Hemodynamic Determinants of Myocardial B-Type Natriuretic Peptide Release
Relative Contributions of Systolic and Diastolic Wall Stress
Although B-type natriuretic peptide (BNP) is widely used as a biomarker for heart failure, the in vivo mechanical stimulus for its cardiac release remains poorly defined. We aimed to characterize the hemodynamic determinants of the transcardiac BNP gradient as a measure of myocardial BNP release by performing a detailed hemodynamic assessment in subjects with a broad spectrum of systolic and diastolic left ventricular dysfunction. Forty-two subjects underwent a detailed transthoracic echocardiographic study, right heart catheterization, and simultaneous BNP measurement in arterial and coronary sinus plasma. The transcardiac BNP gradient was lowest in subjects with normal left ventricular ejection fraction/high peak early diastolic annular velocity (n=11), intermediate in those with normal left ventricular ejection fraction/low peak early diastolic annular velocity (n=13), and highest in those with low left ventricular ejection fraction/low peak early diastolic annular velocity (n=18; 29 ng/L (range: 15 to 78 ng/L) versus 88 ng/L (range: 34 to 172 ng/L) versus 1566 ng/L (range: 624 to 2349 ng/L; P<0.001). Across the range of patients, left ventricular end-systolic wall stress (r2=0.51) and peak systolic mitral annular velocity (r2=0.47) showed the strongest correlation with higher transcardiac BNP gradient. In contrast, the transcardiac BNP gradient was weakly related to indices of diastolic load, including pulmonary capillary wedge pressure (r2=0.27) and left ventricular end-diastolic wall stress (r2=0.21). Across this spectrum of pathophysiology, left ventricular end-systolic wall stress appears to be the key mechanical stimulus influencing cardiac BNP release.
Because of its diagnostic value in patients with shortness of breath and its prognostic value in patients with heart failure (HF), B-type natriuretic peptide (BNP) has become a widely used biomarker in clinical practice.1 BNP is a 32-amino acid peptide derived from the precursor peptide prepro-BNP, which is synthesized by atrial and ventricular cardiomyocytes.1–3 Generation of mature BNP requires the cleavage of a signal peptide and the biologically inactive N-terminal proBNP (NT-proBNP).2 Although many aspects of the biochemistry of BNP biosynthesis and processing are well known, the physiological regulation of cardiac BNP release is still not well understood. The most commonly held view is that stretch of ventricular cardiomyocytes is the key stimulus for prepro-BNP expression.1 Although measurement of peripheral BNP has clear cross-sectional use,1 its applicability as a reproducible, reliable serial index of ventricular function appears more limited. This may reflect the fact that BNP concentrations in peripheral blood not only reflect cardiac BNP release but also BNP clearance by the kidneys4,5 and other mechanisms.2
To interpret plasma BNP, it is necessary to understand in detail the in vivo hemodynamic determinants of its cardiac release. This is particularly important now as BNP measurement moves into diagnostic algorithms used in the assessment of patients with suspected HF with a normal left ventricular (LV) ejection fraction (LVEF).6,7 In HF with reduced LVEF, previous studies have demonstrated a modest relationship between the transcardiac BNP gradient, that is, the difference between BNP concentrations in arterial and coronary sinus blood (ΔA-CSBNP) as an invasive measure of myocardial BNP release, and pulmonary capillary wedge pressure (PCWP), LV end-diastolic pressure, and LV dimensions.4,8,9 However, to date no studies of transcardiac BNP dynamics have included subjects with normal LVEF and impaired LV diastolic function, and as such the biological plausibility of extending BNP as a biomarker to this group of patients is lacking.
The aim of the present study, therefore, was to comprehensively investigate the hemodynamic determinants of ΔA-CSBNP in subjects with impaired LVEF and subjects with a broad range of systolic and diastolic functions to assess to most important hemodynamic determinants of cardiac BNP release.
Patients and Protocol
We studied 24 subjects with normal LVEF and 18 subjects with impaired LVEF. Subjects with normal LVEF underwent right heart catheterization for evaluation of exertional dyspnea (n=15) or suspected pulmonary hypertension on echo in the absence of current symptoms (n=1) or were asymptomatic and volunteered for the study (n=8). Subjects with impaired LVEF consisted of patients with advanced HF undergoing an assessment with regard to cardiac transplantation. No participant reported chest pain. All of the participants with normal LVEF had undergone a noninvasive test without evidence of inducible myocardial ischemia. All of the HF patients with impaired LVEF had previously undergone evaluation regarding the suitability for myocardial revascularization. Subjects on dialysis, patients with acutely destabilized HF, and patients with prosthetic valves were excluded.
