Dynamics of an experimental microbial invasionFrancisco Acostaa,b, Richard M. Zamora,b,1, Fares Z. Najarc,d, Bruce A. Roec,d, and K. David Hambrighta,b,2

aProgram in Ecology and Evolutionary Biology, Department of Biology, University of Oklahoma, Norman, OK 73019; bPlankton Ecology and LimnologyLaboratory, Department of Biology, University of Oklahoma, Norman, OK 73019; cAdvanced Center for Genome Technology, University of Oklahoma,Norman, OK 73019; and dDepartment of Chemistry and Biochemistry, University of Oklahoma, Norman, OK 73019

Edited by Daniel S. Simberloff, The University of Tennessee, Knoxville, TN, and approved August 3, 2015 (received for review March 14, 2015)

The ecological dynamics underlying species invasions have been amajor focus of research in macroorganisms for the last five decades.However, we still know little about the processes behind invasion byunicellular organisms. To expand our knowledge of microbial in-vasions, we studied the roles of propagule pressure, nutrient supply,and biotic resistance in the invasion success of a freshwater invasivealga, Prymnesium parvum, using microcosms containing naturalfreshwater microbial assemblages. Microcosms were subjected to afactorial design with two levels of nutrient-induced diversity andthree levels of propagule pressure, and incubated for 7 d, duringwhich P. parvum densities and microbial community compositionwere tracked. Successful invasion occurred in microcosms receivinghigh propagule pressure whereas nutrients or community diversityplayed no role in invasion success. Invaded communities experienceddistinctive changes in composition compared with communitieswhere the invasion was unsuccessful. Successfully invaded microbialcommunities had an increased abundance of fungi and ciliates, anddecreased abundances of diatoms and cercozoans. Many of thesechanges mirrored the microbial community changes detected duringa natural P. parvum bloom in the source system. This role of propa-gule pressure is particularly relevant for P. parvum in the reservoir-dominated southern United States because this species can formlarge, sustained blooms that can generate intense propagule pres-sures for downstream sites. Human impact and global climatechange are currently causing widespread environmental changes inmost southern US freshwater systems that may facilitate P. parvumestablishment and, when coupled with strong propagule pressure,could put many more systems at risk for invasion.

microbial ecology | diversity | invasion resistance | propagule pressure |Prymnesium

Microbial species invasions, thought to occur worldwide interrestrial and aquatic systems and involve both patho-

genic and free-living taxa, represent an emerging challenge toour understanding of the interplay between biodiversity andecosystem function, particularly under pressures of global envi-ronmental change (1). Despite assertions of limitless dispersalcapability, sensu Baas-Becking (2), many microorganisms seemnot to be cosmopolitan, and biogeographic studies suggest asignificant effect of ecological drift and dispersal limitation intheir distributions (3, 4).Invasions of microbial species are hard to track because small,

inconspicuous species are routinely overlooked in most assess-ments of invasive species (5) and are detected only when theyhave conspicuous impacts, such as the formation of blooms. In-vasions from pathogenic microbes have been relatively well-studied because they are comparatively easy to track, in part, dueto strong, observable impacts (1). A particularly well-studiedexample is the invasion of the human gut microbiota by patho-genic bacteria (6, 7). We know, however, much less about in-vasion by nonpathogenic microbes, even though we have evidencethat they do occur in nature. For example, in the last two decades,a number of invasions by aquatic microbial species have beendocumented (8–11). Although there are some studies on the eco-logical impacts of these invasions (8), we still know little about theirpotential consequences.

