Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Network analysis suggests a potentially ‘evil’ alliance of opportunistic pathogens inhibited by a cooperative network in human milk bacterial communities

Network analysis suggests a potentially ‘evil’ alliance of opportunistic pathogens inhibited by a... OPEN Network analysis suggests a potentially ‘evil’ alliance of opportunistic pathogens SUBJECT AREAS: MICROBIOLOGY inhibited by a cooperative network in MEDICAL RESEARCH METAGENOMICS human milk bacterial communities COMPUTATIONAL BIOLOGY AND BIOINFORMATICS 1 1 1,2 3 4 4 Zhanshan (Sam) Ma , Qiong Guan , Chengxi Ye , Chengchen Zhang , James A. Foster & Larry J. Forney Received Computational Biology and Medical Ecology Lab, State Key Laboratory of Genetic Resources & Evolution, Kunming Institute of 14 September 2014 Zoology, Chinese Academy of Sciences, Kunming, 650223, China, Department of Computer Science University of Maryland Accepted College Park, MD 20740, USA, School of Public Health Columbia University 722 West 168th Street New York, NY 10032, USA, 2 January 2015 Institute of Bioinformatics and Evolutionary Studies & Department of Biological Sciences University of Idaho Moscow, ID 83843, USA. Published 5 February 2015 The critical importance of human milk to infants and even human civilization has been well established. Yet our understanding of the milk microbiome has been limited to cataloguing OTUs and computation of community diversity. To the best of our knowledge, there has been no report on the bacterial interactions Correspondence and within the milk microbiome. To bridge this gap, we reconstructed a milk bacterial community network requests for materials based on Hunt et al. Our analysis revealed that the milk microbiome network consists of two disconnected should be addressed to sub-networks. One sub-network is a fully connected complete graph consisting of seven genera as nodes and Z.S.M. (samma@ all of its pair-wise interactions among the bacteria are facilitative or cooperative. In contrast, the interactions in the other sub-network of eight nodes are mixed but dominantly cooperative. Somewhat surprisingly, the uidaho.edu) only ‘non-cooperative’ nodes in the second sub-network are mutually cooperative Staphylococcus and Corynebacterium that include some opportunistic pathogens. This potentially ‘evil’ alliance between Staphylococcus and Corynebacterium could be inhibited by the remaining nodes that cooperate with one another in the second sub-network. We postulate that the ‘confrontation’ between the ‘evil’ alliance and ‘benign’ alliance and the shifting balance between them may be responsible for dysbiosis of the milk microbiome that permits mastitis. uman milk is generally considered the best source of nutrients for infants, and its health benefits such as prebiotics, immune proteins, and the microbiome of human milk itself, have been increasingly recog- 1–5 H nized . Similar to other habitats in or on the human body such as the gut and skin, human milk is not sterile at all and it hosts extensive bacterial communities that are posited to possess important health implications. In general, traditional literature on human milk has been focused on pathogenic bacteria, and our understanding on commensal bacteria is still very limited in spite of the rapid advances in metagenomic technology and expanding studies of the human microbiome in recent years. For example, Heikkila & Saris (2003) investigated potential inhibition of Staphylococcus aureus by the commensal bacteria of breast milk . Staphylococcus aureus is known as a food-poisoning agent and a common cause of infections including serious antibiotic-resistant hospital 6,7 8,9 infections . In addition it has been implicated in SIDS (Sudden Infant Death Syndrome) as well as infectious 6,10,11 mastitis that affects 20–30% lactating women . There have been several studies that applied metagenomic sequencing technology to characterize human milk 12–17 bacterial communities and a recent one by Hunt et al (2011) provides the largest data set of 16S rRNA sequences from human milk samples . Hunt et al (2011) collected 47 samples from 16 breastfeeding women (3 samples from all but one individual) who self-reported as healthy and between 20–40 yr of age . Their study revealed that the most abundant genera in the milk samples were Staphylococcus, Streptococcus, Serratia, and Corynebacteria, while eight other genera had relative abundances exceeding 1%. Besides characterizing the composition of milk bacterial community, Hunt et al. (2011) for the first time described the within-individual variation and the among-individuals variation of milk bacterial communities. The within-individual variation, which can be thought as a measure of the stability of individual milk bacterial communities, differed between SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 1 www.nature.com/scientificreports foundation, was not triggered until the publication of two independ- individuals. In other words, temporal variation of community mem- bership or stability of individual milk communities varied signifi- ent seminal papers published by Watts & Strogatz (1998) on the dynamics of ‘‘small-world’’ networks in the journal Nature, and cantly between women. For example, from the samples of ‘‘Subject #5,’’ Staphylococcus occupied either the first or second position in by Barabasi and Albert (1999) on the emergence of scaling (i.e., scale free networks) in random networks in the journal Science, respect- terms of the relative abundance (22–59%); but the samples of ‘‘Subject #1’’, Staphylococcus only contributed less than 5% to the ively. A commonality of both the papers is the extension of basic random graph models so that they can better fit the patterns exhib- community abundance. The among-individual variations in the rela- tive abundances of bacteria were as large as six-fold . ited by many empirically observed networks in social, technological and natural networks. One of the most active application fields of Although milk microbiome was apparently missing in the initial network science is biology, thanks to the vast datasets available from US-NIH HMP roadmap, its critical importance to human health and genomic and metagenomic research. The case for applying network diseases is evident. The importance is even more obvious from the analysis to investigate the human microbiome was argued convin- perspective of its relationships with the microbiome in other body cingly by Foster et al. (2008) and since then several network ana- sites because human microbiome is location specific, but not isolated 5,18 lyses have been successfully performed with human microbiome from one another at all . Zaura et al. (2014) hypothesized that 31–33 data . One huge advantage of network analysis is its power to development of fetal tolerance toward the microbiome of the mother visualize multivariate relationships generated from big data sets such during pregnancy is a major factor in the successful acquisition of a as genomic and metagenomic sequence data. Furthermore, various normal microbiome . Jeurink et al. (2013) proposed a mechanism parameters computed with network analysis software packages (e.g., for the formation of breast milk micriobiome, which involves 34 35 Cytoscape , Gephi ) offer informative insights on the patterns in immune cell education by the pregnancy hormone progesterone biological data. Complex network alignment algorithms and soft- leading to the transportation of bacteria from the mother to her ware (e.g., Graphcrunch2 ) can further be utilized to compare bio- mammal glands . Guts of breastfed infants showed significantly logical networks under different treatments. higher counts of bifidobacteria and Lactobacillus and lower counts of Bacteroides, Clostridium coccoides group, Staphylococcus, and Methods Enterobacteriaceae, as compared with formula-fed infants . The The 16S rRNA sequence data sets of human milk were collected by Hunt et al (2011) . pioneering colonizers such as Bifidobacterium longum, which carries Specifically, the V1-V2 region of the bacterial 16S rRNA gene was amplified from several gene clusters dedicated to the metabolism of human milk genomic DNA using universal primers and approximately 300,000 reads were gen- oligosaccharides (HMOs) allows infants to digest breast milk and erated from the barcoded pyrosequencing of amplicons from 47 samples. After ´ quality control, the data set was reduced to approximately 160,000 reads, with a mean possibly some simple vegetal food such as rice. Gonzalez et al. of 3400 sequences per sample. The sequence data were assigned to the most likely (2013) found that women with HIV RNA in breast milk have a bacterial genera using the RDP Bayesian classifier. A table of the 15 most abundant different pattern of microbial composition, compared with milk genera in each sample was supplied in the Supporting Information (Table S1) of Hunt without HIV RNA, indicating specific immunological phenomena et al (2011) and was used for our network analysis. in HIV-infected women . They also argued that breast milk and infant gut microbiota are essential for the maturation and protection Results of infant’s immune system. A metagenomic study conducted by The pair-wise relationships among 15 genera were measured by Ward et al (2013) confirmed the benefits of breast milk ingestion Spearman rank correlation coefficients with p-value of 0.05, and to the microbial colonization of the infant gut and immunity . The the computed values of Spearman correlation coefficients with R- latter is demonstrated by the existence of immune-modulatory statistics package (www.r-project.org) were feed into Cytoscape net- motifs in the metagenome of breast milk . These recent studies 34 35 36 work analysis software and Gephi . The Graphcrunch2 software exhibited the significant importance of breast milk microbiome in was applied to further compare the reconstructed milk microbiome health and diseases. network with several standard models of complex networks. The Obviously, Hunt et al (2011) and other previous culture-inde- results of network analysis are exhibited in Figure 1 and Table 1. 