Access to this full-text is provided by Wiley.
Content available from MicrobiologyOpen
This content is subject to copyright. Terms and conditions apply.
MicrobiologyOpen. 2019;8:e780.
|
1 of 11
https://doi.org/10.1002/mbo3.780
www.MicrobiologyOpen.com
1 | INTRODUCTION
Plant pathogens are a critical th reat to global food security (Bebber &
Gurr, 2015), the conservation of natural ecosystems, and the future
resilience and sustainability of ecosystem ser vices (Bever, Mangan,
& Alexander, 2015). Because of their importance, there is a huge in‐
terest to biomonitor plant pathogens cost‐effectively at large scales
without the need of culturing and before possible disease outbreaks.
Received:3October2018
|
Revised:14November2018
|
Accepted:15Novembe r2018
DOI: 10.1002/mbo3.780
ORIGINAL ARTICLE
Biases in the metabarcoding of plant pathogens using rust fungi
as a model system
Andreas Makiola1,2 | Ian A. Dickie3 | Robert J. Holdaway4 |
Jamie R. Wood4 | Kate H. Orwin4 | Charles K. Lee5 | Travis R. Glare2
This is an op en access article under t he terms of the Creat ive Commons Attributio n License, which permits use, dist ribution and reproduc tion in any medium,
provide d the orig inal work is proper ly cited .
© 2018 The Aut hors. MicrobiologyOpen published by John Wiley & Sons Ltd.
1Agroécologie, AgroSup Dijon,
INRA ,Unive rsitéBourgogne ,Université
Bourgogne Franche‐Comté, Dijon, France
2Bio‐Protection Research Centre, Lincoln
University,Linco ln,NewZealand
3Bio‐Protection Research Centre, School
ofBiologicalSciences,Universi tyof
Canter bury,NewZe aland
4Manaaki Whenua – Landcare Research,
Lincoln ,NewZealand
5WaikatoDNASequen cingFacility,Schoo l
ofScience,Univer sityofWaikato,Hami lton,
NewZealand
Correspondence
Andreas Makiola, Agroécologie, AgroSup
Dijon,INRA ,Univer sitéBourgogne,
UniversitéBour gogneFran che-Comté,
Dijon, France.
Email: Andreas.Makiola@inra.fr and
Ian A. Dickie, Bio‐Protection Research
Centre, School of B iological Sciences,
UniversityofCanterbu ry,NewZealand.
Email: ian.dickie@canterbury.ac.nz
Funding information
Ministry for Business Innovation and
Employment,Grant/AwardNumber:
C09X1411;TertiaryEducationCommission
Abstract
Plant pathogens such as rust fungi (Pucciniales) are of global economic and ecological
importance. This means there is a critical need to reliably and cost‐effectively detect,
identify, and monitor these fungi at large scales. We investigated and analyzed the
causes of dif ferences between next-generation sequencing (NGS) metabarcoding
approachesandtraditionalDNAcloninginthedetectionandquantificationofrecog‐
nized species of rust fungi from environmental samples. We found significant differ‐
ences between observed and expected numbers of shared rust fungal operational
taxonomicunits(OTUs)amongdifferentmethods.However,therewasnosignificant
differenceinrelativeabundanceofOTUsthatallmethodswerecapableofdetecting.
Differences among the methods were mainly driven by the method's ability to detect
specificOTUs,likelycausedbymismatcheswiththeNGSmetabarcodingprimersto
some Puccinia species. Furthermore, detection ability did not seem to be influenced
bydifferences insequence lengthsamongmethods,themostappropriatebioinfor‐
matic pipeline used for each method, or the ability to detect rare species. Our find‐
ings are important to future metabarcoding studies, because they highlight the main
sources of difference among methods, and rule out several mechanisms that could
drive these differences. Furthermore, strong congruity among three fundamentally
differentandindependentmethods demonstratesthe promising potentialofNGS
metabarcodingfortrackingimportanttaxasuchasrustfungifromwithinlargerNGS
metabarcodingcommunities.OurresultssupporttheuseofNGSmetabarcodingfor
thelarge-scaledetectionandquantificationofrustfungi,butnotforconfirmingthe
absence of species.
KEYWORDS
cloning,Illumina,IonTorrent,next-generationsequencing,plantpathogens,Pucciniales
2 of 11
|
MAKIOLA et A L.
Rust fungi (Pucciniales) constitute one of the largest groups
of plant pat hogens, with abo ut 7,800 d escribed spe cies (Helfer,
2014), and some rus t species c an have large econ omic and eco‐
logical impacts. For example, myrtle rust (Austropuccinia psidii) is
currently decimating a wide range of Myrtaceae around the world
(Fernandez Winzer, Carnegie, Pegg, & Leishman, 2018; Glen,
Alfenas,Zauza, Wingfield, & Mohammed, 20 07), suchasthe en‐
demic Eugenia koolauensisinHawai‘i and theendemicRhodamnia
rubescens in native forests in Australia (Carnegie et al., 2016).
Coffee leaf rust (Hemileia vastatrix) is substantially damaging
Coffee plantations worldwide (Talhinhas et al., 2017). Similarly,
wheat leaf rusts like Puccinia triticina, Puccinia recondite, and
Puccinia striiformis are causing serious production losses for one of
theworld'sbigges tfoo dcro ps(McC allum,Hiebe rt,Hu er t a- Es pi no ,
& Cloutier, 2012).
While many studies focus on rust fungi as perceived pests, they
actually constitute a vital component of natural ecosystem function‐
ing. In contrast to agroecosystems, rusts in their natural ecosystems
are less well studied, and some species are threatened by extinction
due to globalchange (Helfer,2014).Becauseofthe economic and
ecological impor tance of plant pathogens, such as rust fungi, new,
reliable, and cost‐effective tools are urgently needed to monitor
them at large scales.
Next-generation sequencing metabarcoding has the potential
to develop into an effective method for the molecular identification
of multiple plant pathogens from environmental samples (Merges,
Bálint, Schmitt, Böhning-Gaese, & Neuschulz, 2018; Taberlet,
Coissac,Hajibabaei,&Rieseberg,2012).DNAmetabarcodingseems
especially promising for the monitoring of potential plant pathogens
(hereafter pathogens), because it bypasses the need for cultivation
and isolation of species, and is able to detect plant pathogenic spe‐
cies when they occur asymptomatically (Malcolm, Kuldau, Gugino,
&Jiménez-Gasco,2013;Stergiopoulos&Gordon,2014)oratbarely
discerniblelevels.While DNA metabarcodingholdsgreatpotential
for detecting and monitoring fungi in their environment (Durand
etal.,2017;Miller, Hopkins,Inward, &Vogler,2016; Schmidtetal.,
2013), it has not yet been widely applied to pathogens specific ally
(Abde lfatta h, Nicosia, C acciola, D roby,& S chena, 2015; Me rges et
al., 2018). It is therefore crucial to more fully understand the poten‐
tial limitations of this new approach.
Two limitations that frequently arise in NGS metabarcoding
studiesare the abilitytoquantify the abundances ofdifferenttaxa
(Deiner et al., 2017; Elbrecht & Leese, 2015), and the introduction
of false positives/negatives by PCR amplification, library prepara‐
tion, an d sequencin g (Coissac, R iaz, & Puillan dre, 2012). Here, we
address thesetwopossiblelimitationsofNGSmetabarcodingusing
the group of rust fungi as a model system. We investigate possi‐
ble differencesbetweenNGS met abarcoding and more traditional
cloning approaches in the detection and abundance of rust fungal
species. We also investigate what causes these differences. We use
two primer pairs because our objective in this study was to compare
methods using the best available and most appropriate approaches
foreachmethod.FortheNGSmetabarcodingapproach,weusetwo
fundamentally different sequencing technologies (Illumina MiSeq
andIonTorrentPGM)an dfung alN GSmet abarcodi ngp rimer stode‐
tect rust fungi from within a larger fungal community. We compare
these results to a cloning approach, targeting the same gene region
but focusing cloning on rust fungi using a rust fungi‐specific primer
pair.
