Ordination and significance testing of microbial community compositionderived from terminal restriction fragment length polymorphisms:application of multivariate statisticsGavin N. Rees1,*, Darren S. Baldwin1, Garth O. Watson1, Shane Perryman1,2and DarylL. Nielsen11Murray Darling Freshwater Research Centre and Cooperative Research Centre for Freshwater Ecology, POBox 921, Albury, NSW 2640 Australia;2Water Studies Centre, School of Chemistry, Monash University,Clayton Victoria 3800 Australia;*Author for correspondence (e-mail: [email protected]; phone: 61 2 60582356; fax: 61 2 6043 1626)Received 13 Februari 2004; Accepted in revised form 2 June 2004Key words: Microbial community analysis, Multivariate statistics, T-RFLPAbstractTerminal restriction fragment length polymorphism共T-RFLP兲is increasingly being used to examine microbialcommunity structure and accordingly, a range of approaches have been used to analyze data sets. A number ofpublished reports have included data and results that were statistically flawed or lacked rigorous statistical test-ing. A range of simple, yet powerful techniques are available to examine community data, however their use isseldom, if ever, discussed in microbial literature. We describe an approach that overcomes some of the problemsassociated with analyzing community datasets and offer an approach that makes data interpretation simple andeffective. The Bray-Curtis coefficient is suggested as an ideal coefficient to be used for the construction of simi-larity matrices. Its strengths include its ability to deal with data sets containing multiple blocks of zeros in ameaningful manner. Non-metric multi-dimensional scaling is described as a powerful, yet easily interpretedmethod to examine community patterns based on T-RFLP data. Importantly, we describe the use of significancetesting of data sets to allow quantitative assessment of similarity, removing subjectivity in comparing complexdata sets. Finally, we introduce a quantitative measure of sample dispersion and suggest its usefulness in describ-ing site heterogeneity.Abbreviations: T-RFLP – Terminal restriction fragment length polymorphism; MDS – non-metric multidimen-sional scaling; ANOSIM – analysis of similarity; SIMPER – similarity percentageIntroductionMicrobial ecologists have long strived to unravel thehighly diverse microbial communities that exist innature. Classical culture-based techniques to studymicrobiological communities have allowed microbi-ologists to identify many culturable organisms ingiven samples, but current application of culturetechniques are recognized as having limited scope forstudying microbial diversity in most environments.The application of DNA-based molecular tools hasgreatly enhanced understanding of microbial diversityas DNA techniques circumvent culturing problems bydetermining the sequence diversity in genes presentwithin given samples. Terminal restriction fragmentlength polymorphism共T-RFLP兲is a DNA-based mo-lecular technique共Avaniss-Aghajani et al. 1994; Liuet al. 1997兲that has been applied to activated sludgeAntonie van Leeuwenhoek 86: 339–347, 2004.© 2004 Kluwer Academic Publishers. Printed in the Netherlands.339共Hiraishi et al. 2000兲, anaerobic sediments共Lüde-mann et al. 2000兲, contaminated sediments共Flynn etal. 2000兲, soils共Lucklow et al. 2000兲and componentsof different animal guts共Gong et al. 2002; Sait et al.2003兲. In this method, genomic DNA extracted froma sample is used as a template for the polymerasechain reaction共PCR兲, in which at least one of theprimers is labeled with a fluorescent dye. Amplifiedproducts are digested with one or more restriction en-zymes and the size of the terminal fragment, with thefluorescent label, is determined. Traces, termed elec-tropherograms, typically exhibit a range of terminalfragments peak sizes of variable fluorescence inten-sity.It is very important to chose an appropriate com-munity analysis method for use with T-RFLP. Themost simplistic approach has been to compare elec-tropherograms of different samples and visually com-pare traces for the presence or absence of differentpeaks. Such an approach is valid, but lacks the ben-efits of a quantitative analysis. Many approaches havebeen used to analyze T-RFLP data sets quantitatively.Some of these include principal components analysis共PCA兲共Clement et al. 1998兲, cluster analysis共Hiraishi et al. 2000; Urakawa et al. 2000; Sessitschet al. 2002; Sait et al. 2003兲and self organizing neu-ral networks共Dollhopf et al. 2001兲. Unfortunately,approaches used for T-RFLP data analysis have notalways made use of techniques that are ideally suitedto this type of data共see below兲. It is apparent fromour reading of the literature, the T-RFLP analysis re-quires a more standardized approach; in particular,one that has statistical rigor and that is easy to carryout.Animal and plant ecologists examining communitypatterns have developed and applied a range of toolsto analyze community structure共Lambshead 1986;Shiel 1990; Clarke 1993; – and references there in;Nielsen et al. 1997; Quinn and Keogh 2002兲. In par-ticular, multivariate statistical methods have been ex-tremely valuable in analyzing complex data sets.Multivariate techniques also have the advantage thatdatasets can comprise a variety of variables and neednot be limited to species lists. In this paper we de-scribe the rationale and application of several of mul-tivariate statistical procedures to analyze T-RFLP datasets. To this end, we selected three sites along a 1kmreach of an ephemeral stream and examined the bac-terial communities in the sediment. The purpose inselecting this stream was that it provided us with amodel system where we could sample three contrast-ing sediment types that occurred in close proximity.In our stream, differences in types of sediment, waterdepth and degree of impact from saline groundwaterintrusions meant each of the three study sites differedin their physical and chemical characteristics. We de-scribe the use of non-metric multidimensional scaling共MDS兲as a simple method for preparing visual inter-pretations of sediment community data. Importantly,we also describe the use of the Analysis of Similarity共ANOSIM兲procedure, which allows significancetesting of our data groups. We used two exploratorytools to examine our data set. In the first instance, weused similarity percentage analysis共SIMPER兲to ex-plain aspects of the