UBC Faculty Research and Publications

Linear growth faltering in infants is associated with Acidaminococcus sp. and community-level changes… Gough, Ethan K; Stephens, David A; Moodie, Erica E; Prendergast, Andrew J; Stoltzfus, Rebecca J; Humphrey, Jean H; Manges, Amee R Jun 13, 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
52383-40168_2015_Article_89.pdf [ 2.31MB ]
Metadata
JSON: 52383-1.0223880.json
JSON-LD: 52383-1.0223880-ld.json
RDF/XML (Pretty): 52383-1.0223880-rdf.xml
RDF/JSON: 52383-1.0223880-rdf.json
Turtle: 52383-1.0223880-turtle.txt
N-Triples: 52383-1.0223880-rdf-ntriples.txt
Original Record: 52383-1.0223880-source.json
Full Text
52383-1.0223880-fulltext.txt
Citation
52383-1.0223880.ris

Full Text

RESEARCHLinear growth faltering in inni, AUndernutrition in early childhood underlies 45 % of mortal-ity in children aged under 5 years worldwide, resulting inmorbidity and mortality [2], leads to poor motor devel-opment and cognition, and reduces educational andiddle-incometal potentiale period fromof deficits inGough et al. Microbiome  (2015) 3:24 DOI 10.1186/s40168-015-0089-2course. Social, economic, and educational factors, as wellBritish Columbia, 137-2206 East Mall, Vancouver V6T 1Z3, BC, CanadaFull list of author information is available at the end of the articlelinear growth beyond that period is limited. Interventionsto prevent stunting are therefore required early in the life-* Correspondence: amee.manges@ubc.ca7Faculty of Medicine, School of Population and Public Health, University ofchronic malnutrition, respectively, and are often measuredin terms of z-scores (i.e., deviations in attained growth froma reference population median). Children whose length- orheight-for-age z-scores (LAZ or HAZ) is more than 2standard deviations below the reference population medianstunted in 2011 [1], representing almostchildren in this age group in low- and mcountries (LMICs), hindering developmenand human capital of entire societies.Most linear growth faltering occurs in thconception to 2 years of age, and restoration3.1 million deaths annually [1]. Ponderal and linear growthfaltering in children are viewed as indicators of acute andeconomic attainment over the life-course [1–3]. An es-timated 165 million children under 5 years old wereone third ofBackground: Chronic malnutrition, termed stunting, is defined as suboptimal linear growth, affects one third ofchildren in developing countries, and leads to increased mortality and poor developmental outcomes. The causesof childhood stunting are unknown, and strategies to improve growth and related outcomes in children have onlyhad modest impacts. Recent studies have shown that the ecosystem of microbes in the human gut, termed themicrobiota, can induce changes in weight. However, the specific changes in the gut microbiota that contribute togrowth remain unknown, and no studies have investigated the gut microbiota as a determinant of chronic malnutrition.Results: We performed secondary analyses of data from two well-characterized twin cohorts of children from Malawi andBangladesh to identify bacterial genera associated with linear growth. In a case-control analysis, we used the graphicallasso to estimate covariance network models of gut microbial interactions from relative genus abundances andused network analysis methods to select genera associated with stunting severity. In longitudinal analyses, we determinedassociations between these selected microbes and linear growth using between-within twin regression models to adjustfor confounding and introduce temporality. Reduced microbiota diversity and increased covariance network density wereassociated with stunting severity, while increased relative abundance of Acidaminococcus sp. was associated with futurelinear growth deficits.Conclusions: We show that length growth in children is associated with community-wide changes in the gut microbiotaand with the abundance of the bacterial genus, Acidaminococcus. Larger cohorts are needed to confirm these findingsand to clarify the mechanisms involved.Keywords: Microbiota, Microbiome, Intestinal, Stunting, Growth, Statistical learning, NetworksBackground are termed stunted. Stunting has short-term effects onwith Acidaminococcus sp. alevel changes in the gut mEthan K. Gough1, David A. Stephens2, Erica E.M. Moodie1Jean H. Humphrey4,6 and Amee R. Manges7*Abstract© 2015 Gough et al. This is an Open Access a(http://creativecommons.org/licenses/by/4.0),provided the original work is properly creditedcreativecommons.org/publicdomain/zero/1.0/Open Accessfants is associatedd community-crobiotandrew J. Prendergast3,4, Rebecca J. Stoltzfus5,rticle distributed under the terms of the Creative Commons Attribution Licensewhich permits unrestricted use, distribution, and reproduction in any medium,. The Creative Commons Public Domain Dedication waiver (http://) applies to the data made available in this article, unless otherwise stated.Gough et al. Microbiome  (2015) 3:24 Page 2 of 10as infectious diseases and poor diet in early childhood allcontribute to linear growth faltering [1, 4–7]. Further-more, a number of studies have shown that small intes-tinal inflammation and permeability are associated withpoor linear growth [8–11]. This subclinical gut pathologyhas been termed environmental enteric dysfunction (EED)and is acquired early in life among children living in un-sanitary conditions [5, 12–15]. Reduced intestinal barrierfunction caused by EED enables bacterial translocationto occur, leading to chronic systemic inflammation, whichis associated with reduced insulin-like growth factor 1(IGF-1) and linear growth faltering [16]. However, thepathophysiology of stunting is not well understood, andcurrently available interventions, which focus mostly ondietary supplementation and prevention of diarrhea, haveonly a modest impact [17]. Mechanisms underlying stunt-ing therefore need to be better defined so that tractablepathways for intervention can be identified.Recent studies suggest a role of the intestinal microbiotain child growth. The intestinal microbiota is an ecosystemof gut microbes that helps to modulate nutrient harvestingfrom the diet, mucosal inflammation, and the immuneresponse in the gut [18–22]. Observational studies inhumans [23–26] have demonstrated a relationship betweenthe intestinal microbiota and severe acute malnutrition(SAM). A causal effect of the intestinal microbiota onweight has also been shown using experimental animalmodels [27, 28]. However, the specific changes in themicrobiota that contribute to growth remain unclear,and no studies to date have investigated the intestinalmicrobiota as a determinant of linear growth.We performed a secondary analysis of publicly availabledata from two twin cohorts of undernourished childrenfrom low-income settings (Malawi and Bangladesh)[25, 27], to identify bacterial