Traditional macroorganism-based ecological theory suggeststhat invasion success tends to be highest in exotic species char-acterized by high dispersal abilities, growth rates, and resourceefficiencies (12), and in native communities characterized by lowspecies diversity and high disturbance levels, including fluctua-tion of resources (13). Once an invasion occurs, it can inducesignificant change in the invaded community, including modifi-cation of community structure and loss of species and ecosystemfunction (13).Although the applicability of macrobial principles to microbial

systems is debatable (1, 4), they constitute a general framework ofstudy and readily testable hypotheses, such as whether microbialinvasions are limited by dispersal or whether high diversity innative communities provides resistance to invasion by exotics.Indeed, few available experimental studies of microbial invasionshave been designed to directly test such hypotheses, but, in-terestingly, they tend to suggest that interspecific interactions,rather than diversity per se, play critical roles in invasion resistance(14–16), although nutrient supply (17) and propagule pressure(18) have also been shown to facilitate microbial invasions.Unfortunately, there are limitations in these previous studies,

such as the use of artificial communities with very low speciesnumbers. By not taking into account the enormous diversity, vari-ability, and stochasticity inherent to natural communities, which

Significance

Current models for biological invasions are predominantlybased on macroorganisms. Few invasion model assumptionshave been validated for microbial systems. Further research onmicrobial invasion dynamics is required to determine whethermacrobial models are appropriate for microbes, as well as tounderstand present and future distributions of invasive mi-croorganisms, particularly in the face of contemporary envi-ronmental changes. We studied the establishment of aninvasive protist in natural microbial assemblages in replicateexperimental microcosms and found that, under adequate en-vironmental conditions, invasion success was determined bythe number of invading propagules rather than resourceavailability and the diversity of the invaded communities. Thisstudy is among the first to test invasibility hypotheses using anactual invasive microbial species in natural communities.

Author contributions: R.M.Z. and K.D.H. designed research; R.M.Z. performed research;F.Z.N. and B.A.R. contributed new reagents/analytic tools; F.A., R.M.Z., and K.D.H. ana-lyzed data; F.A. and K.D.H. wrote the paper; and R.M.Z., F.Z.N., and B.A.R. conductedcommunity sequencing.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The sequences reported in this paper have been deposited in the NationalCenter for Biotechnology Information (NCBI) Sequence Read Archive under BioProjectPRJNA271537 and Biosamples SAMN03274828–SAMN03274857.1Present address: Ecosystems and Lake Management, Grand River Dam Authority, Langley,OK 74350.

2To whom correspondence should be addressed. Email: dhambright@ou.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1505204112/-/DCSupplemental.

11594–11599 | PNAS | September 15, 2015 | vol. 112 | no. 37 www.pnas.org/cgi/doi/10.1073/pnas.1505204112

can have thousands of interacting bacterial and protistan species, aswell as viruses, it is unclear how applicable these earlier experi-mental studies are with respect to actual invasion dynamics. Ad-ditionally, these studies have used arbitrarily selected invaders thatmight lack the specific ecological and life history traits that arethought to facilitate microbial invasion (12).A potential model organism for studying these processes is the

golden alga, Prymnesium parvum.Despite being of marine origin,P. parvum has apparently invaded and successfully established inUS freshwater ecosystems in the last 30 y and is currently presentin inland systems in up to 20 states (19). This species is remarkablyadaptable, being capable of growing in a wide range of tempera-ture and salinity conditions (20). It is a mixotroph, acting both as acompetitor of other algal species, and as a predator of bacteria(21) and other protists (22). It also produces toxic metabolites,which are thought to provide an advantage against competingspecies of algae (ref. 23; but see ref. 24). The generalist lifestyle,metabolic flexibility, and toxicity of P. parvum are among the maincharacteristics expected of an invasive species (1).Also contributing to the use of P. parvum as a model for

studying microbial invasions is a substantial body of literaturedealing primarily with ecotoxicological aspects of P. parvum (25,26). However, there are several recent observational (27–29) andexperimental field studies (30, 31) examining the potential rolesof immigration and nutrient availability in population dynamicsand impacts of P. parvum. Although they provide good first stepsin addressing the recent P. parvum range expansion, these earlierstudies are limited in design. The field studies are correlationaland provide good characterizations of P. parvum’s niche space,but they do not allow for inference into mechanisms or impactsof P. parvum population establishment. Background presence ofP. parvum and unknown initial microbial community diversity inthe field experiments prohibit isolation and identification ofpossible propagule pressure effects. Additionally, the field ex-periments do not address the response of invaded communitiesto successful invasion, nor are these studies placed within thebroader framework of ecological and invasion theory.In this study, we used a robust, replicated experiment com-