We hypothesize that the three methods (Illumina, Ion Torrent,
and cloning):
1. differ in their detection of rust species (i.e., observed from
expected number of detected rust species)
2. differintheirabilitytoquantif yrelativeabundancesofrustfungal
species.
If one or both of the hypotheses are supported, we would then test
hypotheses for the mechanisms driving differences among methods.
Specifically, we hypothesize that differences among methods are due
to:
1. dif ferences in sequence lengths among methods
2. differences in the most appropriate bioinformatic pipelines for
each method
3. taxonomic biases of the methods
4. differentabilitiesofmethodstodetectrarespecies.
2 | METHODS
2.1 | Study site and sampling
We sampled thirty 20 × 20 m grassland plots. All plots were
based on an 8 × 8 km grid that is used extensively for national
biodiversity monitoring in New Zealand (Allen, Bellingham, &
Wiser, 2003) and positioned following the standard protocol of
Hu rs tan dA ll en(20 07 ).Thepl ot swerese le c te db as edont he ou t‐
put of the geographic information system and stratified random
sampling (Figure S1). We limited our sampling to grassland plots
located at altitudes <1,000 m. All sampling was carried out under
dry weather conditions between November 2014 and March
2015.
At each plot, samples were collected using a sterilized leaf
puncher within 64min (4min for eachof six teen 5×5m subplots)
to ensure balanced sampling of the whole plot . Every identifiable
plant par t (e.g., healthy leaves, leaves with lesions, bryophytes,
grass stems, lichens, bark, seeds), including healthy as well as dis‐
eased plant material, was sampled to get all variants and to maximize
rust fungal diversity. Since most of these samples represent above‐
ground herbaceous material (mainly leaves), we hereafter refer to
these samples simply as “leaf samples.” The leaf samples were imme‐
diately pooled by plot, stored in a 50‐ml Falcon tube containing ster‐
ilized DMSO -NaCl solution (20% DMSO, 0. 25M disodium-EDTA,
andNaCltosaturation,pH7.5),sealedwithParafilmM,andkeptat
4°Cuntillaborator yprocessing.
|
3 of 11
MAKIO LA et AL .
2.2 | DNA extraction
TheDNA extractionfromthepooled leaf samples ofeach plot was
car r i e dou t u sing t h eMa c h e rey - N age l N ucl e o S pin9 6Pl a n tII k i t( r o bot
extraction) following the manufac turer's protocol. We used both
provided lysis buffers separately (cetrimonium bromide [CTAB] lysis
buffer PL1 and a sodium dodecyl sulfate [SDS]‐based lysis buffer PL 2)
toenhanceth eamountofextractedD NA .F ivemicro liter sofproduct
wasquantifiedusingaQubit2.0fluorometer(LifeTechnologies)and
the broad‐range assay kit following the manufacturer's protocol be‐
foreequallypoolingtheextractsfromthesameplot.
2.3 | Preparation of next‐generation
sequencing libraries
WepreparedNGSlibrariesinaone-stepPCR(ImmolaseMoTASPpro‐
tocol) to avoid the risk of contamination, following Clarke, Czechowski,
Soubrier, Stevens, and Cooper (2014). We used the fungal primers
fITS7: GTGARTCATCGAATCTTTG (Ihrmark et al., 2012) and ITS4:
TCCTCCGCT TATTGATATGC (White , Bruns, Lee, & Taylor, 1990), ampli‐
fying the highly variable internal transcribed spacer region 2 (ITS2) with
uni versallinkers equence satthe5'endf orfIT S7:TCGTCGG CAGCG TC
andforITS4:GTCTCGTGGGCTCGG.Illuminaadaptersequenceswith
indexsequencesandcomplementarylinkersequenceswereasfollows:
F: AATGATACGGCGACCACCGAGATCTACAG‐8nt index‐TCGT
CGGCAGCGTC,.
R: CA AGCAGAAGACGGCATACGAGAT‐8nt index‐GTCTCGTG
GGCTCGG . Ion Torrent adapter seq uences with inde x sequences
andbarcodeadaptersequenceswereasfollows:
F: CCATCTC ATCCCTGCGTGTC TCCGACTCAG‐10nt ind ex‐GAT,.
R: CCACTACGCCTCCGCTT TCCTC TCTATGGGC AGTCGGTGAT
The universal fITS7 primer has been noted to exclude cer tain
Ascomycota (Penicillium, Orbiliales) and most Mucorales (Ihrmark
et al., 2012), but was chosen because it is more fungi‐specific com‐
pared to other universal primers (e.g., fITS9 or gITS7, which match
some plants because they are degenerated at two positions, poten‐
tially over whelming any fungal signal in leaf substrates). Moreover,
the prim er pair fITS7 and ITS 4 is believed to capt ure most of the
Basidiomycetes, including rust fungi, and its amplicon lengths are well
suited to next-generation sequencing (average of 258.5±27.3bp
forAscomycotaand 309.8±35.6bp forBasidiomycot a)(Bokulich&
Mills, 2013; Ihrmark et al., 2012). Purification and size selection (280–
520 bp) were performed using a PippenPrep s ystem to exclude primer
dimersandhighmolecularweightDNA ,beforepaired-endsequenc‐
ingthesa mpleswiththeIllum inaMiSeqplatform(250 cyclePE)atthe
Australian Genome Research Facility Ltd, Melbourne, Australia, and
withtheIonTorrentPGMplatform (400bpSE)attheWaikatoDNA
SequencingFacility,UniversityofWaikato,Hamilton,NewZealand.
2.4 | Preparation of clone libraries
The use of a rust fungi‐specific primer was necessary to focus the
cloning procedure on Pucciniales and to get to species resolution.
Weamplifiedanapproximately1,40 0-bptargetregionwiththerust
fungi‐specific forward primer Rust2inv:
GATGAAGA ACACAGTGAA A (Aime, 2006) and reverse primer
LR6:CGCC AGTTCTGCTTACC(Vilgalys&Hester,1990), star ting in
the 5.8S subunit and spanning the highly variable ITS2 region and
the three most divergent domains (D1, D2, D3) of the large subunit
(LSU, 28S).WeperformedPCRsforthetwoDNAextracts ofeach
plot using t he TaKaRa Ex Taq DNA poly merase kit ( 25µl reacti on
volumes , containing 2. 5µl 10XE x Taq buf fer,2µl d NTP mixture
(2.5mM each), 5µl 10µg/ml rabbit serum albumin (RSA), 0.6µl
10µMof each upstream and downstream primer,0.125µl TaKaRa
ExTaq,1µlDNAtemplate,and13.175µlofsterilizeddistilledwater).
PCR conditions consisted of an initial denaturation step of 2 min at
94°C,35cyclesof30sat94°C,1minat57°C ,and1.5minat72°C,
andafinalextensionof7minat72°C,asinitiallydescribedbyAime
(2006)butusingfewercycles.Wepooled1µlofPCRproductorigi‐
natingfromtheCTABand1µlfromtheSDS-basedlysisbuf ferDNA
extractions per plot, and cloned using the Strataclone PCR cloning
kit (Agilent, Stratagene), following the manufacturer's protocol, with
blue‐white screening of colonies. We conducted a preliminary re‐
striction fragment leng th polymorphism (RFLP) to determine suf‐
ficient sampling depth. The rarest pattern observed occurred five
times out of 100 colonies within a plot. On that basis, we picked 50
colonies p er plot (1,50 0 overall), resul ting in a 91.47% probab ility
ofdetecting the rarest OTU. Weperformed colonyPCRs withthe
plasmid‐specific primer pair M13–20: GTAAAACGACGGCCAG and
M13RSP: CAGGAA ACAGCTATGACCAT (Wood et al., 2012), using
the TaKaRa Ex Taq DNA po lymerase kit (15µl rea ction volumes,
containing1.5µl 10X Ex Taq buffer,1.2µl dNTP mixture (2.5mM
each), 0.6µl 10µg /ml rabbit serum albumi n (RSA), 0.24µl 10µM
ofeach upstreamanddownstream primer,0.075µlTaKaRaExTaq,
colony DNA template, and 10.15µl of sterilized distilled water).
