and a “hard” inflection point (blue “H”) occurs when the direction changes from “+1” to “−1” or vice versa. C: series of HR acceleration directions (shown in A) can be analyzed in successive groups of four beats, termed “words” to identify patterns occurring at or around the respiratory frequency (3–5 s). Within each “word,” there are either 3, 2, 1, or 0 inflection points, represented by W₃ (D), W₂ (E), W₁ (F), and W₀ (not shown), respectively. A greater number of inflection points per word reflects greater fragmentation.

To analyze patterns occurring at or around the respiratory frequency (3–5 s), we next used symbolic sequences of four beats, termed “words,” as previously described (29). Each successive word is extracted by moving the frame (length 4) to the next beat (Fig. 1C). Each word has 0, 1, 2, or 3 inflection points, belonging to words groups W₀, W₁, W₂, and W₃, respectively (Fig. 1, D–F). The distribution of word groups is then determined, with each group reported as a percentage (%). The degree of fragmentation increases with an increasing number of inflection points (i.e., W₃ is more fragmented than W₂). Thus, words with only 0 or 1 inflection point are considered more “fluent,” whereas words with 2 or 3 inflection points are more fragmented. Consistent with its initial description, we considered W₂ words to be intermediate in fluency and focused on W₃ words to be a more robust indicator of fragmentation as previous studies showed two inflection point word subgroups less discriminatory of fragmentation versus fluency (29).

All HRF indices were analyzed over the 24-h period. In addition, daytime (11:00 am–8:00 pm) and nighttime were analyzed (12:00 am–5:00 am) separately. These time windows were chosen to best capture periods where we expected all participants to be asleep/awake. Table 2 displays the definitions of HRF and traditional HRV metrics used in this study.

Table 2.

Definitions of abbreviations for HRV and HRF metrics

Abbreviations	Definitions
Traditional heart rate variability (HRV)
HR	Heart rate (in beats/min)
pNN50	Percentage of successive NN intervals that differ by >50 ms (%)
HF power	Absolute power of the high-frequency band (0.15–0.4 Hz) (ms2)
SD1	Short axis of Poincaré plot (ms)
Sample Entropy (SampEn)	Measure of regularity and complexity of the NN interval time series
SDNN	Standard deviation of NN intervals (ms)
Heart rate fragmentation (HRF)
PIP	Percentage of inflection points in the NN interval time series
PIPH and PIPS	Percentage of beats that represent “hard” (“−1” to “+1” or vice versa) and “soft” (to or from “0”) inflection points in the NN interval time series (%)
W0–3	Percentage of words in the NN interval time series that contain 0, 1, 2, or 3 inflection points, respectively (%)
Novel metric
SARD	In an NN interval time series, a drop in HR by ≥20% from the previous beat that is sustained for four beats

SARD: A Novel HRV Metric

We introduce a novel metric called significant acute rate drop (SARD), which measures the percentage of beats in an NN interval time series that represents a decrease in HR by 20% from the previous beat and is sustained for at least two beats. Supplemental Fig. S2 displays a representative ECG illustrating this concept. The percentage of SARDs was calculated over the full 24-h period and for daytime and nighttime segments.

Statistical Analysis

Statistical analyses were performed using GraphPad Prism (v.9.4.1 for Mac) and R software (RStudio, v.1.1.463). For patient demographics (Table 1), differences between WBS subjects and healthy controls were assessed using Mann–Whitney tests and χ² tests. Differences in HRF indices between the two groups were evaluated using Mann–Whitney tests. A P value < 0.05 was considered statistically significant. Scatterplots are displayed with bars signifying medians and interquartile ranges [25th–75th percentile].

Because WBS is a rare disease, our sample size is necessarily small and there are many predictor variables under consideration. Therefore, two different statistical approaches were used: logistic regression and principal component analysis (37, 38). Although these two methods do not yield identical results, the results should be considered exploratory since they put forth hypotheses that can be tested in the future. We anticipate a prospective multicenter study will be performed where variables identified in this current analysis can be more fully and properly validated. Multivariable logistic regression models were used to assess the discrimination of case status by individual HRF parameters given HRV parameters, age, and sex. To reduce overfitting, stepwise backward selection was subsequently applied to the models until all remaining variables had P ≤ 0.2. The cutoff was chosen to include relevant parameters while still reducing the number of variables in the final models in this small cohort.

