Author + information
- Received January 16, 2011
- Revision received February 16, 2011
- Accepted March 7, 2011
- Published online July 19, 2011.
- Bonnie Ky, MD, MSCE⁎,†,⁎ (, )
- Benjamin French, PhD⁎,†,
- Kosha Ruparel, MS†,
- Nancy K. Sweitzer, MD, PhD‡,
- James C. Fang, MD§,
- Wayne C. Levy, MD∥,
- Douglas B. Sawyer, MD, PhD¶ and
- Thomas P. Cappola, MD, ScM⁎,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Bonnie Ky or Dr. Thomas P. Cappola, 3400 Spruce Street, 9054 Gates 34th and Civic Center Boulevard, 2PCAM, Philadelphia, Pennsylvania 19104
Objectives We sought to evaluate placental growth factor (PlGF) and soluble Fms-like tyrosine kinase 1 (sFlt-1) as clinical biomarkers in chronic heart failure (HF).
Background Vascular remodeling is a crucial compensatory mechanism in chronic HF. The angiogenic ligand PlGF and its target receptor fms-like tyrosine kinase 1 modulate vascular growth and function, but their relevance in human HF is undefined.
Methods We measured plasma PlGF and sFlt-1 in 1,403 patients from the Penn Heart Failure Study, a multicenter cohort of chronic systolic HF. Subjects were followed for death, cardiac transplantation, or ventricular assist device placement over a median follow-up of 2 years.
Results The sFlt-1 was independently associated with measures of HF severity, including New York Heart Association functional class (p < 0.01) and B-type natriuretic peptide (p < 0.01). Patients in the 4th quartile of sFlt-1 (>379 pg/ml) had a 6.17-fold increased risk of adverse outcomes (p < 0.01). This association was robust, even after adjustment for the Seattle Failure Model (hazard ratio: 2.54, 95% confidence interval [CI]: 1.76 to 2.27, p < 0.01) and clinical confounders including HF etiology (hazard ratio: 1.67, 95% CI: 1.06 to 2.63, p = 0.03). Combined assessment of sFlt-1 and B-type natriuretic peptide exhibited high predictive accuracy at 1 year (area under the receiver-operator characteristic curve: 0.791, 95% CI: 0.752 to 0.831) that was greater than either marker alone (p < 0.01 and p = 0.03, respectively). In contrast, PlGF was not an independent marker of disease severity or outcomes.
Conclusions Our findings support a role for sFlt-1 in the biology of human HF. With additional study, circulating sFlt-1 might emerge as a clinically useful biomarker to assess the influence of vascular remodeling on clinical outcomes.
Although heart failure (HF) is primarily a disorder of the myocardium, abnormal vascular function has a major impact on HF progression and cardiac remodeling (1,2). Angiogenic growth factors, such as the vascular endothelial growth factor (VEGF) family of proteins, govern numerous aspects of vessel homeostasis (3,4). In the setting of ischemic heart disease, alterations in angiogenic growth factors contribute to endothelial cell dysfunction and impaired revascularization (5). Even in the setting of nonischemic heart disease, angiogenic factors regulate myocardial capillary density as the heart hypertrophies (6,7), exert antiapoptoic and protrophic effects in dilated cardiomyopathy (8), and influence peripheral vascular load (9).
Placental growth factor (PlGF) is a member of the VEGF family of angiogenic proteins and is expressed in placental, cardiac, and lung tissue (10–12). Placental growth factor activates the Fms-like tyrosine kinase receptor 1 (Flt-1) and is expressed in numerous cell types, including endothelial cells, monocytes, and renal mesangial cells. The Flt-1 receptor has affinity for PlGF, VEGF-A, and VEGF-B. In animal models, these growth factors exert pleiotrophic effects, including potentially beneficial effects, such as the promotion of angiogenesis, and potentially harmful pro-inflammatory effects, such as the promotion of atherogenesis (10–12). Hence, the overall effect of PlGF/Flt-1 signaling in cardiovascular disorders might vary according to disease state and comorbidities.
