Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

In-silico simulated prototype-patients using TPMS technology to study a potential adverse effect of sacubitril and valsartan

  • Guillem Jorba ,

    Contributed equally to this work with: Guillem Jorba, Joaquim Aguirre-Plans

    Roles Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing

    ‡ These authors are joint senior authors on this work.

    Affiliations Anaxomics Biotech SL, Barcelona, Catalonia, Spain, Structural Bioinformatics Group, Research Programme on Biomedical Informatics, Department of Experimental and Health Science, Universitat Pompeu Fabra, Barcelona, Catalonia, Spain

  • Joaquim Aguirre-Plans ,

    Contributed equally to this work with: Guillem Jorba, Joaquim Aguirre-Plans

    Roles Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing

    ‡ These authors are joint senior authors on this work.

    Affiliation Structural Bioinformatics Group, Research Programme on Biomedical Informatics, Department of Experimental and Health Science, Universitat Pompeu Fabra, Barcelona, Catalonia, Spain

  • Valentin Junet,

    Roles Data curation, Writing – review & editing

    Affiliations Anaxomics Biotech SL, Barcelona, Catalonia, Spain, Institute of Biotechnology and Biomedicine, Universitat Autònoma de Barcelona, Cerdanyola del Vallès, Catalonia, Spain

  • Cristina Segú-Vergés,

    Roles Data curation, Formal analysis, Investigation, Writing – original draft, Writing – review & editing

    Affiliation Anaxomics Biotech SL, Barcelona, Catalonia, Spain

  • José Luis Ruiz,

    Roles Data curation, Writing – review & editing

    Affiliation Anaxomics Biotech SL, Barcelona, Catalonia, Spain

  • Albert Pujol,

    Roles Formal analysis, Methodology, Software, Writing – review & editing

    Affiliation Anaxomics Biotech SL, Barcelona, Catalonia, Spain

  • Narcís Fernández-Fuentes,

    Roles Conceptualization, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review & editing

    Affiliation Department of Biosciences, U Science Tech, Universitat de Vic-Universitat Central de Catalunya, Vic, Catalonia, Spain

  • José Manuel Mas ,

    Roles Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Resources, Software, Supervision, Validation, Writing – review & editing

    baldo.oliva@upf.edu (BJ); xouse@anaxomics.com (JMM)

    Affiliation Anaxomics Biotech SL, Barcelona, Catalonia, Spain

  • Baldo Oliva

    Roles Conceptualization, Formal analysis, Investigation, Methodology, Supervision, Validation, Writing – review & editing

    baldo.oliva@upf.edu (BJ); xouse@anaxomics.com (JMM)

    Affiliation Structural Bioinformatics Group, Research Programme on Biomedical Informatics, Department of Experimental and Health Science, Universitat Pompeu Fabra, Barcelona, Catalonia, Spain

In-silico simulated prototype-patients using TPMS technology to study a potential adverse effect of sacubitril and valsartan

  • Guillem Jorba, 
  • Joaquim Aguirre-Plans, 
  • Valentin Junet, 
  • Cristina Segú-Vergés, 
  • José Luis Ruiz, 
  • Albert Pujol, 
  • Narcís Fernández-Fuentes, 
  • José Manuel Mas, 
  • Baldo Oliva
PLOS
x

Abstract

Unveiling the mechanism of action of a drug is key to understand the benefits and adverse reactions of a medication in an organism. However, in complex diseases such as heart diseases there is not a unique mechanism of action but a wide range of different responses depending on the patient. Exploring this collection of mechanisms is one of the clues for a future personalized medicine. The Therapeutic Performance Mapping System (TPMS) is a Systems Biology approach that generates multiple models of the mechanism of action of a drug. Each molecular mechanism generated could be associated to particular individuals, here defined as prototype-patients, hence the generation of models using TPMS technology may be used for detecting adverse effects to specific patients. TPMS operates by (1) modelling the responses in humans with an accurate description of a protein network and (2) applying a Multilayer Perceptron-like and sampling strategy to find all plausible solutions. In the present study, TPMS is applied to explore the diversity of mechanisms of action of the drug combination sacubitril/valsartan. We use TPMS to generate a wide range of models explaining the relationship between sacubitril/valsartan and heart failure (the indication), as well as evaluating their association with macular degeneration (a potential adverse effect). Among the models generated, we identify a set of mechanisms of action associated to a better response in terms of heart failure treatment, which could also be associated to macular degeneration development. Finally, a set of 30 potential biomarkers are proposed to identify mechanisms (or prototype-patients) more prone of suffering macular degeneration when presenting good heart failure response. All prototype-patients models generated are completely theoretical and therefore they do not necessarily involve clinical effects in real patients. Data and accession to software are available at http://sbi.upf.edu/data/tpms/

Introduction

Systems biology methods are an increasingly recurring strategy to understand the molecular effects of a drug in complex clinical settings [1]. Some of these methods apply computer science techniques and mathematical approaches to simulate the responses of a drug. In 2005, the Virtual Physiological Human initiative was founded with the objective of developing computational models of patients [2]. Later, they defined the concept of In Silico Clinical Trials as “the use of individualized computer simulation in the development or regulatory evaluation of a medicinal product, medical device, or medical intervention” [3]. Since then, In Silico Clinical Trials have been adopted in several occasions in preclinical and clinical trials [1].

However, current methodologies do not consider the inter-patient variability intrinsic to pharmacological treatments, missing relevant information that should be incorporated into the models. Indeed, there are many parameters influencing the Mechanisms of Action (MoA) in such therapies, including demographic data of the patient, co-treatments or clinical history. Thus, by modelling all molecular mechanisms affected by the drug, the diversity of responses observed in patients during or after the treatment could be explained.

The Therapeutic Performance Mapping System (TPMS) [4] is a method used to elucidate all the possible MoAs that could exist between an input drug and a pathology or adverse effect. It is a systems biology approach based on the simulation of patient-specific protein-protein interaction networks. TPMS incorporates data from different resources and uses the information from the drugs and diseases under study to generate multiple models of potential MoAs. In the last years, TPMS has been broadly used in different clinical areas and with different objectives [512], in some cases being validated in the posterior experiments [6,11,12]. Our working hypothesis is that a set of MoAs can represent the different responses to a drug in cells and that a real population of patients is the result of a myriad of cell responses. Thus, we define a prototype-patient as an abstract case with all cells responding to a single MoA.