PCR conditions consisted of an initial denaturation step of 12 min
at94°C,35cyclesof20sat94°C ,10sat55°Cand1.5minat65°C,
and a fina l extension of 10min at 65°C , following the m ethod of
Woodet al. (2012) butdoublingthe annealing time at 65°C . After
a gel visuali zation, seq uencing of colo nyP CR produc ts in the for‐
ward direction was conducted with the Rust2inv primer at the Bio‐
Protection sequencing facility, Lincoln University, New Zealand.
Reversesequencingwasnotconductedbecausethegeneregionsof
interest (ITS2, D1, D2, D3) lie within the first 750 bp of the forward
sequencingread.
2.5 | Bioinformatics
Wetrimmedlow-qualitybasesattheclonelibrarysequence begin‐
n i n g s a n de n d s , a n d r e m o v e d p r i m e r an d v e c t o r s e q u e n c e s . W ea l i g n e d
the sequ ences using t he MUSCLE ver sion 3.8. 31a lgorithm (Edg ar,
2004)andtrimmedthebeginning,sotheystartatthesamepointof
thegen eregionasth esequ ences fromIonTor re nta ndI llumi nau sing
thefITS7primer.Identicalsequenceswerede-replicatedandN-pad‐
dedtothesamelength.N-padding(i.e.,addingNs,whichrepresent
anynucleotide)totheendofeachsequenceuntiltheyhavethesame
4 of 11
|
MAKIOLA et A L.
lengths was needed because the clustering algorithm used consid‐
ers terminal gaps to be absolute differences. However,N-padding
only was necessaryfortwo shortclone sequences. Not N-padding
ofthesetwosequenceswouldhaveresultedintwoadditionalOTUs
but would not have changed the overall results. We clustered the se‐
quencestoa97%similaritythresholdwithoutusingsingletonsusing
UPARSE algorithm(Edgar,2010).This threshold represents theITS
barcode gap for the over whelming majority of fungal species, includ‐
ing the subdivision Pucciniomycotina (Schoch et al., 2012).
The forward and reverse Illumina reads were merged using
the fastq_mergepairs command of USE ARCH version 9.0.2132,
andsequenceswithmore than one expected errorandlessthan
175bpwereremoved.IonTorrentsequenceswereonlyusedifthe
forward and the reverse primer complement could be found within
thesequenceandifthesequencewasatleast175bplong.Wedis‐
ca rd edIonTo rrents equ en ce sw it hm orethantw oe xp ect ed er rors
(EE). We set a higher EE threshold because the mean expected
errorrateoftheIonTorrentrunsatthesequencelengthof300bp
was two. We tri mmed non-biol ogical (prim er) sequences , allow‐
ing 10% bp misma tch using the Py thon tool “cutadap t” version
1.13 (Martin, 2011) if the forward primer or the reverse primer
complementcould be found at the sequenceends.Identicalse‐
quences were de-replicated. Illumina and Ion Torrentdata were
independentlyclusteredto97%similaritythresholdwithoutusing
singletons,usingthe UPARSEgreedyclusteringalgorithm(Edgar,
2013).
We constructed a reference database from UNITE and INSD
(accessed20.11.2016)andmatchedtherepresentativesequenceof
eachOTUtothisdatabase using BLASTversion 2.5.0+(Altschulet
al.,1997).WeconsideredanOTUtorepresenttheorderPucciniales
if it matched Pucciniales sequences in the database >80% iden‐
tity over atleast 150bp (Nguyenetal.,2016;Schochetal.,2012).
Extraction blanks, and positive and negative controls, were checked
for contamination. Tag jumping (false combinations of tags and
samples,which cause incorrectassignmentofsequences) (Schnell,
Bohmann, & Gilber t, 2015) was accounted for by using a regres‐
sion of the abundance of contaminants versus the maximum of
total abundances in all other samples. The coefficient estimate for
the 90thquantileregression was thenused to subtractthat many
sequencesfromallOTUs. Hence,thistag-jumping correction takes
into account thefact thatmore abundant OTUs are morelikely to
dotagjumping.WeblastedOTUsobtainedfromthethreedifferent
methods against each other and considered them to be the same
OTUiftheymatchedat>98.5%similarity,whichcorrespondstoap‐
proximately3%clusteringoftheNGSdatausingthedistance-based
greedyclu steringUPARS Eal gor ithm(Edga r,2013),butallowsdiffer‐
entsequencelengthsas opposed to matchingwithUSEARCHver‐
sion 9.0.2132 (Altschul et al., 1997; Edgar, 2010, 2013).
2.6 | Statistical analyses
WeusedRversion3.4.1(R CoreTeam, 2017)for conduc tinganaly‐
ses and creating graphs if not stated otherwise. To test whether a
method d etected more or f ewer shared/unique r ust fungal OTUs
than expected by chance, we used the “permatswap” function of the
R package “vegan” version 2.0–7 (Oksanen et al., 2017) to create a
null expectation. The simulated community matrices are based on
MonteCarloiterations,wherebythetot alnumberofOTUsperplot
andtotalabundancewithinOTUwerekeptconstant.Wetestedfor
differencesinOTUabundancesamongmethodsusingageneralized
additive model (GAM) of the package “mgcv” version 1.8–18 (Wood,
2001). A GAM was selected because: (a) it allows beta distribution
for the response variable, which in this case was the appropriate dis‐
tributionfortheproportionalabundanceofeachOTUfoundwithin
aplot(toaccount for different sequencing depths);and(b)theap‐
proach allowstestingforOTUand plot as random effects,and in‐
teractionbetweenmethodandOTU.Datawererescaledtoexclude
zeros and ones, as suggested by Smithson and Verkuilen (20 06).
Wald test was used to test the significance of each parametric and
smooth term (Wood, 2012). To see whether perceived rust fungi
communities differ among methods, we converted the obtained
communit y data into Jaccard distance matrices using Wisconsin
doublestandardization.FourplotswithzeroOTUs,aswellasunique
communities, had to be discarded because of a dissimilarity of one.
We displayed the dissimilarities with nonlinear multidimensional
scaling and tested for significance between the configurations
using Procrustes rotation and the “protest” function part of the
“vegan” package, and the “mantel.test” function of the “ape” pack‐
age(Paradis,Claude, &Strimmer,2004).We testedwhether abias
amongmethodswascausedbydifferentsequencelengthsorbioin‐
formaticpipelines,applyingthesamesequencelength(248bp)and/
or an identical bioinformatic pipeline to all methods. To look for a
taxonomic bias in detecting the different methods, we constructed
aneighbor-netphylogeny(Bryant&Moulton,2004)usingSplitstree
4.0(Huson, Kloepper, &Bryant,2008)andusedchi-squaretestto
test whether taxonomic clusters are independent of methods. We
tested whether a possible difference is due to the detection of rare
anddominantOTUsby rerunning all tests using thetopandlower
50%oftherank abundanceofeachmethod.Species identities are
based on the best BL AST match and were displayed as networks
using the “ig raph” package ve rsion 1.0.1 (Csard i & Nepusz, 20 06)
with edge width representing relative species abundance within
method.
3 | RESULTS
3.1 | Differences among methods in detection of
OTUs
There weresevenrust fungalOTUs shared amongthethreemeth‐
ods, which was much less than would be expec ted by random sam‐
pling (17.05±0.33). The dif ference was driven b y OTUs uniquely
detectedby singlemethods (Figure1),thatis,Illumina (one unique
OTU) and Ion Torrent (t wo unique OTUs), and cl oning (10 unique
OTUs). The thre e methods (i.e ., cloning, Il lumina, and I on Torrent)
hencedifferedindetectionofrustfungalOTUs.
|
5 of 11
MAKIO LA et AL .