To better understand PlGF/Flt-1 signaling in the setting of human disease, investigators have capitalized on the observations that both PlGF and the circulating form of the Flt-1 receptor—soluble Fms-like tyrosine kinase receptor 1 (sFlt-1)—can be easily quantified, providing a method to conveniently gauge overall PlGF and sFlt-1 activity in patients with cardiovascular disorders. During pregnancy, changes in circulating PlGF and sFlt-1 reflect impaired endothelial and glomerular function and predict preeclamptic risk (13,14). In patients with chest pain and acute coronary syndromes, higher PlGF levels are seen in those with myocardial infarction (MI) and are associated with an increased risk of short- and long-term adverse outcomes (12,15,16). Studies of circulating sFlt-1 have demonstrated conflicting results, with some studies noting higher levels during acute MI compared with control patients (11) and others noting lower plasma levels in patients during the acute phase of MI compared with control subjects (17,18).
Although PlGF and sFlt-1 might be important disease modifiers in chronic human HF, neither factor has been comprehensively studied in this population. The largest published experience on PlGF was a cross-sectional study of 98 patients that showed a positive relationship between PlGF levels and New York Heart Association (NYHA) functional class in ischemic HF but not in nonischemic disease (19). Circulating sFlt-1 has not been studied in human HF.
The purpose of our study was to evaluate circulating PlGF and sFlt-1 as clinical biomarkers in a multicenter cohort of 1,403 ambulatory HF outpatients. Our goals were: 1) to determine the factors that independently affect baseline levels of PlGF and sFlt-1 in chronic HF; and 2) to test the hypotheses that PlGF and sFlt-1 predict the combined outcome of ventricular assist device (VAD) placement, cardiac transplantation, or death.
The Penn Heart Failure Study is a multicenter prospective cohort study of outpatients with primarily chronic systolic HF recruited from referral centers at the University of Pennsylvania (Philadelphia, Pennsylvania), Case Western University (Cleveland, Ohio), and the University of Wisconsin (Madison, Wisconsin) (20). The primary inclusion criterion is a clinical diagnosis of HF. Participants are excluded if they have a noncardiac condition resulting in an expected mortality of <6 months, as judged by the treating physician, or if they were unable or unwilling to provide informed consent.
At time of study entry, detailed clinical data were obtained with a standardized questionnaire administered to the patient and physician, with verification via medical records. Venous blood samples were obtained at enrollment, processed, and stored at −80°C until time of assay. Follow-up events, including all-cause mortality, cardiac transplantation, and VAD placement, were prospectively ascertained every 6 months.
All participants provided written, informed consent, and the Penn Heart Failure Study protocol was approved by participating institutional review boards.
All biomarkers were measured from the same aliquot from a banked plasma sample that was obtained at time of study entry. The PlGF and sFlt-1were measured with prototype ARCHITECT immunoassays (Abbott Laboratories, Abbott Park, Illinois). The sFlt-1 immunoassay measures both free and bound sFlt-1, with an assay range of 15 to 50,000 pg/ml. The intra- and inter-assay coefficients of variation (CV) ranged from 1.3% to 5.2% and 1.9% to 5.9%, respectively. The PlGF immunoassay measures the free and not bound PlGF-1 with approximately 20% cross-reactivity with the PlGF-2 isoform, with an assay range from 1 to 1,500 pg/ml. The intraassay and interassay CV ranged from 1.4% to 6.7% and 1.8% to 6.7%, respectively.
The B-type natriuretic peptide (BNP) was measured with the ARCHITECT BNP chemiluminescent microparticle immunoassay (Abbott Laboratories) as previously described (21). The assay range was from 10 to 5,000 pg/ml. The intraassay and interassay CV ranged from 0.9% to 5.6% and 1.7% to 6.7%, respectively.
Baseline characteristics were summarized for the entire cohort with standard descriptive statistics. Independent determinants of sFlt-1 and PlGF levels were ascertained with a multivariable linear regression model for each marker. Adjustment variables were selected from a saturated model that included all variables listed in Online Tables 2 and 3. The inclusion of adjustment variables was based on a stepwise model-selection procedure that chose the subset of variables that minimized the Akaike information criterion. The following variables were retained on the basis of clinical/biological judgment, regardless of their impact on Akaike information criterion: age, sex, race, NYHA functional class, cardiomyopathy etiology, estimated glomerular filtration rate (eGFR), and BNP.