12–17,22–23 pendent studies have deepened our understanding of bac- Before discussing our results further, it should be noted that like terial communities in human breast milk. In a recent study, Guan many other studies of biological networks, our network was built and Ma (2014) applied Taylor’s power law and neutral theory to based on the correlation between OTUs. Our usage of terminology investigate the abundance distribution pattern and the maintenance such as ‘cooperative’, ‘non-cooperative’, and ‘evil alliance’ are by mechanism of milk microbial community diversity, respectively by analogy. Correlation is not equivalent with causation. For example, reanalyzing the existing data on milk microbiome . The analysis the correlation between two OTUs may be due to indirect ‘facili- with Taylor’s power law model indicated that bacterial population tation’ or ‘inhibition’ by a third player. Nevertheless, at this stage, abundance in human milk microbiome is aggregated, rather than correlation data are the only available data type for the human milk random, and it was found that neutral theory did not fit to any of microbiome. The correlation network therefore offers the best appar- the 47 samples (communities), suggesting non-random interactions atus available today to tackle the tangled microbiota in human milk. in community assembly and diversity maintenance . Nevertheless, It should also be noted that the OTU data we used resolves taxa at due to the limitation of the analytical approaches used in previous the level of genera and a network based on species data may reveal studies, we still have little knowledge on the bacterial interactions, different patterns from the networks we obtained. Hence, the total beyond their non-random nature, within milk microbiome. Indeed, validity of the conclusions we draw below, ultimately, should be most statistical approaches are not powerful enough to reveal the subject to testing in future biomedical research. Nevertheless, given 25,26 interspecies interactions within a microbial community . In this the fundamental importance of studying milk microbiome, we article, we take advantage the power of network analysis in studying believe that preliminary analyses such as this are warranted. interspecies or inter-OTU (Operational Taxonomic Unit) interac- From the above results in Figure 1 and Table 1, we draw the tions in a complex network setting such as a microbial community. following conclusions: Erdos and Renyi (1960) seminal research on random graph the- (i) The network of the breast milk bacterial community consists of ory opened one of the most exciting new fields in combinatorial two sub-networks, that correspond to two disconnected components th mathematics in the 20 century, and their random graph theory or communities (Figure 1a, Figure 1b, and Table 1). One component attracted extensive studies during the subsequent decades. Still, the (Figure 1b) consists of seven nodes in which all of the nodes are fully avalanche of approaches to network analysis and the emergence of connected (forming a complete graph) and they interact coopera- network science, of which random graph theory forms a theoretic tively (positive correlations). Another component consists of 8 nodes SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 2 www.nature.com/scientificreports a b cd Figure 1 | Bacterial network of the human milk microbiome reconstructed using the data sets of Hunt et al. (2011): Figure (1a) and (1b) show the two disconnected sub-networks (components) of the breast milk bacterial network; Figure (1c) and (1d) are the same components, corresponding to (1a) and (1b), respectively, assuming the Staphylococcus and Corynebacterium nodes were eliminated. The green line represents a positive correlation (cooperative interaction) while the red line represents a negative correlation (non-cooperative interaction). Obviously, when the two mutually cooperative players Staphylococcus and Corynebacterium are removed, the whole network becomes totally cooperative (Figure 1c & 1d). and all of the interactions (edges) are cooperative except those As mentioned previously, Staphylococcus aureus is a food-poisoning that involve Staphylococcus and Corynebacterium. Therefore, if both agent and a common cause of infections including serious antibiotic- 6,7 Staphylococcus and Corynebacterium were removed, then the remain- resistant hospital infections , and the bacterium is also implicated in 8,9 6,10,11 ing nodes in the sub-network are fully cooperative. Furthermore, the SIDS (Sudden Infant Death Syndrome) and infectious mastitis . relationship between Staphylococcus and Corynebacterium are coop- Mastitis is inflammation of the breast with or without infection, and erative, although they do not cooperate with other taxa in the network. Staphylococcus aureus has traditionally been believed to be the patho- Obviously, the milk bacterial network is dominantly cooperative, and gen that is typically associated with infectious mastitis . There are the ratio of cooperative vs. non-cooperative interactions is 451. studies that reveal some other species of Staphylococcus such as Table 1 | Topological properties of human milk bacterial network Number of nodes Number of edges Avg. number of neighbors Clustering coefficient Connected components Network diameter 15 45 6 0.944 2 2 Average path length Network density Modularity Number of communities Small-world network Scale-free network 1.082 0.429 0.498 2 Yes No SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 3 www.nature.com/scientificreports Staphylococcus epidermidis may play a prevalent role in mastitis work (complete graph). Obviously, the lower density is because there 2,5,37,38 infections . Although the etiology of mastitis may vary there is no edge (connection) between the two disconnected components, is evidence from animal studies and clinical trials that suggests which significantly lowers the network density. certain strains of Lactobacillus can produce anti-inflammatory and anti-bacterial factors that inhibit adhesion and internalization Discussion of Staphylococcus spp. For example, Arroyo et al. (2010) reported In the study done by Hunt et al (2011) universal primers were used that treatment with probiotic strains from human milk (containing for amplification of the 16S rRNA gene, which increased sequencing Lactobacillus strains) produced a greater reduction in the bacterial coverage and helped them to obtain the most comprehensive experi- counts of Staphylococcus epidermidis, Staphylococcus aureus and mental survey of milk microbiome to date. Their study confirmed that Staphylococcus mitis, as well as a greater reduction in pain than Staphylococcus and Corynebacterium, identified here appear to exist the treatment with antibiotics . as an ‘‘evil alliance’’ in the milk microbiome based on our network The genus Corynebacterium includes Gram-positive, rod-shaped 28 analysis even though they are typically present on adult skin too . bacteria that are largely innocuous and widely distributed in nature. Hunt et al (2011) had recognized the possibility of contamination However, some species such as C. diphtheriae may cause human from the skin microbiome and took extraordinary caution during disease. Extensive studies on the health implications of breast sample collection. Furthermore, they compared the compositions of milk include protecting infants from diarrheal and respiratory dis- both milk and skin communities and concluded that bacterial com- 1,40,41 eases . Our finding from network analysis that every other munities in milk cannot simply be a result from skin contamination . ‘player’ in the milk microbiome collectively ‘opposes’ or ‘inhibits’ Some species of Staphylococcus and Corynebacteria are opportunistic Staphylococcus and Corynebacterium possibly explains the health infectious agents, and Staphylococcus aureus is associated with lacta- effects of milk despite the presence of potential pathogens. In other tional mastitis. Other studies have also demonstrated the occurrence 1,22,42,43 words, from the perspective of a lactating mother and baby, it may be of Staphylococcus aureus in human milk . It has been reported the cooperative and collective efforts of the other community mem- that, during the course of lactation, up to 30% of women suffer from bers that suppress the ‘evil’ alliance from opportunistically causing breast infections or inflammation (mastitis) that often lead to fever, unhealthy consequences such as mastitis. 