Exercise cardiac magnetic resonance imaging: a feasibility study and meta-analysis

Abstract

Cardiac stress testing improves detection and risk assessment of heart disease. Magnetic resonance imaging (MRI) is the clinical gold-standard for assessing cardiac morphology and function at rest; however, exercise MRI has not been widely adapted for cardiac assessment because of imaging and device limitations. Commercially available magnetic resonance ergometers, together with improved imaging sequences, have overcome many previous limitations, making cardiac stress MRI more feasible. Here, we aimed to demonstrate clinical feasibility and establish the normative, healthy response to supine exercise MRI. Eight young, healthy subjects underwent rest and exercise cinematic imaging to measure left ventricular volumes and ejection fraction. To establish the normative, healthy response to exercise MRI we performed a comprehensive literature review and meta-analysis of existing exercise cardiac MRI studies. Results were pooled using a random effects model to define the left ventricular ejection fraction, end-diastolic, end-systolic, and stroke volume responses. Our proof-of-concept data showed a marked increase in cardiac index with exercise, secondary to an increase in both heart rate and stroke volume. The change in stroke volume was driven by a reduction in end-systolic volume, with no change in end-diastolic volume. These findings were entirely consistent with 17 previous exercise MRI studies (226 individual records), despite differences in imaging approach, ergometer, or exercise type. Taken together, the data herein demonstrate that exercise cardiac MRI is clinically feasible, using commercially available exercise equipment and vendor-provided product sequences and establish the normative, healthy response to exercise MRI.

INTRODUCTION

Exercise stress testing provides important prognostic information for assessing cardiovascular risk and detection of heart disease (4, 16, 19, 20, 39, 42, 54). Combining exercise stress testing with cardiac imaging, such as echocardiography, nuclear imaging, or positron emission tomography, significantly improves diagnostic sensitivity and specificity (6, 27, 29, 35, 60). Accordingly, exercise stress imaging is commonly performed world-wide (12, 14, 15, 52, 62).

Cardiac magnetic resonance (MR) imaging (cMRI) is the clinical gold-standard for assessing cardiac morphology and global systolic function (40, 46, 59). Despite its excellent spatial resolution and unique potential for additional structural and functional quantification (e.g., fibrosis assessment and/or myocardial energetics), cMRI is not currently utilized for cardiac exercise stress imaging. Although many factors may contribute to the underutilization of stress cMRI, the general lack of MR-compatible exercise equipment and inadequate imaging sequences likely play a major role (36).

To overcome these limitations, several investigators have advocated exercising outside of the MRI followed by a brief transition to the scan table and subsequent imaging (1, 2, 18, 30, 32, 44, 45, 56, 58). Under this paradigm, with much familiarization and the use of innovative body molds (to ensure anatomical image registration), image acquisition can be performed within 60 s of peak exercise (45). Marked hemodynamic recovery during the transition from exercise to imaging (18, 30, 32, 44, 45, 58) and the technical training and familiarization required of both the patient and imaging staff to achieve rapid and accurate body placement has prevented wide-spread adoption of this approach. Accordingly, several investigators have focused on MRI-compatible cycle ergometers, which allow patients to exercise inside the bore of the magnet (3, 24, 26, 31, 33, 38, 41, 47, 55, 61). Although several studies have evaluated cardiac output dynamics using velocity-encoding approaches (5, 7, 21, 25, 37, 50, 57, 61), only a few studies have attempted to evaluate ventricular morphology and function (8–10, 23, 24, 31, 34, 38, 43, 47–49, 51, 55), the vast majority of which have utilized a real-time, ungated free-breathing pulse sequence (8–10, 31, 34, 51). Although this latter approach increases overall feasibility by eliminating ECG-gating artifacts, patient breath-holds, and minimizes the impact of chest movement, postprocessing is incredibly complex and time consuming, negating the possibility of “real-time” image reconstruction during the cardiac stress test (and thus real-time physician feedback). Such limitations have also prevented the wide-spread adoption of this approach.

With this background and the goal of making exercise cMRI clinically translatable, the primary aim of this study was to test the feasibility of using a commercially available MR-compatible ergometer, combined with vendor-provided MRI sequences, to assess cardiac morphology and function. To assess the impact of movement on image quality, we compared images acquired during exercise, with images acquired immediately (0–4 s) postexercise. To assess the impact of a prolonged pause between exercise and imaging—similar to that encountered when exercise is performed outside of the bore of the MRI—we compared subject hemodynamics, cardiac morphology, and cardiac function between exercise and following a 60-s pause. A secondary aim of this study was to perform a comprehensive literature review and meta-analysis of published studies reporting cardiac volumes during exercise cMRI to establish the normative response.

