All subjects participated in an ongoing study on haemodynamics (clinical trials registration NCT01742702). An announcement for recruitment was distributed at the Tampere University Hospital, University of Tampere, local occupational health care providers, Varala Sports Institute, and two announcements were published in a newspaper. By the time of the present analysis, 694 subjects had been recruited. The participants were interviewed and examined by a physician. Subjects with diabetes, coronary artery disease, cardiac insufficiency, atherosclerotic vascular disease, cerebrovascular disease, renal disease, or medication influencing CV status were excluded.

All participants gave written informed consent and the study was approved by the ethics committee of Tampere University Hospital District (study number R06086M). The investigation conforms to the principles outlined in the Declaration of Helsinki.

Blood and urine samples were obtained after about 12 h of fasting. Plasma sodium, potassium, calcium, glucose, creatinine, triglyceride, and total, high-density and low-density lipoprotein cholesterol concentrations were determined using Cobas Integra 700/800 or Cobas 6000, module c501 (F. Hoffmann-LaRoche Ltd, Basel, Switzerland), and blood cell count by ADVIA 120 or 2120 (Siemens Healthcare GmbH, Erlangen, Germany). Estimated glomerulus filtration rate (eGFR) was calculated using the Modification of Diet in Renal Disease (MDRD) formula [20].

An automatic tonometric sensor on the left wrist was used to continuously capture radial BP and pulse wave form (Colin BP-508 T, Colin Medical Instruments Corp., USA). The radial BP signal was calibrated every 2–4 min by contralateral brachial BP measurements [12, 17, 21, 22].

Changes in body electrical impedance during cardiac cycles were recorded using whole-body impedance cardiography (CircMon^R, JR Medical Ltd., Tallinn, Estonia) to determine HR, stroke volume, and cardiac output, as previously described [23‐25]. Cardiac output and stroke volume were related to estimated body surface area to derive cardiac index (l/min/m²) and stroke index (ml/m²). Systemic vascular resistance index (SVRI, dyn*s/cm⁵*m²) was calculated from tonometric BP signal and cardiac index. In supine position and during head-up tilt, the cardiac output values of CircMon^R are in good agreement with values measured using thermodilution [23, 24].

PWV was determined using whole-body impedance cardiography [25, 26]. As the impedance-based method slightly overestimates PWV when compared with Doppler ultrasound method, a validated equation was applied to calculate values that correspond to the ultrasound method (PWV = (PWVimpedance *0.696) + 0.864) [25].

The method for PWV measurement using whole body ICG has been described previously [25, 26]. Briefly, the distal impedance is recorded from popliteal artery at knee joint level, and the active electrode is placed on the lateral side of the knee and the reference electrode on the calf below knee. When the pulse pressure wave enters the aortic arch and the diameter of the aorta changes, the whole-body impedance decreases, and this is measured by the voltage electrodes. To calculate the PWV value, the CircMon software measures the time difference between onset of the decrease in impedance in the whole-body impedance signal and the popliteal artery signal (aortic-popliteal PWV). The aortic-popliteal impedance cardiography measurement of PWV has been validated against aortic-popliteal PWV measurements using Doppler ultrasound [25]. The PWV results obtained using CircMon show good repeatability, and normal values for PWV in 799 individuals (age 25–76 years) have been published [26]. We have also shown that the determination of stroke volume using impedance cardiography versus 3-dimensional echocardiography show good correlation [19]. In addition, we have performed carotid-femoral PWV measurements using applanation tonometry and compared them with ICG measurements of aortic-popliteal PWV, and the correlation between these methods was excellent (r = 0.82).

HRV analysis was used to assess cardiac autonomic tone. The electrocardiograms were recorded by the CircMon^R device at 200 Hz sampling rate, and data analyzed using Matlab software (MathWorks Inc., Natick, Massachusetts, USA). Normal R-R intervals were recognized, and a beat was considered ectopic if the interval differed more than 20 % from the previous values. The artifacts were processed using the cubic spine interpolation method [27]. The following frequency domain variables were calculated using the Fast Fourier Transformation method: i) power in low frequency (LF) range (0.04–0.15 Hz), ii) power in high frequency (HF) range (0.15–0.40 Hz), and iii) LF/HF ratio [28].

