The use of the Heter-LP algorithm to reposition antibiotics for managing E. coli mastitis in dairy cattle

Mastitis, a disease with high incidence worldwide, is the most prevalent and costly disease in the dairy industry. Environmental mastitis pathogens, such as Escherichia coli (E. coli), are major etiological agents of bovine mastitis in well-managed dairy farms. However, there is still a need to develop more ecient, safe, and economical treatments for mastitis. In the current research, Heter-LP, a new system biology-based method of drug repositioning, was applied to potentially identify novel therapeutic avenues for the treatment of E. coli mastitis. On-line data repositories relevant to known diseases, drugs, and gene targets along with other specialized biological information for E. coli mastitis, including key genes with robust bio-signatures, drugs and related diseases were used as input data for analysis with the Heter-LP algorithm. Our analyses identied novel drugs such as Glibenclamide, Ipratropium, Salbutamol, and Carbidopa as possible therapeutics that could be used against E. coli mastitis. Predicted relationships can be used by pharmaceutical scientists or veterinarians to nd commercially ecacious medicines or combination of two or more active compounds to treat mastitis.


Introduction
Clinical mastitis, an ongoing problem for dairy producers, results in considerable economic losses and has led to increased risk of culling and death in dairy cows [4][5][6]. Mastitis control programs with impact on prevalence of contagious mastitis pathogens, led to a reduction in the incidence of Staphylococcus aureus and Streptococcus agalactiae mastitis, and as a result the increase in the relative impact of environmental mastitis pathogens such as Escherichia coli (E. coli) [6][7][8][9]. E. coli infection can cause either subclinical infection of the mammary gland or a severe systemic disease. Although, intramammary E. coli infections with acute in ammation may be self-healing by spontaneously eradicated by host defenses, however in extreme case it can be fatal [6,[10][11][12]. Unfortunately, self-care is often associated with signi cant economic damage due to the longer duration of infection, lower milk yield, and the potential for pathological changes to the mammary gland [6,13]. The rate of clearance of bacterial pathogens that may be goverend by mammary gland responses within hours of initial infection, is a key determinant of the outcome of intra-mammary gland infection [6,14].
The therapeutic success of bovine mastitis depends mainly on accurate diagnosis of kind of pathogen, it will contribute to improvement of clinical and microbiological e cacy and helps to prevent of emergence and spread of resistant microorganisms. Despite the development prospects for bovine mastitis diagnosis with a focus on speci c pathogens at an early stage fast together with the e cient devices can offer a "cow-side" and "on-site" with high sensitivity and speci city [1,6,[15][16][17]; the most e cient, safe, and economical treatments for mastitis are still topics of scienti c debate [6,18,19]. Generally, narrow and/or broad spectrum antibiotics are generally used for the treatment of E. coli mastitis, but treatment studies have shown controversial results. Given the problems associated with antibiotic therapy, including emergence of antibiotic-resistant strains, and the concern about antibiotics entering the food chain, efforts are being made to substitute the alternative strategies for new antimicrobial agents including bacteriophage, vaccination, nanoparticles, cytokines, homeopathy, natural compounds from plants, animals, and bacteria or the discovery of new drugs that are effective against mastitis pathogens [6,[20][21][22].
Novel computational systems biology tools, such as meta-analysis, the pathways analysis, data mining and machine learning have provided good opportunities to understand the molecular mechanisms associated with diseases [23][24][25][26][27]. Currently, the rst step to drug development is the use of previously known drugs; this is known as drug repositioning. This approach has attracted a lot of interest in recent years because of the increased speed of this process, drug safety concerns, and its lower cost. The integration of drug, disease and gene target information, in addition to how they affect and function in the body, can have a signi cant impact on drug repositioning and the possible identi cation of disease treatments.
The primary goal of this research was to identify some drugs that might be used for the treatment of E. coli mastitis (most prevalent agent in clinical cases). We used a recently introduced computational approach to facilitate drug repositioning, called Heter-LP [3]. Heter-LP is a systems biology approaches that integrate different types of data at different levels. Its ability to detect drug-disease relation has been proven by various analyses [3]. Heter-LP was selected because of some more of its advantages such as accuracy, no need to negative samples, its ability to predict trivial and non-trivial relationships between drugs, diseases and protein targets, and also its ability to use heterogenous data. Data resources in the public repository relevant to diseases, drugs, and gene targets along with other specialize biological information for E. coli mastitis were used as input data network to Heter-LP algorithm. Key genes with robust bio-signatures achieved from our recent meta-analysis work, related disease constructed by Pathway Studio web tool and drugs extracted from the literature review were used as complementary biological information to add to dataset gathered from the public repository.