We reported Spearman correlation coefficients for cases and controls separately to address the pairwise association of SARD and HRF parameters with conventional HRV measures. Graphical methods are used extensively to show case-control comparisons and their directionality.

In addition, we conducted multivariable linear regression in combination with principal component analysis (PCA) for the log-transformed and standardized parameters (37, 38). We fitted linear regression models of the transformed HRF parameters against principal components of the transformed HRV parameters for cases and controls separately. Models were adjusted for age and sex. To avoid overfitting, we reduced the dimension of the multivariate linear regression models by backward selection. Components with a P value > 0.1 were thereby successively removed. From histograms and q–q plots, we found a reasonable fit for all regression models.

We also investigated the linear association of SARD and the HRV parameters with univariate linear regression models, adjusted for age and sex. Log transformation of HRF and HRV parameters resulted in an improved linear fit of the models.

RESULTS

Demographics

The demographics for subjects and controls are summarized in Table 1 (7). CT surgery was performed in three subjects (16.7%). Supplemental Table S1 (all Supplemental tables are available at https://doi.org/10.6084/m9.figshare.25588089.v1) provides additional details regarding subject group features. All participants had normal cardiac systolic function (ejection fraction, EF ≥ 60%). Of the participants with WBS, three had history of aortic surgery (with no significant residual stenosis, i.e., no patient with discrete SVAS > 40 mmHg) and one had a pulmonary artery surgery. Vessel stenosis was absent in controls. So far, no patients in this study have exhibited ventricular tachycardia, ventricular fibrillation, or high-grade atrioventricular block by ambulatory ECG we studied.

HRF in a 24-h Period

The differences in the percentage of inflection points (PIP, PIP^S, and PIP^H) between WBS subjects and controls over a 24-h recording period are shown in Fig. 2 (subjects with a history of surgery for SVAS, i.e., moderate to severe discrete stenosis, marked as triangles). Those with WBS had a higher PIP compared with controls, signifying a higher degree of HR fragmentation (Fig. 2A). Further classification of inflection points as either “soft” (PIP^S) or “hard” (PIP^H) reveals that those with WBS have a significantly higher PIP^S compared with controls (Fig. 2B). A higher PIP^S signifies that those with WBS have more changes in HR acceleration to and from “zero” accelerations (“−1” or “+1” to/from “0”). There is no significant difference in PIP^H (i.e., accelerations “−1” to “+1” or “+1” to “−1”) between groups (Fig. 2C).

Figure 2.

Percentage of inflection point (PIP) analyses demonstrate increased heart rate (HR) fragmentation in subjects with Williams–Beuren syndrome (WBS) compared with age- and sex-matched controls. A: PIP is significantly higher in those with WBS compared with controls. B: percentages of soft inflection points in NN time series (PIP^S, change to or from “0”) is higher in those with WBS compared with controls. C: percentages of hard inflection points in NN time series [PIP^H, “−1” to “+1” or vice versa) is not significantly different between the two groups, however. Data were collected from 24-h ECG of WBS subjects (n = 18) and age- and sex-matched controls (n = 18). Open triangles represent patients who have undergone CT surgery.

Figure 3 displays the percentages of each “word” type, where a “word” is a group of four successive beats with 0 (W₀), 1 (W₁), 2 (W₂), or 3 (W₃) inflection points (Fig. 1). An increasing number of inflection points corresponds to an increase in the amount of fragmentation, or loss of “fluency,” in a word. W₀ (no fragmentation), shows no significant difference between WBS subjects and controls (Fig. 3A). W₁ and W₂ are significantly lower in those with WBS compared with controls (Fig. 3, B and C). W₃ is significantly higher in those with WBS compared with controls (Fig. 3D). Further comparison of the distribution of word groups shows that WBS subjects have a greater proportion of highly fragmented words (W₃) and a smaller proportion of words with minimal fragmentation (W₁) when compared with matched controls (Fig. 3, E and F).

Figure 3.

Distribution of word types in Williams–Beuren syndrome (WBS) features more highly fragmented words (W₃) and fewer minimally fragmented words (W₁) compared with controls. Each word in the NN time series (length 4 beats) has 0, 1, 2, or 3 inflection points, where an increase in the number of inflections corresponds to increased fragmentation. A: no significant difference in W₀ (percentage of words with no inflection points) between groups. B and C: W₁ (minimal fragmentation) and W₂ (moderate fragmentation) are higher in controls compared with WBS subjects. D: W₃ (high fragmentation) is higher in those with WBS compared with controls. E and F: respective distribution of word types for those with WBS subjects (E) and controls (F). Percentages are reported as median of each word group. Open triangles represent patients who have undergone CT surgery.