Caffeine containing products, smoking and a heavy meal for at least 4 h, and alcohol for at least 24 h prior to the investigation were to be avoided. Data was captured in a temperature-controlled laboratory by research nurses [12, 17, 22]. The left arm with the tonometric sensor was abducted 90 degrees in an arm support, which held it at the level of the heart in supine and upright positions. The recording comprised continuous capture of data during three consecutive 5-min periods (total 15 min): 5 min supine, 5 min of head-up tilt to 60 degrees, 5 min supine. The repeatability and reproducibility of the supine and upright measurements are good [22].

SVRI and cardiac index were chosen for the phenotypic classification, since peripheral arterial resistance and cardiac output are the principal determinants of BP [17, 23, 29]. To characterize the reaction to upright posture, the differences in SVRI and cardiac index between supine values (5th minute average) and upright values (8th and 10th minute averages) were examined. Two values during the head-up tilt (8th and 10th minute) were chosen to take into account possible changes during upright posture. To avoid scaling effects, all differences were standardized to have a mean value of 0 and a variance of 1. Hierarchical clustering was performed using Ward’s criterion for squared Euclidean distances: at first each subject served as a cluster, and at each step two clusters closest to each other were combined, until only one cluster remained (R software version 2.14.1 [30]). Based on the dendrogram, the 3 main clusters were analysed (Fig. 1; classification to clusters I, II and III). We also formed a practical rule for clustering, which was tested in a small validation population (Additional file 1).

Due to skewed distribution, PWV and HRV parameters were logarithmically transformed before statistical analyses. To evaluate the impact of phenotype on PWV, a L1-regression was fitted using sex, categorized BMI, categorized age, categorized hypertension, phenotype, interaction between sex and phenotype (Model 1); and fasting plasma lipids and glucose, and haematocrit (Model 2, with variables of Model 1 included) as variables. In addition, Model 1 was used to evaluate the impact of phenotype on HRV parameters. BMI, age, and presence of hypertension were categorized due to lack of linearity in the data. We performed additional analyses using continuous variables without categorization to examine the association of changes in SVRI and cardiac index with PWV.

There is no accepted cut-off level to define elevated BP during tilt-table measurements, while office BP is usually higher than home BP [5, 31]. As the supine BP during haemodynamic measurements was on average 8/12 mmHg (systolic/diastolic) lower than the seated office BP (n = 470), we applied the accepted cut-off level for home and ambulatory daytime measurements, i.e. values ≥135/85 mmHg [31], to define hypertension during the recordings. The same approach was applied previously, with testing of different cut-off levels [17].

As summary statistics, median and 25th to 75th percentile, or mean and standard deviation (SD) or 95 % confidence intervals (CI) were reported. Kruskal-Wallis rank sum test, Fisher’s exact test and Wilcoxon rank sum test were used to compare haemodynamics and demographics between the phenotypes. All analyses were performed using R software version 2.14.1 [30].

The study population consisted of 470 (240 female and 230 male) individuals aged 20–72 years (average 46 years). Average systolic and diastolic BP, SVRI and cardiac index in all subjects during the 15-min measurement protocol are shown in Fig. 2. In response to upright posture, systolic BP slightly decreased and diastolic BP increased, while SVRI increased by about 15 %, and cardiac index decreased by about 15 %. Both SVRI and cardiac index were stabilized during the time-points chosen for clustering into phenotypes (at 5 min of recording while supine, and at 8 and 10 min of recording while upright) (Fig. 2).

The classification to phenotypes was based on the changes in systemic vascular resistance and cardiac output in response to passive head-up tilt. The cluster that was characterised by the greatest increases in SVRI (+45 %) and the greatest decreases in cardiac index (−27 %) in response to upright posture was named ‘constrictor’ (Fig. 3a, d). The cluster characterised by the smallest changes in SVRI (+2 %) and cardiac index (−2 %), so that these variables were sustained on corresponding levels in supine and upright positions, was named ‘sustainer’ (Fig. 3c, f). The cluster with the intermediary changes (+22 %, -13 %, respectively) was named ‘intermediate’ (Fig. 3b, e). The mean (±SD) head-up tilt -induced changes in SVRI were +944 (±361), +554 (±197) and +60 (±254) dyn*s/cm⁵*m² in the constrictor, intermediate and sustainer phenotypes, respectively (Additional file 2). The mean changes in cardiac index were -0.89 (±0.51), -0.37 (±0.29) and -0.06 (±0.46) l/min/m², respectively (Additional file 2).