genera whose relative abun-dances explain linear growth. Previous analyses from thesecohorts showed that acute malnutrition was associated withdifferences in gut microbiota functional gene abundances[27] and maturation [25]. Our analyses aimed to deter-mine changes in gut microbiota networks and relativeabundance associated with stunting status, in order to iden-tify potential microbiota members that contribute to lineargrowth faltering (i.e., chronic malnutrition). We hypothe-sized that differences in the relative abundance of identifiedgenera are independently associated with prospective defi-cits in linear growth between siblings.Results and discussionCohort descriptionData were provided for 44 children in the Malawi co-hort, who were median 10.2 months (interquartile range(IQR) 4.6, 14.5) old at baseline and followed for median9.7 months (IQR 4.1, 14.5). Baseline HAZ and weight-for-height z-scores (WHZ) were −2.95 (IQR −3.70, −2.18)and −0.46 (IQR −0.87, −0.13), respectively. Anthropomet-ric, epidemiological, and DNA whole genome shotgunsequencing data were provided for median 7 (IQR 4, 8)follow-up visits per child, for a total of 308 longitudinalobservations (Additional file 1: Table S1). Data wereavailable for 25 children in the Bangladesh birth cohort,who were 0.3 months (IQR 0.19, 0.63) old at baselineand followed for median 14.5 months (IQR 11.9, 20.7).Baseline HAZ and WHZ were −3.75 (IQR −4.54, −2.68)and −0.57 (IQR −1.51, 0.35), respectively. Anthropometric,epidemiological, and relative abundance data were pro-vided for median 17 (IQR 13, 22) follow-up visits perchild. Randomly excluding one child from the set of trip-lets for between-within regression analyses provided 429longitudinal observations.Description of cases and controlsIn the Malawi cohort, 13 children had a follow-up visitthat met incident case criteria for severe stunting, and11 had a follow-up visit that met control criteria forstunting (see “Methods” for details on case and controldefinitions). Six eligible cases were co-twins, and six eligiblecontrols were also co-twins. In the Bangladesh cohort, eightchildren had a follow-up visit that met incident case cri-teria, and ten had a follow-up visit that met control criteria.Four eligible cases were co-twins, and ten eligible controlswere co-twins. For each pair of co-twins that both met casecriteria, we randomly chose one sibling as a case to avoidwithin-group correlations [29]. The same was done forpairs of co-twins that both met control criteria. Thisprovided ten cases and eight controls from Malawi, andsix cases and five controls from Bangladesh (Fig. 1).Cases from the Malawi cohort had lower HAZ (−3.08v −2.45, p < 0.01) and were younger compared to controls(10.8 v 19.6 months, p = 0.05). Similarly, in the Bangladeshcohort, case HAZ was −3.17 v −2.63 for controls, p < 0.01,and age was 2.9 v 11.0 months, p < 0.01. WHZ was alsohigher in Bangladesh cases compared to controls (0.53v −0.64, p = 0.05) (Additional file 1: Table S1).Genus relative abundance and microbiota diversityRoche 454 shotgun whole genome sequence data wereprovided for median 76,700 (IQR 55,200, 103,000) readsper sample in the entire Malawi cohort, while relativeabundance data from the Bangladesh cohort were quan-tified from a median 20,192 (IQR 16, 155, 24,632) reads.In both cohorts, a similar number of reads were avail-able for cases and controls (Additional file 1: Table S1).In the Malawi cohort, Bifidobacterium (42.8 %) andPrevotella (22.7 %) were the most abundant genera iden-tified, followed by Bacteroides (3.7 %), Faecalibacterium(3.14 %), Collinsella (1.0 %), Lactobacillus (0.6 %), andBlautia (0.6 %). In the Bangladesh cohort, Bifidobacterium(46.2 %), Streptococcus (4.8 %), Lactobacillus (2.6 %), andl hoGough et al. Microbiome  (2015) 3:24 Page 3 of 1022 twin pairs (44 children), with median 7 visits per child (308 child visits)8 controls13 children had a visit that met incident case criteria for severe stunting6 siblings (3 twin pairs)11 children had a visit that met control criteria for stunting7 non-siblings6 siblings (3 twin pairs)7 children included as cases3 children randomly selected as cases(1 per pair)3 children randomly selected as controls(1 per pair)10 cases5 non-siblings5 children included as controls22 twin pairs (44 children) providing 308 child visits included in longitudinaanalysesFig. 1 Flow chart of case and control selection from the Malawi twin coselection from the Bangladesh twin cohort for network analysis (right)Escherichia/Shigella (1.8 %) were the most abundant gen-era, followed by Collinsella (0.5 %). These were also themost prevalent genera identified in fecal samples collectedduring follow-up (Additional file 2: Table S2) and are con-sistent with the literature on microbiota in infants andwith different diets [30–35]. In the Malawi cohort, Prevo-tella (18.1 v 42.9, p = 0.06), Bacteroides (1.9 v 7.4, p =0.01), Eubacterium (0.0 v 2.4, p < 0.01), and Blautia (0.6 v2.4, p = 0.03) showed the largest decrease in relative abun-dance in cases v controls (Additional file 3: Table S3). Inthe Bangladesh cohort, Lactobacillus (0.1 v 8.7, p < 0.01),Olsenella (0.0 v 0.8, p < 0.01), Dorea (0.0 v 0.7, p = 0.05),Blautia (0.0 v 0.2, p < 0.01), and unclassified genera in theCoriobacteriaceae (0.0 v 0.3, p < 0.01) and Enterococcaceae(0.0 v 0.1, p = 0.08) families showed the largest decreasein relative abundance in cases v controls. Lesser, butstatistically significant depletion of Anaerococcus, Dialister,Faecalibacterium, Megamonas, Weissella, Megasphaera,and unclassified genera in the Lachnospiraceae, Lactoba-cillaceae, and Veillonellaceae families were also observedin Bangladesh cases (Additional file 4: Table S4). Casemicrobiota were less diverse than controls in both cohorts(Malawi: 0.5 v 0.7, p = 0.02; Bangladesh: 0.5 v 0.7, p = 0.05)(Additional file 1: Table S1).Network indicesNetwork density (i.e., the probability that two randomlyselected microbes co-vary) was greater in case compared11 twin pairs and 1 set of triplets (25 children), with median 17 visits per child (448 child visits)5 controls8 children had a visit that met incident case criteria for severe stunting4 siblings (2 twin pairs)10 children had a visit that met control criteria for stunting4 non-siblings10 siblings (5 twin pairs)4 children included as cases2 children randomly selected as cases(1 per pair)5 children randomly selected as controls(1 per pair)6 cases11 twin pairs and  2 randomly chosen from the set of triplets (24 children,  providing 448 child visits included in longitudinal analysesrt for network analysis (left) and flow chart of case and controlto control networks in both cohorts (Malawi: 0.56 v0.25, p = 0.08; Bangladesh: 0.56 v 0.33, p = 0.42), indicat-ing a greater potential for information flow in casemicrobiota. We also observed that the density of edgesfrom