HRF: Daytime versus Nighttime

HRF analyses of daytime (11:00 am–8:00 pm) and nighttime segments (12:00 am–5:00 am) are shown in Supplemental Fig. S3. No significant differences were observed between subjects’ and controls’ daytime and nighttime patterns of metric expression.

Discrimination of Case Status Using HRF and HRV

Table 3 displays the results of a logistic regression model for the prediction of case status (WBS vs. control) given all HRV parameters (and age and sex) and individual HRF parameters. After backward selection, four HRF parameters remained in the model: PIP, PIP^H, PIP^S, and W₃. PIP^H, PIP^S, and W₃ showed a statistically significant ability to distinguish WBS cases from controls with HRV parameters included in the model (P < 0.05, Table 3), whereas PIP showed a trend toward significance (P = 0.061, Table 3).

Table 3.

Logistic regression models of case status (WBS vs. control) given HRV and individual HRF parameters

Parameter	Coefficient	P Value
PIP	0.344	0.061
PIPH	0.319	0.030
PIPS	−1.727	0.021
W 0	Not in final model
W 1	Not in final model
W 2	Not in final model
W 3	0.350	0.014

Values are coefficients and P values from logistic regression models predicting cases status given heart rate variability (HRV) [heart rate (HR), percentage of successive NN intervals that differ by >50 ms (%) (pNN50), high-frequency (HF) power, short axis of Poincaré plot (ms) (SD1), sample entropy (SampEn), and standard deviation of NN intervals (SDNN)], age and sex, and individual heart rate fragmentation (HRF) parameters. After stepwise backward selection, W₀, W₁, and W₂ were not included in the model. PIP, percentage of inflection points in NN time series; PIP^H, percentages of hard inflection points in NN time series; PIP^S, percentages of soft inflection points in NN time series; W_j, percentage of “words” with j inflection points; WBS, Williams–Beuren syndrome.

Supplemental Fig. S4 displays histograms illustrating the differentiation of cases and controls using both HRF (and SARD) and traditional HRV measures.

Relationship between HRF Indices and Traditional HRV Measures

The pairwise associations between HRF and HRV parameters were first assessed using Spearman correlations for cases and controls separately (Fig. 4). In both groups, parasympathetic markers HF power and SD1 showed a moderate-to-very strong negative correlation with HRF indices PIP, PIP^S, and W_3, (marker of increased fragmentation) and a weak-to-moderate positive correlation with W₁ (marker of low fragmentation). The third parasympathetic marker, pNN50, showed a similar relationship with the HRF markers in the control group. However, in those with WBS, pNN50 showed the opposite relationship with the same HRF markers (positive correlation with PIP and W₃). Both groups showed a positive correlation of HR with PIP^S (moderate to strong) and W₃ (weak). The remaining correlations between HRF and HR were inconsistent in direction and strength between controls and those with WBS. HRF did not show any strong correlations with SampEn (nonlinear HRV marker) or SDNN (marker of overall HRV) in either group. Overall, Spearman correlations showed mostly weak-to-moderate correlations with HRV. Supplemental Table S2 displays the r_s values with accompanying P values, which varied in significance. In both groups, correlations with PIP^S showed the strongest evidence for true correlations (lowest P values). Supplemental Fig. S5 displays scatterplots of HRF markers PIP, W₁, and W₃ versus individual HRV parameters with fitted linear regression lines to illustrate pairwise associations for each group.

Figure 4.