3.2 | No differences among methods in relative
abundances of shared OTUs and in perceived
community composition
Therewasnoevidenceofdifferencesinquantificationofrelative
abundances among the three methods (i.e., cloning, Illumina, and
Ion Torrent) amo ng OTUs that all met hods were c apable of de‐
tecting. A likelihood ratio test between models with and without
aninteract ionterm(method×OTU)wasnotsignificant(χ2 = 7.6 2 ,
df = 12, p = 0.81). In general, rust communities perceived by the
three methods did not result in largely different community pat‐
terns, as visualized by theoverlap of thecommunities in NMDS
(Figure 2). Mantel test and Procrustes analysis confirmed simi‐
larity (p < 0.05) for Ion Torrent/cloning (abundance data), and
Ion Torrent/cloning and Illumina/Ion Torrent (presence/absence
data).
3.3 | Mechanisms driving OTU detection
differences among methods
Differences in detection among methods seemed not to be due
to sequen ce length dif ferences a mong method s. After tri mming
all sequences to thesame length (248bp), which is the shortest
common sequence of all methods, and rerunning the analysis,
the number of observed (seven) compared to randomly expected
(17) shared rus t OTUs stayed unchan ged. Differe nces in detec‐
tion among methods also seemed not to be due to differences in
the most appropriate bioinformatic pipelines for each method.
Usinganidenticalbioinformaticpipelineforallmethodsmadedif‐
ferenceseven moreextreme,withonlyfourOTUs shared among
methods, compared to seven (with the most appropriate pipelines)
or 17 (expected). Differences in detection among methods were
dueto ataxonomicbiasof themethods.Neighbor-netphylogeny
FIGURE 1 (a) Observed and (b)
expected number of rust fungal
operationaltaxonomicunits(OTUs)per
method.OTUswereconsideredtobe
identicalamongmethodswhen>98.5%
BLAST similarity. Expectations were
based on Monte Carlo random sampling
(100iterations)anddisplayedwith95%
confidence intervals
Observed OTUs Expected OTUs
(a) (b)
FIGURE 2 Multidimensional scaling of rust communities (using abundance and presence/absence data) as perceived by three different
methods:Illumina(green,squares),IonTorrent(blue,circles),cloning(orange,triangles).Fourplotsweredroppedbecauseoflackofany
detected rust communities in these plots
NMDS (Abundances) NMDS (Presence/Absence)
(a) (b)
6 of 11
|
MAKIOLA et A L.
(Figure 3) indicates three taxonomic clusters. Cluster 1 could
equally bedetectedbyall methods;cluster2 wasonly detected
using Illumina; cluster 3 was only detected using cloning. The chi‐
square test for independencewassignificant( χ2 = 17.536, df=4,
p<0.01)andconfirmedthatclusters werenotequallyformedby
the different methods.
Speciesidentitiesofcluster3(i.e.,uniquelydetectedbycloning)
andcluster 2(i.e., uniquelydetectedby Illumina)weredisplayedin
a co-occurrence network (Figure 4). While Illumina'suniquely de‐
tected species is from the genus Kuehneola,uniquelydetectedspe‐
cies from cloning and Ion Torrent are from the genus Puccinia. The
taxonomic bias seemed not to be driven by poor detection of rare
FIGURE 3 Neighbor-netphylogenyof
rust fungal operational taxonomic units
(OTUs)detectedbythedif ferentmethods:
Illumina(squares),IonTorrent(circles),
cloning (triangles)
FIGURE 4 Networkrepresentingsharedanduniquerustfungaloperationaltaxonomicunits(OTUs)amongmethods.Edgewidth
representsproportionalabundanceofanOTUwithinmethod.SpeciesidentitiesarebasedontheirbestBL ASTmatch.OTUsfoundineach
methodareconsideredtobeidenticalwhenshowing>98.5%sequencesimilarit y
|
7 of 11
MAKIO LA et AL .
OTUs. The sa me clusters occ ur when only consid ering the uppe r
50% of rank a bundance, h ereafter c alled domin ant OTUs (Figures
S2andS3),andwhenonlyconsideringthelower50%ofrankabun‐
dance, hereafter rare OTUs (Figures S4 and S5). The number of
observedshareddominant(six)andrare(two)OTUsstilldifferssig‐
nificantlyfromrandomly expected(11.08±0.36OTUs)sharedrust
OTUs.Thisdifferenceinobservedfromexpectedisstillmainlydue
totheuniquelydetectedOTUsfromcloning (cluster2ofFigureS2
andcluster3ofFigureS 4).
Differences in detection among methods seemed to be caused
by base pair mismatches of the NGS metabarcoding primer pair.
Table 1 shows selec ted species that were detected by cloning but
notbyNGSmetabarcodingandhadatleastonebasepairmismatch
totheNGSmetabarcodingprimers.
4 | DISCUSSION
This stu dy demonstrat es that NGS metab arcoding is an eff ective
techniquefor large-scale detection ofrustfungus plant pathogens,
but that taxonomic biases due to primer selection are a potential
limitation. To the best of our knowledge, this is the first study with
areal-worldapplicationandcomparisonofcloningandNGSmeta‐
barcoding to survey Pucciniales. We found dif ferences in the detec‐
tion of rust fungus species among Illumina and Ion Torrent platforms,
andcloningfollowedbySangersequencing.However,wefoundno
significant difference in the relative abundances of the rust fungus
species that all methods were capable of detecting. The mechanism
driving detection differences among methods seemed to be due to
a taxonomic bias, which was very likely caused by base pair mis‐
matches of th e NGS metabarc oding primer pai r to some Puccinia
species. Otherwise, the consistency among fundamentally different
andindependentmolecularmethodsshowsthatNGSmetabarcoding
and cloning are on a par. Altogether, the results support the applica‐
tion of NGS m etabarcodin g for the large-sc ale detectio n of plant
pathogens (presences) and contradict it s application for inferring ab‐
sence of species, depending on the primer pairs. These findings are
important to future metabarcoding studies because they highlight
the main source of difference among methods and rule out several
mechanisms that could drive differences.
Themaindifferencebetweenthemethods(NGSmetabarcod‐
ing and cloning) was due to their biases in species detection, not
quantification.This suggests thatprevious problems whenusing
quantitativenext-generationsequencing data (Elbrecht& Leese,
2015; Piñol, Mir, Gomez‐Polo, & Agustí, 2015) were probably
induced by PCR, and not by the methodorsequencing platform
per se. Furthermore, this is in line with the finding that the dif‐
ference in detection between NGS metabarcoding and cloning
shows a ta xonomic bias. Both t he NGS metabarc oding and the
cloning primers have either a perfect match or only a maximum of
two base pair mismatches to all detected rust fungi in this study.
Moreover,theNGSmet abarcodingprimers were thoughttocap‐
ture most of the Basidiomycetes (Ihrmark et al., 2012; White et al.,
1990),inclu di ngrustfungi .Consequent ly,theNG Sm et ab arcodin g
and the cloning primers would be expected to detec t a similar as‐
semblageofrustfungi.However,thebasepairmismatchesofthe
NG Smet abarcodingpri me ro cc ur insp ec ie st hata re on lydetec te d
by cloning, and the cloning primer had no mismatches in these spe‐
cies. The lower specificityof the “universal”NGSmetabarcoding
primers is therefore more likely to discriminate against the ampli‐
ficationofthosespecieswhenexposedto 100%matching other
fungalsequencetemplates(Bellemainetal.,2010).Loweringthe
annealing temperatures might help remedy these mismatch biases
for this group in the future, par ticularly as none are very close
to the 3' end of primers (Table 1). Although taxonomically clus‐
tered, the Pucciniaspe ciesw it hthebas epairmi smatc hofth eNGS
Species
5’‐fITS7 (forward primer)
GTGARTCATCGA ATCTTTG
3’‐ITS4 (rever se primer)
GCATATCAATAAGCGGAGGA
Puccinia calcitrapaea
GTGA ATCATTGA ATCTTTG GCATATCA ATAAG CAGAGGA
Puccinia nishidanab
....A T C A T TGAATCTTTG G CATATCA ATAAGC AGAGGA
Puccinia balsamorrhizaec
......C A T TGA ATCTTTG GCATATC AATA AGCAGAGGA
Puccinia komaroviid
GTGA ATCATTGA ATCTTTG GCATATCA ATAAG CAGAGGA
Puccinia hieraciie
......C A T C G A A T C T T T G GCATATCA ATAAG CAGAGGA
Note. Mismatches are highlighted (bold and underlined).