prising natural aquatic microbial communities to isolate the rolesof propagule pressure and community resistance to invasion inthe establishment success of P. parvum in an otherwise envi-ronmentally suitable habitat. We also characterized the responseof the invaded microbial community. We hypothesized that (i)microcosms with high nutrients and low diversity would be moreeasily invaded, having more resources available and potentially alower biotic resistance, (ii) increased propagule pressure wouldfacilitate invasion, and (iii) successful invasion of P. parvumwould cause distinctive changes in the richness and diversity ofinvaded microbial communities.

ResultsCommunity Diversity Manipulation. Microbial assemblages werecreated from composite mixtures of lake water collected fromthree different sites in Lake Texoma (Oklahoma and Texas,United States) during the summer non-P. parvum bloom season.Salinity and temperature were adjusted to 2.3 parts per thousand(ppt) and 15 °C to simulate the ambient winter lake conditionsconducive to P. parvum blooms (29). Three days later, commu-nity resistance to invasion was manipulated in half of the mi-crocosms by supplementing nitrogen and phosphorus levels toreduce community diversity and increase resource availability.Nutrient manipulation resulted in clear differences in experi-

mental microcosms after 5 d (day 8) of incubation (Table 1). Totalchlorophyll was higher [generalized linear model (GLM), F-test,P < 0.001] whereas richness and alpha diversity of eukaryoticcommunities were lower (GLM, F-test, P = 0.008 and P < 0.001,respectively) (Table S1) in microcosms that received nutrient ad-ditions. Communities contained 6,543 bacterial and 1,079 eukaryoteoperational taxonomic units (OTUs, here defined as rRNA se-quences with a 97% similarity), with community composition sig-nificantly affected by nutrients [permutational multivariate analysisof variance (PERMANOVA), F-test, P < 0.012 for both commu-nities] (Table S2). Nonmetric multidimensional scaling (NMDS)ordination showed that microcosm assemblages for both eukaryotesand bacteria clustered by nutrient treatment (Fig. S1, blue symbols).It was under these conditions in which P. parvum was inoculatedinto the microcosms (day 8).

Invasion Success. Propagule pressure was manipulated by addingP. parvum from laboratory cultures at three invading populationsizes: 12,860,000 cells (6,430 cells per mL), 1,286,000 cells (643cells per mL), and 128,000 cells (64 cells per mL). P. parvumestablishment success was unaffected by the diversity and rich-ness of the receiving microbial eukaryote community but wasdirectly proportional to propagule pressure (P < 0.001) (Fig. 1and Table S3). After addition of the highest propagule pressure,P. parvum population mean (±SD) density was 7,764 ± 1,616cells per mL on day 15. In the medium propagule treatment,P. parvum was still present in three microcosms at very low densities(mean = 333 ± 422 cells per mL) on day 11, but remained in onlyone microcosm on day 15 (mean = 14 ± 34 cells per mL). In the lowpropagule pressure treatment, P. parvum was detected in one mi-crocosm on day 11 but, by day 15, had become undetectable in allsix microcosms. Therefore, high propagule pressure overwhelmedany effects of reduced resistance in the form of increased nutrientsand lower community diversity.

Effects of Invasion in the Microbial Community. The microbial com-munities at the end of the experiment (day 15) for all microcosmscombined contained 7,691 bacteria and 1,064 eukaryote OTUs.