Spearman correlation analyses generally show a weak-to-moderate association between heart rate (HR) variability (HRV) and HR fragmentation (HRF) parameters. A: In subjects with Williams–Beuren syndrome (WBS), HR (sympathetic marker) had a moderately positive correlation with percentage of inflection points in NN time series (PIP), percentage of soft inflection points in NN time series (PIP^S), and W₃, as well as a moderately negative correlation with W₀. percentage of successive NN intervals that differ by >50 ms (%) (pNN50) had a moderately positive correlation with PIP and W₃. High-frequency (HF) power and short axis of Poincaré plot (ms) (SD1) had a moderate-to-strong negative correlation with PIP, PIP^S, and W₃. Sample entropy (SampEn) had a moderately negative correlation with PIP, PIP^S, and W₃. Standard deviation of NN intervals (SDNN) showed a moderately negative correlation with PIP^S. Significant acute rate drops (SARD) had a strong positive correlation with HF power and SD1. B: in healthy controls, HR positively correlated with PIP^S. W₃ and negatively correlated with W₂. pNN50 shows moderate-to-strong negative correlations with PIP, PIP^S, and W₃. HF power and SD1 showed moderate-to-strong negative correlations with PIP, PIP^S, and W₃. SampEn shows a moderate positive correlation with percentage of hard inflection points in NN time ser (PIP^H) and W₂, as well as a moderate negative correlation with W₀. The following criteria were used for reporting the strength of correlations: |r_s| < 0.4, weak; 0.4 ≤ |r_s| < 0.7, moderate; 0.7 ≤ |r_s| < 0.85, strong; and |r_s| ≥ 0.85, very strong, where r_s is Spearman correlation coefficient. W_j, percentage of “words” with j inflection points.

To investigate the association between HRF and HRV parameters more generally, we applied PCA and multivariable linear regression analyses. Supplemental Table S3 displays the PCA composition and the proportion of the variance explained by each. Supplemental Table S4 displays the PCA composition of the linear regression after backward selection. Table 4 shows the adjusted R² values for linear regression models that were fitted to explain HRF parameters by the principal components of conventional HRV parameters. The R² values indicate the extent to which each parameter is explained by HRV. ∼62% of the variation in PIP is explained by the principal components of HRV in those with WBS (moderate association), potentially implying a sizable signal not covered by conventional HRV metrics phenomenon (38%), but is captured by PIP. This compares with 33% of the variation in PIP explained by HRV in controls. For 4 out of 7 HRF parameters (PIP^H, W₀, W₂, W₃), less than 50% of the variation in the HRF parameter is explained by conventional HRV, suggesting a weak-to-moderate association (all R² < 0.5) of HRF with HRV. However, PIP^S and W₁ are exceptions: >90% of the variation in PIP^S in both groups is explained by conventional HRV, suggesting a strong association between PIP^S and HRV. W₁ also displays a strong association with HRV in controls (R² = 0.812).

Table 4.

Adjusted R² (coefficient of determination) for linear regression models fit to explain specific HRF parameters by the principal components of HRV

HRF Parameter	R 2
	WBS	Controls
PIP	0.620	0.331
PIPH	0.437	0.257
PIPS	0.955	0.952
W 0	0.481	0.460
W 1	0.812	0.154
W 2	0.225	0.299
W 3	0.431	0.392

Heart rate variability (HRV) and heart rate fragmentation (HRF) parameters were log-transformed. Models were adjusted for age and sex. PIP, percentage of inflection points in NN time series; PIP^H, percentages of hard inflection points in NN time series; PIP^S, percentages of soft inflection points in NN time series; W_j, percentage of “words” with j inflection points; WBS, Williams–Beuren syndrome.

SARD: New Metric

The prevalence of SARDs in those with WBS and healthy controls over 24 h, daytime, and nighttime is shown in Fig. 5. Those with WBS show a significantly lower percentage of SARDs compared with healthy controls during all three time periods.

Figure 5.

Those with Williams–Beuren syndrome (WBS) exhibit a higher percentage of significant acute rate drops (SARDs) compared with healthy controls. Those with WBS have a significantly higher percentage of SARDs compared with healthy controls during the 24-h period (A), daytime (B), and nighttime (C). SARD is defined as a decrease in heart rate (HR) by 20% compared with the previous beat that is sustained for at least 2 beats. Daytime, 11:00 am–8:00 pm; nighttime, 12:00 am–5:00 am.

Univariate linear regression models of SARD versus HRV metrics are shown in Table 5 and are illustrated with scatterplots in Fig. 6. Overall, SARD showed variable association with the different parasympathetic markers (pNN50, HF power, SD1). On the one hand, SARD was strongly associated with the nonlinear general HRV and parasympathetic marker SD1 in both cases and controls (R² = 0.70 and 0.82), with scatterplots illustrating a positive linear relationship (Fig. 6D). However, SARD showed weaker associations with HF power and pNN50 in both groups (R² = 0.11 to 0.47) (Table 5). SARD was negatively associated with pNN50 in those with WBS, whereas it was positively associated with pNN50 in controls (Fig. 6B). SARD showed a weak association with SDNN (marker of overall HRV) in WBS and close to no association in controls.

Table 5.