All of the subjects underwent transthoracic echocardiography and right heart catheterization with arterial and coronary sinus blood sampling. In all of the subjects with normal LVEF, echocardiography was performed immediately before cardiac catheterization. In patients with impaired LVEF, echocardiography was predominantly performed within 48 hours of the invasive study (range: 0 to 34 days). The study was approved by the ethics committee of the Alfred Hospital, and all of the participants provided written informed consent. Procedures were performed in accordance with institutional guidelines.
Echocardiography and Definition of the Study Groups
Transthoracic echocardiograms were obtained by experienced echocardiographers, using standard views in accordance with current guidelines for LV chamber quantification10 and assessment of LV diastolic function.11 Measurements were performed offline by a single reader blinded to hemodynamic data and BNP and NT-proBNP results. Measurements were averaged from 3 cycles. Dimensions were indexed to height or body surface area as appropriate. LVEF was calculated using the modified biplane Simpson method. Peak systolic (s′) and peak early diastolic annular (e′) velocities were measured by pulsed wave tissue Doppler at the medial and lateral mitral annulus, and readings were averaged.11
An LVEF cutoff of 50% was used to categorize patients as having a normal or impaired LVEF. For descriptive purposes, subjects with a LVEF ≥50% were further categorized as having normal or abnormal LV diastolic function based on a cutoff for the averaged e′ of 9 cm/s. This is a reflection of the current recommendations,11 where a medial e′ <8 cm/s and a lateral e′ <10 cm/s are considered abnormal, and where averaging of medial and lateral readings is recommended. Thus, we studied the following 3 groups: patients with LVEF ≥50% and e′ ≥9 cm/s (normal LVEF/high e′ group; n=11), patients with LVEF ≥50% and e′ <9 cm/s (normal LVEF/low e′ group; n=13), and patients with LVEF <50% (low LVEF group; n=18).
A 3F arterial line (Cook) was placed in a radial (n=40) or brachial artery (n=2). A 6F coronary sinus catheter (Cordis) was inserted via an introducer sheath placed in the right internal jugular (n=12) or brachial (n=30) vein and positioned under fluoroscopic control. The tip of the catheter was positioned ≥2 cm proximal to the orifice of the coronary sinus, as confirmed by contrast injection. Blood samples were simultaneously taken from the arterial line and the coronary sinus catheter. A balloon-tipped 7F thermodilution catheter (Edwards Lifesciences) was inserted for measurement of right atrial pressure, right ventricular pressure, pulmonary artery pressure, and PCWP. The wedge position was confirmed by fluoroscopy and pressure wave form, and the mean PCWP was measured at end expiration. Cardiac output was measured using thermodilution with measurements taken in triplicate. Measurements were indexed to body surface area, as appropriate. LV end-diastolic wall stress and LV end-systolic wall stress (LVESWS) were calculated using the following formula: wall stress=0.334*LV pressure*LV internal diameter/(posterior wall thickness*[1+posterior wall thickness/LV internal diameter]), where LV internal diameter and posterior wall thickness at end diastole or end systole were used.12 PCWP was used as a surrogate for LV end-diastolic pressure, and systolic arterial pressure was used as a substitute for LV end-systolic pressure.
Blood samples were collected in plastic tubes containing ethylene-diamine-tetra-acetate. BNP and NT-proBNP concentrations in arterial (BNPA and NT-proBNPA) and coronary sinus (BNPCS and NT-proBNPCS) plasma were measured using the ARCHITECT BNP assay (Abbott)13 and Roche Elecsys proBNP assay on the E170 analyzer (Roche).14 The within and total coefficients of variation for ARCHITECT assay and the Elecsys proBNP assay are ≤5.3% and <3.0%, respectively.13,14 Transcardiac BNP (ΔA-CSBNP) and NT-proBNP (ΔA-CS NT-proBNP) gradients were calculated as concentration in coronary sinus blood minus concentration in arterial blood. In the majority of patients, venous blood samples taken from the introducer sheath in the internal jugular vein or from an antecubital vein were analyzed for BNP (BNPV; n=35) and NT-proBNP (NT-proBNPV; n=40). Estimated glomerular filtration rate was calculated based on a serum creatinine measurement performed on the day of catheterization using the abbreviated Modified Diet in Renal Disease study equation.15 All of the analyses were performed in the clinical laboratory of the Alfred Hospital by technicians unaware of any clinical data.
