Companion diagnostic for the chloroquine use in the treatment of COVID-19: systems biology report of candidate markers.

diagnostic for the use of chloroquine for the treatment of COVID-19 patients. Some of the markers, e.g. albumin, soluble endoglin and amyloid precursor protein have records of clinical correlations of their expression and cardiac adverse effects. Other proteins are candidates for companion diagnostic in clinical trials of chloroquine in the treatment of COVID-19 infection.


Background
The use of chloroquine for the treatment of COVID-19 patients has been under discussion (1)(2)(3). To be effective, chloroquine has to act on its targets that would lead to a therapeutic response. To discriminate between responding, non-responding and adverse effects-prone patients, there is a need of a companion diagnostic for chloroquine. Such markers are routine in oncology (4,5). These markers inform clinicians whether a drug would be useful for a given patient. Without these markers, an effect of drugs is frequently non-conclusive when evaluated at the population level. Companion diagnostics allows selection of responsive patients and prediction of the disease development.
Chloroquine has been used since the 1940th. Studies of this remedy generated information about molecular mechanisms of its action. An international DrugBank depository (drugbank.ca) is an example of a curated and proven drug target database (6). The studies of COVID- 19 are not yet as extensive as studies of chloroquine, but there are already reports of COVID-19 targets in human cells (7)(8)(9). Identi ed targets re ect molecular mechanisms engaged by chloroquine and COVID-19, and systems biology allows identi cation of these regulatory processes. A number of network building tools and high-quality databases are available for systemic analysis of molecular mechanisms engaged by COVID-19 and chloroquine (7,(10)(11)(12)(13). An analysis of regulatory networks is the most comprehensive way to explore mechanisms that are initiated or dependent on the targets of COVID-19 and chloroquine. Comprehensiveness is ensured by the incorporation of the experimental data from hundreds to thousands of reports. For example, UniProt database contains 562,755 records of experimental data (uniprot.org) (14). This a rich source for systemic network analysis.
COVID-19 infection manifests in many different clinical symptoms (15)(16)(17). It indicates that the virus employs different molecular mechanisms and attacks different types of cells. Here we report an identi cation of potential markers to evaluate the e cacy of chloroquine in the treatment of COVID-19 patients. Our systemic analysis identi ed 266 nodes, i.e. genes and proteins that represent common molecular mechanisms engaged by chloroquine and COVID-19. An example of cardiac arrhythmia showed 19 potential companion diagnostic markers for chloroquine use and prediction of cardiac adverse effects.

Methods
The datasets for building networks were collected as follows, and are listed in Supplementary Table S1. For chloroquine, the targets were retrieved from the Drug Bank depository (drugbank.ca) (6). For COVID-19 interacting proteins, 322 interactors were reported by Gordon et al., and ACE2 and TMPRSS2 were used (7,18). For arrhythmia, markers described by Bose et al. were used (19).
The networks building and analysis was performed in Cytoscape (10). The signi cance for the inclusion of nodes and edges was set to p < 0.05. For the building of the networks, we used the UniProt database (14). For extraction of intersections, the "Network Analysis" tool of Cytoscape was used. Statistical signi cance of network building (inclusion of nodes and con dence of edges) was set on p < 0.05. BiNGO tool was used for the analysis of affected biological processes. For statistical signi cance, the level was set at p < 0.05, and the hypergeometric statistical test was used, with Benjamini and Hochberg false discovery rate correction.
A cross-validation analysis of identi ed nodes with published reports about their clinical values and a role in physiology was performed. We searched PubMed with the Medical Subject Headings (MeSH) of a node and words "COVID-19", "chloroquine", and "heart". Retrieved publications were scrutinized for information about clinical values of the nodes as markers and for involvement of the nodes in molecular mechanisms and biological processes of relevance for a virus infection, predictive marker value, correlation with clinical outputs and adverse effects, and a role in crucial intracellular regulatory mechanisms, e.g. proliferation, death and differentiation of cells.