Linear regression of SARD given HRV parameters

	WBS				Controls
	β	SEE	P value	R 2	β	SEE	P value	R 2
Y = log(SARD)
log(HR)	−0.395	0.248	0.133	−0.003	−0.369	0.299	0.044	0.238
log(pNN50)	−0.495	0.229	0.048	0.112	0.859	0.260	0.005	0.353
log(HF power)	0.666	0.211	0.007	0.307	0.804	0.201	0.001	0.465
log(SampEn)	0.681	0.198	0.004	0.358	−0.361	0.308	0.261	−0.046
log(SDNN)	0.659	0.212	0.008	0.298	0.469	0.315	0.159	0.008
log(SD1)	0.859	0.134	0.000	0.698	1.038	0.119	0.000	0.821

Values are β-coefficients, standard errors of the estimate (SEE), P values, and coefficients of determination (R²). Models were adjusted for age and sex. HR, heart rate; HRV, heart rate variability; HF power, high-frequency power; pNN50, percentage of successive NN intervals that differ by >50 ms (%); SARD, significant acute rate drop; SampEn, sample entropy; SDNN, standard deviation of NN intervals; SD1, short axis of Poincaré plot (ms); WBS, Williams–Beuren syndrome.

Figure 6.

Scatterplots of significant acute rate drop (SARD) vs. individual heart rate (HR) variability (HRV) parameters demonstrate a positive correlation between SARD and parasympathetic activity. Log transformation of SARD and HRV improved the linear fit of models. Linear regression lines are displayed for each group [red “x”, Williams–Beuren syndrome (WBS), blue “o”, control]. A: SARD had a negative relationship with HR in both groups. B: SARD showed a negative association with percentage of successive NN intervals that differ by >50 ms (%) (pNN50, parasympathetic marker) in those with WBS and a positive association with pNN50 in controls. C and D: SARD showed a positive association with parasympathetic markers short axis of Poincaré plot (ms) (SD1) and high-frequency (HF) power in both groups. E: SARD showed a positive association with sample entropy (SampEn) in those with WBS and a slightly negative association with SampEn in controls. F: SARD showed a positive association with standard deviation of NN intervals (SDNN, marker of overall HRV) in both groups.

Spearman analyses further support a strong positive correlation with parasympathetic markers HF power and SD1 in both cases and controls (|r_s| > 0.7, Fig. 4). SARD and pNN50 were positively correlated in controls (r_s = 0.63) and negatively correlated in those with WBS (r_s = 0.47).

DISCUSSION

This is the first study to investigate HRF in individuals with WBS. Several observations arise from this analysis. First, we found that those with WBS demonstrate a significantly higher degree of fragmentation as represented by PIP, PIP^S, and W₃ compared with controls. If we extrapolate from previous studies correlating increased fragmentation with cardiovascular risk (24, 29, 30), we may hypothesize that increased HRF seen in this WBS cohort may also be predictive of increased cardiovascular risk in WBS. Second, multivariable analyses exploring the relationship between HRF and conventional HRV showed that variations in HRF parameters are not well explained by HRV and that HRF may improve case status discrimination. Third, we explored a novel metric, we term “SARD,” based on the HRF approach designed to quantify acute vagally mediated compensatory events that are commonly identified clinically as sinus arrhythmia, sinus pauses, or sinus bradycardia events. Consistent with the diminished vagal activity previously demonstrated in WBS, SARDs occurred less frequently in those with WBS compared with healthy controls. Unlike currently used autonomic markers, SARD may provide the advantage of a connection to a known biological correlate.

Those with WBS Demonstrate Increased Fragmentation That May Be Predictive of Increased SCD Risk

Those with WBS demonstrate increased fragmentation compared with healthy controls.

Those with WBS demonstrate significantly higher PIP and PIP^S compared with controls during all time periods (24 h, day, night), corresponding to a higher degree of fragmentation (Fig. 2 and Supplemental Fig. S3). The increased PIP in those with WBS is consistent with prior HRF studies that showed a higher PIP in those with coronary artery disease (CAD) and a positive association between PIP and cardiovascular events (CVEs) and CV death (24, 30). This may suggest that PIP could also serve as a biomarker for increased SCD risk in subjects with WBS, although further study is needed to formally validate this. A prior study by others also showed a higher PIP^S in CAD versus controls in models adjusted for HR (29). Our data suggest that PIP and PIP^S may quantify a type of HRV that correlates with increased cardiovascular risk. A relatively small proportion of PIP variability, in particular, was found to be attributable to conventional HRV phenomenon (Table 4), suggesting other mechanisms may be at work.