Sequenceswereselected fromtheNational CenterforBiotechnology Information(NCBI) tocover
the gene region of cloning and metabarcoding primers when possible.
Dot indic ates no entry of base pair in the database.
Accession numbers are given as footnotes. Accession numbers:
aJN2 04183 .1 bHM 02 2141.1cJ N204182.1dKC466553.1eHQ 317515.1
TABLE 1 Metabarcoding primer
mismatches to selected species that were
detected by cloning but not by
metabarcoding
8 of 11
|
MAKIOLA et A L.
metabarcoding primer seemed not to fall into a known taxonomic
cluster,likeasubgenus(VanderMerwe,Ericson,Walker,Thrall,&
Burdon, 2007).
Numerous NGS metabarcoding studie s have pointed out that
NGS metabarcoding primers can discriminate against certain ta xa
(Bellemainet al.,2010; Clarke et al., 2014;Elbrecht&Leese,2015;
Schmidt e t al., 2013). Some stud ies have tried to limit t his bias to some
extent by us ing quantit ative PCR and cor rection f actors ( Thomas,
Deagle, Eveson,Harsch, &Trites,2016),primermixes( Tedersoo et
al.,2015),orblockingoligonucleotidestonon-targetDNA(Piñolet
al., 2015). Ficetola et al. (2010) proposed an “electronic PCR” ap‐
plication to measure barcode coverage and specificit y. This in silico
approach has proven useful to identify the appropriate barcode gene
regions and when comparing different primers for fungi (Bellemain
etal.,2010)andvertebrates(Valentinietal.,2016).Theresultsfrom
this study and from the literature, taken together, highlight the im‐
portance of primer choice for NGS metabarcoding studies. NGS
metabarcoding studies should therefore carefully examine in silico
what taxa their primers might discriminate against in order to select
appropriateNGSmetabarcodingmarkersandaidtheinterpretation
of results.
This study also ruled out several mechanisms that could possibly
drivedetectiondifferencesbetweenNGSmet abarcodingandclon‐
ing.We found noevidence thatsequencelength, most appropriate
bioinformatic pipeline, or ability to detect rare species caused any
differencesamongmethods.Wefoundthatshorteningallsequences
tothelengthoftheshortest sequence(248bp)didnot change the
interpretation of the overall results and resulting phylogeny. Min and
Hickey(2007)andHanetal.(2013)showedthatreducingsequence
lengthcan haveeffectsonthe accuracyof phylogenieswhenDNA
barcoding fungi. They also showed that despite some loss of phylo‐
geneticsignal,shortersequencescanstillresolvetheterminalnodes
ofthephylogeny quite efficiently in most cases.Currentnex t-gen‐
eration sequencing technologies still require the amplification of
short sequences,andsomebarcoderegions(e.g.,the ITSregionfor
fungi) can lack the necessar y resolution for particular fungal taxa
(Gazis, Rehner & Chaverri, 2011). Despite these challenges, short
sequencesprovideenoughresolutionatagenusandof tenawithin-
genus level for the majority of fungi (Blaalid et al., 2013). While short
sequenceshavebeenrepeatedlyshowntobesufficientforgenus-
or even species‐level identifications (Blaalid et al., 2013; Bokulich &
Mills,2013),futurenext-generationsequencingtechnologiesshould
be able to overcome the current leng th limitations and provide the
field ofNGS metabarcodingwithevenbetter speciesdelimitations
(Goodwin, McPherson, & McCombie, 2016).
Bioinformatic pipelines can have profound effects on the outcome
of NGS metabarcoding studies (Flynn, Brown, Chain, MacIsaac, &
Cristes cu, 2015). In this study, the error rate s trongly differ ed between
Illumina,IonTorrent,andSanger sequencing runs. Using an identical
bioinformaticpipeline,suchasidenticalqualityfilteringandclustering,
resultedinamu chl owe rnumberofsharedOTUsamo ngt hem ethod s.
These results justify using the most appropriate bioinformatic pipeline
for each method. Moreover, we did not find any effec t of rare species
on detection ability among methods. The same t axonomic bias among
the methods occurred when only looking at the dominant or only look‐
ingattherareOTUs.RareOTUsinNGSmetabarcodingdataaregen‐
erally more prone to errors due to the accumulation of errors (Dickie,
2010), tag jumping (Schnell et al., 2015), chimera formation (Edgar,
Haas, Clemente,Quince, & Knight,2011),or falsepositive/negatives
(Ficetola etal., 2010). However, previous studies have shown thatif
these problems associated with rareOTUsareovercome,theability
ofNGS metabarcodingtodetect rarespeciesisequal to or exceeds
non-molecularmethods(Valentinietal.,2016;Zhanetal.,2013).
Next-generationsequencingmetabarcodingseems appropriate
for the large‐scale detection of rust fungi and less appropriate for
inferring absence of species. For example, the species Puccinia sorghi
wasinitiallypresent intheraw data ofall threemethods.However,
onlytwosequencesofthisspecieswerepresentintheIlluminaraw
data. Th ese two seq uences exh ibited a point m utation or a p ossi‐
blesequencingerrorintheirreversesequencereadandgottreated
as unique sequences (singletons) after merging. Hence, although
initially present in the Illumina raw data, these two sequences
could not fo rm an OTU. This phen omenon of speci es getting lo st
duringmergingofpaired-endsequencinghasbeennotedearlierby
Nguyen,Smith,Peay,andKennedy(2015)andwasgenerallycaused
by the usual ly poorer qualit y of reverse sequen cing reads of the
Illumina MiSeq platform. The problem of missing extremely rare
species, however, is not method specific, as the case of Kuehneola
uredinisdemonstrates.Thisrarespecieshadatotalof47sequences
intheIlluminadataandwasinitiallypresentasasinglesequencein
the raw data of the clone libraries. Because singletons got discarded
regardless of the method, Kuehneola uredinis got discarded from the
clone data. The fact that the cloning primer pair had a perfect match
to Kuehneola uredinis and that this species got picked up once clearly
shows that the detection of rare species does not rely on the applied
method butratheronsequencingdepthand bioinformaticassump‐
tions. Picking a greater number of clones would probably have re‐
sultedinatleastanothersequenceofKuehneola uredines, and hence
detection of this species. Despite failing to detect two rare species
by some methods, other rare species, such as Uromyces dactylidis
and Puccinia hordei, could be detected regardless of the method.
Another way of easily missing species when merging paired‐
endsequencing readsistolose“toolong”sequences,sincethese
would not overlap. This can be simply tested by not merging the
reads and using forward and reverse reads separately. In this
study, we found no rust fungus species getting lost during merging
asares ultof“toolong”s eq uence s. Theac tualIllum inase quencin g
process, however, is known for discriminating against longer ampl‐
icons (Allen et al., 2016). Although less likely than, for instance, a
primer mismatch, the Puccinia species that could not be detected
byNGS metabarcoding butcould by cloningcould possiblyhave
beenmissedduringthenext-generationsequencingprocessdue
to slightly longer amplicons. We did not compare abundance data
to a field survey or biomass, but found no significant difference in
relativeabundancesofOTUson plotlevelamongNGSmetabar‐
codingandcloning.Thissuggeststhatanybiasesinquantification
|
9 of 11
MAKIO LA et AL .
using molecular techniques are not method dependent. Despite
issues arising from PCR (yet common for most molecular methods)
suchasthedif ferencein rRNA copy numbers,several studiesdo
show NGS me tabarcoding t o be successful f or semiquantit ative
abundance estimation of, for example, feather mite communi‐
ties in birds (Diaz‐Real, Serrano, Piriz, & Jovani, 2015), fish and
amphibians in freshwater ecosystems (Evans et al., 2016), plant–
pollinator interactions (Pornon et al., 2016), the biomass of mac‐
roinvertebrates (Elbrecht & Leese, 2015), and fungi (Taylor et al.,
2016). These studies suggest that if obstacles associated with PCR
biasescanbeovercome,NGSmetabarcodingholdspromisingpo‐
tentialnotonlyforthedetectionbutalsoforthequantificationof
species.Moreover,PCR-freetechniquesmayremedyprimerand
amplification biases (including chimera formation) in the near fu‐
ture. Dif ferent gene copy numbers still pose a significant challenge
for biomass estimates but could be overcome with the growing
number of whole genome databases.