Categorical data are given as counts and percentages, and comparisons between groups were performed using χ2 tests. Distribution of continuous data was assessed using Shapiro-Wilk tests, and data are presented as mean±SD or median (interquartile range), as appropriate. Comparisons among the normal LVEF/high e′, normal LVEF/low e′, and low LVEF groups were performed using ANOVA followed by Bonferroni post hoc tests. Given that BNP concentrations were not normally distributed, data were log-normal transformed. BNPA and BNPCS were compared using paired t tests. To assess the major determinants of ΔA-CSBNP and ΔA-CSNT-proBNP, exponential models (y=y0+aeb*x) to achieve optimal curve fit were used, and all of the models are characterized by their r2 values. We decided not to use linear models with log-normal transformed BNP, because this approach a priori implies a linear relationship, which does not sufficiently reflect the biological relationship between natriuretic peptides and hemodynamics. A P value <0.05 was considered statistically significant. Analysis was performed using commercially available software packages (SPSS, version 15.0, SPSS, Inc, and SigmaPlot, version 10.0, Systat Software Inc).
For descriptive purposes, the clinical characteristics of the normal LVEF/high e′, normal LVEF/low e′, and low LVEF groups are show in Table 1.
LV Structure and Function
As shown in Table 2, the low LVEF group was characterized by dilated ventricles with eccentric hypertrophy and low LVEF, whereas the normal LVEF/low e′ group was characterized by a smaller left ventricle with more concentric remodeling and normal LVEF. As also shown in Table 2, s′, as a measure of LV systolic long-axis function, differed among all 3 of the groups, being highest in the normal LVEF/high e′ group and being lowest in the low LVEF group.
Mean pulmonary artery pressure, PCWP, LVESWS, and LV end-diastolic wall stress were higher and mean arterial pressure, cardiac index, and stroke volume index were lower in the low LVEF group compared with the other 2 groups (Table 2). However, there was no significant difference in hemodynamics between the normal LVEF/high e′ and normal LVEF/low e′ groups.
Plasma Concentrations and Transcardiac Gradients of Natriuretic Peptides
As shown in Table 3, BNPA in the low LVEF group was higher than in the other 2 groups, but there was no significant difference in BNPA between the normal LVEF/high e′ and the normal LVEF/low e′ groups. There was a highly significant step-up in BNP across the heart in all 3 of the groups (Figure 1). In contrast to BNPA, ΔA-CSBNP differed among all 3 of the groups: ΔA-CSBNP was lowest in the normal LVEF/high e′ group and was highest in the low LVEF group (Table 3). Similar findings were obtained for NT-proBNP (Table 3).
Overall, there was a close correlation between BNPA and ΔA-CSBNP (r2=0.85; P<0.001) and NT-proBNPA and ΔA-CSNT-proBNP (r2=0.90; P<0.001). However, if subjects with LVEF ≥50% and those with LVEF <50% were analyzed separately, the associations between BNPA and ΔA-CSBNP (LVEF ≥50%: r2=0.61; LVEF <50%: r2=0.54; Figure 2) and NT-proBNPA and ΔA-CSNT-proBNP (LVEF ≥50%: r2=0.59; LVEF <50%: r2=0.77) were weaker.
Hemodynamic Determinants of ΔA-CSBNP and ΔA-CSNT-proBNP
In Table 4, models for the relationship between various hemodynamic parameters and ΔA-CSBNP are shown. The models with the highest r2 were obtained for higher ΔA-CSBNP and lower LVEF, higher LVESWS, and lower s′. The correlations between ΔA-CSBNP and LVESWS and s′, respectively, are shown in Figure 3. To better understand the association between ΔA-CSBNP and LVEF, the close inverse association between higher LVESWS and lower LVEF (r2=0.78) is shown in Figure 4A. There was also a significant inverse correlation between higher LVESWS and lower s′ (r2=0.56; Figure 4B).