Table 1. Chlorophyll and diversity conditions for microcosmsbefore inoculation of propagule treatments

IndicatorLow nutrienttreatment

High nutrienttreatment P

Chlorophyll a 32.2 ± 2.84 98.3 ± 7.02 <0.001Eukaryotic richness 703 ± 12.4 593 ± 37.2 0.008Eukaryotic alpha diversity 60.1 ± 4.49 28.4 ± 4.34 <0.001Bacterial richness 4162 ± 197 3892 ± 489 0.425Bacterial alpha diversity 479 ± 35.6 408 ± 85.6 0.259

Values are mean ± SD; n = 9 per treatment for chlorophyll (μg/L), n = 3 pertreatment for richness (Chao1) diversity (inverse Simpson) estimates. Boldfont indicates a significant difference between low and high nutrient treat-ments; for more detail on statistical analyses, consult Table S1.

Log

P. p

arvu

m d

ensi

ty, c

ells

/mL

0 8 11 15

1

2

3

4

5

Day

Fig. 1. Population densities of P. parvum in microcosms during the exper-iment in the three invading propagule pressure treatments (high, 6,430 cellsper mL, red stars; medium, 640 cells per mL, blue squares; and low, 64 cellsper mL, gray circles). Points for days 11 and 15 are offset for clarity.

Acosta et al. PNAS | September 15, 2015 | vol. 112 | no. 37 | 11595

ECOLO

GY

Although the nutrient treatments still were evident (Fig. 2 andTable S4) (linear mixed-effects model, nutrient effect, P < 0.02 forbacterial diversity and eukaryote richness and diversity), taxo-nomic richness and diversity were substantially lower on day 15compared with day 8 across all propagule pressure and nutrienttreatments (compare Table 1 and Fig. 2; Table S4) (time effect, allP ≤ 0.02). There was no effect of propagule pressure nor anyinteraction between propagule pressure and nutrient treatment onthe diversity or richness of any of the communities (Table S4)(propagule effect, all P > 0.05). Communities on days 8 and 15were clearly different, not only in diversity values, but also incommunity composition, as visualized by NMDS ordination (Fig.S1), regardless of treatment.Day-15 eukaryote and bacteria communities were different across

nutrient and propagule pressure treatments (Fig. 3 and Table S5)(PERMANOVA, F-test, all P < 0.013), but no interaction betweentreatments was detected. Among the propagule pressure treatments,the successfully invaded microcosms (i.e., the high propagule pres-sure treatment) clustered separately from all other treatments (grayellipses in Fig. 3) (Table S6) (PERMANOVA post hoc analysis, P ≤0.047 for all pairwise comparisons with the high propagule pressuretreatment). Control communities, which received no P. parvum in-ocula or nutrient additions, did not differ in composition fromnoninvaded communities (Fig. 3, orange squares). In the case of theeukaryotic communities, the inoculation of P. parvum cells as part ofthe propagule pressure treatment was not the source of significantdifferences. Analysis of community composition after removing allOTUs identified from the order Prymnesiales revealed no differ-ences from analysis of the complete communities in terms of NMDSordination or grouping.

Characteristic Species.Regarding community composition, changes inthe abundances of eukaryotic phyla (rank 3) or bacterial classes

(Fig. S2) were limited to higher proportions of fungi and ciliateOTUs in invaded microcosms and more diatom and cercozoanOTUs in the noninvaded microcosms. However, indicator valueanalysis (32) revealed that the primary distinction between in-vaded and noninvaded communities was driven by changes inabundances of taxa at lower taxonomic levels and specific OTUs(Tables S7 and S8). Noninvaded communities were character-ized by 26 eukaryotic and 14 bacterial OTUs. The most abundantindicator eukaryotes in noninvaded communities were diatoms(seven OTUs) comprising mostly centric diatoms from the sub-phyla Bacillariophytina and Coscinodiscophytina. Ciliates, chlor-ophytes, and dinoflagellates were also present (six, four, and fourOTUs, respectively). Remaining groups in noninvaded communitiesincluded cercozoans, centrohelids, chrysophytes, and an unknowneukaryotic OTU. Indicator bacterial taxa were predominantly fromthe phylum Actinobacteria (10 OTUs), mostly from the orderActinomycetales. Other indicator bacterial OTUs were from thephyla Cyanobacteria, Planctomycetes, and Chloroflexi.By contrast, invaded communities were characterized by 20