Here, we propose the application of TPMS and protein-network approaches in the specific case study of the drug combination sacubitril/valsartan, used for the treatment of Heart Failure (HF). HF is becoming a major health problem in the western world due to its increasing hospitalization rates [13], with a prevalence being influenced by many factors like age, nutritional habits, lifestyles or genetics. This complicates the development of treatments and the identification of universal biomarkers to stratify the population. To facilitate this segmentation, it is necessary to understand the molecular details of the treatment and the pathology. Sacubitril/valsartan (marketed by Novartis as Entresto®) is a drug combination that shows better results than conventional treatments by reducing cardiovascular deaths and heart failure (HF) readmissions [14]. In pharmacological terms, it is an angiotensin receptor-neprilysin inhibitor. Consequently, it triggers the natriuretic peptide system by inhibiting neprilysin (NEP) and inhibits renin-angiotensin-aldosterone system by blocking the type-1 angiotensin II receptor (AT1R) [15]. In a previous work, TPMS was already applied to unveil the MoA of sacubitril/valsartan synergy, revealing its effect against two molecular processes [9]: the left ventricular extracellular matrix remodeling, mediated by proteins like gap junction alpha-1 protein or matrix metalloproteinase-9; and the cardiomyocyte apoptosis, through modulation of glycogen synthase kinase-3 beta. However, several publications warned about the potential long-term negative implications of using a neprilysin inhibitor like sacubitril [1519]. Neprilysin plays a critical role at maintaining the amyloid-β homeostasis in the brain, and the alteration of amyloid-β levels has been linked to a potential long-term development of Alzheimer’s disease or Macular Degeneration (MD) [15,17,1921]. During the clinical trials PARADIGM-HF and PARAGON-HF with sacubitril/valsartan no serious effects were detected [14,22]. Still, their patient follow-up was relatively short and not specialized in finding neurodegenerative specific symptoms. For this reason, in a forthcoming PERSPECTIVE trial (NCT02884206) a battery of cognitive tests was taken [18]. In line with this, the application of systems biology methods may shed light to the potential relationship between the treatment and the adverse effect.

In this study, we used TPMS and GUILDify v2.0 to analyze the relationship between sacubitril/valsartan, HF and MD in entirely theoretical models. Because these are theoretical models it is important to note that they are not associated with clinical effects in real patients, they only point on potential mechanisms to explain potential adverse effects. We analyzed a population of MoAs that describe the possible protein links from a sacubitril/valsartan treatment to HF and MD phenotypes. We clustered the MoAs in groups according to their response intensity and labelled them as high or low efficacy of treating HF and possibility of causing MD. We then compared these sets of MoAs and proposed a list of biomarkers to identify potential cases of MD when using sacubitril/valsartan. Simultaneously, we used GUILDify v2.0 web server [23] as an alternative approach to compare the biomarkers proposed by TPMS and reinforce the results.

Materials and methods

1. Biological Effectors Database (BED) to molecularly describe specific clinical conditions

Biological Effectors Database (BED) [5,24] describes more than 300 clinical conditions as sets of genes and proteins (effectors) that can be “active”, “inactive” or “neutral”. For example, in a metabolic protein-like network, an enzyme will become “active” in the presence of a catalyst, or become inactivated when interacting with an inhibitor (see further details in supplementary material).

2. TPMS modelling

The Therapeutic Performance Mapping System (TPMS) is a tool that creates mathematical models of the protein pathways underlying a drug/pathology to explain a clinical outcome or phenotype [410]. These models find MoAs that explain how a Stimulus (i.e. proteins activated or inhibited by a drug) produces a Response (i.e. proteins active or inhibited in a phenotype). In the present case study, we applied TPMS to the drug-indication pair sacubitril/valsartan and HF. Regarding the drug, we retrieved the sacubitril/valsartan targets from DrugBank [25], PubChem [26], STITCH [27], SuperTarget [28] and hand curated literature revision. As for the indication, we retrieved the proteins associated with the phenotype from the BED [5,24].

2.1. Building the Human Protein Network (HPN).

To apply the TPMS approach and create the mathematical models of MoAs, a Human Protein Network (HPN) is needed beforehand. In this study, we used a protein-protein interactions network created from the integration of public and private databases: KEGG [29], BioGRID [30], IntAct [31], REACTOME [32], TRRUST [33], and HPRD [34]. In addition, information extracted from scientific literature, which was manually curated, was also included and used for trimming the network. The resulting HPN considers interactions corresponding to different tissues to take into account the effect of the Stimulus in the whole body.

2.2. Defining active/inactive nodes.

We define the state of human proteins as active or inactive for a particular phenotype, including its expression (as active) or repression (as inactive) extracted from the GSE57345 gene expression dataset [35] as in Iborra-Egea et al [9] (see further details in supplementary material).

2.3. Description of the mathematical models.

The algorithm of TPMS takes as input signals the activation (+1) and inactivation (-1) of the drug target proteins, and as output the BED protein states of the pathology. It then optimizes the paths between both protein sets and computes the activation and inactivation values of all proteins in the HPN. Each node of the protein network receives as input the output of the incoming connected nodes and every link is given a weight (ωl). The sum of inputs is transformed by a hyperbolic tangent function that generates a score for every node, which becomes the “output signal” towards the outgoing connected nodes. The ωl parameters are obtained by optimization, using a Stochastic Optimization Method based on Simulated Annealing [36]. The models are then trained by using the general restrictions (i.e. defined as edges and nodes with the property of being active or inactive) and the specific conditions set by the user. Details of the approach are shown in Fig 1 and supplementary material.

thumbnail
Fig 1. Scheme of how to apply TPMS to find the Mechanisms of Action (MoA) of a drug.

(a) Scheme of the method, transmitting information over the Human Protein Network (HPN) using a Multilayer Perceptron-like and sampling. (b) After a given number of iterations, we obtain a collection of Mechanisms of Actions (MoA). Rows represent the MoAs and columns the output signal values of the proteins (nodes of the network). The final column shows the accuracy of the model as a percentage of the number restrictions accomplished. (c) 200 MoAs are selected (coloured in the slide) and sorted by TSignal. The first quartile is defined as the Low-disease group, and the fourth quartile as High-disease group. The distribution of the output signals of the two groups of MoA are shown in (d) (High-disease in red and Low-disease is in blue).

https://doi.org/10.1371/journal.pone.0228926.g001

3. Measures to compare sets of MoAs

To understand the relationships between all potential mechanisms we defined some measures of comparison between different sets of solutions. We expect that a drug will revert the conditions of a disease phenotype; subsequently, a drug should inactivate the active protein effectors of a pathology-phenotype and activate the inactive ones. In this section we describe the measures used in the present study to analyze and compare sets of MoAs from different views (see further details in supplementary material).