44,45 redness, swelling and breast pains . In Hunt et al (2011) study, the (ii) The milk bacterial network is a ‘small-world’ network with a milk donors were self-reported as healthy, but at least one of the network diameter of only 2 and average minimal path length of subjects showed symptoms of mastitis . Therefore, at least, existing p51.082 (Table 1). This is in accord with the criterion for judging literature on milk microbiome referenced above indicates the existence small world networks , in which p51.082,log(N)5log(15)52.708. of potentially opportunistic pathogens in milk of both healthy and The criterion means that the typical distance p between two ran- diseased (mastitis) women. But why do those potential pathogens domly chosen nodes grows proportionally to the logarithm of N, often seem harmless to lactating mothers and infants? In contrary, which is even smaller than the linear growth. In biological terms the existing literature documented the benefits of milk microbiome the interactions amongst the nodes (genera) are very tight. such as the protective effects of breastfeeding against diarrheal and Nevertheless, the milk microbiome network is not scale-free,which respiratory disease as well as reduced risk of developing obesity in is evidenced by the fact that the degree distribution of the network infants . There is not an existing theory with sufficient evidence to does not fit to a power law distribution (p-value50.323.0.05 from explain the natural phenomenon in the existing literature of human 2l fitting power-law model, i.e., p(k) / k ). A scale-free network has milk microbiome. two properties: growth and preferential attachment . Growth means A recent study by Urbaniak et al. (2014) investigated the existence that the number of nodes in the network increases over time and of microbiome within mammary tissue by using 16S rRNA sequen- preferential attachment means that the more connected a node is, cing and culture . They analyzed the breast tissue from 81 women the more likely it is to receive new links. This absence of scale-free with and without cancer in Canada and Ireland, and confirmed the property may be due to the fact that our network is built with bacterial existence of both health-conferring bacteria such as Lactobacillus genera as the node units, which might be relatively constant over time and Bifidobacterium, as well as taxa known for pathogenesis such and therefore the growth of nodes could be insignificant. We also as Enterobacteriaceae, Pseudomonas, and Streptococcus agalactiae. utilized the network alignment software GraphCrunch2 to compare Yet, none of the 81 women recruited had any clinical signs or symp- our milk microbiome network with 50 random instances of each of toms of breast infection. This echoes the phenomenon of the pres- the following network models with the same size as microbiome net- ence of opportunistic pathogens in breast milk revealed by our work, respectively: ER (Erdos-Renyi random graphs), ER_DD (Erdos- network analysis in this article. Re´ nyi random graphs with the same degree distribution as the data), Our network analysis revealed the possible existence of an ‘‘evil GEO (Geometric Random Graphs), GEO-GD (Geometric Gene Dup- alliance’’ between Staphylococcus and Corynebacterium, and this alli- lication Models), SF (Scale-free Baraba´ si-Albert Preferential Attach- ance is collectively ‘opposed’ or inhibited by the other members of ment Models), SF-GD (Scale-free Gene Duplication Models), and breast milk bacterial community (network). Our finding from net- STICKY (Stickiness-index Based Models). The results from graph- work analysis therefore offers a piece of concrete evidence to support crunch2 software also demonstrated that a scale-free network model the following hypothesis: Similar to natural ecosystems, the ecosystem was among the worst performing models. of human milk microbiome, which consists of the milk microbial (iii) The other topological parameters of milk bacterial network community and its environment (i.e., the human body or host), con- shown in Table (1) also offer some interesting information about the tains microbial species of various characteristics or functions, being network characteristics. Both the numbers of connected components beneficial, harmful, or neutral from a human health perspective. This and the number of communities in the network are two, correspond- dynamic balance in the milk microbiome ecosystem depends on the ing to the two sub-networks (Figure 1a & 1b). In general, the values of species interactions within the microbiome bacterial community as these two parameters are not necessarily equal, but their equality in well as the host environment such as immune system. The states of the the case of milk microbiome further strengthens the evidence that the milk microbiome (network), which could correspond to healthy or milk microbiome network is divided into two separate sub-networks. disease states of human body, depend on the interactions within the The high clustering coefficient of 0.944 signals the high aggregation microbiome network (as exhibited by Figure 1) as well as the host, tendency of network nodes. Moreover, a network density of 0.429 which may have her unique genomic, immunological, physiological suggests that the number of edges in the network is about 43%, i.e., and demographic properties. Specifically, we postulate that in healthy less than half of all possible edges with a completely connected net- state, the adverse consequences of the ‘‘evil alliance’’ of Staphylococcus SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 4 www.nature.com/scientificreports 10. Jonsson, S. & Pulkkinen, M. O. Mastitis today: incidence, prevention and and/or Corynebacteria is ‘contained’ or inhibited by the collective treatment. Ann Chir Gynaecol Suppl 208, 84–87 (1994). cooperative defense of other community members. In the face of a 11. Semba, R. D. et al. Mastitis and immunological factors in breast milk of lactating disturbance of ‘evil alliance’ cannot be contained, and the balance may women in Malawi. Clin. Diagn. Lab. Immunol. 6, 671–674 (1999). be disrupted or shifted to an alternative state, which may correspond 12. Perez, P. F. et al. Bacterial imprinting of the neonatal immune system: lessons from maternal cells? Pediatrics 119, e724–732 (2007). to disease such as mastitis, whose underlying causes or triggers are not 13. Martı n, R. et al. Cultivation-independent assessment of the bacterial diversity of known yet. In other words, the ‘‘evil alliance’’ of Staphylococcus and breast milk among healthy women. Res. Microbiol. 158, 31–37 (2007). Corynebacterium are opportunistic and they cause disease to their host 14. Martı´ n, R. et al. Isolation of bifidobacteria from breast milk and assessment of the only if the collective containment by other members of the micro- bifidobacterial population by PCR-denaturing gradient gel electrophoresis and biome coupled with the defense of host immune system are weakened quantitative real-time PCR. Appl. Environ. Microbiol. 75, 965–969 (2009). 15. Gueimonde, M., Laitinen, K., Salminen, S. & Isolauri, E. Breast milk: a source of to such extent that their disturbance cannot be ‘absorbed’ or tolerated. bifidobacteria for infant gut development and maturation? Neonatology 92, 64–66 In a recent study, Ward et al. (2013) postulated that it might be the (2007). diversity or genetic traits of the milk metagenome that confer benefits 16. Hamady, M., Walker, J. J., Harris, J. K., Gold, N. J. & Knight, R. Error-correcting to infants, rather than the action of any one bacterial genus or species . barcoded primers for pyrosequencing hundreds of samples in multiplex. Nat. Methods 5, 235–237 (2008). We argue that our hypothesis of dynamic balance of milk microbiome 17. Collado, M. C., Delgado, S., Maldonado, A. & Rodrı´ guez, J. M. Assessment of the ecosystem is consistent with the ideas postulated by Ward et al. 4 bacterial diversity of breast milk of healthy women by quantitative real-time PCR. (2013) for the following two reasons: (i) the diversity and/or se- Lett. Appl. Microbiol. 48, 523–528 (2009). quencesofDNA within themetagenomereferredby Wardetal. 18. Costello, E. K., Stagaman, K., Dethlefsen, L., Bohannan, B. J. M. & Relman, D. A. (2103) are carried by the microbiome, and they are the ‘surrogates’ The application of ecological theory toward an understanding of the human microbiome. Science 336, 1255–1262 (2012). of the species or OTUs within microbiome; (ii) ecological theory tells 19. Zaura, E., Nicu, E. A., Krom, B. P. & Keijser, B. J. F. Acquiring and maintaining a us, that diversity and stability, are tangled together, even if stability is normal oral microbiome: current perspective. Front Cell Infect Microbiol 4,85 not determined by diversity. Indeed, the diversity-stability paradigm (2014). debates have been going on for more than a half-century now, but 20. Jeurink, P. V. et al. Human milk: a source of more life than we imagine. Benef nobody denies their close relationship. In fact, a recent consensus has Microbes 4, 17–30 (2013). 21. Matamoros, S., Gras-Leguen, C., Le Vacon, F., Potel, G. & de La Cochetiere, M.