There was no significant association between ΔA-CSBNP and age, sex, or epidermal growth factor receptor (P>0.05 for all). There was a significant association between higher ΔA-CSBNP and lower hemoglobin (r=−0.40; P=0.009). However, as shown in Table 1, lower hemoglobin was a surrogate for the low LVEF group.
The r2s for the ΔA-CS NT-proBNP models were generally weaker than those for the ΔA-CS BNP models (Table 4), but results were very similar to those for ΔA-CS BNP. The models with the highest r2 were obtained for higher ΔA-CSNT-proBNP and lower LVEF, higher LVESWS, and lower s′.
This is the first study assessing ΔA-CSBNP, and its hemodynamic determinants were assessed using a comprehensive noninvasive and invasive approach in subjects with a broad range of LV function. We demonstrated that ΔA-CSBNP increased with progressive LV dysfunction, and we found that lower LVEF, lower s′, and higher LVESWS had the strongest associations with higher ΔA-CSBNP.
Experimental data indicate that BNP gene expression in cardiac myocytes can be upregulated by numerous stimuli through several different pathways.2,16 Apart from endothelin 1, angiotensin II, and α- and β-adrenergic agonists, “mechanical stretch” is regarded as a key mechanism.2,3,16 It has been suggested that LV end-diastolic wall stress is the most important hemodynamic parameter regulating the levels of circulating BNP.12,17 Of note, circulating BNP reflects the balance between BNP release and elimination. In contrast, our data suggest that LVESWS is the key loading parameter for ΔA-CSBNP, that is, cardiac BNP release, indicating that LV geometry and afterload rather than filling pressures are the most important mechanical determinants of cardiac BNP release. The strength of our study compared with previous ones evaluating the relationship between plasma BNP levels and wall stress12,17 lies in the fact that we studied subjects with a broad range of LV dysfunction, including a group with normal LVEF and normal LV diastolic function. Most importantly, in our study we measured ΔA-CSBNP rather than peripheral BNP.
Similar to many cross-sectional studies revealing an association between higher plasma BNP and lower LVEF,18 we found a strong association between lower LVEF and higher ΔA-CSBNP. Physiologically, however, LVEF does not reflect a specific mechanical load that would contribute to the regulation of BNP gene expression or release by cardiomyocytes. Rather, we demonstrated that LVEF was strongly negatively correlated with higher LVESWS, which is a more likely regulator of BNP release. Therefore, we propose that impaired LVEF is a surrogate for LV dilatation and, thus, increased wall stress and, hence, higher ΔA-CSBNP and thereby also circulating BNP.
Although LVESWS in our model accounted for ≈50% of the variability of ΔA-CSBNP, clearly other hemodynamic or nonhemodynamic3,17 factors also play important roles. For example, in the LVEF/high e′ and normal LVEF/low e′ groups a significant difference in ΔA-CSBNP was evident, but LVESWS did not differ. As another novel finding, we report an increase in ΔA-CSBNP with decreasing s′, a measure of longitudinal LV systolic function, which, as shown by the present and previous studies,19 can be reduced despite normal LVEF. A previous study had shown a relationship between higher s′ and lower peripheral BNP.20 It is also possible that regional variability in release of BNP could occur in the context of local variation in mechanical stress, as might be present in the setting of mechanical dyssynchrony.
Similar to LVEF, s′ is also dependent on afterload,21 and this may also reflect structural alterations of the myocardium, such as fibrosis. Notably, s′ has been shown to be an independent predictor of mortality in a large community-based study,22 and elevated peripheral NT-proBNP was related to lower s′ in that population.23 In addition to the recognized natriuretic effects of BNP, it has also been suggested that BNP exerts local paracrine effects, including the inhibition of fibrosis.24 For example, data from a study in mice suggest that changes in ventricular BNP expression parallel changes in cardiac fibrosis, possibly in an effort to regulate the extent of fibrosis.25 Thus, we speculate that alterations in the structure of the LV myocardium, including fibrosis and cellular hypertrophy, for which s′ may be a surrogate, may go in parallel with alterations in the expression pattern of BNP.