3.1. TSignal.

To quantify the intensity of the response of a MoA, we defined TSignal as the average signal arriving at the protein effectors (equation in supplementary material).

3.2. Distance between two sets of MoAs.

We used the modified Hausdorff distance (MHD) introduced by Dubuisson and Jain [37] as the distance between two or more sets of MoAs in order to determine their similarity. Details of the equations are explained in the supplementary material.

3.3. Potential biomarkers extracted from MoAs.

In order to extract potential biomarkers when comparing sets of MoAs, we first defined the best-classifier proteins. These are proteins inside the HPN that allow to better classify between groups of models and are identified following a Data-Science strategy (see supplementary material). Best-classifier proteins are usually strongly related to the intensity of a response and are proteins with values differently distributed between the groups of MoAs analyzed. For this study, and for the sake of simplicity, we focused only on the 200 proteins (or pair of proteins) showing the higher classification accuracy. Assuming the hypothesis that the selected MoAs are representative of individual prototype-patients, these proteins could be used as biomarkers to classify a cohort of patients.

Then, we applied the Mann-Whitney U test to compare the distributions of the best-classifier proteins values between the groups and selected those proteins with significant difference (p-value< 0.01). We also restricted the list to proteins having an average value with opposite sign among groups (i.e. positive vs. negative or vice versa) and named them as differential best-classifier proteins. By following this strategy, we can identify two groups of differential best-classifier proteins: those active in the first group (positive output signal in average) and inactive in the other (negative output signal in average), and the opposite.

Results and discussion

We applied TPMS to the HPN using as input signals the drug targets of sacubitril/valsartan (NEP / AT1R) and as output signals the proteins associated with HF extracted from the BED. Out of all MoAs found by TPMS, we selected the 200 satisfying the largest number of restrictions (and at least 80% of them) to perform further analysis.

Note that TPMS was only executed once, optimizing the results to satisfy the restrictions on HF data. The values of MD are obtained by measuring the signal arriving at the MD effectors, which are part of the HPN and also receive signal. This procedure was chosen because we defined HF as the indication of the drug (sacubitril/valsartan), while MD is a potential adverse effect.

1. Stratification of MoAs

In order to compare models related to a good or bad response to the treatment, or those more prone to lead towards potential MD adverse effect, we stratified the MoAs. For HF, or treatment response, MoAs were ranked by their TSignal and then split in four quartiles. The first quartile (top 25%) contains MoAs with higher intensity of the response, which in turn corresponds to lower values of the effectors associated with HF phenotype (we named them as “Low”-disease MoAs). On the contrary, the fourth quartile (bottom 25%) collects MoAs with lower intensity of response (thus, we named as “High”-disease MoAs) (S1 File). On the other hand, for MD, the first quartile (top 25%) contains MoAs with higher intensity, which as an adverse event, correspond to models with high values of the effectors associated to MD (we named them as High-adverseEvent MoAs). The fourth quartile (bottom 25%) collects MoAs with lower intensity of response (thus, we named as Low- adverseEvent MoAs) (S1 File). Note that, in the following steps and because HF and MD groups were extracted from the same 200 set of models, common MoAs between different HF and MD-defined sets could be expected.

2. Comparison of MoAs with high/low TSignal associated to HF or MD

We calculated the modified Hausdorff distance between the groups of MoAs (High-MD, Low-MD, High-HF and Low-HF) to elucidate their similarity values (S1 File). In this sense, the higher the distance between the groups is, the more different they are. We used these distances to calculate a dendrogram tree (see S1 File) showing that MoAs associated with a bad response to sacubitril/valsartan for HF (high-HF) are more similar (i.e. closer) to MoAs linked to a stronger MD adverse effect (high-MD). It is remarkable that the distances between Low- and High-HF and between Low- and High-MD are larger than the cross distances between HF and MD. However, by the definition of distance (equation 3 in supplementary material), it cannot account for the dispersion among the MoAs within and between each group. Therefore, for each set we calculated the mean Euclidean distance between all the points and its center, defined by the average of all points (see S1 File). As a result, all groups showed very similar dispersion values.

In order to have a global and graphical view of the distance between the individual MoAs, we generated a multidimensional scaling (MDS) plot calculated using MATLAB (see Fig 2). MDS plots display the pairwise distances in two dimensions while preserving the clustering characteristics (i.e. close MoAs are also close in the 2D-plot and far MoAs are also far in 2D). Focusing on the Low-HF group depicted in blue circles, we observe that there is no clear tendency to cluster with any of the MD groups. There are few cases of Low-HF MoAs coinciding in the space with Low- or High-MD MoAs. This implies that a good response to sacubitril/valsartan of HF patients would not be usually linked to the development of MD. Moreover, no clear distinction is found when plotting only the MD MoAs within the Low-HF group (see S1 File). However, regarding the set of High-HF MoAs, we can differentiate two clusters of MoAs: one related to the High-MD group (green crosses); and the other close to MoAs of the Low-MD group (black crosses) (see S1 File).

thumbnail
Fig 2. Multidimensional scaling plot of the distances between the Mechanisms of Action (MoA) of the four groups defined.

Each point represents a MoA. Axes are defined by the most representative dimensions.

https://doi.org/10.1371/journal.pone.0228926.g002

Assuming the hypothesis that different MoAs correspond to distinct prototype-patients, we conclude that for the specific set of patients for which sacubitril/valsartan works best reducing HF, it would be more difficult to differentiate between those presenting MD and those who do not. Instead, for the High-HF group, patients having MD could indeed be easily distinguished from those not presenting MD as side effect. However, because Low-HF group has more relevance to the clinics, specific functional analyses were performed in this specific group, as seen in following sections. Finally, we highlight that, as these distinct groups of prototype-patients are theoretical simulations, they don’t reflect the clinical effects of real patients.

3. Identification and functional analysis of potential biomarkers

For this section, we identified the nodes (i.e. proteins) significantly differentiating two groups of models (using a Mann-Whitney U test) for which the average of output signals have opposite signs (see methods in 3.3). After that, the function of the identified proteins was extracted from Gene Ontology (GO).

