Circulatory factors associated with function and prognosis in patients with severe heart failure

Abstract

Background: Multiple circulatory factors are increased in heart failure (HF). Many have been linked to cardiac and/or skeletal muscle tissue processes, which in turn might influence physical activity and/or capacity during HF. This study aimed to provide a better understanding of the mechanisms linking HF with the loss of peripheral function.

Methods and results: Physical capacity measured by maximum oxygen uptake, myocardial function (measured by echocardiography), physical activity (measured by accelerometry), and mortality data was collected for patients with severe symptomatic heart failure an ejection fraction < 35% (n = 66) and controls (n = 28). Plasma circulatory factors were quantified using a multiplex immunoassay. Multivariate (orthogonal projections to latent structures discriminant analysis) and univariate analyses identified many factors that differed significantly between HF and control subjects, mainly involving biological functions related to cell growth and cell adhesion, extracellular matrix organization, angiogenesis, and inflammation. Then, using principal component analysis, links between circulatory factors and physical capacity, daily physical activity, and myocardial function were identified. A subset of ten biomarkers differentially expressed in patients with HF vs controls covaried with physical capacity, daily physical activity, and myocardial function; eight of these also carried prognostic value. These included established plasma biomarkers of HF, such as NT-proBNP and ST2 along with recently identified factors such as GDF15, IGFBP7, and TfR, as well as a new factor, galectin-4.

Conclusions: These findings reinforce the importance of systemic circulatory factors linked to hemodynamic stress responses and inflammation in the pathogenesis and progress of HF disease. They also support established biomarkers for HF and suggest new plausible markers.

Keywords: Multiplex immunoassay; New york heart association (NYHA) functional classification; Orthogonal projections to latent structures discriminant analysis (OPLS-DA); Principal component analysis (PCA).

Fig. 1

Correlation matrix illustrating mutual correlation amongst all biomarkers. Red and blue denote statistically significant positive and negative correlations, respectively. A large number of biomarkers are highly positively correlated and visual inspection indicates that there are clusters of variables with very high degree of correlation apparent, shown as bright red squares on the correlation matrix

Fig. 2

For the case–control analysis, we used OPLS which is similar to PCA, but developed to handle classification rather than correlation: an OPLS model will try to find the multidimensional direction in the X space that explains the maximum multidimensional variance direction in the Y space. OPLS regression is particularly suited when the matrix of predictors has more variables than observations and when there is multicolinearity among X values. The OPLS model differentiating patients form controls class correctly classified 84% (R²Y = 0.84) of the observations in the data set, with a predictive (Q₂ value) of 0.71 after cross validation

Fig. 3

PCA and biplot on physiological variables and individual observations, where red dots denote patients suffering an event and the size of the dot is proportional to the event, larger dots denote earlier events. There was a high overall correlation between all variables and 40% of the overall variance could be captured with the first two principal components. Variables related to daily physical activity had highest loading on the 1st component and echocardiographic variables on the second. Exercise capacity and heart rate contributed to both PC1 and PC2. The biplot indicates, as expected, the prognostic utility of the investigated physiological variables: Cox proportional hazard-ratio calculated on high vs low loading on PC1 and PC2 showed an HR of 1.8 and 2.8, respectively. With both components combined the hazard ratio (HR) for patients in lower left quadrant of the PCA was 4.0 (highest risk) compared with patients in upper right quadrant (lowest risk)

Fig. 4

Network inference edges a denote significant correlations and the length of each edge is inversely proportional to the strength of the correlation. Thus, nodes appearing closely together share higher number of significant edges, and large nodes indicate key markers with many significant connections. The network analysis identified 17 biomarkers, b relating to the majority of the clinical components and one, GDF15 had significant connections to all components. These key markers contained both classical biomarkers considered to reflect cardiac stretch (ST2 and BNP) but also a large number of inflammatory components and factors related to metabolism such as IGFBP7

Fig. 5

Biplot (a) and kaplan–meier curve (b) on the 16 key-network hubs vs mortality and individual observations, where red dots denote patients suffering an event and the size of the dot is proportional to the event, larger dots denote earlier event. There was a high overall correlation between all variables, and 61% of the overall variance could be captured with the first two principal components. The prognostic utility of the investigated physiological variables: Cox proportional hazard ratio calculated on high vs low loading on PC1 showed a HR of 3.5 (p < 0.001)