We observed a reasonable but not perfect correlation between BNPA and ΔA-CS BNP highlighting the importance of clearance of BNP or NT-proBNP from the circulation by the kidneys and other mechanisms, which we did not assess in this study. Interpretation of the plasma BNP or NT-proBNP levels must be considered in both terms of the rate of cardiac release and the rate of clearance. It has been shown recently that NT-proBNP is cleared not only by the kidneys but by multiple tissues.26 In this context, previous studies have shown that BNP levels change to varying degrees in response to therapy,27,28 possibly reflecting the relative balance of alterations in the release and clearance of BNP and the resultant effect on peripheral steady-state BNP concentration. As such, our data highlight the fact that, whereas plasma BNP measurement may aid in the diagnosis of HF,1 it is not necessarily an index of cardiac BNP turnover.
BNP assays not only detect circulating BNP but also unprocessed proBNP. In patients with very advanced HF, hardly any circulating processed BNP was found,29,30 and the high BNP levels, as measured by standard assays in fact reflected proBNP,30 which has a lower biological activity compared with the processed BNP.30 This, together with a resistance to BNP on the receptor and postreceptor pathways,3 has been suggested as a mechanism for how the BNP system fails in HF.31 Given that we were interested in the input signals for cardiac BNP release, it seems to be appropriate, however, to use a conventional BNP assay, because both BNP and proBNP reflect BNP production.
Our study was limited by the relatively modest sample size; however, this study, for the first time, included subjects with a broad range of LV dysfunction, including asymptomatic volunteers. In addition, we did not measure coronary sinus flow, and, thus, the transcardiac gradients are only an estimate of myocardial BNP release. In some patients, in the low LVEF group there was a time interval between catheterization and echocardiography, although patients remained clinically stable during this interval. Importantly, echocardiography was not used to estimate pressures. To avoid catheterization of the left ventricle, we did not measure LV end-diastolic pressure directly but estimated by PCWP. Finally, we report only univariate associations as the number of subjects and the statistical technique limit the feasibility of a multivariable analysis.
Although the measurement of plasma BNP is widely performed as a biomarker for the presence of HF, the underlying driving stimulus for cardiac release of BNP has been poorly understood. Our study shows that systolic wall stress rather than diastolic wall stress is a major contributor to the release of BNP from the human heart. These observations may provide an impetus for the application of therapeutic manipulations that target afterload rather than preload in the setting of HF syndromes with elevated BNP when appropriate.
The excellent technical assistance by Donna Vizi, Jenny Starr, Liz Dewar, and Sofie Karapanagiotidis is greatly appreciated.
Sources of Funding
This study was supported by a Program Grant for the National Health and Medical Research Council of Australia to D.M.K., a joint National Health and Medical Research Council and National Heart Foundation of Australia Postgraduate Medical Scholarship (grant 317845) to J.A.M., and the Swiss National Science Foundation (grant PBZHB-121007 to M.T.M.).
- Received May 12, 2010.
- Revision received June 12, 2010.
- Accepted July 19, 2010.
Maisel A, Mueller C, Adams K Jr, Anker SD, Aspromonte N, Cleland JG, Cohen-Solal A, Dahlstrom U, DeMaria A, Di Somma S, Filippatos GS, Fonarow GC, Jourdain P, Komajda M, Liu PP, McDonagh T, McDonald K, Mebazaa A, Nieminen MS, Peacock WF, Tubaro M, Valle R, Vanderhyden M, Yancy CW, Zannad F, Braunwald E. State of the art: using natriuretic peptide levels in clinical practice. Eur J Heart Fail. 2008; 10: 824–839.
Clerico A, Recchia FA, Passino C, Emdin M. Cardiac endocrine function is an essential component of the homeostatic regulation network: physiological and clinical implications. Am J Physiol Heart Circ Physiol. 2006; 290: H17–H29.
Tsutamoto T, Wada A, Sakai H, Ishikawa C, Tanaka T, Hayashi M, Fujii M, Yamamoto T, Dohke T, Ohnishi M, Takashima H, Kinoshita M, Horie M. Relationship between renal function and plasma brain natriuretic peptide in patients with heart failure. J Am Coll Cardiol. 2006; 47: 582–586.
van Kimmenade RR, Januzzi JL Jr, Bakker JA, Houben AJ, Rennenberg R, Kroon AA, Crijns HJ, van Dieijen-Visser MP, de Leeuw PW, Pinto YM. Renal clearance of B-type natriuretic peptide and amino terminal pro-B-type natriuretic peptide a mechanistic study in hypertensive subjects. J Am Coll Cardiol. 2009; 53: 884–890.