Results
Identi cation of common targets of COVID-19 and chloroquine For chloroquine, there have been reported 11 direct targets, i.e. GSTA2, TNF, TLR9, GST, HMGB1, GSTM1, CYP2C8, CYP3A4, CYP3A5, CYP2D6 and CYP1A1 (Supplementary Table S1). Chloroquine impact on these targets may lead to engagement of a regulatory network containing 1,336 nodes and 2,526 edges (Supplementary Figure S1; Supplementary File S1, network "Chloroquine_UniProt"). The network was built with the retrieval of interaction data from the UniProt database. The same database was used to build networks of angiotensin-converting enzyme 2 (ACE2) and type 2 transmembrane serine protease (TMPRSS2) and COVID-19 interactors that are listed in Supplementary Table S1. The structure of the networks are shown in Supplementary Figures S2 and S3, and the networks are presented in Supplementary File S1 (networks "Cov_UniProt" and "ACE2TMPRSS2_UniProt"). The ACE2/TMPRSS2 network contains 15 nodes and 19 edges, and the COVID-19 network contains 828 nodes and 1,545 edges. These 3 networks represent molecular mechanisms engaged by chloroquine and COVID-19 directly or via ACE2-TMPRSS2. Note that the graphical presentation of the networks is to illustrate structure of the networks. Cytoscape Session le (Supplementary File S1) provides access to the networks and allows exploration of the networks, zooming on identi ers, perform selection of sub-networks, clustering and search for biological processes of clinical relevance.
To identify mechanisms shared by COVID-19 and chloroquine, we searched for intersections between these 3 networks. The intersection of the chloroquine and ACE2/TMPRSS2 networks extracted only 2 nodes, i.e. albumin and 14-3-3 zeta/delta. This shows that chloroquine has rather a narrow impact on ACE2 and TMPRSS2-dependent mechanisms. The intersection of the chloroquine and COVID-19 target networks extracted 266 nodes interconnected by 347 edges ( Figure 1A; Supplementary Table S2, Supplementary File S1, network "Intersection_ChloroqUniProt_CovUniProt_.."). This large number of common nodes indicates a signi cant molecular cross-talk between chloroquine and COVID-19. One hundred nine of these nodes were also detected in the human plasma (Table 1). These intersections identify mechanisms of chloroquine interference with COVID-19 action and list potential plasma markers ( Figure 2). The intersection nodes may represent markers of companion diagnostic for chloroquine use. If these nodes are affected in a patient infected with the virus, then the chloroquine prescription may be of help, as chloroquine would markers act on/via these affected nodes.

COVID-19 and cardiac arrhythmia markers
To evaluate whether the intersection nodes would lead to the identi cation of clinically relevant markers, we used an example of cardiac arrhythmia. Markers of arrhythmia were used to generate a network (Supplementary Figure S4). The arrhythmia markers are OPN, ANXA5, GDF15, MPO, LGALS3, TNNT2, TNNI3, ANFB, REN, IL6 and CRP (Supplementary Table S1) (19). The arrhythmia network was explored further for the intersection with common nodes of chloroquine and COVID-19 regulatory mechanisms ( Figure 1B; Supplementary File S1 network "Intersection_Arhythmia_Cov19_..")). There were no edges retrieved between these nodes and amyloid precursor protein was retrieved with 3 different accession numbers. We identi ed 19 nodes linking arrhythmia markers to chloroquine and COVID-19 ( Table 2). Analysis of these 19 nodes showed an engagement of processes affecting the heart and regulation of cell death and proliferation.
Detection of proteins in serum or plasma suggest their suitability as makers for repeatable sampling by blood collection. We used a database of proteins detected in plasma (http://www.plasmaproteomedatabase.org) and retrieved 13 proteins (Table 2). Then, we searched for reports of clinical applications of these 13 proteins as markers of cardiac conditions. Levels of human serum albumin (ALB), amyloid proteins (APP) and soluble endoglin (ENG) correlate with cardiovascular diseases ( Figure 2). Albumin concentration below 10 g/L correlates with cardiovascular diseases (20). Levels of amyloid precursor protein (APP) higher than 150 pg/mL correlate with cardiomyopathy (21). Amyloid-beta (1-40) protein was associated with the incidence of coronary heart failure (22). Two of other identi ed by us proteins, i.e. microtubule-associated protein tau (MART) and prion protein (PRNP) are also associated with the onset of cellular degeneration (23)(24)(25). Endoglin is involved in the development and regulation of vasculature. Elevated levels of soluble endoglin in plasma correlate with enhanced left ventricular lling pressure (26). 14-3-3zeta/delta (YWHAZ) is one of the 10 genes enhanced in ischemic stroke (27).
The systems biology approach allowed us to explore published original experimental data in the search for companion diagnostic markers for chloroquine. Reported here 109 nodes represent a pool of these markers.
The example of the search for markers to guide the use of chloroquine and preventing cardiac arrhythmia identi ed 19 candidates. Four of these were reported to correlate with adverse effects, thus con rming the clinical value of our approach.