3.1. Identification of best-classifier proteins differentiating HF responses.

After comparing High- vs Low- HF groups, we found a total of 45 differential best-classifier proteins associated with the treatment response (6 Low-HF-active/High-HF-inactive and 39 Low-HF-inactive/High-HF-active) (see Fig 3A and S1 File). To pinpoint the biological role of these proteins, we first identified the GO enriched functions (see S1 File) and then searched in the literature for evidences linking them with HF. As a result, we found that the differential best-classifier proteins Low-HF-active/High-HF-inactive point towards an important role for actin nucleation and polymerization mechanisms in drug response (reflected by the functions regulation of actin nucleation, regulation of Arp2/3 complex-mediated actin nucleation, SCAR complex, filopodium tip, or dendrite extension). In fact, the alteration of actin nucleation and polymerization mechanisms has been reported in heart failure [3840]. Interestingly, a role for the activation of another differential best-classifier candidate, ATGR2, has been proposed to mediate some of the beneficial effects of angiotensin II receptor type 1 antagonists, such as valsartan [41,42]. On the other hand, the results of the differential best-classifier proteins Low-HF-inactive/High-HF-active are linked to phosphatidylinositol kinase mediated pathways (phosphatidylinositol-3,4-bisphosphate 5-kinase activity) and MAP kinase mediated pathways (MAP kinase kinase activity, best classifier proteins MAPK1, MAPK3, MAPK11, MAPK12 or MAPK13). In this case, both signaling pathways have been associated to cardiac hypertrophy and subsequent heart failure [43,44]. These outcomes clearly lead towards the idea that High-HF models are a representation of prototype-patients with a worst response to the treatment, while Low-HF models are related to more beneficial response to the medication. A more detailed explanation can be found in the supplementary material.

thumbnail
Fig 3. Scatter plot of the mean signal values of Low and High-“disease” Mechanisms of Action (MoA).

Scatter plot of the mean signal values of Low-“disease” and High-“disease” MoAs for each protein using as disease Heart Failure (HF) in (a) and Macular Degeneration (MD) in (b). The average of the output signal of each protein in High-group is presented versus its value in Low-group. Differential signals (Diff., shown as triangles) are defined as those with opposite sign when comparing High versus Low average, and a p-value < 0.01 when calculating the Mann-Whitney U test between the two distributions of signals. Best-classifier proteins (BCP) are colored in red, otherwise they are blue. Sizes of markers are proportional to p-values of the Mann-Whitney U test.

https://doi.org/10.1371/journal.pone.0228926.g003

3.2. Identification of best-classifier proteins differentiating MD responses.

We identified 57 differential best-classifier proteins of MD (28 Low-MD-active/High-MD-inactive and 29 Low-MD-inactive/High-MD-active) (see Fig 3B and S1 File). Again, we searched for relationships between these proteins and MD by identifying the GO enriched functions (see S1 File) and searching for links in the literature. Some of the proteins and functions highlighted in the current analysis had been related to MD in previous works. The presence of dendritic spine development and dorsal/ventral axon guidance related proteins emphasizes the role of sacubitril/valsartan in dendritic and synaptic plasticity mechanisms, which had been previously linked to MD [45]. Furthermore, valsartan treatment has been reported to promote dendritic spine development in other related neurodegenerative diseases, such as Alzheimer’s disease [46]. Other enriched functions are implicated in growth factor related pathways, which are known to be involved in wet MD pathogenesis [47]. Moreover, neovascularization in the wet variant of MD has been linked to the signaling of some of the growth factors detected as sacubitril/valsartan-associated MD classifiers in this study, including FGF1 [47] and PDGF [48,49]. A more detailed explanation can be found in the supplementary material.

3.3. Identification of potential biomarkers differentiating MD responses in Low-HF.

Because of its clinical relevance, we decided to focus on analyzing the special case of prototype-patients in which the treatment reduces HF (Low-HF) but produces MD adverse effect (High-HF). In order to find these prototype-patients, we: (i) identified 13 Low-HF ∩ Low-MD MoAs and 12 Low-HF ∩ High-MD MoAs; and (ii) compared the protein signal of the two groups and proposed 30 potential biomarkers (Table 1). Among the proposed biomarkers, we found 16 proteins active in Low-HF ∩ Low-MD MoAs but inactive in Low-HF ∩ High-MD (15 of them shared with MD best-classifier proteins). On the other hand, 14 proteins were identified as inactive in Low-HF ∩ Low-MD and active in Low-HF ∩ High-MD MoAs (12 of them were MD best-classifier proteins). We calculated the GO enriched functions of these two groups and observed that “phosphatidylinositol bisphosphate kinase activity” is enriched among proteins that are active in Low-HF ∩ Low-MD MoAs. Instead, “fibrinolysis” was found to be enriched among proteins active in Low-HF ∩ High-MD MoAs (Table 2). With this, we conclude that among the group of prototype-patients for which sacubitril/valsartan improves HF treatment response, the modulation of fibrinolysis could play a role at inducing the MD adverse effect. Moreover, we propose 12 best-classifier proteins that may be considered as biomarkers for good prognosis of the side effect.

thumbnail
Table 1. Potential biomarker proteins, with opposite signal in Low-HF ∩ Low-MD and Low-HF ∩ High-MD MoAs.

https://doi.org/10.1371/journal.pone.0228926.t001

thumbnail
Table 2. Top 10 Gene Ontology functions enriched from proteins with opposite signal in Low-HF ∩ Low-MD and Low-HF ∩ High-MD MoAs.

https://doi.org/10.1371/journal.pone.0228926.t002

In fact, since neovascular MD development is characterized by subretinal extravasations of novel vessels derived from the choroid (CNV) and the subsequent hemorrhage into the photoreceptor cell layer in the macula region [51], it might be reasonable to think that the modulation of fibrinolysis and blood coagulation pathways could play a role. The reported implication of some fibrinolysis related classifiers, such as FGB, SERPINE1 (PAI-1), and SERPING1, in neovascular MD development seems to support this hypothesis [5254]. Besides, valsartan might be implicated in this mechanism, since it has been reported to modulate PAI-1 levels and promote fibrinolysis in different animal and human models [55,56]. In addition, the presence of several other MD related classifiers in this list, such as IRS2 [57], PTGS2 [58], DCN [59] and FGF1 [60], further supports the interest of the classifiers as biomarkers of MD development in sacubitril/valsartan good responders. Still, we would like to highlight that the biomarkers have been proposed using a theoretical approach, and that the clinical effects studied may not be present in real patients.