aerobes to anaerobes was greater in the case net-work in both populations (Figs. 2 and 3).In the Malawi cohort, differences in degree centralitywere observed for Acidaminococcus (0.6 v 0.0, p = 0.06),Bacteroides (0.6 v 0.2, p = 0.03), Brachyspira (0.6 v 0.0,p = 0.09), Haemophilus (0.6 v 0.2, p = 0.07), and unclassifiedgenera in the Neisseriaceae (0.6 v 0.2, p = 0.08) and Chla-mydiaceae (0.6 v 0.0, p = 0.05) families in case v controlnetworks (Additional file 3: Table S3). In the Bangladeshcohort, Acinetobacter (0.5 v 0.0, p = 0.03), Anaerococcus(0.7 v 0.2, p = 0.09), Blautia (0.7 v 0.2, p = 0.08), Coprococ-cus (0.5 v 0.0, p = 0.03), Geobacillus (0.6 v 0.0, p = 0.09),Lactococcus (0.6 v 0.0, p = 0.02), Micrococcus (0.5 v 0.0,p = 0.05), Proteus (0.6 v 0.0, p = 0.09), and Sarcina (0.6v 0.0, p = 0.09) were more central in the case network(Additional file 4: Table S4).Between-within modelsThirty of 164 genera identified across both populationswere selected, based on statistically significant differencesin relative abundance or centrality, to estimate their asso-ciation with future HAZ using multivariable between-within regression models. Acidaminococcus, of the phylumFirmicutes, was the only genus associated with HAZ inFig. 2 Graphical models of Malawi case and control microbiota networks constructed using glasso. (Top) Case networks. (Bottom) Control networks.(Left to right) Associations found in both groups, cases only and controls only. Solid and dotted edges indicate positive and negative associations. Blueindicates associations among aerobic and facultative anaerobic genera. Orange indicates associations among anaerobic genera. Gray indicatesassociations from aerobic/facultative anaerobic to anaerobic genera. Node size is proportional to median abundanceFig. 3 Graphical models of Bangladesh case and control microbiota networks constructed using glasso. (Top) Case networks. (Bottom) Controlnetworks. (Left to Right) Associations found in both groups, cases only and controls only. Solid and dotted edges indicate positive and negativeassociations. Blue indicates associations among aerobic and facultative anaerobic genera. Orange indicates associations among anaerobic genera.Gray indicates associations from aerobic/facultative anaerobic to anaerobic genera. Node size is proportional to median abundanceGough et al. Microbiome  (2015) 3:24 Page 4 of 10longitudinal analyses of both cohorts. In the Malawicohort, a 0.1 % difference in the relative abundance ofthis genus between co-twins was associated with a 0.08lower height-for-age z-score (90 % confidence interval(CI) −0.12, −0.04) at the subsequent study visit in theco-twin who had the greater Acidaminococcus abundancecompared to their sibling. In the Bangladesh cohort, a0.1 % difference in the relative abundance of this genusbetween co-twins was associated with a 0.19 lower HAZ(90 % CI −0.25, −0.13) at the subsequent visit in the co-twin with the greater Acidaminococcus abundance. Theseassociations remained significant after controlling for mul-tiple hypothesis testing (Table 1).The literature on Acidaminococcus sp., with which wecan infer its role in the human gut and its potential im-pact on linear growth in children, is sparse. Only twospecies in this genus have been described [36, 37]. Onenotable characteristic of these described species is theirability to consume glutamate as their sole source of car-bon and energy. In porcine models, dietary glutamate isan essential oxidative fuel for the intestinal epithelium[38, 39], which undergoes a continuous process of re-generation and has high energy demands. Estimates forthe amount of glutamate completely metabolized in thehas been observed using in vitro cell lines [40–42], aswell as in animal models of glutamate supplementation[43–46]. Glutamate is an important precursor and inter-mediate in the synthesis and metabolic recycling of otheramino acids, and with the urea cycle, in the gut[38, 39, 47, 48]. Amino acids closely interlinked withglutamate metabolism include arginine, which also con-tributes to epithelium restitution, preserves barrier func-tion, prevents accumulation of ammonia in the gut, andattenuates intestinal tissue damage [49–51], and gluta-thione, which protects the epithelium from damage byoxidative stress [52, 53]. Altogether, major functions of glu-tamate in the gut appear to be its role as a key intermediatein gut amino acid metabolism and nitrogen cycling, main-tenance of epithelial integrity, and preservation of barrierfunction. Biomarkers of intestinal injury and repair havebeen associated with lower HAZ in LMICs [54]. Impairedgut barrier function is characteristic of EED, which is alsoassociated with poor linear growth [8–11].This evidence led us to pose the a posteriori hypothesisthat glutamate fermentation by microbes is negatively asso-ciated with future HAZ. We tested this hypothesis usingKEGG enzyme abundance data provided for the Malawicohort. We fitted between-within regression models whereesty bereGough et al. Microbiome  (2015) 3:24 Page 5 of 10gut range from 64 [39] to 90 % [38]. As such, glutamateis important to gut epithelium restitution. The beneficialeffect of glutamate on restoration of gut barrier functionTable 1 Relative genus abundance associations with future HAZmodels for genera with a significant difference in degree centralitMalawiGenus AbundancedifferenceaCoefficient (90 % CI) p valueAcidaminococcus 0.40 −0.080 (−0.124, −0.037) <0.01AcinetobacterbAnaerococcusbBacteroides 4.51 0.000 (−0.001, 0.001) 0.67Blautia 2.51 −0.001 (−0.003, 0.002) 0.64Brachyspira 1.03 0.003 (−0.002, 0.007) 0.32Chlamydiaceae_uncl 0.37 −0.012 (−0.054, 0.030) 0.65Coprococcus 0.35 −0.006 (−0.061, 0.049) 0.87GeobacillusbHaemophilus 0.76 0.001 (−0.009, 0.010) 0.92LactococcusbMicrococcusbNeisseriaceae_uncl 0.22 −0.027 (−0.103,0.048) 0.56ProteusbSarcinabCoefficients are expressed as the average difference in future HAZ per 0.1 % diff90 % CI 90 % confidence interval, HAZ height-for-age z-scoreaMedian difference in relative abundance between siblings in a twin pairbModels could not be fit in the Malawi cohort because these genera were only identhe relative abundance of critical genes utilized in glutamatefermentation pathways by microbes [55] was the exposureof interest. We found that the abundance of genes encodingimated using multivariable between-within twin regressionetween cases and controlsBangladeshAdjustedp valueAbundancedifferenceaCoefficient (90 % CI) p value Adjustedp value0.02 0.30 −0.191 (−0.253, −0.129) <0.01 <0.010.00 −0.032 (−0.159, 0.094) 0.68 0.890.01 −0.182 (−0.915, 0.551) 0.68 0.890.89 0.29 −0.001 (−0.002, 0.001) 0.63 0.890.89 5.00 0.001 (0.000, 0.001) 0.07 0.450.890.890.92 4.33 −0.003 (−0.010, 0.003) 0.38 0.890.01 0.266 (−0.154, 0.685) 0.30 0.890.920.04 −0.002 (−0.007, 0.004) 0.59 0.890.46 −0.107 (−2.183, 0.169) 0.16 0.940.89 0.01 0.001 (−0.001, 0.004) 0.46 0.640.00 −0.002 (−0.037, 0.033) 0.94 0.945.00 0.000 (0.000, 0.001) 0.54 0.89nce in abundance between siblingstified in ≤2 samplesGough et al. Microbiome  (2015) 3:24 Page 6 of 10glutamate dehydrogenase and α-keto-glutarate reductasewas negatively associated with future HAZ. For glutamatedehydrogenase and α-keto-glutarate reductase, respectively,a one unit greater gene abundance in one co-twin com-pared to their sibling was associated with a −0.17 (90 %CI −0.29, −0.04, p = 0.03) and −0.08 (90 % CI −0.16, −0.01,p = 0.07) smaller HAZ in that co-twin at the subsequentstudy visit. These are the first two enzymes involved in thehydroxyglutarate fermentation pathway used by Acidami-nococcus fermentans for glutamate fermentation; somespecies in the Peptoniphilus, Fusobacterium, and Clos-tridia families can also utilize this pathway [55, 56].In the Bangladesh cohort, we also observed a −0.003(90 % CI −0.004, −0.002) lower HAZ and a 0.001 (90 %CI 0.000, 0.001) greater HAZ at the subsequent visit inco-twins who had a 0.1 % greater abundance of Weissellaor Blautia, respectively, compared to their siblings (Table 1and Additional file 5: Table S5). The association withBlautia was not statistically significant after controllingfor multiple hypothesis testing.DiscussionIn these analyses, we show that less diverse gut microbiotawith greater covariance network density are associated withstunting severity, and an increase in the relative abundanceof Acidaminococcus sp. is associated with lower futurelinear growth in two very different, well-characterizedcohorts of children living in low-income settings. Weapplied a novel approach, utilizing a statistical learningmethod combined with network analysis and a permutationtest to determine differences between microbiota com-munities of stunted and severely stunted children fromthese cohorts, and applied longitudinal epidemiologicalanalysis methods to investigate whether changes in thegenera identified were associated with future linear growth.In our longitudinal models, greater abundance of Acid-aminococcus was associated with a future deficit in HAZbetween co-twins in both cohorts. Acidaminococcus sp.can utilize glutamate as their sole source of carbon andenergy. Greater abundance of genes encoding the first twoenzymes in the hydroxyglutarate pathway for glutamatefermentation was also associated with a future HAZ def-icit. Overgrowth of bacteria that can ferment glutamatemay have a deleterious effect on linear child growth, po-tentially as a result of glutamate’s importance in aminoacid metabolism, nitrogen balance, and barrier function.This observation may also reflect the state of malnutritionin these cohorts of children, as the microbiota turns tohost-associated proteins for energy. The weak negative as-sociation between Weissella and future HAZ observed inthe Bangladesh cohort was not detected in the Malawichildren and needs to be confirmed in other studies.The impact of Acidaminococcus on growth may alsoinvolve its microbial relationships. Network analysisprovides a useful framework for identifying importantbacteria by their number of relationships [57–59]. Onestudy used correlation network centrality measures toidentify bacteria that successfully promote the growthconditions of a previously uncultivable microorganism[59]. In the Malawi cohort, Acidaminococcus showed alarge increase in degree centrality in cases, indicating a po-tential increase in its influence on microbiota composition.The possibility that rare commensals can promote patho-logical states based on their relationships with othermicrobes, despite their low abundance, has been proposed[60] and is in line with the notion of keystone organisms[60–62]. Although an increase in Acidaminococcus cen-trality was not observed in the Bangladesh cases, ran-dom sampling error introduced by selecting cases andcontrols from such a small population (n = 25), lackingtruly healthy control subjects of normal length, couldbias how representative the case and control exposurehistories were in that cohort. Larger epidemiologicaland experimental investigations are needed to confirmthese findings and the mechanisms involved.Finally, in both populations, we observed greater dens-ity in case networks that was only statistically significantin the Malawi cohort and a larger proportion of connec-tions from aerobes to anaerobes in cases. An increase inthe average number of connections with worsening nu-tritional status was also reported in children with SAMusing correlation networks [23], and greater connectivitybetween aerobic and anaerobic bacteria was reportedfor the microbiota correlation network of children withmoderate-to-severe diarrhea compared to non-diarrhealcontrols [63]. Simulation studies suggest that increaseddensity may provide greater resource flow to nodes thatare normally of low importance and may reduce the effi-ciency of resource flow out of the system [64, 65].In construction of our graphical models, we adjusted forpotential confounders that were reported (e.g., age andWHZ) but could not control for confounding whencomparing case and control network indices. These dif-ferences may, therefore, still be confounded by age orby other unreported factors such as infant diet, maternal,or environmental variables, since controls were older thancases in both populations, and microbiota compositionand structure may relate to the timing of complementaryfood introduction or environmental exposures. We cannotdismiss the possibility of spurious associations in ourgraphical models due to compositional effects [66], re-sidual confounding by diet or other factors, and smallsample size. The resulting “noise” limited our ability todetect differences between case and control networks,and we must exercise caution in interpreting pairwiseassociations as true ecological interactions.The between-within multivariable regression models,however, control for unreported confounders that arefour fecal samples were available from each co-sibling.bacterial taxon in each stool sample) were estimatedGough et al. Microbiome  (2015) 3:24 Page 7 of 10shared between co-twins (e.g., fetal, maternal, and envir-onmental), other factors that are identical between twinssuch as age at each visit and length of follow-up, as wellas reported confounders that differ between siblings(e.g., diarrhea and infant sex). Data on any antibioticuse and diet at each visit were only provided for theBangladesh cohort. Including antibiotic use and breast-feeding (without the use of formula or solid foods) inthe between-within models did not change the results.The association between Acidaminococcus and lineargrowth was reproduced in both populations, suggestingthat residual confounding due to other unreported factorsthat may differ between siblings, such as HIV status (thesedata were not available from either