Although increased PIP and PIP^S support our hypothesis that those with WBS have increased fragmentation, PIP^H does not show a significant difference between cohorts. This contrasts with a prior study showing significantly higher PIP^H in CAD versus healthy controls (29). It is possible that PIP^H may correlate with myocardial infarction-specific disease phenomena, for example, as opposed to a process unrelated to infarction (as we speculate is the case in WBS-related SCD). Alternatively, PIP^H may reflect age-dependent mechanisms that align more closely with populations studied by others, compared with this cohort with a mean age of 22 (39).

Dynamic approach to HRF further supports increased fragmentation in those with WBS.

In addition to measuring fragmentation over the 24-h recording period, we employed the symbolic dynamic method introduced by Costa et al. to identify patterns occurring at or around the respiratory frequency (3–5 s, or ∼4 beats). This is assumed to reflect the time course necessary for afferent and efferent nerve conduction to initiate vagal modulation (24, 29, 40). This approach revealed significant differences between subjects with WBS and controls: W₁, representing less fragmented sequences, was significantly lower in those with WBS, whereas W₃, representing highly fragmented sequences, was significantly higher in subjects (Fig. 3, B and D). This distribution of word types, skewed toward highly fragmented words (i.e., W₃), aligns with findings by others who correlated increased fragmentation with other pathological states including MI and stroke (40, 29). Although W₂ did not follow this distribution pattern (W₂ was higher in controls), W₂ represents an intermediate level of fragmentation and thus does not yield a definitive conclusion about differences in fragmentation between those with WBS and controls (29).

HRF may be a useful marker in predicting SCD risk.

The increased fragmentation in those with WBS highlights an important consideration regarding the interpretation of HRV as an indicator of heart rhythm regulatory health or robustness. Conventionally, diminished variability due to autonomic dysregulation is thought to confer increased risk (7). However, others have observed that there are some circumstances when greater variability may correlate with increased SCD risk: this is the case with highly fragmented or “erratic” (i.e., more variable) rhythms. Of importance, despite some series showing increased fragmentation, cohorts with high fragmentation are frequently found to have diminished variability overall when traditional HRV metrics are applied (24, 27, 41, 42). Thus, HRF suggests nuance to prior dogma as demonstrated by this form of increased short-term variability that appears to correlate with increased CV risk. Previous CAD, atherosclerosis, and post-MI cohorts displayed increased fragmentation and low HRV, both of which correlated with increased SCD risk. Although MI, stroke, and atherosclerosis are considered polygenic diseases, WBS is associated with a well-defined loss of 25–27 coding genes, but it remains unclear if one specific feature or gene abnormality drives HRF abnormalities. Nonetheless, a similar pattern of increased fragmentation and diminished HRV in the WBS cohort may suggest that both HRF and traditional HRV metrics are useful markers in WBS when assessing SCD risk. Again, we anticipate carrying out prospective studies to further validate this (12, 24, 29, 30, 42). In a study of MESA participants, the Costa group saw that although HRF was associated with CVE risk and death, conventional HRV measures (rMSSD, pNN50, HF power) did not predict these outcomes (30). Thus, HRF indices, particularly those that were higher in those with WBS (PIP, PIP^S, W₁, and W₃), may be helpful in future studies designed to identify those at greatest risk of SCD in the WBS population.

We found W₃ particularly intriguing because these same stereotyped HR inflection changes have been previously identified by others as preceding important arrhythmia induction events, and mechanisms underlying arrhythmia induction. Specifically, W₃ captures the “short-long-short” RR interval sequence (“+1, −1, +1”) that commonly induces some ventricular tachycardia and ventricular fibrillation episodes (torsade de pointes (TdP)) (43). Long-short RR intervals have also been shown to precede TdP (44–46). Beats timed in this configuration may confer increased arrhythmia risk based on two separate mechanisms. First, irregularly timed beats may increase the dispersion of refractoriness, lowering the threshold for arrhythmia induction (47–49). Second, increased RR length results in increased sarcoplasmic reticulum calcium loading, which, particularly in the context of compromised calcium handling, may precipitate dysregulated calcium leak that triggers an arrhythmic event. Although some of these events are initiated by premature ventricular contractions that would not be quantified by conventional HRV or HRF methods, events captured by W₃ may offer a window into a subset of proarrhythmic events that may yield an increased understanding of arrhythmia and SCD mechanisms (46, 50–52). Again, W₃ was significantly elevated in those with WBS (Fig. 3) and is not well explained by conventional HRV (Table 4). In sum, W₃ may prove to be a particularly useful measure in identifying those at risk for SCD among the WBS population.