-F. been that network analysis should play a critical role in investigating Development of intestinal microbiota in infants and its impact on health. Trends the paradigm. We hope this relatively simple network analysis re- Microbiol. 21, 167–173 (2013). ported here will induce more extensive applications of network ana- 22. Gonzalez, R. et al. Breast milk and gut microbiota in African mothers and infants lysis approaches to the study of human milk microbiome, especially from an area of high HIV prevalence. PLoS ONE 8, e80299 (2013). 23. Xuan, C. et al. Microbial dysbiosis is associated with human breast cancer. PLoS the stability (balance) of the milk ecosystem, which should greatly ONE 9, e83744 (2014). enhance our understanding of the health and disease implications of 24. Guan, Q. & Ma, Z. S. Ecological analysis of the human milk microbiome. Chinese milk bacterial microbiota. Science Bulletin 59, 2205–2212 (2014). There remain some important follow-up questions to this network 25. Ma, Z. S., Abdo, Z. & Forney, L. J. Caring about trees in the forest: incorporating analysis, and the answers will need efforts from the biomedical frailty in risk analysis for personalized medicine. Personalized Medicine 8, 681–688 (2011). research community. Indeed, the insights from our network analysis 26. Ma, Z. S., Geng, J. W., Abdo, Z. & Forney, L. J. [A Bird’s Eye View of Microbial should only be a starting point for thoroughly understanding the Community Dynamics.] Microbial Ecological Theory: Current Perspectives health and disease implications of the human milk microbiome. [Ogilvie, L. A. and Hirsch, P. R. (ed.)][pp57–71] ( Caister Academic Press London, Experimental and clinical studies of human milk microbiome are 2012). crucial, but also understandably difficult to conduct. We realize that 27. Erdo˝s, P. & Re nyi, A. On the Evolution of Random Graphs. Publication of the Mathematical Institute of the Hungarian Academy of Sciences 5, 17–61 (1960). even the basic research of the species identification of human milk 28. Watts, D. J. & Strogatz, S. H. Collective dynamics of ‘small-world’ networks. microbes is still at the very preliminary stage. Both our analysis and Nature 393, 440–442 (1998). the study done by Hunt et al (2011) , from which we obtained data 29. Barabasi, A. L. & Albert, R. Emergence of scaling in random networks. Science 286, for reconstructing the milk microbiome network, only reported the 509–512 (1999). 30. Foster, J. A., Krone, S. M. & Forney, L. J. Application of ecological network theory genera of bacteria present, but not the species within these genera. to the human microbiome. Interdiscip Perspect Infect Dis 2008, 839501 (2008). This data limitation strongly constrained our capability to infer fur- 31. Bapteste, E., Bicep, C. & Lopez, P. Evolution of genetic diversity using networks: ther information on the bacterial interaction within human milk the human gut microbiome as a case study. Clin. Microbiol. Infect. 18 Suppl 4, microbiome. Given the self-evident, critical importance of human 40–43 (2012). milk microbiome, we hope this study will be a stepping stone for 32. Greenblum, S., Turnbaugh, P. J. & Borenstein, E. Metagenomic systems biology of the human gut microbiome reveals topological shifts associated with obesity and more advanced network analysis based on more comprehensive bio- inflammatory bowel disease. Proc. Natl. Acad. Sci. U.S.A. 109, 594–599 (2012). medical data sets generated by future studies of the human milk 33. Faust, K. et al. Microbial co-occurrence relationships in the human microbiome. microbiome. PLoS Comput. Biol. 8, e1002606 (2012). 34. Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003). 1. Hunt, K. M. et al. Characterization of the diversity and temporal stability of 35. Bastian, M., Heymann, S. & Jacomy, M. Gephi: An open source software for bacterial communities in human milk. PLoS ONE 6, e21313 (2011). exploring and manipulating networks. Proceedings of the Third International st 2. Ferna´ ndez, L. et al. The human milk microbiota: origin and potential roles in Conference on Weblogs and Social Media (2009) 2pp, Paris, France, March 1 , health and disease. Pharmacol. Res. 69, 1–10 (2013). 3. Gregory, K. E. & Walker, W. A. Immunologic Factors in Human Milk and Disease 36. Kuchaiev, O., Stevanovic´ , A., Hayes, W. & Przˇ ulj, N. GraphCrunch 2: Software Prevention in the Preterm Infant. Curr Pediatr Rep 1, 222–228 (2013). tool for network modeling, alignment and clustering. BMC Bioinformatics 12,24 4. Ward, T. L., Hosid, S., Ioshikhes, I. & Altosaar, I. Human milk metagenome: a (2011). functional capacity analysis. BMC Microbiol. 13, 116 (2013). 37. Delgado, S., Arroyo, R., Martı´ n, R. & Rodrı´ guez, J. M. PCR-DGGE assessment of 5. Latuga, M. S., Stuebe, A. & Seed, P. C. A review of the source and function of the bacterial diversity of breast milk in women with lactational infectious mastitis. microbiota in breast milk. Semin. Reprod. Med. 32, 68–73 (2014). BMC Infect. Dis. 8, 51 (2008). 6. Heikkila¨, M. P. & Saris, P. E. J. Inhibition of Staphylococcus aureus by the 38. Delgado, S. et al. Staphylococcus epidermidis strains isolated from breast milk of commensal bacteria of human milk. J. Appl. Microbiol. 95, 471–478 (2003). women suffering infectious mastitis: potential virulence traits and resistance to 7. Farr, B. M., Salgado, C. D., Karchmer, T. B. & Sherertz, R. J. Can antibiotic- antibiotics. BMC Microbiol. 9, 82 (2009). resistant nosocomial infections be controlled? Lancet Infect Dis 1, 38–45 (2001). 39. Arroyo, R. et al. Treatment of infectious mastitis during lactation: antibiotics 8. Zorgani, A. et al. Detection of pyrogenic toxins of Staphylococcus aureus in sudden versus oral administration of Lactobacilli isolated from breast milk. Clin. Infect. infant death syndrome. FEMS Immunol. Med. Microbiol. 25, 103–108 (1999). Dis. 50, 1551–1558 (2010). 9. Blackwell, C. C. et al. Toxigenic bacteria and sudden infant death syndrome 40. Brown, K. H., Black, R. E., Lopez de Roman˜ a, G. & Creed de Kanashiro, H. Infant- (SIDS): nasopharyngeal flora during the first year of life. FEMS Immunol. Med. feeding practices and their relationship with diarrheal and other diseases in Microbiol. 25, 51–58 (1999). Huascar (Lima), Peru. Pediatrics 83, 31–40 (1989). SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 5 www.nature.com/scientificreports 41. Lo´ pez-Alarco´ n, M., Villalpando, S. & Fajardo, A. Breast-feeding lowers the Program of Yunnan Province’’ ( ‘‘ ’’ ‘‘ ’’) and frequency and duration of acute respiratory infection and diarrhea in infants National Science Foundation of China (NSFC Grants No. 61175071, No. 71473243). under six months of age. J. Nutr. 127, 436–443 (1997). 42. Osterman, K. L. & Rahm, V. A. Lactation mastitis: bacterial cultivation of breast Author contributions milk, symptoms, treatment, and outcome. J Hum Lact 16, 297–302 (2000). Z.M. conceived the study, and wrote the manuscript; Q.G., C.Z., C.Y. & Z.M. performed the 43. Fetherston, C. Mastitis in lactating women: physiology or pathology? Breastfeed data analysis and interpretation; J.F. & L.F provided advices to the data interpretation and Rev 9, 5–12 (2001). revised the manuscript; all authors reviewed the manuscript. 44. Barbosa-Cesnik, C., Schwartz, K. & Foxman, B. Lactation mastitis. JAMA 289, 1609–1612 (2003). 45. Foxman, B., D’Arcy, H., Gillespie, B., Bobo, J. K. & Schwartz, K. Lactation mastitis: Additional information occurrence and medical management among 946 breastfeeding women in the Competing financial interests: The authors declare no competing financial interests. United States. Am. J. Epidemiol. 155, 103–114 (2002). How to cite this article: Ma, Z.S. et al. Network analysis suggests a potentially ‘evil’ alliance 46. Urbaniak, C. et al. Microbiota of human breast tissue. Appl. Environ. Microbiol. of opportunistic pathogens inhibited by a cooperative network in human milk bacterial 80, 3007–3014 (2014). communities. Sci. Rep. 5, 8275; DOI:10.1038/srep08275 (2015). This work is licensed under a Creative Commons Attribution-NonCommercial- Acknowledgments NoDerivs 4.0 International License. The images or other third party material in We appreciate Prof. Zhang Yaping, Academician and Vice President of the Chinese this article are included in the article’s Creative Commons license, unless indicated Academy of Science (CAS), for reviewing the manuscript and for his insightful comments otherwise in the credit line; if the material is not included under the Creative and suggestions, which played a significant role in our preparing for this submission. Z. Commons license, users will need to obtain permission from the license holder Ma’s research received funding from the following grants: ‘‘CAS One-Hundred Talented PI in order to reproduce the material. To view a copy of this license, visit http:// Program’’( ), ‘‘Exceptional Scientists Program and Overseas’ Top Scholar creativecommons.org/licenses/by-nc-nd/4.0/ SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 6 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Scientific Reports Springer Journals

Network analysis suggests a potentially ‘evil’ alliance of opportunistic pathogens inhibited by a cooperative network in human milk bacterial communities

Download PDF
6 pages

Loading next page...