Material And Methods
Our approach consisted of two steps. The rst step was to construct a heterogeneous network by using input data including different kinds of nodes and edges. There were three types of nodes in the proposed heterogeneous network: targets, drugs and diseases. There were six different kinds of edges with represented one type of similarity or association: target similarity, drug similarity, disease similarity, known drug-target interaction, known drug-disease association, and known disease-target interaction.
Then, at the second step, we tried to predict important links which were not identi ed in the input data by using of Heter-LP. As we used the Heter-LP for the second step, we had to arrange our network (step 1) according to its input structures, which consisted of six separate matrices: (1) drug similarities, (2) disease similarities, (3) targets similarities, (4) drug-target relations, (5) disease-target relations and (6) drug-disease relations [3].
We tried to nd a specialized dataset for mastitis to integrate with the aforementioned data.
Unfortunately, no dataset is publically available. It seems that all available data and information related to animal diseases, genes targets, and drugs are only embedding in the publication and there are no comprehensive datasets or repositories for them. Lack of access to this data however did not negatively impact the current analysis because of the generality of the input datasets due to the similarity of this disease in humans and other animals. To specialize the results for E. coli mastitis we added three parts of information to our generated datasets (cases 2 to 4 in the following).
Based on the above descriptions, data sets that were used to construct the input data network for Heter-LP algorithm included: 1. Previously known drugs information gathered by public repositories. 2. Key genes that had a robust bio-signature in response to mastitis specially in E. coli infection. These genes are the results of our recent research ndings based on meta-analyses to detect up/down regulated genes by utilizing machine learning algorithms [1] and network analysis [2,28], 3. Functionally related diseases or biological processes related to bovine mastitis illustrated by Pathway Studio web tool. 4. Relevant drugs and antibiotics to E. coli mastitis gathered by literature analysis.
Finally, by the integration of these data a heterogeneous network is constructed and used as input for the Heter-LP algorithm.
A brief description and the data preparation methods of each part are presented in next subsections ( Fig. 1). Different parts of this work ow were described in its caption and in more details in sub-Sect. 2.1 to 2.5.

Previously known drugs information
As mentioned before six different kinds of data including (1) Drug similarity, (2) Disease similarity, (3) Target similarity, (4) Known drug-disease associations, (5) Known drug-target interactions, and (6) Known disease-target associations, were necessary as input data to develop the network for Heter-LP. The base of these data was gathered previously [3], from some important databases. The data resources are summarized in the Table 1. In order to have an updated version of these data, the last version of data resources to generate the six matrices are provided according to the methods described at GitHub (https://github.com/MLot SH/Heter-LP) and DKR site (http://dkr.iut.ac.ir/projects) and in detail in our previously published research [3]. Table 1 Resources of data related to each sub-network and the number of nodes in each one.

Subnetwork
Using Targets:2563 Table 2 The key genes or regulators with robust bio-signature in response to E. coli mastitis reported to our previous meta-analysis based microarray studies [1,2] Row Gene symbol Functional group Gene name (alias)

Disease similarity data
One of the problems that arises when examining diseases is the use of different names or identi ers for the same disease. In the case of the disease in the current research, bovine mastitis was used for dairy cattle, and mastitis was used for human and other mammals in the literature. The Pathway Studio web tool 12.0.1.5 was used to construct a network of disease or cell processes that were functionally associated with mastitis or bovine mastitis. Pathway Studio as a pathway analysis tool incorporates some commercial and public databases such as BIND [29], KEGG [30], and GO [30] that utilizes the ResNet Mammal database. Moreover, it also uses the powerful text-mining tool MedScan to seek the latest information from PubMed and other public sources (Elsevier-Ariadne Genomics, Rockville, MD) [31].
For more con dence, all relationships which were reported by more than two references were selected. All relations between mastitis or bovine mastitis with other diseases or cell process are indicated in Table 3.
Additional details and references are provided in Additional le 1. As shown, most of the cases related to mastitis or bovine mastitis are the same and demonstrated the similarity of this disease in all mammals. This information has been added to disease similarity part of dataset shown in Table 1.

Drugs disease
With a review of the literature, we were able to develop a comprehensive list of drugs or antibiotics that have been used to treat E. coli mastitis (see Table 4). This information has been added to drug-disease relation part of dataset shown in Table 1.

Integration of data
The nal heterogeneous network model was constructed by integration of these four mentioned data which were discussed in previous sections. It is necessary to mention that Heter-LP input data could be incoincident in different parts of the heterogeneous network. This means that a complete list of drugs, diseases and proteins/genes (as targets) will be achieved by union of similar typed items in each subnetwork.