4. Analysis of proposed biomarkers with GUILDify

In the previous section, we proposed 30 proteins that could potentially help to identify HF patients at risk of developing MD. To corroborate these biomarkers, we tested how many of them are found using a different approach also based on the use of functional networks. For this purpose, we used GUILDify v2.0 [23], a web server that extends the information of disease-gene associations through the protein-protein interactions network. GUILDify scores proteins according to their proximity with the genes associated with a disease (seeds). Using this web server, we identify a list of top-scoring proteins that are critical on transmitting the perturbation of disease genes through the network. The network used by GUILDify is completely independent from the HPN used in the TPMS, becoming an ideal, independent context to test the potential biomarkers.

Thus, we used GUILDify to indicate which of the potential biomarkers identified by TPMS may have a relevant role in the molecular mechanism of the drug. We ran GUILDify using the two targets of sacubitril/valsartan (NEP, AT1R) as seeds, and selected the top 2% scored nodes (defined as the “top-drug” set). We did the same with the phenotypes of HF and MD, using as seeds the 124 effectors of HF and 163 effectors of MD from the BED database. We merged the top scored sets of HF, MD and top-drug (“top-drug ∪ top-HF ∪ top-MD”) and studied the overlap with the set of 30 biomarkers proposed in the previous section. 10 of the candidate biomarkers are found in the merged set “top-drug ∪ top-HF ∪ top-MD” and are consequently significant (see S1 File).

Some of these candidates can be functionally linked to both diseases and the drug under study. For example, among these 10 classifiers, AGER has been implicated in both HF [61], through extracellular matrix remodeling, and MD development [62], through inflammation, oxidative stress, and basal laminar deposit formation between retinal pigment epithelium cells and the basal membrane; furthermore, this receptor is known to be modulated by AT1R [63], valsartan target. Similarly, FGF1 has been proposed to improve cardiac function after HF [64], as well as to promote choroid neovascularization leading to MD [47]. Moreover, FGF1 is regulated by angiotensin II through ATGR2 [65], another protein suggested as classifier in the current analysis that is known to mediate some of the effects of AT1R antagonists, such as valsartan [41,42]. Another candidate, NRG1, has been linked to myocardial regeneration after HF [66] and is known to lessen the development of neurodegenerative diseases such as Alzheimer’s disease [67], which shares similar pathological features with MD [68]. NRG1 is also linked to the expression of neprilysin [67], sacubitril target. ITGB5 has been identified as risk locus for HF [69] and its modulation has been linked to lipofucsin accumulation in MD [70]. Interestingly, ATGR1 inhibitors have been reported to modulate ITGB5 expression in animal models [71]. Finally, IL1A has been proposed as an essential mediator of HF pathogenesis [72,73] through inflammation modulations, and serum levels of this protein have been found increased in MD patients [74]. In addition, as described in previous sections, classifiers FGB, SERPINE1, and SERPING1 have been linked to MD [5254] and are also known to play a role in HF development [7578]. According to these findings, the 10 potential biomarkers proposed by TPMS and identified with GUILDify might be prioritized when studying good responder HF patients at risk of MD development.

Limitations

Although TPMS returns the amount of signal from the drug arriving to the rest of the proteins in the HPN, this signal is only a qualitative measure. We are not using data about the dosage of the drug or the quantity of expression of the proteins. However, we are already working to make TPMS move towards the growing tendency of Quantitative Systems Pharmacology. The quantification of the availability of drugs in the target tissue for each patient opens the opportunity to have an accurate patient simulation to do in silico clinical trials.

Conclusions

It exists an increasing need for new tools to get closer to real life clinical problems and the Systems Biology-based computational methods could be the solution needed. The specific case of sacubitril/valsartan stands out because of the amount of resources invested in the safety of the drug and the concern on the possible risk of inducing amyloid accumulation-associated conditions, such as macular degeneration (MD), in the long term. In this study, we applied TPMS technology to uncover different Mechanisms of Action (MoAs) of sacubitril/valsartan over heart failure (HF) and reveal its molecular relationship with MD. For this approach, we hypothesize that each MoA would correspond to a prototype-patient. The method is then used to generate a wide battery of MoAs by performing an in silico trial of the drug and pathology under study. TPMS computes the models by using a hand curated Human Protein Network and applying a Multilayer Perceptron-like and sampling method strategy to find all plausible solutions. After analyzing the models generated, we found different sets of proteins able to classify the models according to HF treatment efficacy or MD treatment relationship. The sets include functions such as PI3K and MAPK kinase signaling pathways, involved in HF-related cardiac hypertrophy, or fibrinolysis and coagulation processes (e.g. FGB, SERPINE1 or SERPING1) and growth factors (e.g. FGF1 or PDGF) related to MD induction. Furthermore, we propose 30 biomarker candidates to identify patients potentially developing MD under a successful treatment with sacubitril/valsartan. Out of this 30, 10 biomarkers were also found in the alternative, independent molecular context proposed by GUILDify, including some HF and MD effectors such as AGER, NRG1, ITGB5 or IL1A. Further studies might prospectively validate the herein raised hypothesis.

We notice that the models generated with TPMS are completely theoretical and thus, they are not associated with clinical effects of real patients. Consequently, the biomarkers proposed on the basis of these models are also theoretical and would require an experimental validation. Still, TPMS represents a huge improvement for studying the hypothetical relationship between a drug and an adverse effect. Until now, there were not enough tools that allow to perform an exhaustive study on the MoAs of an adverse effect. Now, with the MoAs and biomarkers proposed by TPMS, we provide the tools for this type of research.

Supporting information

S1 File. Extended version of materials and methods; S1-S5 Figs; S1-S13 Tables.

https://doi.org/10.1371/journal.pone.0228926.s001

(DOCX)