METHODS

Participants

Eight recreationally active, young, healthy subjects were recruited from the Dallas-Fort Worth community. Participants were eligible for inclusion if they were 18–35 yr old, physically active, and had no history of cardiovascular, metabolic, or neurological disease. Exclusion criteria included contraindications to MRI and physical limitations precluding exercise.

All procedures were performed in accordance with the principles of the Helsinki Declaration; all participants provided written informed consent and the study was approved by the University of Texas at Arlington and University of Texas Southwestern Medical Center Institutional Review Boards.

Protocol

To address our primary aim, each subject completed three visits. The first visit was used for screening, familiarization with the MRI-compatible ergometer (Ergospect Cardio-Stepper, Ergospect, Austria), and determination of target workload and heart rate for exercise cMRI. The second visit was used to characterize the metabolic cost and hemodynamic differences between supine exercise with our MR-compatible ergometer and upright exercise with a traditional cycle ergometer (Lode Corival, Lode, The Netherlands). The third visit consisted of a resting and exercise stress cMRI.

Supine cardiopulmonary exercise test.

Subjects underwent a supine, incremental maximal exercise test, starting at 15 W at a cadence of 50 revolutions/min and progressing by 15 W every 2 min until volitional exhaustion or significant and continued loss of cadence (≥10 revolutions/min despite verbal encouragement to continue). Heart rate (Polar H1, Polar), expired gas analysis (TrueOne 2400, Parvomedics), and work-rate (Ergospect Cardio-Stepper, Ergospect) were recorded continuously.

Upright cardiopulmonary exercise test.

Subjects underwent an incremental ramped maximal exercise test, starting at 50 W at a self-selected cadence (50–90 revolutions/min) and progressing by 20 W every 2 min until volitional exhaustion. Heart rate (Polar H1, Polar), expired gas analysis (TrueOne 2400, Parvomedics), and work-rate (Lode Corival, Lode) were recorded continuously.

Cardiac MRI.

Imaging was conducted on a Phillips Achieva 3T scanner. Following resting imaging, subjects exercised at the workload determined from study visit 1 required to achieve four metabolic equivalents (4 metabolic equivalents ≈14 ml·kg⁻¹·min⁻¹), the minimal threshold for independent living (53). High resolution long-axis (4-chamber and 2-chamber) and short-axis (midventricular, papillary muscle level) cine images were acquired using balanced fast-field echo at rest, during exercise, immediately (0–4 s) postexercise, and 60 s postexercise (to mimic patient transfer from an external exercise device). Typical image parameters include repetition time = 3.4 ms; echo time = 1.7 ms; flip angle = 45°; acquisition matrix = 195 × 195; field of view = 295 × 295 mm; 8-mm slice thickness; 20–30 phases/cardiac cycle, with 1 slice acquired per breath hold. Breath-hold time ranged between 7 and 9 s at rest and 3–5 s during exercise.

cMRI data were analyzed using commercially available software (CVI⁴², Circle Cardiovascular Imaging). Left ventricular volumes were calculated by Simpson’s Biplane method using standard 4- and 2-chamber slices, and cross-sectional area was measured in the short-axis plane. All volumes were indexed to body surface area calculated by the Dubois-Dubois formula to account for body size variation (13).

Comprehensive Literature Review

To accomplish our secondary aim, we performed a comprehensive literature review, searching articles on PubMed from 1985 to March 2018 using the search terms “exercise AND cardiac MRI,” hand-searched references, and contacted experts in the field. A priori, we included studies reporting left-ventricular cardiac volumes for healthy controls aged 16–36 to best match our participant inclusion criteria. We extracted mean resting and exercise end-diastolic, end-systolic, and stroke volumes, heart rates, and ejection fractions from study text, tables, and figures (WebPlotDigitizer, v. 4.1, Rohatgi). If data were not readily extractable, authors were contacted directly and asked to provide the missing or unattainable data.

Statistical Analysis

Aim 1: feasibility study.