Cox regression models were used to determine the unadjusted association between sFlt-1 and PlGF and time to the combined outcome of all-cause death, cardiac transplantation, or VAD placement. Biomarkers were modeled as continuous variables and according to quartiles. Adjusted models included covariates on the basis of statistical evidence for confounding and clinical judgment. Statistical evidence was defined by a univariable association with the combined outcome at a significance level <0.10 and a >10% change in the estimated regression coefficient for each biomarker. Covariates that met these criteria were: age, sex, race, NYHA functional class, history of hypertension, history of diabetes, tobacco use, cardiomyopathy etiology, cardiac resynchronization, defibrillator, angiotensin-converting enzyme inhibitor/angiotensin receptor blocker use, aldosterone use, aspirin use, beta-blocker use, 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitor use, and body mass index. Age exhibited nonproportional hazards and was thus adjusted for with a time-varying covariate, which was obtained by multiplying age by a linear term for time. Adjustment for NYHA functional class was achieved by stratifying the baseline hazard function.
Clinical judgment included the consideration of candidate mediators of the observed association between biomarker and outcome on the basis of the known biology of these vascular growth factors. We decided a priori to not adjust for peripheral vascular disease, ejection fraction, pulse pressure, eGFR, and sodium, given the concern that each of these measures might represent causal pathway mediators of the association between vascular growth factors and adverse outcomes. These hypotheses were based upon the established biological effects of sFlt-1 and PlGF on renal dysfunction, vascular disease, and cardiac remodeling (7,8,22–26). We fit additional multivariable models to comprehensively assess the independence and predictive value of our observed associations in the context of validated clinical models by adjusting for the Seattle Heart Failure Model score, a standard risk prediction algorithm in HF (27).
The joint effects of sFlt-1 and BNP were evaluated by dividing the cohort into groups on the basis of the median level of each marker. In addition, time-dependent receiver-operator characteristic (ROC) curves were used to compare the ability of ln-transformed sFlt-1 and BNP to classify patients with regard to death, cardiac transplantation, or VAD placement at 1 year (28). Confidence intervals (CIs) for the area under the receiver-operator characteristic curve (AUC) were obtained from 1,000 bootstrapped samples, and AUCs were compared with Wald tests. All statistical analyses were completed with R software (version 2.11.0, R Foundation for Statistical Computing, Vienna, Austria), including the MASS, survival, and survivalROC packages (29–32).
Biomarker data were available for 1,535 subjects. Twenty-four subjects whose sFlt-1 or PlGF was greater than the 99th percentile were excluded a priori from all analyses, given that the levels in these patients are most likely to be indicative of the influence of non-HF disease states (e.g., pregnancy, infection, inflammation, lupus, recent surgery, or cancer) (13,33–38). Of these 24 patients, there were 6 without an identifiable non-HF cause of highly elevated biomarker levels. Inclusion of these 6 patients did not substantially change the results. Of the remaining 1,511 patients, complete data on all baseline characteristics and outcomes were available for 1,403 (93%) subjects. For each characteristic with any missing data, the amount of missingness averaged 1.5% and was no more than 1.7%. Those patients with any missing data did not differ systematically from the remainder of the cohort (Online Table 1).
The clinical characteristics of the 1,403 patients with complete data are shown in Table 1. The majority of the patients were male (67%) and Caucasian (74%), with a mean age across the cohort of 56 years. There were 423 patients (30%) with an ischemic cause of HF, 397 (28%) patients with a history of diabetes, and 817 (58%) with a history of hypertension.
Independent determinants of baseline sFlt-1 and PlGF levels
Across the cohort, the distributions of sFlt-1 and PlGF were approximately normal, with slightly heavier positive tails than would be expected if their distributions were truly normal. The sFlt-1 levels ranged from 115 to 2,012 pg/ml, with a mean ± SD of 348 ± 181 pg/ml. The median level was 308 (interquartile range: 258 to 379) pg/ml. The PlGF levels ranged from 0.7 to 42.3 pg/ml, with a mean ± SD of 19.4 ± 6.2 pg/ml. The median level was 18.6 (interquartile range: 15.0 to 22.7) pg/ml.