cohort), is unlikely.We also lagged these models so that changes in exposurepreceded changes in growth. The temporality adds cred-ibility to our main findings that an increase in Acidamino-coccus and glutamate-fermenting microbes are associatedwith future growth deficits. Measurement error in quanti-fication of relative abundance is unavoidable in microbiotastudies. Since any such error is unlikely to be systemat-ically related to future growth deficits between siblings,measurement error in these analyses would attenuatetrue associations with growth, further reducing our powerin these small cohorts. Finally, the original cohort studieswere not designed to investigate stunting. The averagechild in these populations already suffered from severegrowth restriction at study entry, and these data may notelucidate the potential negative effect of microbiota dysbio-sis or the protective effect of certain genera in children whoare of normal length but still at risk of becoming stunted.This may apply particularly to the case-control analyses, forwhich there were no healthy, non-stunted controls.ConclusionsOur study applied a novel use of statistical learning andnetwork methods to identify and interpret changes ingraphical models of microbiota covariance patterns. Theysuggest that reduced microbiota diversity and changes incovariance network density are associated with stuntingseverity and that overgrowth of Acidaminococcus, andpossibly other glutamate-fermenting microbes, may con-tribute to future growth deficits in already malnourishedchildren. Our findings demonstrate the potential role thatcertain types of commensals in the gut may have on lineargrowth deficits. Larger primary studies in other settings,designed specifically to evaluate stunting in infants, areneeded to confirm these findings, and experimental stud-ies are needed to clarify the mechanisms involved.MethodsStudy sampleDemographic, clinical, and anthropometric data from acohort of 22 twin pairs from Malawi, and a second cohortfrom shotgun reads using MetaPhlan [68]. Relative op-erational taxonomic unit (OTU) abundance data fromthe Bangladesh cohort were used as provided at http://gordonlab.wustl.edu/SuppData.html and were analyzedat the genus level. Extraction of genomic DNA fromfecal samples, DNA sequencing, processing and filtering ofreads, and, for Bangladesh data, OTU picking and taxonassignment have been described [25, 30]. The Simpson di-versity index was calculated as a measure of alpha diversityin all samples using vegan [69]. Simpson diversity measuresthe probability that two randomly selected microbes in asample will be from different taxa and provides a measureof the number of different types of bacteria present.Statistical analysesAnalyses were performed separately for the Malawi andBangladesh cohorts using two approaches. We first con-ducted an analysis of unmatched cases and controls se-lected from each cohort in order to identify changes inmicrobiota networks and relative genus abundance associ-ated with stunting status and to select genera for inclusionThe 12 sets of siblings from Bangladesh were selectedfrom among mothers with multiple pregnancies at a childhealth and family planning clinic in Dhaka and werefollowed up for longitudinal gut microbiota evaluation. Inboth twin cohorts, at each follow-up visit, length/heightand weight were measured, and fecal samples werecollected along with data on age in months and diarrheain the 7 days prior to or at the visit for Malawi andBangladesh, respectively. Anthropometric measures wereprovided as height-for-age and weight-for-height z-scores.In the Malawi cohort, if at least one co-twin developedSAM, as defined using WHO criteria [67], both weretreated with ready-to-use therapeutic food (RUTF).Whole genome sequencing and annotationWhole genome sequence datasets from the Malawi cohortwere made available through the European BioinformaticsInstitute at http://www.ebi.ac.uk/ena/data/view/ERP001911&display=html [27] and MG-RAST (http://metagenomics.anl.gov/) [30]. Relative genus abundances (expressed asa percentage of the total amount of DNA assigned to aof 11 twin pairs plus one set of triplets from Bangladesh,were made available at http://gordonlab.wustl.edu/Supp-Data.html. Details are provided in Smith et al. [27] andSubramanian et al. [25]. In brief, 22 twin pairs ages birthto 3 years were selected from among 317 available pairs infive rural communities in Malawi for longitudinal analysesof their gut microbiota. Twin pairs were selected if at leastin longitudinal analyses. Next, in longitudinal analyses, wefitted multivariable regression models, using data availableby permutation. Specifically, children were randomlyGough et al. Microbiome  (2015) 3:24 Page 8 of 10reallocated between the case and control groups 1000times. For each permutation, one network was estimatedper group and distributions of the difference in networkat all follow-up visits for the entire cohort of children, tocontrol for confounding and to introduce temporality.Case-control network analysesChildren in the Malawi and Bangladesh twin cohortshad median baseline HAZ of −2.96 (IQR −3.68, −2.18)and −3.75 (IQR −4.54, −2.68), respectively, indicatingthat the majority were severely stunted at study entry(Additional file 6: Figure S1). For the case-control analyses,linear growth status was therefore defined as severelystunted (HAZ ≤ −3, cases) or stunted (−3 < HAZ ≤ −2,controls). For cases, the visit where a child first reachedHAZ ≤ −3 was selected, excluding children already se-verely stunted at study entry. The subset of childrenwho were not siblings of cases and who had HAZ > −3but ≤ −2 at the end of follow-up, regardless of theirbaseline z-score, was selected as controls. Spurious in-ferences can arise in the analysis of correlated data[70]. Since methods to address correlations are notavailable for the network analyses methods we imple-mented, if both siblings in a pair met case or control cri-teria, one was randomly chosen to avoid within-groupcorrelations [29], and data from only one visit were usedper child. Differences in anthropometric, demographicand epidemiological measures, alpha diversity, and relativeabundance between cases and controls were evaluatedusing the chi-squared test or by permutation test on themedian, as appropriate.A supplemental approach to diversity indices for in-vestigating the microbiota uses networks of pairwise cor-relations between taxa as a model of microbe-microbeinteractions. In this representation, nodes are genera anda link between two nodes represents a non-zero associ-ation between two genera. This association is used as aproxy for bacterial interaction (see Additional file 7 forfurther information). An alternative to pairwise correla-tions is to estimate an inverse covariance matrix fromgenus abundances as a graphical model of important