Dependent variables were confirmed for normality and homoscedasticity using the Shapiro-Wilk test. Changes in cardiac volumes and global left ventricular function from rest to exercise/postexercise were assessed using a one-way repeated measures ANOVA with Sidak post hoc testing. Hemodynamic and cardiopulmonary data were compared between the supine and upright incremental exercise tests using a paired sample t-test. Statistical analyses were performed using SPSS (version 24 IBM SPSS Statistics, Armonk). All data are expressed as mean (standard deviation) unless otherwise stated, and statistical significance was considered when P ≤ 0.05.

Aim 2: comprehensive literature review.

Previously published results from the literature review were pooled and analyzed by a random-effect analysis to create a single, more precise estimate of the effect size. The rest/exercise correlation was assumed to be moderate (50%; 17). Analysis and graphical presentation were performed in R 3.4.2 using the “metaphor” package (R Core Team 2016, R Foundation for Statistical Computing, Vienna, Austria). Effect sizes were calculated using both standardized mean differences (SMD) and weighted mean differences (WMD) between rest and exercise conditions. A SMD of 0.25 was considered as a small effect size, 0.5 as a medium effect size, and 0.8 and higher as a large effect size. Heterogeneity of studies was explored using the Cochrane’s Q test of heterogeneity (P < 0.05 was considered statistically significant). Inconsistency in the results of the studies was assessed by I², which described the percentage of total variation across studies that was due to heterogeneity. When I² was ≥ 50%, there was more than moderate inconsistency (11, 28).

RESULTS

Aim 1: Feasibility Study

Eight healthy men and women completed the feasibility study (male/female, 3/5; age, 25 ± 3 yr; body mass index, 22.3 ± 2.7 kg/m²; BSA, 1.62 ± 0.14 m²).

Supine versus upright exercise.

The metabolic and hemodynamic data for both the supine and upright cardiopulmonary stress tests are displayed in Table 1. Peak oxygen consumption was markedly lower during supine exercise compared with upright cycling, reflective of the smaller muscle mass being utilized with the MR-compatible ergometer. Similarly, peak heart rate, peak respiratory exchange ratio, and peak ventilation were also markedly lower. None of the subjects met the criteria for true maximal O₂ consumption(V̇o_2max) during supine exercise, defined as achieving 1) age-predicted maximal heart rate, 2) respiratory exchange ratio > 1.10, or 3) a plateau in oxygen consumption. In contrast, all of the subjects during the upright cycling ramp test met at least two of these criteria. The primary reason for stopping the supine exercise test was an inability to overcome the change in resistance; none of the subjects reported maximal perceived exertion or breathlessness as a reason for stopping.

Table 1.

Metabolic and hemodynamic comparison between supine and upright exercise

Parameter	Supine	Upright	Percentage of Upright	P Value
V̇o2peak, ml·kg−1·min−1	26 (3)	39 (5)	67 (10)	0.0007
V̇o2peak, l/min	1.24 (0.35)	2.31 (0.53)	67 (10)	0.0004
Ventilationpeak, l/min	57 (15)	106 (24)	56 (13)	0.0003
RERpeak	1.07 (0.06)	1.23 (0.09)	87 (6)	0.0007
Heart rate, beats/min	153 (19)	188 (9)	82 (9)	0.0011

Values are means (SD). RER, respiratory exchange ratio; RER_peak, peak RER; Ventilation_peak, peak ventiliation; V̇o_2peak, peak oxygen consumption.

cMRI feasibility.

Cardiac imaging was possible during exercise in all of the subjects. None of the subjects reported any discomfort or experienced any adverse events. As expected, image quality was improved postexercise (0–4 s, 60 s), when upper body and chest coil movements were eliminated (Fig. 1).

Fig. 1.

A: representative high resolution cardiac magnetic resonance (cMR) images at rest (black circle), during exercise (yellow square), immediately postexercise (0–4 s, green triangle), and 60 s postexercise (red diamond). Left ventricular (LV) volumes were indexed to body surface area for determining LV end-diastolic volume (EDV_i) (B), end-systolic (ESV_i) (C), and stroke volume (SV_i) (D) and multiplied by heart rate (E) to generate cardiac index (I). LV cavity areas were measured from a mid-short-axis image (papillary muscle level) and indexed to body surface area for determination of end-diastolic (EDA_i) (F) and end-systolic (ESA_i) (G) cavity areas. Ejection fraction (H) was calculated from the biplane volumes. Data presented as means and SE. †Statistically significant (P < 0.05) difference from rest; ‡statistically significant from exercise. CI, cardiac index; EF, ejection fraction; HR, heart rate.