HRF May Capture Unique HR Modulators Compared with Conventional HRV

HRF improves discrimination of those with WBS from healthy controls.

Three of the seven HRF parameters (PIP^S, PIP^H, and W₃) were significantly associated with WBS after adjusting for all HRV parameters (a 4th, PIP, trended toward significance with P = 0.061) (Table 3). These findings suggest that HRF may provide additional information about the HRV profile in WBS, independent of what is captured by conventional HRV measures.

HRF shows a weak-to-moderate association with HRV.

The association between HRF and HRV was explored using Spearman correlations and multivariable linear regressions. Spearman correlations generally showed weak-to-moderate pairwise associations between HRF and HRV (Fig. 4). Although larger studies are needed to confirm these findings, our data suggest the possibility that HRF may partially reflect phenomena not captured by conventional HRV measures. Linear regression models evaluated the extent to which the variation in HRF can be explained by the principal components of HRV; 4 out of 7 HRF parameters (PIP^H, W₀, W₂, W₃) demonstrated a weak-to-moderate association with HRV in both subjects and controls (R² < 0.5), supporting our hypothesis that HRF conveys data beyond what is substantially quantifiable by conventional HRV measures. However, PIP^S (both cases and controls) and W₁ (WBS cases only) were notable exceptions: both showed strong associations with HRV (R² = 0.81 to 0.95). This observation aligns with the fact that, of all HRF parameters, PIP^S had the strongest association with HRV using Spearman correlations. The strong association between PIP^S and HRV suggests that PIP^S in particular may not add additional utility in characterizing the HRV profile. We suspect other HRF metrics, which include full reversals in HR acceleration termed “hard inflections” (−1 to +1 or vice versa), might be more reflective of the HR fluctuations not captured by traditional HRV metrics. HRF metrics PIP and W₃ in particular may prove to be a valuable addition to the HRV toolkit because they both 1) showed minimal association with HRV, 2) seemed to add to WBS case status discrimination (Table 3), and 3) were significantly different between those with WBS and controls (Figs. 2 and 3).

HRF quantifies the degree to which an NN time series exhibits an erratic rhythm (24), which led us to hypothesize that HRF may overlap with what is captured by sample entropy (SampEn). SampEn measures the degree of “randomness” of a signal. Despite this, SampEn did not strongly correlate with HRF indices (Fig. 4). This suggests that the two approaches may measure distinct HRV characteristics. For example, the highly fragmented signal “+1, 0, +1, 0” exhibits low complexity (low SampEn), demonstrating that a fragmented signal can have varying degrees of randomness. Thus, although HRF quantifies the erratic behavior of HR over time, it is not a surrogate for the level of entropy.

HRF may partially reflect the activity of nonautonomic modulators.

We sought to explore whether HRF offered a means to quantify and study arrhythmia and rhythm modulators not currently measured by commonly used metrics. Spearman correlations and exploratory multivariable quantitative analyses (described before) that included four known autonomic markers (HR, pNN50, HF power, SD1) suggest that HRF may contain pertinent information relevant to WBS case status that is not captured by autonomic markers. Although autonomic measures are useful in identifying rhythm-related abnormalities, HRF may better characterize sinus node dysfunction and proarrhythmic states (24, 39). Stated another way, a breakdown of conventional feedback regulatory loops renders autonomic measures less reliable markers of cardiac risk (24, 27). HRF may capture and reflect this state quantitatively.

Several different physiological and pathophysiological processes may underlie nonautonomic modulators: Atrial stretch, inflammation, fibrosis, and intracellular coupling likely all contribute to shaping SAN regulatory function in both health and disease. These same factors certainly inform arrhythmogenesis as well (24, 29, 30, 53–56). Fragmentation due to these mechanisms may serve as markers of arrhythmia and SCD risk. Future studies with a larger sample size will permit a more precise assessment of the true association between HRF and HRV in those with WBS and healthy controls.