References

  1. 1. Pappalardo F, Russo G, Tshinanu FM, Viceconti M. In silico clinical trials: concepts and early adoptions. Brief Bioinform. 2018; 1–10.
  2. 2. Viceconti M, Clapworthy G. The virtual physiological human: Challenges and opportunities. 2006 3rd IEEE International Symposium on Biomedical Imaging: From Nano to Macro—Proceedings. 2006. pp. 812–815. https://doi.org/10.1109/isbi.2006.1625042
  3. 3. Viceconti M, Henney A, Morley-Fletcher E. In silico clinical trials: how computer simulation will transform the biomedical industry. Int J Clin Trials. 2016;3: 37.
  4. 4. Anaxomics Biotech SL. TPMS technology [Internet]. 2018. Available: http://www.anaxomics.com/tpms.php
  5. 5. Pujol A, Mosca R, Farrés J, Aloy P. Unveiling the role of network and systems biology in drug discovery. Trends Pharmacol Sci. 2010;31: 115–123. pmid:20117850
  6. 6. Herrando-Grabulosa M, Mulet R, Pujol A, Mas JM, Navarro X, Aloy P, et al. Novel Neuroprotective Multicomponent Therapy for Amyotrophic Lateral Sclerosis Designed by Networked Systems. PLoS One. 2016;11: e0147626. pmid:26807587
  7. 7. Gómez-Serrano M, Camafeita E, García-Santos E, López JA, Rubio MA, Sánchez-Pernaute A, et al. Proteome-wide alterations on adipose tissue from obese patients as age-, diabetes- and gender-specific hallmarks. Sci Rep. 2016;6: 1–15.
  8. 8. Perera S, Artigas L, Mulet R, Mas JM, Sardón T. Systems biology applied to non-alcoholic fatty liver disease (NAFLD): treatment selection based on the mechanism of action of nutraceuticals. Nutrafoods. 2014;13: 61–68.
  9. 9. Iborra-Egea O, Gálvez-Montón C, Roura S, Perea-Gil I, Prat-Vidal C, Soler-Botija C, et al. Mechanisms of action of sacubitril/valsartan on cardiac remodeling: a systems biology approach. npj Syst Biol Appl. 2017;3: 1–8.
  10. 10. Romeo-Guitart D, Forés J, Herrando-Grabulosa M, Valls R, Leiva-Rodríguez T, Galea E, et al. Neuroprotective Drug for Nerve Trauma Revealed Using Artificial Intelligence. Sci Rep. 2018;8: 1879. pmid:29382857
  11. 11. Lorén V, Garcia-Jaraquemada A, Naves JE, Carmona X, Mañosa M, Aransay AM, et al. Anp32e, a protein involved in steroid-refractoriness in ulcerative colitis, identified by a systems biology approach. J Crohn’s Colitis. 2019;13: 351–361. pmid:30329026
  12. 12. Iborra-Egea O, Santiago-Vacas E, Yurista SR, Lupón J, Packer M, Heymans S, et al. Unraveling the Molecular Mechanism of Action of Empagliflozin in Heart Failure With Reduced Ejection Fraction With or Without Diabetes. JACC Basic to Transl Sci. 2019;4: 831–840. pmid:31998851
  13. 13. Van Riet EES, Hoes AW, Wagenaar KP, Limburg A, Landman MAJ, Rutten FH. Epidemiology of heart failure: The prevalence of heart failure and ventricular dysfunction in older adults over time. A systematic review. Eur J Heart Fail. 2016;18: 242–52. pmid:26727047
  14. 14. McMurray JJV, Packer M, Desai AS, Gong J, Lefkowitz MP, Rizkala AR, et al. Angiotensin–Neprilysin Inhibition versus Enalapril in Heart Failure. N Engl J Med. 2014;371: 993–1004. pmid:25176015
  15. 15. Singh JSS, Burrell LM, Cherif M, Squire IB, Clark AL, Lang CC. Sacubitril/valsartan: Beyond natriuretic peptides. Heart. 2017;103: 1569–1577. pmid:28689178
  16. 16. Ponikowski P, Voors AA, Anker SD, Bueno H, Cleland JGF, Coats AJS, et al. 2016 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure. Eur Heart J. 2016;37: 2129–2200m. pmid:27206819
  17. 17. Feldman AM, Haller JA, DeKosky ST. Valsartan/sacubitril for heart failure: Reconciling disparities between preclinical and clinical investigations. JAMA—J Am Med Assoc. 2016;315: 25–26. pmid:26641736
  18. 18. Riddell E, Vader JM. Potential Expanded Indications for Neprilysin Inhibitors. Current Heart Failure Reports. 2017. pp. 134–145. pmid:28281174
  19. 19. Zhang Z lu, Li R, Yang FY, Xi L. Natriuretic peptide family as diagnostic/prognostic biomarker and treatment modality in management of adult and geriatric patients with heart failure: Remaining issues and challenges. J Geriatr Cardiol. 2018;15: 540–546. pmid:30344534
  20. 20. Baranello RJ, Bharani KL, Padmaraju V, Chopra N, Lahiri DK, Greig NH, et al. Amyloid-Beta Protein Clearance and Degradation (ABCD) Pathways and their Role in Alzheimer’s Disease. Curr Alzheimer Res. 2015;12: 32–46. pmid:25523424
  21. 21. Ohno-Matsui K. Parallel findings in age-related macular degeneration and Alzheimer’s disease. Prog Retin Eye Res. 2011;30: 217–238. pmid:21440663
  22. 22. Solomon SD, Rizkala AR, Gong J, Wang W, Anand IS, Ge J, et al. Angiotensin Receptor Neprilysin Inhibition in Heart Failure With Preserved Ejection Fraction: Rationale and Design of the PARAGON-HF Trial. JACC Hear Fail. 2017;5: 471–482. pmid:28662936
  23. 23. Aguirre-Plans J, Piñero J, Sanz F, Furlong LI, Fernandez-Fuentes N, Oliva B, et al. GUILDify v2.0: A Tool to Identify Molecular Networks Underlying Human Diseases, Their Comorbidities and Their Druggable Targets. J Mol Biol. 2019; 30117–2. pmid:30851278
  24. 24. Anaxomics Biotech SL. Biological Effectors Database [Internet]. 2018. Available: http://www.anaxomics.com/biological-effectors-database.php
  25. 25. Wishart DS, Feunang YD, Guo AC, Lo EJ, Marcu A, Grant JR, et al. DrugBank 5.0: A major update to the DrugBank database for 2018. Nucleic Acids Res. 2018;46: D1074–D1082. pmid:29126136
  26. 26. Kim S, Thiessen PA, Bolton EE, Chen J, Fu G, Gindulyte A, et al. PubChem substance and compound databases. Nucleic Acids Res. 2016;44: D1202–D1213. pmid:26400175