Discussion
Systemic network analysis becomes a potent and e cient tool for the investigation of correlations and molecular mechanisms (8,12,13). Well-developed and curated databases contain large volumes of original experimental data. This data are available for analysis with a number of tools. Here, we used Cytoscape that allows retrieval of molecular interactions, functional dependencies, correlation and clinical data (10). Used by us the UniProt database contains more than 500,000 curated entries (14). This rich source of data in combination with the e cient analysis tool, i.e. Cytoscape, leads to unveiling novel dependencies. Two However, changes in expression and/or activity of many of these nodes may also have undesirable consequences, leading to adverse effects of chloroquine.
This manuscript reports the identi cation of potential companion markers of chloroquine. As an example of applicability of our data, we report 19 marker candidates for guiding chloroquine treatment of COVID-19infected patients and monitoring for cardiac arrhythmia ( Table 2). Four of these markers are already known to affect cardiac conditions. The decrease in albumin to concentrations below 10 g/L correlates with cardiac adverse effects (20). Albumin levels have been recommended for clinical monitoring of COVID-19 patients (20,(28)(29)(30)(31). Hypoalbuminemia with the albumin levels lower than 35 g/L was associated with the 2-time higher risk of the long-term mortality in heart failure (31). Chloroquine was described as a drug against prion and Alzheimer's diseases (32). Prion protein and amyloid beta peptide are likely to be components of the innate immune system (33). Amyloid-beta protein association with coronary heart disease and amyloidosis-related heart disease was reported (21,22). Identi cation of amyloid precursor protein, microtubuleassociated tau and prion proteins indicate a link of cell damage and degeneration to cardiac conditions.
These examples show that the identi ed nodes have a high probability to be markers for a companion diagnostic. The 19 markers annotated in Table 2 are the example of using the pool of 266 common nodes of COVID-19 and chloroquine. Our report provides a basis for further clinical studies of the potential markers.
Reported by us results can be used in clinical practice already now, as some of identi ed by us nodes are used in routine clinical diagnostics, e.g. albumin, soluble endoglin and amyloid precursor protein.
Repurposing of their use for COVID-19 patients treated with chloroquine can be applied now. For example, a higher risk of adverse cardiac effects would be indicated by downregulation of albumin and up-regulation of amyloid precursor protein, tau protein, prion protein and soluble endoglin (21,22,26,43).

Conclusion
Presented here network analysis describes potential markers of a companion diagnostic for the use of

Declarations
Ethics approval and consent to participate: Not required for this work.
Consent to publication: Not required for this work.
Availability of data and material: All data are freely available. The .cys le of the networks and analysis is also available without restrictions (Supplementary File S1). The .cys le is available for download at https:// gshare.com/articles/online_resource/SupplementaryFileS1_Cytoscape_DataNetwork_cys/12793580 . This le would allow further clinical use of data reported in this publication. The systemic network analysis le (.cys le) may be useful for clinicians who want to use this publication for clinical applications. The .cys le allows clinicians to do own analysis of reported here data, e.g. search for markers and evaluate clinical conditions of their interest.
Competing interests: The authors declare no con ict of interest in relation to the subject matter and nancial interests of this publication. Contribution from Oranta CancerDiagnostics AB was pro bono.  Tables   Table 1 List of nodes common for COVID-19 and chloroquine that have been observed in the human plasma. These 109 nodes are candidate plasma or serum markers for assessment of chloroquine efficacy in treating COVID-19 infection.
At the end of the table are listed 35 nodes that were not observed in the human plasma. Figure 2 Work ow of selection of potential companion diagnostic markers. Two hundred sixty-six common COVID-19 and chloroquine nodes were evaluated for representation of biological functions and relevance to adverse effects. Retrieved with BiNGO tool biological processes and the nodes of the relevance to the heart arrhythmia markers are annotated.