Both heart rate and stroke volume increased with exercise, resulting in a twofold increase in cardiac index (CI; Fig. 1). The increase in stroke volume was mediated entirely by a reduction in end-systolic volume (ESV), as end-diastolic volume (EDV) was unchanged with exercise. Accordingly, ejection fraction was also significantly increased with exercise.

As compared with imaging during exercise, imaging immediately postexercise had no significant impact on chamber volumes, ejection fraction or CI. In contrast, imaging 60 s postexercise resulted in a significant increase in ESV and a significant decrease in heart rate, ejection fraction, and CI (Fig. 1).

Aim 2: Literature Review and Data Synthesis

Our search produced 1,787 articles; 17 studies reported rest and exercise cardiac volumes. Of the 17 studies, data were extracted from 14 studies (Table 2). One study (33) reported median and interquartile range that could not be used in our synthesis. Two studies (22, 47) were confirmed to have used previously reported data sets and were therefore excluded. When multiple exercise intensities were reported, the exercise intensity closest to that chosen in our feasibility study was reported.

Table 2.

Characteristics of exercise cMRI studies reporting LV volumes

Study	n, m/f	Age	Resting HR	Image HR	Exercise Intensity	Field Strength	Exercise Type	Image Type
Present Study	3/5	25 (3)	67 (8)	115 (9)	Submax	3.0T	Supine, Ergospect CardioStepper	Gated, breath-hold, SAX, biplane LAX
Claessen 2014(a) (9)	14/0	36 (6)	66 (7)	143 (7)	Max*	1.5T	Supine, Lode MRI cycle	Real-time free-breathe SAX/LAX stacks
Claessen 2014(b) (8)	8/1	35 (12)	NR	NR	Max*
Claessen 2015 (10)	14/0	36 (15)	56 (8)	149 (11)	Max
Gusso 2012 (24)	4/4	25 (4)	67 (15)	110 (3)	Submax	1.5T	Supine, custom MRI cycle	Gated breath-hold SAX stack, 3 LAX
Gusso 2012 (23); 2017 (22)	10/12	17 (1)	75 (14)	109 (5)	Submax
La Gerche 2012 (31)	13/2	35 (9)	59 (14)	164 (10)	Max*	1.5T	Supine, Lode MRI cycle	Real-time free-breathe SAX/LAX stacks
Lafountain 2016 (31)	8/8	26 (5)	62 (NR)	166 (NR)	Max*	1.5T	Upright, custom MRI treadmill	Ungated, free-breathe SAX stack and biplane LAX slices
Lurz 2009 (34)	5/7	33 (29–42)	74 (3)	154 (13)	Max*	1.5T	Supine, Lode MRI cycle, flutter kicking motion	Real-time free-breathe SAX stacks
Oosterhof 2005 (38)	8/6	27 (4)	66 (8)	121 (8)	Submax	1.5T	Supine, custom MRI cycle	Turbo field echo planar imaging SAX stack
Pinto 2014 (43)	10/9	17 (2)	70 (10)	109 (4)	Submax	1.5T	Supine, custom MRI cycle	Gated breath-hold SAX stack, 3 LAX
Roest 2001 (48); 2002 (47)	8/8	18 (2)	71 (10)	121 (14)	Submax	1.5T	Supine, Custom MRI ergometer	Gated breath-hold SAX stack
Roest 2004 (49)	7/7	25 (5)	67 (8)	122 (8)	Submax
Schnell 2017(a) (51)	15/0	35 (8)	59 (12)	155 (18)	Max	1.5T	Supine, Lode MRI cycle	Real-time free-breathe SAX/LAX stacks
Schnell 2017(b) (51)	12/3	35 (14)	66 (7)	150 (12)	Max
Steding-Ehrenborg 2013 (55)	20/6	30 (9)	60 (11)	94 (13)	Submax	1.5T	Supine, Custom MRI ergometer	Gated breath-hold SAX stack and LAX slice

cMRI, cardiac MRI; f, female; HR, heart rate; LAX, long axis; T, Tesla. LV, left ventricle; m, male; MRI, magnetic resonance imaging; NR, not reported; SAX, short axis; T, Tesla.

^*Submaximal data used in meta-analysis, see Fig. 2 legend.