To establish the independent determinants of either biomarker, we first assessed the univariable associations between baseline levels of either biomarker and clinical characteristics (Table 1, Online Tables 2 and 3). Then we used multivariable models to determine clinical factors that independently influenced baseline levels of each biomarker. African-American race, higher NYHA functional class, hypercholesterolemia, and higher plasma BNP were each independently associated with higher levels of sFlt-1 (Table 2). Increasing age, aspirin use, beta blocker use, higher eGFR, and sodium were each independently associated with lower levels of sFlt-1. Given the positive skew of sFlt-1, we considered a sensitivity analysis in which we modeled ln-transformed sFlt-1 levels as the outcome rather than the nontransformed levels. The direction and magnitude of independent associations with ln-transformed sFlt-1 were similar to those provided in Table 2, with the exception that pulse pressure was independently associated with a significant decrease in sFlt-1 levels (1.4% decrease in sFlt-1/10-mm Hg increase in pulse pressure, p = 0.01).
For PlGF, increasing age, male sex, history of diabetes, higher pulse pressure, use of cardiac resynchronization, and higher BNP were each associated with higher levels (Table 3). African-American race, angiotensin-converting enzyme inhibitor or angiotensin receptor blocker use, and higher eGFR were each associated with lower levels of PlGF. Interestingly, ischemic etiology was not independently associated with either biomarker (p = 0.36 for sFlt-1; p = 0.19 for PlGF). Comparing Tables 2 and 3, it is apparent that sFlt-1 levels were associated more strongly with measures of HF severity (e.g., NYHA functional class, BNP), compared with PlGF.
sFlt-1 is independently associated with adverse outcomes in chronic HF
Over a median follow-up time of 2 years, there were 175 deaths, 103 transplants, and 27 VADs implanted. In unadjusted Cox models comparing the 4th versus 1st quartile, those patients with a circulating sFlt-1 level >379 pg/ml had a 6.17-fold increased risk of adverse outcomes (p < 0.01) (Table 4,Fig. 1A). After adjustment for demographic data, HF characteristics, and clinical measures including BNP, this association was attenuated in magnitude but remained statistically significant (hazard ratio [HR]: 1.67, 95% CI: 1.06 to 2.63, p = 0.03, comparing 4th quartile with 1st quartile). After adjustment for established clinical risk scores such as the Seattle Heart Failure Model, this association was attenuated to a lesser degree (HR: 2.54, 95% CI: 1.76 to 2.27, p < 0.01). Similar results were obtained when sFlt-1 was modeled as a continuous variable. In contrast, patients in the highest quartile of PlGF (>22.7 pg/ml) had only a 1.89-fold increased risk, which did not remain significant in adjusted models (Table 4, Fig. 1B). As in our cross-sectional analyses, these findings support a role for sFlt-1 as an independent biomarker of HF severity, whereas PlGF had no independent associations with outcomes.
Associations between vascular growth factors and outcomes might differ on the basis of the underlying etiology of HF, with prior published reports indicating increased relevance in ischemic disease (11,19). To explore these possibilities, we performed secondary analyses that included interaction terms between biomarker levels (modeled continuously) and HF etiology (ischemic or nonischemic). In contrast to previously published reports, there were no significant interactions by HF etiology on the associations between either marker and outcome in our adjusted models (interaction p = 0.18 for sFlt-1; p = 0.41 for PlGF).
Combined use of sFlt-1 and BNP in predicting outcomes
We explored the effects of joint assessment of sFlt-1 and the clinically used biomarker BNP in predicting adverse outcomes. There was a moderate correlation between levels of sFlt-1 and BNP (R = 0.54, p < 0.01), and their combined use was important in risk assessment (Table 5,Online Fig. 1). Compared with the referent group of patients with levels of both markers less than the median, patients with elevations in both sFlt-1 and BNP had a markedly elevated risk—in contrast with either marker alone—and this association remained significant in multivariable adjusted models (HR: 2.87, 95% CI: 1.96 to 4.21, p < 0.01). Furthermore, in the group of patients with high BNP levels, the combination of a high sFlt-1 level was associated with a 1.5- to 2-fold increase in risk, compared with those patients with low sFlt-1 levels (p < 0.01 unadjusted, p = 0.04 adjusted).
In ROC analysis at 1 year (Fig. 2), sFlt-1 and BNP in combination (AUC: 0.791, 95% CI: 0.752 to 0.831) showed greater accuracy in classifying patients who died or required heart transplantation or VAD placement than sFlt-1 alone (AUC: 0.735, 95% CI: 0.689 to 0.781, p < 0.01) or BNP alone (AUC: 0.766, 95% CI: 0.726 to 0.807, p = 0.03). These findings illustrate an improved ability to discern high- and low-risk HF patients at 1 year with both sFlt-1 and BNP, compared with BNP alone.