bac-terial relationships. We generated these graphical modelsseparately for cases and controls using the graphical lasso(glasso) [71]. The covariance associations estimated by theglasso (i.e., the links between genera in each network) areindependent of all other taxa and covariates included inthe model. For each case and control network, we calcu-lated graph density and the normalized degree centralityof each genus [72] using igraph [73]. Differences in net-work indices were assessed for statistical significanceindices between case and control networks were generatedfor statistical inference. Genera with significant differencesin degree centrality or relative abundance between casesand controls were selected for longitudinal analyses.Longitudinal analysesAfter performing microbiota feature selection in the case-control analyses, we fitted between-within regressionmodels [70, 74], using data for all follow-up visits from alltwin pairs in each cohort (regardless of their selection ascases or controls), to investigate whether the relativeabundance of selected genera was associated with lineargrowth. A between-within model allows estimation of theeffect that differences in exposure level (e.g., differences ingenus abundance) between siblings within a twin pair haveon their outcomes, while adjusting for unmeasured con-founders that siblings share, such as fetal, maternal, andenvironmental factors. This is done by including both (i)individual sibling exposure values and (ii) the mean expos-ure value of co-twins as covariates in a regression model.Adjustment for measured confounders not shared be-tween co-twins (e.g., diarrhea) can be made by includingsibling-specific covariates in the model [74].We fitted a separate model for each genus selected,with relative abundance as the exposure and HAZ as theoutcome. Each model was adjusted for reported diarrhea,WHZ, infant sex, and alpha diversity as reported con-founders not shared by co-twins. Age in months andlength of follow-up since baseline were also included aspredictors of the outcome. All covariates were laggedby one visit in order to model their effect on futureHAZ, with the exception of length of follow-up andage. All between-within models were fitted by generalizedestimating equations (GEE) using geepack [75], and mul-tiple hypothesis testing adjustments using the Benjamini-Hochberg method [76] were made. Statistical significancewas determined at α < 0.1 due to the small sample sizeof both cohorts. All analyses were conducted in R ver-sion 3.0.1.Additional filesAdditional file 1: Table S1. Study participant characteristics in eachcohort at the baseline visit and in cases v controls. Summary statisticsdescribing the baseline characteristics of participants enrolled in theMalawi and Bangladesh twin cohorts, and comparison of case versuscontrol characteristics.Additional file 2: Table S2. Genus relative abundance and genuspresence in 308 Malawi and 429 Bangladesh fecal samples collectedduring follow-up. Median, minimum and maximum relative abundanceof each genus identified in 308 Malawi and 429 Bangladesh fecal samplescollected during follow-up, and the number and proportion of samplesin each cohort in which each genus was identified.Additional file 3: Table S3. Relative abundance and normalized degreecentrality of genera identified in severely stunted cases and stuntedcontrols selected from the Malawi cohort. Median, minimum and maximumrelative abundance of each genus identified in 10 case or 8 control samplesselected from the Malawi cohort. P-values were obtained by permutationtest for a difference between case and control medians. Degree centralitieseach network. P-values were obtained by permutation test for a differencefrom either the Malawi or Bangladesh cohorts. Coefficients measure thez-scores, diarrhea, and alpha diversity using multivariable between-withintwin regression, since these factors may differ between co-twins.Gough et al. Microbiome  (2015) 3:24 Page 9 of 10Additional file 6: Figure S1. Histograms of height-for-age z-scoredistributions in Malawi and Bangladesh children at study entry. Figure: (top)Height-for-age z-score distribution in the 44 Malawi children at study entry;(bottom) Height-for-age z-score distribution in the 25 Bangladesh childrenat study entry. Red vertical lines indicate the World Health Organization cut-off for stunting.Additional file 7. Extended Methods. An extended description of thestatistical analysis methods.Competing interestsThe authors declare that they have no competing interests.Authors’ contributionsEKG, DAS, EEMM, AJP, RJS, JHH, and ARM participated in the conception anddesign. EKG, DAS, EEMM, and ARM participated in the analysis and interpretationof data. EKG drafted the manuscript. All authors read and revised the manuscriptcritically for important intellectual content and approved the final version to bepublished.Funding sourcesEthan Gough was supported by the Vanier Canada Graduate Scholarship.Author details1Department of Epidemiology, Biostatistics and Occupational Health, McGillUniversity, Montreal H3A 1A2, QC, Canada. 2Department of Mathematics andStatistics, McGill University, Montreal H3A 2K6, QC, Canada. 3Centre forPaediatrics, Blizard Institute, Queen Mary University of London, London E12AT, UK. 4Zvitambo Institute for Maternal Child Health Research, Harare,Zimbabwe. 5Program in International Nutrition, Division of NutritionalSciences, Cornell University, Ithaca, NY 14853, USA. 6Department ofInternational Health, Johns Hopkins Bloomberg School of Public Health,Baltimore, MD 21205, USA. 7Faculty of Medicine, School of Population andPublic Health, University of British Columbia, 137-2206 East Mall, Vancouveraverage difference in future HAZ between siblings within a pair of twins thatis associated with each 0.1 % difference in relative abundance betweensiblings. Coefficients are also adjusted for infant sex, weight-for-heightbetween case and control degree centralities.Additional file 5: Table S5. Relative genus abundance associationswith future HAZ estimated using multivariable between-within twinregression models for genera with a significant difference in relativeabundance between cases and controls. Associations between relativeabundance and future HAZ for each genus with a statistically significantdifference in median abundance between cases and controls selectedof each genus identified in case or control samples were calculated fromcovariance network models of gut microbial interactions estimated usingthe graphical lasso, and were normalized for the number of genera in eachnetwork. P-values were obtained by permutation test for a differencebetween case and control degree centralities.Additional file 4: Table S4. Relative abundance and normalized degreecentrality of genera identified in severely stunted cases and stunted controlsselected from the Bangladesh cohort. Median, minimum and maximumrelative abundance of each genus identified in 6 case or 5 control samplesselected from the Bangladesh cohort. P-values were obtained by permutationtest for a difference between case and control medians. Degree centralities ofeach genus identified in case or control samples were calculated fromcovariance network models of gut microbial interactions estimatedusing the graphical lasso and were normalized for the number of genera inV6T 1Z3, BC, Canada.Received: 6 March 2015 Accepted: 4 June 2015References1. Black RE, Victora CG, Walker SP, Bhutta ZA, Christian P, de Onis M, et al.Maternal and child undernutrition and overweight in low-income andmiddle-income countries. Lancet. 