 
/lp/springer-journals/network-analysis-suggests-a-potentially-evil-alliance-of-opportunistic-906lZvyBi2

References (51)

Publisher
Springer Journals
Copyright
Copyright © 2015 by The Author(s)
Subject
Science, Humanities and Social Sciences, multidisciplinary; Science, Humanities and Social Sciences, multidisciplinary; Science, multidisciplinary
eISSN
2045-2322
DOI
10.1038/srep08275
Publisher site
See Article on Publisher Site

Abstract

OPEN Network analysis suggests a potentially ‘evil’ alliance of opportunistic pathogens SUBJECT AREAS: MICROBIOLOGY inhibited by a cooperative network in MEDICAL RESEARCH METAGENOMICS human milk bacterial communities COMPUTATIONAL BIOLOGY AND BIOINFORMATICS 1 1 1,2 3 4 4 Zhanshan (Sam) Ma , Qiong Guan , Chengxi Ye , Chengchen Zhang , James A. Foster & Larry J. Forney Received Computational Biology and Medical Ecology Lab, State Key Laboratory of Genetic Resources & Evolution, Kunming Institute of 14 September 2014 Zoology, Chinese Academy of Sciences, Kunming, 650223, China, Department of Computer Science University of Maryland Accepted College Park, MD 20740, USA, School of Public Health Columbia University 722 West 168th Street New York, NY 10032, USA, 2 January 2015 Institute of Bioinformatics and Evolutionary Studies & Department of Biological Sciences University of Idaho Moscow, ID 83843, USA. Published 5 February 2015 The critical importance of human milk to infants and even human civilization has been well established. Yet our understanding of the milk microbiome has been limited to cataloguing OTUs and computation of community diversity. To the best of our knowledge, there has been no report on the bacterial interactions Correspondence and within the milk microbiome. To bridge this gap, we reconstructed a milk bacterial community network requests for materials based on Hunt et al. Our analysis revealed that the milk microbiome network consists of two disconnected should be addressed to sub-networks. One sub-network is a fully connected complete graph consisting of seven genera as nodes and Z.S.M. (samma@ all of its pair-wise interactions among the bacteria are facilitative or cooperative. In contrast, the interactions in the other sub-network of eight nodes are mixed but dominantly cooperative. Somewhat surprisingly, the uidaho.edu) only ‘non-cooperative’ nodes in the second sub-network are mutually cooperative Staphylococcus and Corynebacterium that include some opportunistic pathogens. This potentially ‘evil’ alliance between Staphylococcus and Corynebacterium could be inhibited by the remaining nodes that cooperate with one another in the second sub-network. We postulate that the ‘confrontation’ between the ‘evil’ alliance and ‘benign’ alliance and the shifting balance between them may be responsible for dysbiosis of the milk microbiome that permits mastitis. uman milk is generally considered the best source of nutrients for infants, and its health benefits such as prebiotics, immune proteins, and the microbiome of human milk itself, have been increasingly recog- 1–5 H nized . Similar to other habitats in or on the human body such as the gut and skin, human milk is not sterile at all and it hosts extensive bacterial communities that are posited to possess important health implications. In general, traditional literature on human milk has been focused on pathogenic bacteria, and our understanding on commensal bacteria is still very limited in spite of the rapid advances in metagenomic technology and expanding studies of the human microbiome in recent years. For example, Heikkila & Saris (2003) investigated potential inhibition of Staphylococcus aureus by the commensal bacteria of breast milk . Staphylococcus aureus is known as a food-poisoning agent and a common cause of infections including serious antibiotic-resistant hospital 6,7 8,9 infections . In addition it has been implicated in SIDS (Sudden Infant Death Syndrome) as well as infectious 6,10,11 mastitis that affects 20–30% lactating women . There have been several studies that applied metagenomic sequencing technology to characterize human milk 12–17 bacterial communities and a recent one by Hunt et al (2011) provides the largest data set of 16S rRNA sequences from human milk samples . Hunt et al (2011) collected 47 samples from 16 breastfeeding women (3 samples from all but one individual) who self-reported as healthy and between 20–40 yr of age . Their study revealed that the most abundant genera in the milk samples were Staphylococcus, Streptococcus, Serratia, and Corynebacteria, while eight other genera had relative abundances exceeding 1%. Besides characterizing the composition of milk bacterial community, Hunt et al. (2011) for the first time described the within-individual variation and the among-individuals variation of milk bacterial communities. The within-individual variation, which can be thought as a measure of the stability of individual milk bacterial communities, differed between SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 1 www.nature.com/scientificreports foundation, was not triggered until the publication of two independ- individuals. In other words, temporal variation of community mem- bership or stability of individual milk communities varied signifi- ent seminal papers published by Watts & Strogatz (1998) on the dynamics of ‘‘small-world’’ networks in the journal Nature, and cantly between women. For example, from the samples of ‘‘Subject #5,’’ Staphylococcus occupied either the first or second position in by Barabasi and Albert (1999) on the emergence of scaling (i.e., scale free networks) in random networks in the journal Science, respect- terms of the relative abundance (22–59%); but the samples of ‘‘Subject #1’’, Staphylococcus only contributed less than 5% to the ively. A commonality of both the papers is the extension of basic random graph models so that they can better fit the patterns exhib- community abundance. The among-individual variations in the rela- tive abundances of bacteria were as large as six-fold . ited by many empirically observed networks in social, technological and natural networks. One of the most active application fields of Although milk microbiome was apparently missing in the initial network science is biology, thanks to the vast datasets available from US-NIH HMP roadmap, its critical importance to human health and genomic and metagenomic research. The case for applying network diseases is evident. The importance is even more obvious from the analysis to investigate the human microbiome was argued convin- perspective of its relationships with the microbiome in other body cingly by Foster et al. (2008) and since then several network ana- sites because human microbiome is location specific, but not isolated 5,18 lyses have been successfully performed with human microbiome from one another at all . Zaura et al. (2014) hypothesized that 31–33 data . One huge advantage of network analysis is its power to development of fetal tolerance toward the microbiome of the mother visualize multivariate relationships generated from big data sets such during pregnancy is a major factor in the successful acquisition of a as genomic and metagenomic sequence data. Furthermore, various normal microbiome . Jeurink et al. (2013) proposed a mechanism parameters computed with network analysis software packages (e.g., for the formation of breast milk micriobiome, which involves 34 35 Cytoscape , Gephi ) offer informative insights on the patterns in immune cell education by the pregnancy hormone progesterone biological data. Complex network alignment algorithms and soft- leading to the transportation of bacteria from the mother to her ware (e.g., Graphcrunch2 ) can further be utilized to compare bio- mammal glands . Guts of breastfed infants showed significantly logical networks under different treatments. higher counts of bifidobacteria and Lactobacillus and lower counts of Bacteroides, Clostridium coccoides group, Staphylococcus, and Methods Enterobacteriaceae, as compared with formula-fed infants . The The 16S rRNA sequence data sets of human milk were collected by Hunt et al (2011) . pioneering colonizers such as Bifidobacterium longum, which carries Specifically, the V1-V2 region of the bacterial 16S rRNA gene was amplified from several gene clusters dedicated to the metabolism of human milk genomic DNA using universal primers and approximately 300,000 reads were gen- oligosaccharides (HMOs) allows infants to digest breast milk and erated from the barcoded pyrosequencing of amplicons from 47 samples. After ´ quality control, the data set was reduced to approximately 160,000 reads, with a mean possibly some simple vegetal food such as rice. Gonzalez et al. of 3400 sequences per sample. The sequence data were assigned to the most likely (2013) found that women with HIV RNA in breast milk have a bacterial genera using the RDP Bayesian classifier. A table of the 15 most abundant different pattern of microbial composition, compared with milk genera in each sample was supplied in the Supporting Information (Table S1) of Hunt without HIV RNA, indicating specific immunological phenomena et al (2011) and was used for our network analysis. in HIV-infected women . They also argued that breast milk and infant gut microbiota are essential for the maturation and protection Results of infant’s immune system. A metagenomic study conducted by The pair-wise relationships among 15 genera were measured by Ward et al (2013) confirmed the benefits of breast milk ingestion Spearman rank correlation coefficients with p-value of 0.05, and to the microbial colonization of the infant gut and immunity . The the computed values of Spearman correlation coefficients with R- latter is demonstrated by the existence of immune-modulatory statistics package (www.r-project.org) were feed into Cytoscape net- motifs in the metagenome of breast milk . These recent studies 34 35 36 work analysis software and Gephi . The Graphcrunch2 software exhibited the significant importance of breast milk microbiome in was applied to further compare the reconstructed milk microbiome health and diseases. network with several standard models of complex networks. The Obviously, Hunt et al (2011) and other previous culture-inde- results of network analysis are exhibited in Figure 1 and Table 1. 