Paulus WJ, Tschope C, Sanderson JE, Rusconi C, Flachskampf FA, Rademakers FE, Marino P, Smiseth OA, De Keulenaer G, Leite-Moreira AF, Borbely A, Edes I, Handoko ML, Heymans S, Pezzali N, Pieske B, Dickstein K, Fraser AG, Brutsaert DL. How to diagnose diastolic heart failure: a consensus statement on the diagnosis of heart failure with normal left ventricular ejection fraction by the Heart Failure and Echocardiography Associations of the European Society of Cardiology. Eur Heart J. 2007; 28: 2539–2550.
Yasue H, Yoshimura M, Sumida H, Kikuta K, Kugiyama K, Jougasaki M, Ogawa H, Okumura K, Mukoyama M, Nakao K. Localization and mechanism of secretion of B-type natriuretic peptide in comparison with those of A-type natriuretic peptide in normal subjects and patients with heart failure. Circulation. 1994; 90: 195–203.
Tsutamoto T, Sakai H, Ishikawa C, Fujii M, Tanaka T, Yamamoto T, Takashima H, Ohnishi M, Wada A, Horie M. Direct comparison of transcardiac difference between brain natriuretic peptide (BNP) and N-terminal pro-BNP in patients with chronic heart failure. Eur J Heart Fail. 2007; 9: 667–673.
Lang RM, Bierig M, Devereux RB, Flachskampf FA, Foster E, Pellikka PA, Picard MH, Roman MJ, Seward J, Shanewise JS, Solomon SD, Spencer KT, Sutton MS, Stewart WJ. Recommendations for chamber quantification: a report from the American Society of Echocardiography’s Guidelines and Standards Committee and the Chamber Quantification Writing Group, developed in conjunction with the European Association of Echocardiography, a branch of the European Society of Cardiology. J Am Soc Echocardiogr. 2005; 18: 1440–1463.
Iwanaga Y, Nishi I, Furuichi S, Noguchi T, Sase K, Kihara Y, Goto Y, Nonogi H. B-type natriuretic peptide strongly reflects diastolic wall stress in patients with chronic heart failure: comparison between systolic and diastolic heart failure. J Am Coll Cardiol. 2006; 47: 742–748.
Yu CM, Lin H, Yang H, Kong SL, Zhang Q, Lee SW. Progression of systolic abnormalities in patients with “isolated” diastolic heart failure and diastolic dysfunction. Circulation. 2002; 105: 1195–1201.
Troughton RW, Prior DL, Pereira JJ, Martin M, Fogarty A, Morehead A, Yandle TG, Richards AM, Starling RC, Young JB, Thomas JD, Klein AL. Plasma B-type natriuretic peptide levels in systolic heart failure: importance of left ventricular diastolic function and right ventricular systolic function. J Am Coll Cardiol. 2004; 43: 416–422.
Mogelvang R, Sogaard P, Pedersen SA, Olsen NT, Marott JL, Schnohr P, Goetze JP, Jensen JS. Cardiac dysfunction assessed by echocardiographic tissue Doppler imaging is an independent predictor of mortality in the general population. Circulation. 2009; 119: 2679–2685.
Mogelvang R, Goetze JP, Pedersen SA, Olsen NT, Marott JL, Schnohr P, Sogaard P, Jensen JS. Preclinical systolic and diastolic dysfunction assessed by tissue doppler imaging is associated with elevated plasma pro-B-type natriuretic peptide concentrations. J Card Fail. 2009; 15: 489–495.
Jourdain P, Jondeau G, Funck F, Gueffet P, Le Helloco A, Donal E, Aupetit JF, Aumont MC, Galinier M, Eicher JC, Cohen-Solal A, Juilliere Y. Plasma brain natriuretic peptide-guided therapy to improve outcome in heart failure: the STARS-BNP Multicenter Study. J Am Coll Cardiol. 2007; 49: 1733–1739.
Hawkridge AM, Heublein DM, Bergen HR III, Cataliotti A, Burnett JC Jr, Muddiman DC. Quantitative mass spectral evidence for the absence of circulating brain natriuretic peptide (BNP-32) in severe human heart failure. Proc Natl Acad Sci U S A. 2005; 102: 17442–17447.