2013;382(9890):427–51.2. Victora CG, Adair L, Fall C, Hallal PC, Martorell R, Richter L, et al. Maternaland child undernutrition: consequences for adult health and human capital.Lancet. 2008;371(9609):340–57.3. Adair LS, Fall CH, Osmond C, Stein AD, Martorell R, Ramirez-Zea M, et al.Associations of linear growth and relative weight gain during early life withadult health and human capital in countries of low and middle income:findings from five birth cohort studies. Lancet. 2013;382(9891):525–34.4. Checkley W, Buckley G, Gilman RH, Assis AM, Guerrant RL, Morris SS, et al.Multi-country analysis of the effects of diarrhoea on childhood stunting. IntJ Epidemiol. 2008;37(4):816–30.5. Dewey KG, Mayers DR. Early child growth: how do nutrition and infectioninteract?: How do nutrition and infection interact? Matern Child Nutr. 2011;7:129–42.6. Guerrant RL, Oriá RB, Moore SR, Oriá MO, Lima AA. Malnutrition as anenteric infectious disease with long-term effects on child development. NutrRev. 2008;66(9):487–505.7. Rodriguez L, Cervantes E, Ortiz R. Malnutrition and gastrointestinal andrespiratory infections in children: a public health problem. Int J Environ ResPublic Health. 2011;8(4):1174–205.8. Lin A, Arnold BF, Afreen S, Goto R, Huda TMN, Haque R, et al. Householdenvironmental conditions are associated with enteropathy and impairedgrowth in rural Bangladesh. Am J Trop Med Hyg. 2013;89(1):130–7.9. Kosek M, Haque R, Lima A, Babji S, Shrestha S, Qureshi S, et al. Fecal markersof intestinal inflammation and permeability associated with the subsequentacquisition of linear growth deficits in infants. Am J Trop Med Hyg.2013;88(2):390–6.10. Campbell DI, Elia M, Lunn PG. Growth faltering in rural Gambian infants isassociated with impaired small intestinal barrier function, leading toendotoxemia and systemic inflammation. J Nutr. 2003;133(5):1332–8.11. Lunn PG. The impact of infection and nutrition on gut function and growthin childhood. Proc Nutr Soc. 2000;59:147–54.12. Humphrey JH. Child undernutrition, tropical enteropathy, toilets, andhandwashing. Lancet. 2009;374(9694):1032–5.13. Keusch GT, Rosenberg IH, Denno DM, Duggan C, Guerrant RL, Lavery JV,et al. Implications of acquired environmental enteric dysfunction for growthand stunting in infants and children living in low-and middle-income countries.Food Nutr Bull. 2013;34(3):357–65.14. Korpe PS, Petri WA. Environmental enteropathy: critical implications of apoorly understood condition. Trends Mol Med. 2012;18(6):328–36.15. Prendergast A, Kelly P. Enteropathies in the developing world: neglectedeffects on global health. Am J Trop Med Hyg. 2012;86(5):756–63.16. Prendergast AJ, Rukobo S, Chasekwa B, Mutasa K, Ntozini R, Mbuya MNN,et al. Stunting is characterized by chronic inflammation in Zimbabweaninfants. John-Stewart GC, editor. PLoS One. 2014;9(2):e86928.17. Bhutta ZA, Das JK, Rizvi A, Gaffey MF, Walker N, Horton S, et al. Evidence-based interventions for improvement of maternal and child nutrition: whatcan be done and at what cost? Lancet. 2013;382(9890):452–77.18. Palmer C, Bik EM, DiGiulio DB, Relman DA, Brown PO. Development of thehuman infant intestinal microbiota. PLoS Biol. 2007;5(7):e177.19. Maynard CL, Elson CO, Hatton RD, Weaver CT. Reciprocal interactions of theintestinal microbiota and immune system. Nature. 2012;489(7415):231–41.20. Hooper LV. Do symbiotic bacteria subvert host immunity? Nat RevMicrobiol. 2009;7(5):367–74.21. Chung H, Kasper DL. Microbiota-stimulated immune mechanisms to maintaingut homeostasis. Curr Opin Immunol. 2010;22(4):455–60.22. Tremaroli V, Bäckhed F. Functional interactions between the gut microbiotaand host metabolism. Nature. 2012;489(7415):242–9.23. Ghosh TS, Sen Gupta S, Bhattacharya T, Yadav D, Barik A, Chowdhury A,et al. Gut microbiomes of Indian children of varying nutritional status. ZwickME, editor. PLoS One. 2014;9(4):e95547.24. Monira S, Nakamura S, Gotoh K, Izutsu K, Watanabe H, Alam NH, et al. Gutmicrobiota of healthy and malnourished children in Bangladesh. FrontMicrobiol. 2011;2:1–7.25. Subramanian S, Huq S, Yatsunenko T, Haque R, Mahfuz M, Alam MA, et al.Persistent gut microbiota immaturity in malnourished Bangladeshi children.Nature. 2014;510:417–21.26. Gupta SS, Mohammed MH, Ghosh TS, Kanungo S, Nair GB, Mande SS.Metagenome of the gut of a malnourished child. Gut Pathog. 2011;3(1):7.53. Roediger WE, BABIDGE W. Human colonocyte detoxification. Gut. 1997;41(6):731–4.54. Peterson KM, Buss J, Easley R, Yang Z, Korpe PS, Niu F, et al. REG1B as aGough et al. Microbiome  (2015) 3:24 Page 10 of 1027. Smith MI, Yatsunenko T, Manary MJ, Trehan I, Mkakosya R, Cheng J, et al.Gut microbiomes of Malawian twin pairs discordant for kwashiorkor.Science. 2013;339(6119):548–54.28. Ridaura VK, Faith JJ, Rey FE, Cheng J, Duncan AE, Kau AL, et al. Gut microbiotafrom twins discordant for obesity modulate metabolism in mice. Science.2013;341(6150):1241214.29. Goodrich JK, Waters JL, Poole AC, Sutter JL, Koren O, Blekhman R, et al.Human genetics shape the gut microbiome. Cell. 2014;159(4):789–99.30. Yatsunenko T, Rey FE, Manary MJ, Trehan I, Dominguez-Bello MG, ContrerasM, et al. Human gut microbiome viewed across age and geography. Nature.2012;486:222–8.31. Gueimonde M, Laitinen K, Salminen S, Isolauri E. Breast milk: a source ofbifidobacteria for infant gut development and maturation? Neonatology.2007;92(1):64–6.32. Martin R, Jimenez E, Heilig H, Fernandez L, Marin ML, Zoetendal EG, et al.Isolation of bifidobacteria from breast milk and assessment of thebifidobacterial population by PCR-denaturing gradient gel electrophoresisand quantitative real-time PCR. Appl Environ Microbiol. 2009;75(4):965–9.33. Marcobal A, Sonnenburg JL. Human milk oligosaccharide consumption byintestinal microbiota: human milk oligosaccharide consumption. ClinMicrobiol Infect. 2012;18:12–5.34. De Filippo C, Cavalieri D, Di Paola M, Ramazzotti M, Poullet JB, Massart S,et al. Impact of diet in shaping gut microbiota revealed by a comparativestudy in children from Europe and rural Africa. Proc Natl Acad Sci.2010;107(33):14691–6.35. Wexler HM. Bacteroides: the good, the bad, and the nitty-gritty. Clin MicrobiolRev. 