Results
The repositioning of antibiotics for managing E. coli mastitis in dairy cattle is the main ndings of this study. Based on Heter-LP categorization, there are two kinds of predictions, known and novel [3]. The 30 top predicted drugs and antibiotics associated with E. coli mastitis are presented in Table 5. Most of the drugs listed in Table 4 have been reported in literature as treatments for E. coli mastitis. These results demonstrate that Heter-LP could identify known relations correctly, which indicates that the novel compounds may be realistic predictions. All predicted results of Heter-LP are presented in Additional le 2.

Discussion
The e cacy of antibiotic and/or anti-in ammatory drugs/compounds in the treatment of mastitis disease is not fully speci ed. Given the problems associated with antibiotic therapy, including emergence of antibiotic-resistant strains, and the concern about antibiotics entering the food chain, efforts are being made to substitute the alternative strategies for new antimicrobial agents including bacteriophage, vaccination, nanoparticles, cytokines, homeopathy and natural compounds from plants and animals, and bacteria or the discovery of new drugs that are effective against mastitis pathogens [6,[20][21][22].
While the pharmaceutical industry has explored the use of drug repositioning to identify novel treatments for diseases, this work has been hampered by a lack of a fundamental and systematic approach. In the current research, the biological algorithm Heter-LP was used to reposition antibiotics for managing E. coli mastitis in dairy cattle. The utility of Heter-LP, to discover new drug repositioning to rare diseases in human have been explored previously [43]. Data that was available in the public repositories along with other specialize biological information for E. coli mastitis including crucial genes, antibiotic or drugs used for treatment of E. coli mastitis, and its association with other disease or cell processes were used as input data for the Heter-LP algorithm. By using Heter-LP, we were able to introduce a list of most likely candidate drugs that could be used as therapeutic strategies against the E. coli infection. It is noteworthy that these drugs have been suggested among more than 11000 different drugs, which could help to accelerate and facilitate the drug identi cation process. Certainly, this list of suggested drugs is valubale for pharmaceutical scientists or veterinarians in order to nd a commercial and e cacious medicine or combinations of two or more active compounds. In the following, we have tried to validate and con rm most of these new predictions by review of available scienti c literature.
Penicillin G (also known as Benzylpenicillin), Rifampicin, Cefprozil and Cefadroxil are antibiotic drugs.
Recent research has shown that Rifampicin could be used as a solo medical therapy in humans for chronic mastitis [44]. Cefprozil, a second-generation cephalosporins antibiotic, is approved worldwide strictly for the treatment of mastitis disease in dairy cattle. Cefadroxil, a broad-spectrum cephalosporins, is a rst-generation cephalosporin antibacterial drug that is effective against gram-positive and gramnegative bacterial infections. It is the para-hydroxy derivative of cefalexin that is a bactericidal antibiotic and is used in the treatment of mild to moderate susceptible infections. Lipopoly sacharrides on the the outer membrane of the gram-negative bacteria such as E. coli are an important barrier that provides protection against toxic compounds, which include antibiotics and host innate immune molecules such as cationic antimicrobial peptides. These bacteria use a wide variety of mechanisms to resist antimicrobials [45,46].
Glibenclamide is an antidiabetic drug in a class of medications known as sulfonylureas, closely related to sulfonamide antibiotics. Sulfonamides are also occasionally used to treat septicemia caused by coliform mastitis in dairy cattle [47]. It has been investigated that, effects of in ammation markers (TNFα and NFκB), and activation of cell injury or cell death markers (IgG endocytosis and caspase-3), signi cantly reducing with glibenclamide [48].
In the case of Ipratropium, it has been shown that partially protect the lungs against in ammation by reducing neutrophilic in ltration. This protective effect is associated with a reduction in the MMP-9 activity, which is known to play an important pro-in ammatory role in the acute in ammatory process [49].
It has been demonstrated that hypothyroidism is associated with signs of low-grade in ammation (raised C-reactive protein levels) which may be elicited by the raised level of triglyceride or be an independent effect of an intracellular hypometabolic state or of a combination of them [50]. Also, other research has shown that l-thyroxine treatment of patients with subclinical hypothyroidism can reduce in ammation Salbutamol, the other predicted drug listed in Table 5 Based on these results, it can be concluded that the Heter-LP has successfully predicted drugs/compounds that can be used as suitable alternatives for the treatment of E. coli mastitis.

Conclusions
In the current study, has been shown that the system biology-based algorithm, Heter-LP, can be used to identify repositioned drugs that may be useful for the treatment of important disease. Integration of the biological data and using the Heter-LP algorithm enabled us to introduce novel drugs relevant to E. coli mastitis. Our results provide valuable information for pharmaceutical scientists or veterinarians in the dairy industry to nd a commercial and e cacious medicine or a combination of two or more active compounds.
50. Kvetny J, Heldgaard PE, Bladbjerg EM, Gram J. Subclinical hypothyroidism is associated with a lowgrade in ammation, increased triglyceride levels and predicts cardiovascular disease in males below