Since supine SVRI and cardiac index differed between the clusters, we performed additional analyses in which the subjects were classified to three new clusters merely on the basis of supine levels of SVRI and cardiac index. These new clusters did not systematically or significantly correspond to the constrictor, intermediate and sustainer clusters (data not shown). This indicates that the above phenotypes were not discovered because of differences in the resting levels of SVRI and cardiac index between the groups.

The characteristics and laboratory values of the phenotypes are shown in Table 1. Age did not differ between the groups (p = 0.216). The gender distribution (p < 0.001), prevalence of laboratory BP ≥135 and/or ≥85 mmHg (p = 0.009), BMI (p < 0.001), and office BP values (p < 0.008) were different in the phenotypes, i.e. the sustainers were more likely to be males with elevated BP and higher BMI than in the other phenotypes (Table 1). In addition, total and LDL cholesterol and triglycerides were lowest, and HDL cholesterol highest in the constrictors (Table 1). Fasting plasma glucose values were lowest in the constrictors, but the mean values were within the normal range in all groups. Haematocrit and creatinine concentrations were lowest in the constrictors, probably due to the female predominance, but the eGFR (MDRD-formula) values did not differ between the phenotypes (Table 1). In analyses adjusted for differences in gender distribution, prevalence of laboratory BP ≥135/85, and BMI, the above differences in total, LDL, and HDL cholesterol, triglycerides, and haematocrit were no more significant between the phenotypes (p-values for all >0.12).

The sustainers showed highest supine systolic BP, while upright systolic BP did not differ between the phenotypes (Additional files 3 and 4). Constrictors showed numerically lowest supine and upright diastolic BP. Supine heart rate was lowest and upright heart rate highest in sustainers, while supine stroke index was numerically highest and upright stroke index lowest in the constrictors (Additional file 3). Subsequently, supine cardiac index was highest but upright cardiac index was lowest in the constrictors. Supine SVRI was lowest in the constrictors, while upright SVRI was lowest in the sustainers (Additional file 3, Fig. 3).

Median PWV was different between the phenotypes, sustainers showing the highest values (7.35, 8.00 and 8.65 m/s in the constrictor, intermediate and sustainer phenotypes, respectively) (Fig. 4). In multivariate analysis, sustainer and intermediate phenotypes, high and middle BMI tertiles, high and middle age tertiles, BP ≥135/85 mmHg, and interaction between female sex and intermediate phenotype were significant explanatory factors for supine PWV (Table 2, Model 1). If the laboratory variables that were different between the phenotypes (plasma lipids and glucose, haematocrit) were included in the multivariate model, the outcome was that sustainer and intermediate phenotype, high and middle BMI tertiles, high and middle age tertiles, BP ≥135/85 mmHg, interaction between female sex and intermediate phenotype, and plasma triglyceride level were significant explanatory factors for supine PWV (Table 2, Model 2). In additional analyses using continuous variables without categorization, the supine to upright change in cardiac index was still independently associated with PWV, while the association between the change in SVRI and PWV was not significant (Additional file 5).

The HRV variables in supine and upright position are depicted in Fig. 5. No significant differences were found between the phenotypes in supine or upright power in LF range, a variable predominantly reflecting cardiac sympathetic tone [28, 32]. Supine power in HF range did not differ between the phenotypes, either. However, upright HF power was highest in the constrictors, followed by the intermediate phenotype, while the sustainer phenotype showed lowest level, suggesting reduced cardiac parasympathetic tone [28, 32]. The sympathovagal balance, as reflected by LF/HF ratio, was higher in supine position in the sustainers than in the constrictors, while upright LF/HF ratio was higher in the sustainers than in the two other phenotypes (Fig. 5). In all phenotypes, LF/HF ratio clearly increased in the upright position (p < 0.001). When the analyses were adjusted for differences in sex, BMI, and presence of hypertension, there were no statistically significant differences in the HRV variables between the phenotypes (Additional file 6).