Next-generation sequencing metabarcoding has been increas‐
ingly recognized as a promising tool for biomonitoring species and
complex communities atlarge scales(Holdaway etal.,2017).Inre‐
cent cases, it has been applied to plant pathogenic fungi (Merges et
al., 2018) and oomycetes (Burgess et al., 2017). It is impor tant to un‐
derstandtheadvantages anddisadvantagesofusingNGSmetabar‐
coding for detecting and monitoring important functional groups
at the ecosystems scale. Our study suggests that rust fungi can be
tracked from within a larger NGS metabarcoding dataset, which
should facilitate the future monitoring of this critically import ant
group of fungi.
ACKNOWLEDGEMENTS
We thank landowners and the Department of Conservation for allow‐
ing access and sampling, and Manaaki Whenua – Landcare Research
staff, especially Larry Burrows and Karen Boot, for their extensive
support in the field and laboratory. We acknowledge support pro‐
vided by John Longmore of theWaikatoDNA Sequencing Facility.
ThisresearchwasfundedbytheNewZealandMinistryofBusiness,
Innovation and Employment (MBIE Contrac t No. C09X1411) via
Manaaki Whenua – Landcare Research, and by Centre of Research
Excellence funding to the Bio‐Protection Research Centre.
CONFLICT OF INTEREST
The authors declare that the research was conducted in the ab‐
sence of any commercial or financial relationships that could be
construed as a potential conflic t of interest.
AUTHORS CONTRIBUTION
The samp les used in t his paper were p art of a stu dy led by RH,
IAD,JRW,and KHO. JRW,IAD,RH,andKHOconceived primary
funding, with additional funding from the Bio‐Protection Research
Centre, led by TRG. IAD and TRG advised AM. AM collected the
sampleswithhelpfromRHandothers.AM,CKL,andJRWdevel‐
oped the methods, and AM and CKL carried out molecular char‐
acterization. AM and IAD conducted data analysis. AM produced
the figures and wrote the first draft . All other authors provided
editorial input.
ETHICS STATEMENT
This article does not contain any studies with human participants
or animals performed by any of the authors.
DATA ACCESSIBILITY
The data associated with the paper are available from the Manaaki
Whenuadatarepositoryatht tps://doi.org/10.25898/KK41-CY40.
ORCID
Andreas Makiola https://orcid.org/0000‐0002‐9611‐9238
Ian A. Dickie https://orcid.org/0000-0002-2740-2128
Jamie R. Wood https://orcid.org/0000‐0001‐8008‐6083
Travis R. Glare https://orcid.org/0000‐0001‐7795‐8709
REFERENCES
Abdelfattah,A.,Nicosia,M .G.L.D.,Cacciola,S.O.,D roby,S.,&Schena,
L. (2015). Metabarcoding analysis of fungal diversity in the phyllo‐
sphere and carposphere of olive (Olea europaea). PLoS ONE, 10(7),
e0131069. https://doi.org/10.1371/journal.pone.0131069
Aime, M. C. (2006). Toward resolving family‐level relationships in
rust fungi (Uredinales). Mycoscience, 47(3), 112–122. https://doi.
org /10.1007/S10267‐00 6‐0281‐0
Allen, H. K., Bayles,D.O.,Looft, T.,Trachsel,J., Bass, B.E.,A lt,D.P.,…
Casey, T. A. (2016). Pipeline for amplifying and analyzing amplicons
oftheV1–V3regionofthe16SrRNAgene.BMC Research Notes, 9(1),
380.ht tps://doi.org/10.1186/s13104-016-2172-6
Allen, R . B., Bellingham, P. J., & Wiser, S. K. (20 03). Developing a for‐
estbiodiversitymonitoringapproachforNewZealand.New Zealand
Journal of Ecology, 27(2), 207–220.
Altsc hul, S. F., Madden, T. L., Schäffer, A. A ., Zhang, J., Zhang, Z.,
Miller, W., & Lipman, D. J. (1997). Gapped BLAST and PSI‐BLAST:
A new generation of protein database search programs. Nucleic
Acids Research, 25(17), 3389–34 02. ht tps://doi. org/10.1093/
nar/25.17.3389
Bebber, D. P., & Gurr, S. J. (2015). Crop‐destroying fungal and oomycete
pathogens challenge food security. Fungal Genetics and Biology, 74,
62–64.https://doi.org/10.1016/j.fgb.2014.10.012
Bellemain, E., C arlsen, T., Brochmann, C., Coissac, E., Taberlet , P., &
Kauser ud,H.(2010).ITSasanenvironmentalDNAbarco deforfungi:
An in silico approach reveals potential PCR biases. BMC Microbiology,
10(1) ,189.https://doi.org /10.118 6/1471-2180 -10-189
Bever, J. D., Mangan, S. A ., & Alexander, H. M. (2 015). Maintenance
of plant species diversity by pathogens. Annua l Review of Ecology,
Evolution, and Systematics, 46, 305–325. https://doi.org/10.1146/
annurev-ecolsys-112414-054306
Blaalid, R., Kumar, S., Nilsson , R. H., A barenkov, K., Kirk, P. M., &
Kauser ud, H. (2013). ITS1 versus I TS2 as DNA metabarcodes for
10 of 11
|
MAKIOLA et A L.
fungi. Molecular Ecolog y Resources, 13(2), 218–224. https://doi.
org /10.1111/1755‐0998.120 65
Bokulich,N.A.,&Mills,D.A.(2013).Improvedinternaltranscribedspacer
(ITS) primer selection enables quantitative, ultra-high-throughput
fungal community profiling. Applied and Environmental Microbiology,
79(8), 2519–2526.
Bryant, D., & Moulton, V. (2004). Neighbor-net: An agglomerative
method for the construction of phylogenetic networks. Molecular
Biology and Evolution, 21(2), 255–265. https://doi.org/10.1093/
molbev/ms h018
Burgess, T. I., White, D., McDougall, K. M., Garnas, J., Dunstan, W.
A., C atalà, S ., … Stukely, M. J. (20 17). Distri bution an d diversit y of
Phytophthora across Australia. Pacific Conservation Biology, 23(2),
150–162. https://doi.org /10.1071/ PC16032
Carnegie, A. J., Kathuria, A., Pegg, G. S., Entwistle, P., Nagel, M., &
Giblin, F. R. (2016). Impact of the invasive rust Puccinia psidii (myr‐
tle rust) on native Myrtaceae in natural ecosystems in Australia.
Biological Invasions, 18(1), 127–144. h tt ps://doi.org/10.10 07/
s10530‐015‐0996‐y
Clarke, L . J., Czechowski, P., Soubrier, J., Stevens, M. I., & Cooper, A.
(2014). Modula r tagging of a mplicons usi ng a single PCR f or high-
throughputsequencing.Molecular Ecology Resources, 14(1), 117–121.
https://doi. org/10.1111/1755‐0 998 .12162
Coissac,E.,Riaz,T.,&Puillandre,N.(2012).Bioinformaticchallengesfor
DNAmetabarcodingofplantsandanimals.Molecular Ecology, 21(8),
1834–1847.https://doi.org/10.1111/j.1365-294X.2012.05550.x
Csardi,G.,&Nepusz,T.(2006).Theigraphsoft warepackageforcomplex
network research. InterJournal, Complex Systems, 1695. Retrieved
from http://igraph.org
Deiner,K.,Bik,H.M.,Mächler,E.,Seymour,M.,L acoursière-Roussel,A .,
Alterm att,F.,…Pf ren der,M.E .(2017).Environme ntalDNAme ta bar‐
coding: Trans forming how we survey animal and plant communities.