We report the first comprehensive assessment of PlGF and sFlt-1 as biomarkers in chronic HF. Our results indicate that sFlt-1 is strongly associated with adverse outcomes across a broad spectrum of disease, even after adjusting for existing standards such as the Seattle Heart Failure Model and natriuretic peptide levels. Furthermore, combined use of sFlt-1 and BNP might be superior to classifying patient risk than either biomarker alone. These findings support a role for sFlt-1 in the biology of human HF and suggest that, with additional study, circulating sFlt-1 might emerge as a clinically useful biomarker to assess the influence of vascular remodeling on clinical outcomes. In contrast, we found no evidence to support a role for circulating PlGF in chronic HF.
Studies in animal models suggest several potential mechanisms through which VEGF/Flt-1 signaling might modify the severity and course of HF. Soluble Flt-1 opposes angiogenesis by binding to and sequestering salutary VEGF ligands in the circulation, resulting in endothelial dysfunction and vascular rarefaction that increases mechanical load on the heart (39–41). Excess sFlt-1 might also increase myocardial fibrosis and decrease myocardial capillary density, thereby directly affecting myocardial structure and function (7,26). Soluble Flt-1 also impairs glomerular function and might contribute to unfavorable cardiorenal interactions. Exogenous administration of sFlt-1 to both pregnant and nonpregnant animals induces widespread endothelial dysfunction, hypertension, and renal dysfunction (13,23). We found that, consistent with these observations, sFlt-1 was independently associated with renal dysfunction (22). Discerning which of the effects of sFlt-1 is responsible for our observed clinical associations will require additional laboratory work.
Of note, we found that sFlt-1 was significantly associated with disease severity and clinical outcomes independent of HF etiology. These results provide human data supporting a role for angiogenic growth factors even in nonischemic disease, which has been observed in animal models. For example, Izumiya et al. (7) have shown that inhibiting angiogenic factors in mice subjected to pressure overload impair cardiac growth and accelerate the transition to HF. Similarly, VEGF signaling maintains endothelial cell homeostasis (2) and exerts anti-apoptotic effects in pacing-induced cardiomyopathy (8). Taken together with our findings, these results emphasize the importance of vascular growth remodeling in diverse forms of HF.
Studies of sFlt-1 in other cardiovascular diseases have associated elevated sFlt-1 with worse outcomes, consistent with our findings in HF. Soluble Flt-1 is elevated in patients with acute MI who subsequently develop HF (11) and in pregnant patients who subsequently develop preeclampsia (13,14). As noted in the preceding text, a plausible interpretation of these observations is that increased sFlt-1 reflects an underlying pathogenic process that accelerates endothelial dysfunction, renal dysfunction, vascular disease, and adverse remodeling by sequestering VEGF and associated ligands. However, Kodama et al. (18) have shown elevated sFlt-1 levels in response to atorvastatin treatment in a post-infarction study, and in this setting, a serial increase in sFlt-1 is associated with an improvement in left ventricular function. These findings suggest that the relationship between circulating sFlt-1 and outcomes might be complicated by the presence of acute ischemia and also by pharmacological therapy with statins. Although the associations we identified in chronic HF were independent of pharmacotherapies, we cannot comment on the impact of acute ischemia, which was not present in our study.
Although PlGF levels were associated with a variety of clinical factors in our cohort, including higher levels in renal dysfunction, hypertension, and diabetes, PlGF was not an independent marker of HF severity or adverse outcomes in adjusted models. This stands in contrast to other disease phenotypes in which PlGF is a predictor of disease outcomes, such as preeclampsia, acute coronary syndromes, and cancer (12,13,16,42). Thus, PlGF might be more relevant to disease states that are primarily vascular in etiology, with no current data to support a role for PlGF as a marker of risk in chronic HF.