Characteristics of the studies included in our comprehensive data synthesis are displayed in Table 2. The SMD model was used to account for variance in study participants (sex, age, BSA, exercise capacity) and volumetric reporting (standard vs. normalized volumes). By combining these data sets, we were able to evaluate the normal physiological response to supine exercise in 226 healthy subjects. The pooled analysis supports our proof-of-concept data (reported above), showing a marked and significant increase in ejection fraction during exercise compared with rest (Fig. 2 and Table 3). Moreover, we observed no change in left ventricular EDV, with changes in stroke volume being entirely driven by reductions in ESV (Fig. 2 and Table 3). Similarly, the magnitude of change in cardiac output was therefore predominantly driven by changes in heart rate (Table 2). The Q test revealed significant variation in the exercise response to ejection fraction (P = 0.004), ESV (P = 0.007), and stroke volume (P = 0.005), although less than moderate (I² = 41%, 49%, and 27%). There was no heterogeneity in EDV response (Table 3A).

Fig. 2.

Pooled standardized mean difference [exercise-rest (95% confidence interval)] for left ventricular end-diastolic volume (EDV) (A), end-systolic volume (ESV) (B), stroke volume (SV) (C), and ejection fraction (EF) (D) from the available exercise cardiac magnetic resonance (cMR) studies. Values taken from graphical representations at ¹moderate and ²peak intensity exercise. ³Adolescent populations with ⁴values taken from graphical representations. FE model, fixed effects model; RE model, random effects model.

Table 3.

Effect size, significance, and heterogeneity for pooled analyses

A. Measure	n	SMD [95% CI]	Significance (H0:SMD = 0)	Q P Value	I2, %
EDV, ml	16	−0.14 [−0.31, 0.04]	0.13	1.00	0
ESV, ml/m2	16	−1.18 [−1.42, −0.95]	<0.0001	0.005	41
SV, ml/kg	16	0.87 [0.64, 1.09]	<0.0001	0.006	50
EF, %	15	1.75 [1.48, 2.03]	<0.0001	0.006	26

B. Measure		WMD	(H0:WMD = 0)	Q P Value	I2, %
EDV, ml	11	−0.33 [−0.61, −0.06]	0.027	0.18	28
ESV, ml	11	−3.51 [−4.11, −2.9]	<0.0001	<0.0001	75
SV, ml	11	3.86 [3.07, 4.65]	<0.0001	<0.0001	91
EF, %	15	6.12 [5.26, 6.98]	<0.0001	0.005	95

n, number of comparison. 95% CI, 95% confidence intervals; H₀, null hypothesis; EDV, end-diastolic volume; EF, ejection fraction; ESV, end-systolic volume; SMD, standardized mean difference; SV, stroke volume; WMD, weighted mean difference. SMD calculated by pooling EDV (ESV, SV) for studies reporting measures in ml, ml/m² (BSA normalized) and ml/kg (normalized to fat-free mass). WMD calculated by pooling studies reporting EDV (ESV, SV) in ml.

Visual inspection of Fig. 2 prompted post hoc analysis to determine if the Lafountain study (2016, treadmill exercise) was a statistical outlier. Based on our model, ~5% of the externally standardized residuals would be expected to exceed the bounds ± 1.96. Externally standardized residuals were calculated for ESV, stroke volume, and ejection fraction as −3.85, 1.92, and 12.24, respectively, indicating the study as a potential outlier. In a sensitivity analysis, removal of Lafountain (2016) from the meta-analysis increased Q test P values (0.08, 0.05, and 0.26) and reduced I² values (38%, 42%, and 26%) for ESV, stroke volume, and ejection fraction, demonstrating a significant reduction in heterogeneity.

DISCUSSION

The major novel findings of this study are threefold. First, we demonstrate feasibility of exercise stress cMR using a commercially available leg ergometer and vendor-provided product sequences at 3 Tesla. Second, we demonstrate hemodynamic similarities between imaging during and immediately postexercise but marked hemodynamic decline when imaging is delayed for at least 60 s postexercise. Finally, by combining our results with previously published data, we were able to define the normal cardiac response to exercise cMRI.