12–17,22–23 pendent studies have deepened our understanding of bac- Before discussing our results further, it should be noted that like terial communities in human breast milk. In a recent study, Guan many other studies of biological networks, our network was built and Ma (2014) applied Taylor’s power law and neutral theory to based on the correlation between OTUs. Our usage of terminology investigate the abundance distribution pattern and the maintenance such as ‘cooperative’, ‘non-cooperative’, and ‘evil alliance’ are by mechanism of milk microbial community diversity, respectively by analogy. Correlation is not equivalent with causation. For example, reanalyzing the existing data on milk microbiome . The analysis the correlation between two OTUs may be due to indirect ‘facili- with Taylor’s power law model indicated that bacterial population tation’ or ‘inhibition’ by a third player. Nevertheless, at this stage, abundance in human milk microbiome is aggregated, rather than correlation data are the only available data type for the human milk random, and it was found that neutral theory did not fit to any of microbiome. The correlation network therefore offers the best appar- the 47 samples (communities), suggesting non-random interactions atus available today to tackle the tangled microbiota in human milk. in community assembly and diversity maintenance . Nevertheless, It should also be noted that the OTU data we used resolves taxa at due to the limitation of the analytical approaches used in previous the level of genera and a network based on species data may reveal studies, we still have little knowledge on the bacterial interactions, different patterns from the networks we obtained. Hence, the total beyond their non-random nature, within milk microbiome. Indeed, validity of the conclusions we draw below, ultimately, should be most statistical approaches are not powerful enough to reveal the subject to testing in future biomedical research. Nevertheless, given 25,26 interspecies interactions within a microbial community . In this the fundamental importance of studying milk microbiome, we article, we take advantage the power of network analysis in studying believe that preliminary analyses such as this are warranted. interspecies or inter-OTU (Operational Taxonomic Unit) interac- From the above results in Figure 1 and Table 1, we draw the tions in a complex network setting such as a microbial community. following conclusions: Erdos and Renyi (1960) seminal research on random graph the- (i) The network of the breast milk bacterial community consists of ory opened one of the most exciting new fields in combinatorial two sub-networks, that correspond to two disconnected components th mathematics in the 20 century, and their random graph theory or communities (Figure 1a, Figure 1b, and Table 1). One component attracted extensive studies during the subsequent decades. Still, the (Figure 1b) consists of seven nodes in which all of the nodes are fully avalanche of approaches to network analysis and the emergence of connected (forming a complete graph) and they interact coopera- network science, of which random graph theory forms a theoretic tively (positive correlations). Another component consists of 8 nodes SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 2 www.nature.com/scientificreports a b cd Figure 1 | Bacterial network of the human milk microbiome reconstructed using the data sets of Hunt et al. (2011): Figure (1a) and (1b) show the two disconnected sub-networks (components) of the breast milk bacterial network; Figure (1c) and (1d) are the same components, corresponding to (1a) and (1b), respectively, assuming the Staphylococcus and Corynebacterium nodes were eliminated. The green line represents a positive correlation (cooperative interaction) while the red line represents a negative correlation (non-cooperative interaction). Obviously, when the two mutually cooperative players Staphylococcus and Corynebacterium are removed, the whole network becomes totally cooperative (Figure 1c & 1d). and all of the interactions (edges) are cooperative except those As mentioned previously, Staphylococcus aureus is a food-poisoning that involve Staphylococcus and Corynebacterium. Therefore, if both agent and a common cause of infections including serious antibiotic- 6,7 Staphylococcus and Corynebacterium were removed, then the remain- resistant hospital infections , and the bacterium is also implicated in 8,9 6,10,11 ing nodes in the sub-network are fully cooperative. Furthermore, the SIDS (Sudden Infant Death Syndrome) and infectious mastitis . relationship between Staphylococcus and Corynebacterium are coop- Mastitis is inflammation of the breast with or without infection, and erative, although they do not cooperate with other taxa in the network. Staphylococcus aureus has traditionally been believed to be the patho- Obviously, the milk bacterial network is dominantly cooperative, and gen that is typically associated with infectious mastitis . There are the ratio of cooperative vs. non-cooperative interactions is 451. studies that reveal some other species of Staphylococcus such as Table 1 | Topological properties of human milk bacterial network Number of nodes Number of edges Avg. number of neighbors Clustering coefficient Connected components Network diameter 15 45 6 0.944 2 2 Average path length Network density Modularity Number of communities Small-world network Scale-free network 1.082 0.429 0.498 2 Yes No SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 3 www.nature.com/scientificreports Staphylococcus epidermidis may play a prevalent role in mastitis work (complete graph). Obviously, the lower density is because there 2,5,37,38 infections . Although the etiology of mastitis may vary there is no edge (connection) between the two disconnected components, is evidence from animal studies and clinical trials that suggests which significantly lowers the network density. certain strains of Lactobacillus can produce anti-inflammatory and anti-bacterial factors that inhibit adhesion and internalization Discussion of Staphylococcus spp. For example, Arroyo et al. (2010) reported In the study done by Hunt et al (2011) universal primers were used that treatment with probiotic strains from human milk (containing for amplification of the 16S rRNA gene, which increased sequencing Lactobacillus strains) produced a greater reduction in the bacterial coverage and helped them to obtain the most comprehensive experi- counts of Staphylococcus epidermidis, Staphylococcus aureus and mental survey of milk microbiome to date. Their study confirmed that Staphylococcus mitis, as well as a greater reduction in pain than Staphylococcus and Corynebacterium, identified here appear to exist the treatment with antibiotics . as an ‘‘evil alliance’’ in the milk microbiome based on our network The genus Corynebacterium includes Gram-positive, rod-shaped 28 analysis even though they are typically present on adult skin too . bacteria that are largely innocuous and widely distributed in nature. Hunt et al (2011) had recognized the possibility of contamination However, some species such as C. diphtheriae may cause human from the skin microbiome and took extraordinary caution during disease. Extensive studies on the health implications of breast sample collection. Furthermore, they compared the compositions of milk include protecting infants from diarrheal and respiratory dis- both milk and skin communities and concluded that bacterial com- 1,40,41 eases . Our finding from network analysis that every other munities in milk cannot simply be a result from skin contamination . ‘player’ in the milk microbiome collectively ‘opposes’ or ‘inhibits’ Some species of Staphylococcus and Corynebacteria are opportunistic Staphylococcus and Corynebacterium possibly explains the health infectious agents, and Staphylococcus aureus is associated with lacta- effects of milk despite the presence of potential pathogens. In other tional mastitis. Other studies have also demonstrated the occurrence 1,22,42,43 words, from the perspective of a lactating mother and baby, it may be of Staphylococcus aureus in human milk . It has been reported the cooperative and collective efforts of the other community mem- that, during the course of lactation, up to 30% of women suffer from bers that suppress the ‘evil’ alliance from opportunistically causing breast infections or inflammation (mastitis) that often lead to fever, unhealthy consequences such as mastitis. 