  27. 27. Szklarczyk D, Santos A, Von Mering C, Jensen LJ, Bork P, Kuhn M. STITCH 5: Augmenting protein-chemical interaction networks with tissue and affinity data. Nucleic Acids Res. 2016;44: D380–D384. pmid:26590256
  28. 28. Hecker N, Ahmed J, von Eichborn J, Dunkel M, Macha K, Eckert A, et al. SuperTarget goes quantitative: update on drug-target interactions. Nucleic Acids Res. 2011;40: D1113–D1117. pmid:22067455
  29. 29. Kanehisa M, Furumichi M, Tanabe M, Sato Y, Morishima K. KEGG: New perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res. 2017;45: D353–D361. pmid:27899662
  30. 30. Chatr-Aryamontri A, Oughtred R, Boucher L, Rust J, Chang C, Kolas NK, et al. The BioGRID interaction database: 2017 update. Nucleic Acids Res. 2017;45: D369–D379. pmid:27980099
  31. 31. Orchard S, Ammari M, Aranda B, Breuza L, Briganti L, Broackes-Carter F, et al. The MIntAct project—IntAct as a common curation platform for 11 molecular interaction databases. Nucleic Acids Res. 2014;42: 358–363. pmid:24234451
  32. 32. Fabregat A, Jupe S, Matthews L, Sidiropoulos K, Gillespie M, Garapati P, et al. The Reactome Pathway Knowledgebase. Nucleic Acids Res. 2018;46: D649–D655. pmid:29145629
  33. 33. Han H, Cho JW, Lee S, Yun A, Kim H, Bae D, et al. TRRUST v2: An expanded reference database of human and mouse transcriptional regulatory interactions. Nucleic Acids Res. 2018;46: D380–D386. pmid:29087512
  34. 34. Keshava Prasad TS, Goel R, Kandasamy K, Keerthikumar S, Kumar S, Mathivanan S, et al. Human Protein Reference Database—2009 update. Nucleic Acids Res. 2009;37: D767–D772. pmid:18988627
  35. 35. Liu Y, Morley M, Brandimarto J, Hannenhalli S, Hu Y, Ashley EA, et al. RNA-Seq identifies novel myocardial gene expression signatures of heart failure. Genomics. 2015;105: 83–9. pmid:25528681
  36. 36. Collet P, Rennard J-P. Stochastic Optimization Algorithms. Intell Inf Technol. 2011; 1121–1137.
  37. 37. Dubuisson M-P, Jain AK. A modified Hausdorff distance for object matching. Proc 12th Int Conf Pattern Recognit. 1994;1: 566–568. https://doi.org/10.1109/ICPR.1994.576361
  38. 38. Patel VB, Wang Z, Fan D, Zhabyeyev P, Basu R, Das SK, et al. Loss of p47phox subunit enhances susceptibility to biomechanical stress and heart failure because of dysregulation of cortactin and actin filaments. Circ Res. 2013;112: 1542–56. pmid:23553616
  39. 39. Karsanov NV, Pirtskhalaishvili MP, Semerikova VJ, Losaberidze NS. Thin myofilament proteins in norm and heart failure I. Polymerizability of myocardial Straub actin in acute and chronic heart failure. Basic Res Cardiol. 1986;81: 199–212. pmid:3741358
  40. 40. Childers RC, Sunyecz I, West TA, Cismowski MJ, Lucchesi PA, Gooch KJ. Role of the Cytoskeleton in the Development of a Hypofibrotic Cardiac Fibroblast Phenotype in Volume Overload Heart Failure. Am J Physiol Heart Circ Physiol. 2018;316: H596–H608. pmid:30575422
  41. 41. Liu YH, Yang XP, Sharov VG, Nass O, Sabbah HN, Peterson E, et al. Effects of angiotensin-converting enzyme inhibitors and angiotensin II type 1 receptor antagonists in rats with heart failure: Role of kinins and angiotensin II type 2 receptors. J Clin Invest. 1997;99: 1926–35. pmid:9109437
  42. 42. Schrier RW, Abdallah JG, Weinberger HHD, Abraham WT. Therapy of heart failure. Kidney Int. 2000;57: 1418–25. pmid:10760077
  43. 43. Aoyagi T, Matsui T. Phosphoinositide-3 kinase signaling in cardiac hypertrophy and heart failure. Curr Pharm Des. 2011;17: 1818–24. pmid:21631421
  44. 44. Ennis I, Aiello E, Cingolani H, Perez N. The Autocrine/Paracrine Loop After Myocardial Stretch: Mineralocorticoid Receptor Activation. Curr Cardiol Rev. 2013;9: 230–40. pmid:23909633
  45. 45. Sullivan RKP, WoldeMussie E, Pow DV. Dendritic and synaptic plasticity of neurons in the human age-related macular degeneration retina. Investig Ophthalmol Vis Sci. 2007;48: 2782–91. pmid:17525213
  46. 46. Sohn YI, Lee NJ, Chung A, Saavedra JM, Scott Turner R, Pak DTS, et al. Antihypertensive drug Valsartan promotes dendritic spine density by altering AMPA receptor trafficking. Biochem Biophys Res Commun. 2013;439: 464–70. pmid:24012668
  47. 47. Frank RN. Growth factors in age-related macular degeneration: Pathogenic and therapeutic implications. Ophthalmic Res. 1997;29: 341–53. pmid:9323725
  48. 48. Glenn JV, Stitt AW. The role of advanced glycation end products in retinal ageing and disease. Biochim Biophys Acta—Gen Subj. 2009;1790: 1109–16. pmid:19409449
  49. 49. Grossniklaus HE, Green WR. Choroidal neovascularization. Am J Ophthalmol. 2004;137: 496–503. pmid:15013874
  50. 50. Berriz GF, Beaver JE, Cenik C, Tasan M, Roth FP. Next generation software for functional trend analysis. Bioinformatics. 2009;25: 3043–3044. pmid:19717575
  51. 51. Nowak JZ. AMD-the retinal disease with an unprecised etiopathogenesis: In search of effective therapeutics. Acta Pol Pharm—Drug Res. 2014;71: 900–16.
  52. 52. Yuan X, Gu X, Crabb JS, Yue X, Shadrach K, Hollyfield JG, et al. Quantitative Proteomics: Comparison of the Macular Bruch Membrane/Choroid Complex from Age-related Macular Degeneration and Normal Eyes. Mol Cell Proteomics. 2010;9: 1031–46. pmid:20177130
  53. 53. Lee AY, Kulkarni M, Fang AM, Edelstein S, Osborn MP, Brantley MA. The effect of genetic variants in SERPING1 on the risk of neovascular age-related macular degeneration. Br J Ophthalmol. 2010;94: 915–7. pmid:20606025