That cardiac imaging was feasible using a commercially available leg ergometer and vendor-provided product sequences, provides great promise for easy and simple adoption of exercise cMRI to clinical settings. Although we observed a marked improvement in image quality by imaging immediately postexercise, we observed no major differences in cardiac volumes or global left ventricular function between these two conditions. Because inclusion of brief pauses will prolong the exercise stress imaging protocol, we believe imaging during exercise will prove to be the protocol of choice for most clinical sites. Of course, if exercise intensity is increased beyond the submaximal level performed herein, imaging during brief periods of exercise cessation may need to be adopted to overcome the anticipated upper body and chest coil movement. Indeed, as movement artifact increases with higher exercise intensities, spatial resolution may degrade despite consistent imaging parameters. Caution is indeed warranted when interpreting data acquired past 4 s postexercise; however, as our results suggest, prolonged periods of rest following exercise are associated with sharp hemodynamic recovery. Although it may be argued that the rapid hemodynamic recovery observed in this study is reflective of the younger, healthy population studied herein, clinical patients exercised outside of scanner bore and transferred for imaging share a comparable decline in heart rate (44, 45).

Our feasibility data, together with our meta-analysis, demonstrate that the increase in cardiac output during submaximal supine exercise is driven predominantly by an increase in heart rate, with minimal contribution from stroke volume. The data overwhelmingly suggest that healthy subjects have no EDV reserve during supine exercise, despite the expected increase in respiratory and muscle pumps. That this observation is consistent across studies that did not perform breath holds (8–10, 31, 32, 51) argues against intrathoracic pressure (i.e., Valsalva maneuver) mediated reductions in preload. These data must therefore reflect the fact that supine positioning, combined with leg elevation (feet in the ergometer stirrups), produces near maximal EDVs.

That we only observed a small reduction in ESV with exercise was surprising, although consistent with other cMRI studies in healthy populations (8–10, 23, 24, 31, 32, 34, 38, 43, 48, 49, 51, 55). Admittedly, this investigation was focused on submaximal exercise, and thus we cannot extend these ESV findings to higher exercise intensities. Indeed, studies that included higher workloads have reported exercise intensity dependent reductions in ESV (8–10, 31, 33, 51). This interstudy difference likely contributed to the heterogeneity of findings in our meta-analysis for ESV, ejection fraction, and stroke volume. Regardless of exercise intensity, however, the absolute change in ESV remains fairly low. This may reflect the fact that supine exercise with MR-compatible ergometers may not recruit sufficient muscle mass to challenge cardiac output enough to demand greater ESV changes. Compared with conventional upright cycling, our data clearly show blunted metabolic demand, also observed in a prior investigation using a different MR-compatible ergometer (33).

This proof-of-concept feasibility study is not without limitation. First, we limited our data collection to young healthy participants. Future studies are needed to extend these data to clinical populations. Second, spatial resolution was purposely sacrificed to shorten breath-hold time. As a result, the cine images were not sufficient for tissue deformation (i.e., strain) analysis. Inclusion of quantifiable regional wall motion measurements will be an important future addition, especially in clinical populations.

Perspectives and Significance

Taken together, the data herein establish the proof-of-concept that exercise cMRI can be easily adapted to a clinical setting, providing gold-standard noninvasive stress images during exercise. Using commercially available MRI-compatible exercise equipment and vendor-provided product sequences, we were able to successfully characterize the cardiac response to submaximal exercise. By combining our feasibility data with previously published results, we also establish the normative cardiac stress response in over 200 healthy subjects. With this background in place, future studies are needed to adapt this approach in clinical populations.

GRANTS

This project was supported by grants from the National Institutes of Health (R01-HL-136601), the American Heart Association (grant nos. 16SDG27260115, 18PRE33960358, and 18POST33990210), and the Harry S. Moss Heart Trust. Dr. Haykowsky receives support from the University of Texas at Arlington, College of Nursing and Health Innovation, Moritz Chair in Gerontology.

DISCLOSURES

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

AUTHOR CONTRIBUTIONS

R.I.B., M.J.H., and M.D.N. conceived and designed research; R.I.B., T.J.S., W.J.T., M.J.H., and M.D.N. performed experiments; R.I.B., J.W., and M.D.N. analyzed data; R.I.B., T.J.S., J.W., W.J.T., M.J.H., and M.D.N. interpreted results of experiments; R.I.B., J.W., and M.D.N. prepared figures; R.I.B. and M.D.N. drafted manuscript; R.I.B., T.J.S., J.W., W.J.T., M.J.H., and M.D.N. edited and revised manuscript; R.I.B., T.J.S., J.W., W.J.T., M.J.H., and M.D.N. approved final version of manuscript.