44,45 redness, swelling and breast pains . In Hunt et al (2011) study, the (ii) The milk bacterial network is a ‘small-world’ network with a milk donors were self-reported as healthy, but at least one of the network diameter of only 2 and average minimal path length of subjects showed symptoms of mastitis . Therefore, at least, existing p51.082 (Table 1). This is in accord with the criterion for judging literature on milk microbiome referenced above indicates the existence small world networks , in which p51.082,log(N)5log(15)52.708. of potentially opportunistic pathogens in milk of both healthy and The criterion means that the typical distance p between two ran- diseased (mastitis) women. But why do those potential pathogens domly chosen nodes grows proportionally to the logarithm of N, often seem harmless to lactating mothers and infants? In contrary, which is even smaller than the linear growth. In biological terms the existing literature documented the benefits of milk microbiome the interactions amongst the nodes (genera) are very tight. such as the protective effects of breastfeeding against diarrheal and Nevertheless, the milk microbiome network is not scale-free,which respiratory disease as well as reduced risk of developing obesity in is evidenced by the fact that the degree distribution of the network infants . There is not an existing theory with sufficient evidence to does not fit to a power law distribution (p-value50.323.0.05 from explain the natural phenomenon in the existing literature of human 2l fitting power-law model, i.e., p(k) / k ). A scale-free network has milk microbiome. two properties: growth and preferential attachment . Growth means A recent study by Urbaniak et al. (2014) investigated the existence that the number of nodes in the network increases over time and of microbiome within mammary tissue by using 16S rRNA sequen- preferential attachment means that the more connected a node is, cing and culture . They analyzed the breast tissue from 81 women the more likely it is to receive new links. This absence of scale-free with and without cancer in Canada and Ireland, and confirmed the property may be due to the fact that our network is built with bacterial existence of both health-conferring bacteria such as Lactobacillus genera as the node units, which might be relatively constant over time and Bifidobacterium, as well as taxa known for pathogenesis such and therefore the growth of nodes could be insignificant. We also as Enterobacteriaceae, Pseudomonas, and Streptococcus agalactiae. utilized the network alignment software GraphCrunch2 to compare Yet, none of the 81 women recruited had any clinical signs or symp- our milk microbiome network with 50 random instances of each of toms of breast infection. This echoes the phenomenon of the pres- the following network models with the same size as microbiome net- ence of opportunistic pathogens in breast milk revealed by our work, respectively: ER (Erdos-Renyi random graphs), ER_DD (Erdos- network analysis in this article. Re´ nyi random graphs with the same degree distribution as the data), Our network analysis revealed the possible existence of an ‘‘evil GEO (Geometric Random Graphs), GEO-GD (Geometric Gene Dup- alliance’’ between Staphylococcus and Corynebacterium, and this alli- lication Models), SF (Scale-free Baraba´ si-Albert Preferential Attach- ance is collectively ‘opposed’ or inhibited by the other members of ment Models), SF-GD (Scale-free Gene Duplication Models), and breast milk bacterial community (network). Our finding from net- STICKY (Stickiness-index Based Models). The results from graph- work analysis therefore offers a piece of concrete evidence to support crunch2 software also demonstrated that a scale-free network model the following hypothesis: Similar to natural ecosystems, the ecosystem was among the worst performing models. of human milk microbiome, which consists of the milk microbial (iii) The other topological parameters of milk bacterial network community and its environment (i.e., the human body or host), con- shown in Table (1) also offer some interesting information about the tains microbial species of various characteristics or functions, being network characteristics. Both the numbers of connected components beneficial, harmful, or neutral from a human health perspective. This and the number of communities in the network are two, correspond- dynamic balance in the milk microbiome ecosystem depends on the ing to the two sub-networks (Figure 1a & 1b). In general, the values of species interactions within the microbiome bacterial community as these two parameters are not necessarily equal, but their equality in well as the host environment such as immune system. The states of the the case of milk microbiome further strengthens the evidence that the milk microbiome (network), which could correspond to healthy or milk microbiome network is divided into two separate sub-networks. disease states of human body, depend on the interactions within the The high clustering coefficient of 0.944 signals the high aggregation microbiome network (as exhibited by Figure 1) as well as the host, tendency of network nodes. Moreover, a network density of 0.429 which may have her unique genomic, immunological, physiological suggests that the number of edges in the network is about 43%, i.e., and demographic properties. Specifically, we postulate that in healthy less than half of all possible edges with a completely connected net- state, the adverse consequences of the ‘‘evil alliance’’ of Staphylococcus SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 4 www.nature.com/scientificreports 10. Jonsson, S. & Pulkkinen, M. O. Mastitis today: incidence, prevention and and/or Corynebacteria is ‘contained’ or inhibited by the collective treatment. Ann Chir Gynaecol Suppl 208, 84–87 (1994). cooperative defense of other community members. In the face of a 11. Semba, R. D. et al. Mastitis and immunological factors in breast milk of lactating disturbance of ‘evil alliance’ cannot be contained, and the balance may women in Malawi. Clin. Diagn. Lab. Immunol. 6, 671–674 (1999). be disrupted or shifted to an alternative state, which may correspond 12. Perez, P. F. et al. Bacterial imprinting of the neonatal immune system: lessons from maternal cells? Pediatrics 119, e724–732 (2007). to disease such as mastitis, whose underlying causes or triggers are not 13. Martı n, R. et al. Cultivation-independent assessment of the bacterial diversity of known yet. In other words, the ‘‘evil alliance’’ of Staphylococcus and breast milk among healthy women. Res. Microbiol. 158, 31–37 (2007). Corynebacterium are opportunistic and they cause disease to their host 14. Martı´ n, R. et al. Isolation of bifidobacteria from breast milk and assessment of the only if the collective containment by other members of the micro- bifidobacterial population by PCR-denaturing gradient gel electrophoresis and biome coupled with the defense of host immune system are weakened quantitative real-time PCR. Appl. Environ. Microbiol. 75, 965–969 (2009). 15. Gueimonde, M., Laitinen, K., Salminen, S. & Isolauri, E. Breast milk: a source of to such extent that their disturbance cannot be ‘absorbed’ or tolerated. bifidobacteria for infant gut development and maturation? Neonatology 92, 64–66 In a recent study, Ward et al. (2013) postulated that it might be the (2007). diversity or genetic traits of the milk metagenome that confer benefits 16. Hamady, M., Walker, J. J., Harris, J. K., Gold, N. J. & Knight, R. Error-correcting to infants, rather than the action of any one bacterial genus or species . barcoded primers for pyrosequencing hundreds of samples in multiplex. Nat. Methods 5, 235–237 (2008). We argue that our hypothesis of dynamic balance of milk microbiome 17. Collado, M. C., Delgado, S., Maldonado, A. & Rodrı´ guez, J. M. Assessment of the ecosystem is consistent with the ideas postulated by Ward et al. 4 bacterial diversity of breast milk of healthy women by quantitative real-time PCR. (2013) for the following two reasons: (i) the diversity and/or se- Lett. Appl. Microbiol. 48, 523–528 (2009). quencesofDNA within themetagenomereferredby Wardetal. 18. Costello, E. K., Stagaman, K., Dethlefsen, L., Bohannan, B. J. M. & Relman, D. A. (2103) are carried by the microbiome, and they are the ‘surrogates’ The application of ecological theory toward an understanding of the human microbiome. Science 336, 1255–1262 (2012). of the species or OTUs within microbiome; (ii) ecological theory tells 19. Zaura, E., Nicu, E. A., Krom, B. P. & Keijser, B. J. F. Acquiring and maintaining a us, that diversity and stability, are tangled together, even if stability is normal oral microbiome: current perspective. Front Cell Infect Microbiol 4,85 not determined by diversity. Indeed, the diversity-stability paradigm (2014). debates have been going on for more than a half-century now, but 20. Jeurink, P. V. et al. Human milk: a source of more life than we imagine. Benef nobody denies their close relationship. In fact, a recent consensus has Microbes 4, 17–30 (2013). 