2007;20(4):593–621.36. Rogosa M. Acidaminococcus gen. n., Acidaminococcus fermentans sp. n.,anaerobic gram-negative diplococci using amino acids as the sole energysource for growth. J Bacteriol. 1969;98(2):756–66.37. Jumas-Bilak E, Carlier J-P, Jean-Pierre H, Mory F, Teyssier C, Gay B, et al.Acidaminococcus intestini sp. nov., isolated from human clinical samples.Int J Syst Evol Microbiol. 2007;57(10):2314–9.38. Reeds PJ, Burrin DG, Stoll B, Jahoor F. Intestinal glutamate metabolism.J Nutr. 2000;130(4):978S–82.39. Burrin DG, Stoll B. Metabolic fate and function of dietary glutamate in thegut. Am J Clin Nutr. 2009;90(3):850S–6.40. Beutheu S, Ghouzali I, Galas L, Déchelotte P, Coëffier M. Glutamine andarginine improve permeability and tight junction protein expression inmethotrexate-treated Caco-2 cells. Clin Nutr. 2013;32(5):863–9.41. Le Bacquer O, Laboisse C, Darmaun D. Glutamine preserves protein synthesisand paracellular permeability in Caco-2 cells submitted to “luminal fasting”. AmJ Physiol Gastrointest Liver Physiol. 2003;285:G128–36.42. Vermeulen M, De Jong J, Vaessen M, van Leeuwen P, Houdijk APJ. Glutamatereduces experimental intestinal hyperpermeability and facilitates glutaminesupport of gut integrity. World J Gastroenterol. 2011;17(12):11569–73.43. Dai Z-L, Li X-L, Xi P-B, Zhang J, Wu G, Zhu W-Y. l-Glutamine regulates aminoacid utilization by intestinal bacteria. Amino Acids. 2013;45(3):501–12.44. Rezaei R, Knabe DA, Tekwe CD, Dahanayaka S, Ficken MD, Fielder SE, et al.Dietary supplementation with monosodium glutamate is safe and improvesgrowth performance in postweaning pigs. Amino Acids. 2013;44(3):911–23.45. Cabrera RA, Usry JL, Arrellano C, Nogueira ET, Kutschenko M, Moeser AJ,et al. Effects of creep feeding and supplemental glutamine or glutamineplus glutamate (Aminogut) on pre- and post-weaning growth performanceand intestinal health of piglets. J Anim Sci Biotechnol. 2013;4(1):29.46. Wu M, Xiao H, Ren W, Yin J, Tan B, Liu G, et al. Therapeutic effects ofglutamic acid in piglets challenged with deoxynivalenol. te Pas MF, editor.PLoS One. 2014;9(7):e100591.47. Newsholme P, Procopio J, Lima MMR, Pithon-Curi TC, Curi R. Glutamine andglutamate—their central role in cell metabolism and function. Cell BiochemFunct. 2003;21(1):1–9.48. Blachier F, Boutry C, Bos C, Tome D. Metabolism and functions of L-glutamatein the epithelial cells of the small and large intestines. Am J Clin Nutr.2009;90(3):814S–21.49. Fritz JH. Arginine cools the inflamed gut. Infect Immun. 2013;81(10):3500–2.50. Morris SM. Arginine metabolism: boundaries of our knowledge. J Nutr.2007;137(6):1602S–9.51. Wu G, Bazer FW, Davis TA, Kim SW, Li P, Marc Rhoads J, et al. Arginine metabolismand nutrition in growth, health and disease. Amino Acids. 2009;37(1):153–68.52. Jefferies H, Bot J, Coster J, Khalil A, Hall JC, McCauley RD. The role of glutathionein intestinal dysfunction. J Invest Surg. 2003;16:315–23.predictor of childhood stunting in Bangladesh and Peru. Am J Clin Nutr.2013;97(5):1129–33.55. Buckel W, Barker HA. Two pathways of glutamate fermentation byanaerobic bacteria. J Bacteriol. 1974;117(3):1248–60.56. Ezaki T, Kawamura Y, Li N, Li Z-Y, Zhao L, Shu S. Proposal of the generaAnaerococcus gen. nov., Peptoniphilus gen. nov. and Gallicola gen. nov. formembers of the genus Peptostreptococcus. Int J Syst Evol Microbiol.2001;51(4):1521–8.57. Fisher CK, Mehta P. Identifying keystone species in the human gutmicrobiome from metagenomic timeseries using sparse linear regression.White BA, editor. PLoS One. 2014;9(7):e102451.58. Berry D, Widder S. Deciphering microbial interactions and detectingkeystone species with co-occurrence networks. Front Microbiol. 2014;5:1–14.59. Duran-Pinedo AE, Paster B, Teles R, Frias-Lopez J. Correlation network analysisapplied to complex biofilm communities. Gilbert JA, editor. PLoS One.2011;6(12):e28438.60. Hajishengallis G, Darveau RP, Curtis MA. The keystone-pathogen hypothesis.Nat Rev Microbiol. 2012;10(10):717–25.61. Mills LS, Doak DF. The keystone-species concept in ecology and conservation.Bioscience. 1993;43(4):219–24.62. Cottee-Jones HEW, Whittaker RJ. Perspective: the keystone species concept:a critical appraisal. Front Biogeogr. 2012;4(3):117–27.63. Pop M, Walker AW, Paulson J, Lindsay B, Antonio M, Hossain MA, et al.Diarrhea in young children from low-income countries leads to large-scalealterations in intestinal microbiota composition. Genome Biol. 2014;15(6):R76.64. Habiba, Berger-Wolf T. Working for influence: effect of network density andmodularity on diffusion in networks. 2014 Sep 29 [cited 2014 Sep 29];Available from: http://compbio.cs.uic.edu/~habiba/Spread_structure_DamNet2011_final_1013.pdf.65. Yamaguchi K. The flow of information through social networks: diagonal-freemeasures of inefficiency and the structural determinants of inefficiency. SocNetworks. 1994;16(1):57–86.66. Friedman J, Alm EJ. Inferring correlation networks from genomic surveydata. PLoS Comput Biol. 2012;8(9):e1002687.67. World Health Organization, United Nations Children’s Fund. WHO childgrowth standards and the identification of severe acute malnutrition ininfants and children: a joint statement by the World Health Organizationand the United Nations Children’s Fund [Internet]. World Health Organization(WHO); 2009. Available from: http://apps.who.int/iris/handle/10665/44129.68. Segata N, Waldron L, Ballarini A, Narasimhan V, Jousson O, Huttenhower C.Metagenomic microbial community profiling using unique clade-specificmarker genes. Nat Methods. 2012;9(8):811–4.69. Dixon P. VEGAN, a package of R functions for community ecology. J VegSci. 2003;14(6):927–30.70. Carlin JB. Regression models for twin studies: a critical review. Int JEpidemiol. 2005;34(5):1089–99.71. Friedman J, Hastie T, Tibshirani R. Sparse inverse covariance estimation withthe graphical lasso. Biostatistics. 2008;9(3):432–41.72. Newman M. Networks: an introduction. New York, USA: Oxford UniversityPress; 2010.73. Csardi G, Nepusz T. The igraph software package for complex networkresearch. InterJournal Complex Syst [Internet]. 2006 [cited 2014 Sep29];1695(5). Available from: http://www.necsi.edu/events/iccs6/papers/c1602a3c126ba822d0bc4293371c.pdf.74. Sjölander A, Frisell T, Öberg S. Causal interpretation of between-withinmodels for twin research. Epidemiol Methods [Internet]. 2012 Jan 29 [cited2014 Aug 4];1(1). Available from: http://www.degruyter.com/view/j/em.2012.1.issue-1/2161-962X.1015/2161-962X.1015.xml.75. Halekoh U, Højsgaard S, Yan J. The R package geepack for generalizedestimating equations. J Stat Softw. 2006;15(2):1–11.76. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practicaland powerful approach to multiple testing. J R Stat Soc Ser B Methodol.1995;57(1):289–300.

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.52383.1-0223880/manifest

Comment

Related Items