The strong and independent associations among sFlt-1, NYHA functional class, and risk of adverse outcomes observed in our study support sFlt-1 as an HF biomarker, but whether sFlt-1 itself plays a causal role in HF progression cannot be determined from our observational study. Intervention studies are necessary to rigorously test this hypothesis and test the biological basis of our findings. Although the assays used were of high quality, we quantified biomarker levels from peripheral plasma, and there might be differences according to sampling site (11). Our results might also not be generalizable to populations of acute HF but represent chronic HF patients in a tertiary referral setting that includes a substantial representation of more severe disease. We currently do not have complete data on transplant urgency, exercise capacity, other biomarkers such as troponin, or quantitative cardiac remodeling data and as such are unable to determine the relationship between these parameters and sFlt-1 or PlGF. Finally, because this work is the first to assess the relevance of circulating sFlt-1 in chronic human HF, additional studies are necessary to validate this work and further define clinical effectiveness (43,44).
sFlt-1 is robustly associated with adverse outcomes in chronic HF, supporting a role for Flt-1 signaling in human HF progression. With further study, assessment of sFlt-1 might emerge as a useful clinical biomarker to enhance our ability to stratify patient risk above and beyond currently used approaches.
For a supplemental figure and tables, please see the online version of this article.
Dr. Ky was supported by the National Institutes of Health (NIH)/Clinical and Translational Science Award KL1 RR024132, NIH K23 HL095661-01, and the Heart Failure Society of America Research Fellowship Award. This work was also supported by NIH HL088577 (Dr. Cappola). Assay support was provided by Abbott Diagnostics. Neither the funding organizations nor Abbott Diagnostics had any role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; or preparation, review, or approval of the manuscript. Dr. Levy has received research support from Thoratec, General Electric, and Heartware; has received honoraria from GlaxoSmithKline and Boehringer Ingelheim; has licensing with the ACC Toolkit, Epocrates, and Seattle Heart Failure Model; has served on the Steering Committee of Amgen and Scios; has served on the Clinical Endpoint Committee of Cardiomems; and has done consulting with stock options for Cardiac Dimensions. Dr. Cappola reports receiving research support from Abbott Diagnostics. Drs. Ky and Cappola are co-inventors on a pending intellectual property application for the use of sFlt-1 as a biomarker in heart failure. All other authors have reported that they have no relationships to disclose.
- Abbreviations and Acronyms
- area under the receiver-operator characteristic curve
- B-type natriuretic peptide
- confidence interval
- coefficients of variation
- estimated glomerular filtration rate
- heart failure
- hazard ratio
- myocardial infarction
- New York Heart Association
- placental growth factor
- receiver-operator characteristic
- soluble Fms-like tyrosine kinase receptor 1
- ventricular assist device
- vascular endothelial growth factor
- Received January 16, 2011.
- Revision received February 16, 2011.
- Accepted March 7, 2011.
- American College of Cardiology Foundation
- Abraham D.,
- Hofbauer R.,
- Schafer R.,
- et al.
- Pepe M.,
- Mamdani M.,
- Zentilin L.,
- et al.
- Noori M.,
- Donald A.E.,
- Angelakopoulou A.,
- Hingorani A.D.,
- Williams D.J.
- Iwama H.,
- Uemura S.,
- Naya N.,
- et al.
- Lenderink T.,
- Heeschen C.,
- Fichtlscherer S.,
- et al.
- Kodama Y.,
- Kitta Y.,
- Nakamura T.,
- et al.
- Ky B.,
- Kimmel S.E.,
- Safa R.N.,
- et al.
- Di Marco G.S.,
- Reuter S.,
- Hillebrand U.,
- et al.
- Onoue K.,
- Uemura S.,
- Takeda Y.,
- et al.
- Giordano F.J.,
- Gerber H.P.,
- Williams S.P.,
- et al.
- Levy W.C.,
- Mozaffarian D.,
- Linker D.T.,
- et al.
- R Development Core Team
- Venables W.N.,
- Ripley B.D.
- Therneau T.,
- Lumley T.
- Heagerty P.J.,
- Saha P.
- Rini B.I.,
- Michaelson M.D.,
- Rosenberg J.E.,
- et al.
- Tammela T.,
- Enholm B.,
- Alitalo K.,
- Paavonen K.
- Willett C.G.,
- Duda D.G.,
- di Tomaso E.,
- et al.
- Hlatky M.A.,
- Greenland P.,
- Arnett D.K.,
- et al.
- Morrow D.A.,
- de Lemos J.A.