21. Matamoros, S., Gras-Leguen, C., Le Vacon, F., Potel, G. & de La Cochetiere, M.-F. been that network analysis should play a critical role in investigating Development of intestinal microbiota in infants and its impact on health. Trends the paradigm. We hope this relatively simple network analysis re- Microbiol. 21, 167–173 (2013). ported here will induce more extensive applications of network ana- 22. Gonzalez, R. et al. Breast milk and gut microbiota in African mothers and infants lysis approaches to the study of human milk microbiome, especially from an area of high HIV prevalence. PLoS ONE 8, e80299 (2013). 23. Xuan, C. et al. Microbial dysbiosis is associated with human breast cancer. PLoS the stability (balance) of the milk ecosystem, which should greatly ONE 9, e83744 (2014). enhance our understanding of the health and disease implications of 24. Guan, Q. & Ma, Z. S. Ecological analysis of the human milk microbiome. Chinese milk bacterial microbiota. Science Bulletin 59, 2205–2212 (2014). There remain some important follow-up questions to this network 25. Ma, Z. S., Abdo, Z. & Forney, L. J. Caring about trees in the forest: incorporating analysis, and the answers will need efforts from the biomedical frailty in risk analysis for personalized medicine. Personalized Medicine 8, 681–688 (2011). research community. Indeed, the insights from our network analysis 26. Ma, Z. S., Geng, J. W., Abdo, Z. & Forney, L. J. [A Bird’s Eye View of Microbial should only be a starting point for thoroughly understanding the Community Dynamics.] Microbial Ecological Theory: Current Perspectives health and disease implications of the human milk microbiome. [Ogilvie, L. A. and Hirsch, P. R. (ed.)][pp57–71] ( Caister Academic Press London, Experimental and clinical studies of human milk microbiome are 2012). crucial, but also understandably difficult to conduct. We realize that 27. Erdo˝s, P. & Re nyi, A. On the Evolution of Random Graphs. Publication of the Mathematical Institute of the Hungarian Academy of Sciences 5, 17–61 (1960). even the basic research of the species identification of human milk 28. Watts, D. J. & Strogatz, S. H. Collective dynamics of ‘small-world’ networks. microbes is still at the very preliminary stage. Both our analysis and Nature 393, 440–442 (1998). the study done by Hunt et al (2011) , from which we obtained data 29. Barabasi, A. L. & Albert, R. Emergence of scaling in random networks. Science 286, for reconstructing the milk microbiome network, only reported the 509–512 (1999). 30. Foster, J. A., Krone, S. M. & Forney, L. J. Application of ecological network theory genera of bacteria present, but not the species within these genera. to the human microbiome. Interdiscip Perspect Infect Dis 2008, 839501 (2008). This data limitation strongly constrained our capability to infer fur- 31. Bapteste, E., Bicep, C. & Lopez, P. Evolution of genetic diversity using networks: ther information on the bacterial interaction within human milk the human gut microbiome as a case study. Clin. Microbiol. Infect. 18 Suppl 4, microbiome. Given the self-evident, critical importance of human 40–43 (2012). milk microbiome, we hope this study will be a stepping stone for 32. Greenblum, S., Turnbaugh, P. J. & Borenstein, E. Metagenomic systems biology of the human gut microbiome reveals topological shifts associated with obesity and more advanced network analysis based on more comprehensive bio- inflammatory bowel disease. Proc. Natl. Acad. Sci. U.S.A. 109, 594–599 (2012). medical data sets generated by future studies of the human milk 33. Faust, K. et al. Microbial co-occurrence relationships in the human microbiome. microbiome. PLoS Comput. Biol. 8, e1002606 (2012). 34. Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003). 1. Hunt, K. M. et al. Characterization of the diversity and temporal stability of 35. Bastian, M., Heymann, S. & Jacomy, M. Gephi: An open source software for bacterial communities in human milk. PLoS ONE 6, e21313 (2011). exploring and manipulating networks. Proceedings of the Third International st 2. Ferna´ ndez, L. et al. The human milk microbiota: origin and potential roles in Conference on Weblogs and Social Media (2009) 2pp, Paris, France, March 1 , health and disease. Pharmacol. Res. 69, 1–10 (2013). 3. Gregory, K. E. & Walker, W. A. Immunologic Factors in Human Milk and Disease 36. Kuchaiev, O., Stevanovic´ , A., Hayes, W. & Przˇ ulj, N. GraphCrunch 2: Software Prevention in the Preterm Infant. Curr Pediatr Rep 1, 222–228 (2013). tool for network modeling, alignment and clustering. BMC Bioinformatics 12,24 4. Ward, T. L., Hosid, S., Ioshikhes, I. & Altosaar, I. Human milk metagenome: a (2011). functional capacity analysis. BMC Microbiol. 13, 116 (2013). 37. Delgado, S., Arroyo, R., Martı´ n, R. & Rodrı´ guez, J. M. PCR-DGGE assessment of 5. Latuga, M. S., Stuebe, A. & Seed, P. C. A review of the source and function of the bacterial diversity of breast milk in women with lactational infectious mastitis. microbiota in breast milk. Semin. Reprod. Med. 32, 68–73 (2014). BMC Infect. Dis. 8, 51 (2008). 6. Heikkila¨, M. P. & Saris, P. E. J. Inhibition of Staphylococcus aureus by the 38. Delgado, S. et al. Staphylococcus epidermidis strains isolated from breast milk of commensal bacteria of human milk. J. Appl. Microbiol. 95, 471–478 (2003). women suffering infectious mastitis: potential virulence traits and resistance to 7. Farr, B. M., Salgado, C. D., Karchmer, T. B. & Sherertz, R. J. Can antibiotic- antibiotics. BMC Microbiol. 9, 82 (2009). resistant nosocomial infections be controlled? Lancet Infect Dis 1, 38–45 (2001). 39. Arroyo, R. et al. Treatment of infectious mastitis during lactation: antibiotics 8. Zorgani, A. et al. Detection of pyrogenic toxins of Staphylococcus aureus in sudden versus oral administration of Lactobacilli isolated from breast milk. Clin. Infect. infant death syndrome. FEMS Immunol. Med. Microbiol. 25, 103–108 (1999). Dis. 50, 1551–1558 (2010). 9. Blackwell, C. C. et al. Toxigenic bacteria and sudden infant death syndrome 40. Brown, K. H., Black, R. E., Lopez de Roman˜ a, G. & Creed de Kanashiro, H. Infant- (SIDS): nasopharyngeal flora during the first year of life. FEMS Immunol. Med. feeding practices and their relationship with diarrheal and other diseases in Microbiol. 25, 51–58 (1999). Huascar (Lima), Peru. Pediatrics 83, 31–40 (1989). SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 5 www.nature.com/scientificreports 41. Lo´ pez-Alarco´ n, M., Villalpando, S. & Fajardo, A. Breast-feeding lowers the Program of Yunnan Province’’ ( ‘‘ ’’ ‘‘ ’’) and frequency and duration of acute respiratory infection and diarrhea in infants National Science Foundation of China (NSFC Grants No. 61175071, No. 71473243). under six months of age. J. Nutr. 127, 436–443 (1997). 42. Osterman, K. L. & Rahm, V. A. Lactation mastitis: bacterial cultivation of breast Author contributions milk, symptoms, treatment, and outcome. J Hum Lact 16, 297–302 (2000). Z.M. conceived the study, and wrote the manuscript; Q.G., C.Z., C.Y. & Z.M. performed the 43. Fetherston, C. Mastitis in lactating women: physiology or pathology? Breastfeed data analysis and interpretation; J.F. & L.F provided advices to the data interpretation and Rev 9, 5–12 (2001). revised the manuscript; all authors reviewed the manuscript. 44. Barbosa-Cesnik, C., Schwartz, K. & Foxman, B. Lactation mastitis. JAMA 289, 1609–1612 (2003). 45. Foxman, B., D’Arcy, H., Gillespie, B., Bobo, J. K. & Schwartz, K. Lactation mastitis: Additional information occurrence and medical management among 946 breastfeeding women in the Competing financial interests: The authors declare no competing financial interests. United States. Am. J. Epidemiol. 155, 103–114 (2002). How to cite this article: Ma, Z.S. et al. Network analysis suggests a potentially ‘evil’ alliance 46. Urbaniak, C. et al. Microbiota of human breast tissue. Appl. Environ. Microbiol. of opportunistic pathogens inhibited by a cooperative network in human milk bacterial 80, 3007–3014 (2014). communities. Sci. Rep. 5, 8275; DOI:10.1038/srep08275 (2015). This work is licensed under a Creative Commons Attribution-NonCommercial- Acknowledgments NoDerivs 4.0 International License. The images or other third party material in We appreciate Prof. Zhang Yaping, Academician and Vice President of the Chinese this article are included in the article’s Creative Commons license, unless indicated Academy of Science (CAS), for reviewing the manuscript and for his insightful comments otherwise in the credit line; if the material is not included under the Creative and suggestions, which played a significant role in our preparing for this submission. Z. Commons license, users will need to obtain permission from the license holder Ma’s research received funding from the following grants: ‘‘CAS One-Hundred Talented PI in order to reproduce the material. To view a copy of this license, visit http:// Program’’( ), ‘‘Exceptional Scientists Program and Overseas’ Top Scholar creativecommons.org/licenses/by-nc-nd/4.0/ SCIENTIFIC REPORTS | 5 : 8275 | DOI: 10.1038/srep08275 6

Journal

Scientific Reports – Springer Journals

Published: Feb 5, 2015

There are no references for this article.