  54. 54. Higgins P. Balancing AhR-Dependent Pro-Oxidant and Nrf2-Responsive Anti-Oxidant Pathways in Age-Related Retinopathy: Is SERPINE1 Expression a Therapeutic Target in Disease Onset and Progression? J Mol Genet Med. 2015;8: 101. pmid:25237384
  55. 55. Miyata M, Ikeda Y, Nakamura S, Sasaki T, Abe S, Minagoe S, et al. Effects of Valsartan on Fibrinolysis in Hypertensive Patients With Metabolic Syndrome. Circ J. 2012;76: 843–51. pmid:22451451
  56. 56. Oubiña MP, De las Heras N, Vázquez-Pérez S, Cediel E, Sanz-Rosa D, Ruilope LM, et al. Valsartan improves fibrinolytic balance in atherosclerotic rabbits. J Hypertens. 2002;20: 303–10. pmid:11821716
  57. 57. Albert-Fort M, Hombrebueno JR, Pons-Vazquez S, Sanz-Gonzalez S, Diaz-Llopis M, Pinazo-Durán MD. Retinal neurodegenerative changes in the adult insulin receptor substrate-2 deficient mouse. Exp Eye Res. 2014;124: 1–10. pmid:24792588
  58. 58. Zhang R, Liu Z, Zhang H, Zhang Y, Lin D. The COX-2-selective antagonist (NS-398) inhibits choroidal neovascularization and subretinal fibrosis. PLoS One. 2016;11: e0146808. pmid:26760305
  59. 59. Wang X, Ma W, Han S, Meng Z, Zhao L, Yin Y, et al. TGF-β participates choroid neovascularization through Smad2/3-VEGF/TNF-α signaling in mice with Laser-induced wet age-related macular degeneration. Sci Rep. 2017;7: 9672. pmid:28852052
  60. 60. Skeie JM, Zeng S, Faidley EA, Mullins RF. Angiogenin in age-related macular degeneration. Mol Vis. 2011;17: 576–82. pmid:21364907
  61. 61. Hegab Z, Gibbons S, Neyses L, Mamas M. Role of advanced glycation end products in cardiovascular disease. World J Cardiol. 2012;4: 90–102. pmid:22558488
  62. 62. Banevicius M, Vilkeviciute A, Kriauciuniene L, Liutkeviciene R, Deltuva VP. The Association Between Variants of Receptor for Advanced Glycation End Products (RAGE) Gene Polymorphisms and Age-Related Macular Degeneration. Med Sci Monit. 2018;24: 190–199. pmid:29317590
  63. 63. Pickering RJ, Tikellis C, Rosado CJ, Tsorotes D, Dimitropoulos A, Smith M, et al. Transactivation of RAGE mediates angiotensin-induced inflammation and atherogenesis. J Clin Invest. 2019;129: 406–421. pmid:30530993
  64. 64. Garbayo E, Gavira JJ, De Yebenes MG, Pelacho B, Abizanda G, Lana H, et al. Catheter-based intramyocardial injection of FGF1 or NRG1-loaded MPs improves cardiac function in a preclinical model of ischemia-reperfusion. Sci Rep. 2016;6: 25932. pmid:27184924
  65. 65. Lakó-Futó Z, Szokodi I, Sármán B, Földes G, Tokola H, Ilves M, et al. Evidence for a Functional Role of Angiotensin II Type 2 Receptor in the Cardiac Hypertrophic Process in Vivo in the Rat Heart. Circulation. 2003;108: 2414–22. pmid:14568903
  66. 66. Galindo CL, Ryzhov S, Sawyer DB. Neuregulin as a heart failure therapy and mediator of reverse remodeling. Curr Heart Fail Rep. 2014;11: 40–9. pmid:24234399
  67. 67. Xu J, De Winter F, Farrokhi C, Rockenstein E, Mante M, Adame A, et al. Neuregulin 1 improves cognitive deficits and neuropathology in an Alzheimer’s disease model. Sci Rep. 2016;6: 31692. pmid:27558862
  68. 68. Kaarniranta K, Salminen A, Haapasalo A, Soininen H, Hiltunen M. Age-related macular degeneration (AMD): Alzheimer’s disease in the eye? J Alzheimer’s Dis. 2011;24: 615–31. pmid:21297256
  69. 69. Verweij N, Eppinga RN, Hagemeijer Y, Van Der Harst P. Identification of 15 novel risk loci for coronary artery disease and genetic risk of recurrent events, atrial fibrillation and heart failure. Sci Rep. 2017;7: 2761. pmid:28584231
  70. 70. Kaarniranta K, Sinha D, Blasiak J, Kauppinen A, Veréb Z, Salminen A, et al. Autophagy and heterophagy dysregulation leads to retinal pigment epithelium dysfunction and development of age-related macular degeneration. Autophagy. 2013;9: 973–84. pmid:23590900
  71. 71. Kawano H, Cody RJ, Graf K, Goetze S, Kawano Y, Schnee J, et al. Angiotensin II Enhances Integrin and α-Actinin Expression in Adult Rat Cardiac Fibroblasts. Hypertension. 2012;35: 273–9. pmid:10642310
  72. 72. Bujak M, Frangogiannis NG. The role of IL-1 in the pathogenesis of heart disease. Arch Immunol Ther Exp (Warsz). 2009;57: 165–76. pmid:19479203
  73. 73. Turner NA. Effects of interleukin-1 on cardiac fibroblast function: Relevance to post-myocardial infarction remodelling. Vascul Pharmacol. 2014;60: 1–7. pmid:23806284
  74. 74. Nassar K, Grisanti S, Elfar E, Lüke J, Lüke M, Grisanti S. Serum cytokines as biomarkers for age-related macular degeneration. Graefe’s Arch Clin Exp Ophthalmol. 2015;253: 699–704. pmid:25056526
  75. 75. Zhang YN, Vernooij F, Ibrahim I, Ooi S, Gijsberts CM, Schoneveld AH, et al. Extracellular vesicle proteins associated with systemic vascular events correlate with heart failure: An observational study in a dyspnoea cohort. PLoS One. 2016;11: e0148073. pmid:26820481
  76. 76. Zaman AKMT, French CJ, Schneider DJ, Sobel BE. A Profibrotic Effect of Plasminogen Activator Inhibitor Type-1 (PAI-1) in the Heart. Exp Biol Med. 2009;234: 246–54. pmid:19144865
  77. 77. Messaoudi S, Azibani F, Delcayre C, Jaisser F. Aldosterone, mineralocorticoid receptor, and heart failure. Mol Cell Endocrinol. 2012;350: 266–72. pmid:21784127
  78. 78. Chakravarthy U, Wong TY, Fletcher A, Piault E, Evans C, Zlateva G, et al. Clinical risk factors for age-related macular degeneration: A systematic review and meta-analysis. BMC Ophthalmol. 2010;10: 31. pmid:21144031