Molecular Ecology, 26( 21), 5872–58 95. h tt ps://doi.org /10.1111/
mec.14350
Diaz-Real , J., Serran o, D., Piriz, A. , & Jovani, R. ( 2015). NGS meta bar‐
coding proves successful for quantitative assessment of symbi‐
ont abundance: The case of feather mites on birds. Experimental
and Applied Acarology, 67(2), 209–218. https://doi. org /10.10 07/
s10493-015-9944-x
Dickie,I. A. (2010). Insidiouseffectsof sequencing errorsonperceived
diversity in molecular surveys. New Phytologist, 188 (4), 916–918.
https://doi.org/10.1111/j.1469-8137.2010.03473.x
Durand,A.,Maillard, F.,Foulon,J.,Gweon, H.S.,Valot ,B., &Chalot, M.
(2017). Environmental metabarcoding reveals contrasting below‐
ground a nd abovegrou nd fungal co mmunities f rom poplar a t a Hg
phytomanagement site. Microbial Ecology, 74(4), 795–80 9. https://
doi.org/10.1007/s00248-017-0984-0
Edgar, R. C . (2004). MUSC LE: Multiple s equence align ment with high
accuracy and high throughput. Nucleic Acids Research, 32(5 ), 179 2–
1797.ht tps://doi.org/10.1093/n ar/gkh34 0
Edgar, R. C . (2010). Search and clu stering orders of magnitu de faster than
BL AST. Bioinformatics, 26(19), 246 0–2461. https ://doi.org /10.10 93/
bioinformatics/btq461
Edgar,R .C.(2013).UPARSE: Highlyaccurate OTU sequences from mi‐
crobial amplicon reads. Nature Methods, 10(10), 996. https://doi.
org/10.1038/nmeth.2604
Edgar, R. C., Haas, B. J., Clemente, J. C., Quince, C., & Knight, R.
(2011). UCHIME im proves sens itivit y and speed o f chimera d etec‐
tion. Bioinformatics, 27(16), 2194–2200. https://doi.org/10.1093/
bioinformatics/btr381
Elbrecht,V.,&Leese,F.(2015).CanDNA-basedecosystemassessments
quantif y species ab undance? Testing pr imer bias and bio mass–se‐
quence relationships with an innovative metabarcoding proto‐
col. PLoS ONE, 10 (7), e0130324. https://doi.org/10.1371/journal.
pone.0130324
Evans, N. T., Olds, B. P., Renshaw, M.A., Turner, C. R., Li, Y., Jerde, C.
L., … Lodge, D. M. (2016). Quantific ation of mesocosm fish and
amphibian species diversity via environmental DNA metabar‐
coding. Molecular Ecology Resources, 16 (1), 29–41. https://doi.
org /10.1111/1755-0998.12433
Fernandez Winzer, L., Carnegie, A. J., Pegg, G . S., & Leishman, M. R.
(2018). Impacts of the invasive fungus Austropuccinia psidii (myrtle
rust) on three Aus tralian Myrt aceae species of coastal swamp wood‐
land. Austral Ecology, 43(1), 56–68.
Ficetola , G. F., Coissac, E ., Zundel, S., R iaz, T., Shehzad, W., Be ssière,
J., … Pompanon, F. (2010). An in silico approach for the evalua‐
tion of DNA barcodes. BMC Genomics, 11(1), 434. ht tps://doi.
org /10.118 6/1471-2164 -11-434
Flynn, J.M.,Brown,E.A.,Chain,F.J.,MacIsaac,H .J.,&Cristescu,M.E.
(2015). Toward accurate molecular identification of species in com‐
plexenvironmentalsamples:Testingthe performance ofsequence
filtering and clustering methods. Ecology and Evolution, 5(11), 2252–
2266.https://doi.org/10.1002/ece3.1497
Gazis, R ., Rehner, S., & Chaverri, P. (2011). Species delimitation in fun‐
gal endophyte diversity studie s and its implications in ecological and
biogeographic inferences. Molecular Ecology, 20(14), 3001–3013.
https://doi. org/10.1111/j .1365-294X. 2011 .05110.x
Glen, M.,Alfenas,A. C.,Zauza, E.A .V.,Wingfield,M.J.,& Mohammed,
C. (2007). Puccinia psidii: A threat to the Australian environment and
economy – a review. Australasian Plant Patholog y, 36(1 ) , 1–16 .
Goodwin, S., McPherson, J. D., & McCombie, W. R. (2016). Coming
of age: Ten years of next-generation sequencing technologies.
Nature Reviews Genetics, 17(6), 333–351. https://doi.org/10.1038/
nrg.2016 .49
Han,J.,Zhu,Y.,Chen,X.,Liao,B.,Yao,H.,Song,J.,…Meng,F.(2013).The
short ITS2 sequenceserves asanef ficienttaxonomic sequencetag
in comparison with the full‐length ITS. BioMed Research International,
2013,741476.https://doi.org/10.1155/2013/741476
Helfer, S.(2014).Rustfungi and globalchange. New Phytologist, 201(3),
770 –780. https://doi. org/10.1111/np h.1 2570
Holdaway,R .J.,Wood,J.R.,Dickie,I.A.,Orwin,K .H., Bellingham,P.J.,
Richardson,S.J.,…Buckley,T.R.(2017).UsingDNAmetabarcoding
toassessNewZealand’sterrestrialbiodiversity.New Zealand Journal
of Ecology, 41(2 ), 251–26 2.
Hurst ,J.M.,&Allen,R.B.(2007).TheReccemethodfordescribingNew
Zealand vegetation:Fieldprotocols.Lincoln,NewZealand:Manaaki
Whenua – Landcare Research.
Huson, D. H., Kloepper, T., & Bryant , D. (200 8). Split sTree 4.0:
Computation of phylogenetic trees and networks. Bioinformatics, 14,
68–7 3 .
Ihrmar k, K., Böde ker,I ., Cruz-Mar tinez, K., Fri berg, H., Kub artova, A .,
Schenck,J.,…Lindahl,B.D.(2012).Newprimerstoamplifythefun‐
galITS2region:Evaluationby454-sequencingofartificialandnatu‐
ral communities. FEMS Microbiology Ecology, 82(3), 666–677. https://
doi .or g/10.1111/ j.15 74-6941.2012.0 1437.x
Malcolm, G. M., Kuldau, G. A., Gugino, B. K., & Jiménez‐Gasco, M . D. M.
(2013).Hiddenhostplantassociationsofsoilbornefungalpathogens:
An ecological perspec tive. Phytopathology, 103(6),538–54 4.https://
doi .org/10.10 94/P HY TO -0 8-12- 0192-LE
Martin, M. (2011). Cut adapt removes adapter sequences from high-
throughputsequencingreads.Embnet. Journal, 17(1), 10. https://doi.
org /10.148 06/ej .17.1.20 0
McCallum,B.,Hiebert,C.,Huerta-Espino,J.,&Cloutier,S.(2012).Wheat
leaf rust. Disease Resistance in Wheat, 1, 33.
Merges,D.,Bálint,M.,Schmitt,I.,Böhning-Gaese,K.,&Neuschulz,E.L.
(2018). Spatial patterns of pathogenic and mutualistic fungi across
the elevational range of a host plant. Journal of Ecology, h t t p s : //d o i .
org /10.1111/1365-2745.12942 .
Miller, K. E., Hopkins, K., Inward, D. J., & Vogler, A. P. (2016).
Metabarcoding of fungal communities associated with bark beetles.
|
11 of 11
MAKIO LA et AL .
Ecology and Evolution, 6(6), 1590–1600. https://doi.org/10.1002/
ece3.1925
Min, X. J ., & Hickey, D. A. (20 07). Asses sing the eff ect of var ying se‐
quence length on DNA barcoding of fungi. Molecular Ecology
Resources, 7(3), 365–373.