NEW METRIC “SARD” MAY BE ABLE TO LINK CLINICAL EVENTS DIRECTLY TO VAGAL ACTIVITY

Those with WBS Exhibit Less Frequent Acute Drops in HR Consistent with Diminished Parasympathetic Activity

In considering HRF methodologies, we postulated a variation of this approach may be valuable for capturing and quantifying variation seen in sinus arrhythmia. We found no previously articulated measure designed to capture this phenomenon specifically, so we sought to develop a metric (“SARD”) to quantify these events, which are broadly understood to be vagally mediated. SARD captures a sudden HR drop (20% drop compared with the previous beat) sustained for at least two beats. Those with WBS had significantly fewer SARDs compared with healthy controls over all periods measured (24 h, daytime, and nighttime) (Fig. 5). This is consistent with a previous study showing decreased parasympathetic activity (lower pNN50, HF power, and SD1) in our WBS cohort (7).

We also investigated the relationship between SARD and currently used parasympathetic markers (pNN50, HF power, and SD1) using univariate logistic regressions (Table 5) and Spearman correlations (Fig. 4). As expected, SARD showed a moderate-to-strong positive association of SARD with SD1 and HF power in both subjects and controls. We expected SARD to also be positively associated with pNN50, but, to our surprise, SARD showed a weak negative association with pNN50 (Table 5 and Fig. 6B). This suggests that SARD may capture something not measured by known parasympathetic markers.

Unlike currently used parasympathetic markers, SARD provides the advantage of directly measuring a specific event well understood by most clinicians. It may also inform arrhythmia risk and provide additional insights into vagally mediated activity in health and disease. It is worth noting that the number of SARDs per patient in this study was lower than expected, as we aimed to capture sinus arrhythmia events. It is possible our criteria were overly stringent in requiring drops in HR to be sustained for two beats, resulting in a very narrow range of low values (<1%).

Limitations

As with most rare disease studies, the study is limited by the small study size. Furthermore, the sample may be skewed in important respects. The cohort is limited to those who could travel to the NIH, skewing the results to represent the healthier patients with WBS. A larger prospective cohort in future studies would increase the power of the results and permit more reliable identification of useful biomarkers. This study is in anticipation of a larger prospective study designed to track CV events in these patients over a significant period, which will allow us to better evaluate the true predictive value of HRF, SARD, and traditional HRV metrics. Furthermore, HRF has been applied to a limited number of CV diseases whose phenotypic features are not identical to those of WBS. Thus, there are insufficient data to determine the relationship between the measures described and their relationship to sudden death or SCD in WBS. We should note, however, that one subject who was excluded from the study because of β-blockade did die suddenly in a manner consistent with ventricular arrhythmia.

Conclusion

HRF offers important insight into arrhythmia and SCD risk and appears to augment those offered by traditional HRV analysis. Specifically, evidence found by us and others suggest that HRF may provide a quantitative means for studying additional, possibly nonautonomic modulators of sinus node function and dysfunction that contribute to arrhythmogenesis and SCD risk. HRF may therefore be a useful addition to the HRV toolkit for SCD risk stratification. The novel metric SARD may further add additional utility to HRV analysis by providing a more direct and specific vagal activity readout that correlates with a clinically identifiable process. Still, these measures must be further validated for use in WBS, ideally through a prospective multicenter clinical trial. The first step in preventing SCD is to translate biological events into quantifiable metrics that can be studied, validated and, if necessary, refined. We hope that the measures forwarded here advance this pursuit.

DATA AVAILABILITY

Source data for this study are openly available at https://doi.org/10.6084/m9.figshare.25719426.v5. Included with these files is a key file that contains a new ID, case status, age, and sex of each subject.

SUPPLEMENTAL MATERIALS

Supplemental Figs. S1–S5: https://doi.org/10.6084/m9.figshare.25719708.v1.

Supplemental Tables S1–S4: https://doi.org/10.6084/m9.figshare.25588089.v1.

GRANTS

This work was supported by National Heart, Lung, and Blood Institute Grants ZIA-HL-006210 and ZIA-HL-006212 (to B.A.K.).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

B.M.C. and M.D.L. conceived and designed research; S.O. and N.R. performed experiments; B.M.C., A.B., J.T., K.S., and M.D.L. analyzed data; B.M.C., B.A.K., and M.D.L. interpreted results of experiments; B.M.C. prepared figures; B.M.C. and M.D.L. drafted manuscript; B.M.C., K.S., N.R., B.A.K., and M.D.L. edited and revised manuscript; B.M.C., K.S., S.O., N.R., B.A.K., and M.D.L. approved final version of manuscript.