Nguyen,N.H.,Smith,D.,Peay,K.,&Kennedy,P.(2015).Parsingecolog‐
ical signal fromnoiseinnext generation ampliconsequencing. New
Phytologist, 205(4), 1389–1393.ht tp s://doi .org/10.1111/n ph .129 23
Nguyen,N.H.,Song,Z.,Bates,S.T.,Branco,S.,Tedersoo,L.,Menke,J.,…
Kennedy,P.G.(2016).FUNGuild:An open annotationtoolforpar s‐
ing fungal community datasets by ecologic al guild. Fungal Ecology, 20,
241–248 .
Oksan en, J., Blanchet , F. G., Friend ly, M., Kin dt, R., Legendre, P., McGlinn,
D.,Wagner,H.(2017). Vegan:Communit yEcologyPackage.Rpack‐
age version 2.4-4. Retrieved from https://CRAN.R-project.org/
package=vegan
Paradi s, E., Clau de, J., & Strim mer, K. (200 4). APE: Anal yses of phylo ‐
genetic s and evolution in R language. Bioinformatics, 20, 289–290.
https://doi.org/10.1093/bioinformatics/btg412
Piñol, J., Mir, G., Gome z-Polo, P., & Agustí, N. (2015). Universal and
blocking primer mismatches limit the use of high‐throughput
DNA sequencing for the quantitative metabarcoding of arthro‐
pods. Molecular Ecology Resources, 15(4), 819–830. https://doi.
org /10.1111/1755‐0998.12355
Pornon,A.,Escaravage,N.,Burrus,M.,Holot a,H.,Khimoun,A.,Mariette,
J., … Vida l, M. (2016). Using m etabarc oding to reve al and quant ify
plant‐pollinator interactions. Scientific Reports, 6, 27282. https://doi.
org /10.103 8/srep27282
R Core Team. (2017). R : A language and enviro nment for statistica l comput-
ing.Vienna,Austria:RFoundationforStatisticalComputing.https://
www.R‐project.org/.
Schmidt, P. A., Bálint, M., Greshake, B., Bandow, C., Römbke, J., &
Schmit t, I. (2013). Illumina metabarcoding of a soil fungal community.
Soil Biology and Biochemistry, 65, 128–132. https://doi.org/10.1016/j.
soilbio.2013.05.014
Schnell, I. B., Bohmann, K., & Gilbert, M. T. P. (2015). Tag jumps illu‐
minated: Reducing sequence-to-sample misidentifications in me‐
tabarcoding studies. Molecular Ecology Resources, 15(6), 1289–1303.
https://doi. org/10.1111/1755-0 998 .124 02
Schoch, C. L., Seifer t, K. A ., Huhndorf, S., Robert, V., Spouge, J. L .,
Levesqu e, C. A. , … Miller, A. N. (2 012). Nuclear r ibosoma l internal
transcribed spacer (ITS) region as auniversal DNA barcode marker
for Fungi. Proceedings of the National Academy of Sciences, 109 (16),
6241–6246 .h tt ps://doi.org /10.1073/pnas .1117018109
Smithson, M., & Verkuilen, J. (2006). A better lemon squeezer?:
Maximum‐likelihood regression with betadistributed dependent
variables. Psychological Methods, 11(1), 54. https ://doi.org /10.10 37/
1082-989X.11.1 .54
Stergiopoulos, I.,& Gordon, T.R.(2014).Cr yptic fungal infections: The
hidden agenda of plant pathogens. Frontiers in Plant Science, 5, 506.
https://doi.org/10.3389/fpls.2014.00506
Taberlet, P., Coissac, E., Hajibabaei, M., & Rieseberg, L. H. (2012).
Environmental DNA. Molecular Ecology, 21(8), 1789–1793. https://
doi .or g/10.1111/ j.13 65-294X. 20 12.055 42.x
Talhinhas, P., Batist a, D., Diniz, I. , Vieira, A ., Silva, D. N. , Loureiro, A .,
… Várzea, V. (2017). Th e Coffee Leaf R ust pathogen Hemileia vas-
tatrix: One and a half centuries around the tropic s. Molecular Plant
Pathology, 18(8), 1039–1051.
Taylor, D. L., Walters , W. A., L ennon, N. J., Bo chicchio, J., Kro hn, A.,
Caporaso, J. G., & Pennanen, T. (2016). Accurate estimation of fun‐
gal diver sity and abundance through improved lineage‐specific
primers optimized for Illumina amplicon sequencing. Applied
and Environmental Microbiology, 82(24), 7217–7226. https://doi.
org/10.1128/AEM.02576‐16
Tedersoo,L.,Anslan,S.,Bahram,M.,Põlme,S.,Riit,T.,Liiv,I.,…Bork,
P. (2015). Shotgun metagenomes and multiple primer pair‐bar‐
code combinations of amplicons reveal biases in metabarcod‐
ing analyses of fungi. MycoKeys, 10, 1. https://doi.org/10.3897/
mycokeys.10.4852
Thomas,A.C., Deagle,B.E., Eveson, J.P.,Harsch, C.H.,& Trites,A .W.
(2016). Quantitative DNA metabarcoding: Improved estimates of
species propor tional biomass using correction factors derived from
control material. Molecul ar Ecology Resources, 16(3),714–726.https://
doi .or g/10.1111/1755-0998.12490
Valentini,A.,Taberlet,P.,Miaud,C.,Civade,R.,Herder,J.,Thomsen,P.F.,
…Gaboriaud,C .(2016).Next- gene ra ti onmonitoringo faquaticb io di ‐
versityusingenvironmentalDNAmetabarcoding.Molecular Ecology,
25(4), 929–942.ht tps://doi .or g/10 .1111/m ec.13428
VanderMerwe,M .,Ericson, L., Walker, J., Thrall, P.H., & Burdon, J. J.
(2007). Evolutionary relationships among sp ecies of Puccinia and
Uromyces (Pucciniaceae, Uredinales) inferred from partial protein
coding gene phylogenies. Mycological Research, 111(2), 163–175.
https://doi.org/10.1016/j.mycres.2006.09.015
Vilgalys,R.,&Hester,M.(1990).Rapidgeneticidentific ationandmapping
ofenzymaticallyamplifiedribosomalDNAfromseveralCryptococcus
species. Journal of Bacteriology, 172(8), 4238–4246. htt ps://doi.
org /10.1128/jb.172 .8 .4238-4246.199 0
White, T. J., Bruns, T., Lee, S. J. W. T., & Taylor, J. L. (1990). Amplification
and direct sequencing of fu ngal ribosomal RNA genes for phylo‐
genetics. PCR Protocols: A Guide to Methods and Applications, 18(1),
315–322 .
Wood,J.R., Wilmshurst ,J.M.,Wagstaff,S.J.,Worthy,T.H., Rawlence,
N. J., & Cooper, A. (2012). High-resolution coproecology: Using
coprolites to reconstructthe habit s and habitats of New Zealand’s
extinct upland moa (Megalapteryx didinus). PLoS ONE, 7(6), e40025.
https://doi.org/10.1371/journal.pone.00 40025
Wood,S.N.(2001).mgcv:Gamsandgeneralizedridgeregressionforr.R
News, 1(2), 20–25.
Wood,S.N.(2012).Onp-valuesforsmoothcomponent sofanextended
generalized additive model. Biometrika, 100 (1), 221–228. https://doi.
org /10.1093/bio met /ass 04 8
Zhan, A ., Hulak, M., Sy lvester, F., Huang, X., A debayo, A. A. , Abbott,
C. L., … MacIsaac, H. J. (2013). High sensitivity of 454 pyrose‐
quencing for detection of rare species in aquatic communities.
Methods in Ecology and Evolution, 4(6), 558–565. https://doi.
org /10.1111/20 41-210X .12037
SUPPORTING INFORMATION
Additional suppor ting information may be found online in the
Suppor ting Information section at the end of the article.
How to cite this article:MakiolaA ,DickieIA,HoldawayRJ,
et al. Biases in the metabarcoding of plant pathogens using
rust fungi as a model system. MicrobiologyOpen.
2019;8:e78 0. https://doi.org/10.1002/mbo3.780
Available via license: CC BY
Content may be subject to copyright.