eukaryotic and 18 bacterial OTUs. Indicator eukaryotes in invadedcommunities included 11 ciliate OTUs from diverse families,chrysophytes from the genera Paraphysomonas and Ochromonas(six OTUs), one pirsonid, and two apicomplexans. Bacterial OTUsincluded the phyla Proteobacteria (13 OTUs from the classesAlphaproteobacteria and Deltraproteobacteria) and Bacteroidetes(five OTUs from the class Flavobacteriales).

DiscussionWe studied an experimental P. parvum invasion in natural mi-crobial assemblages from Lake Texoma, a reservoir with a his-tory of seasonal P. parvum blooms. To characterize the param-eters that drive the successful establishment of this invasivespecies, we manipulated nutrient supply and propagule pressure,

2300

2900

3500

4100

4700

250

350

450

550

650

0 1 2 3 4

Bacterial richness

Euka

ryot

ic ri

chne

ss

Propagule pressure

Euk.Bac.

200

300

400

0

10

20

30

0 1 2 3 4

Bacterial diversity

Euka

ryo�

c di

vers

ity

Low HighNutrients

L M HA

B

Fig. 2. Community conditions of experimental microcosms at the end of theexperiment (day 15) showing (A) taxonomic richness and (B) alpha diversityfor eukaryotes (blue symbols) and bacteria (red symbols) for three propagulepressure treatments: low (circles), medium (triangles), and high (crosses).Richness (Chao1) and diversity (Simpson’s reciprocal) of microbial commu-nities were calculated using 97%-similarity OTU abundances.

A

B

Fig. 3. Composition of (A) eukaryotic and (B) bacterial communities in mi-crocosms at the end of the experiment (day 15) visualized using NMDS ordi-nation. Colors indicate low (blue), medium (green), or high (yellow) propagulepressure; shape indicates either low (triangles) or high (circle) nutrient treat-ments. Communities from control bottles appear as orange squares. High(High PP) and all remaining (Non-high PP) propagule pressure treatments aresurrounded by 95% confidence dispersion ellipses. NMDS stress values are0.145 for bacteria and 0.160 for eukaryotes.

11596 | www.pnas.org/cgi/doi/10.1073/pnas.1505204112 Acosta et al.

allowing us to elucidate the intrinsic importance of each of thesefactors and their potential interactions as they pertain to theinvasion process.Within experimental microcosms, invasion success was de-

termined exclusively by propagule pressure, with successful in-vasions occurring only in high propagule pressure microcosms,which had increased P. parvum cell densities at the end of theexperiment (day 15). Conversely, microcosms with failed invasionsshowed no presence of P. parvum cells, meaning that the alga waseither absent or present at abundances below our detection limitof ∼200 cells per mL. In these cases, it seems that resistance toinvasion of the receiving community was strong enough to preventestablishment and growth of the initial inoculum.In contrast to our initial hypotheses and general invasion theory,

nutrient treatments had no effect on invasion success, even thoughthis disturbance increased the gross resource supply and yielded areduced number and diversity of species, both factors that arethought to facilitate invasions (1, 13). This lack of an effect ofincreased resource availability may reflect the fact that our systemis eutrophic and that ambient nutrient availability was alreadyhigh. Alternatively, it might be attributed to the metabolic flexi-bility of P. parvum, which can obtain energy both autotrophicallyand heterotrophically from a wide variety of sources (22).Bottle effects may have altered available niche space and

influenced the response of the communities to treatments (asevidenced by the effect of time on community composition inFig. S1), possibly causing the observed dissociation betweendiversity and invasibility. However, previous experiments onP. parvum invasion processes showed that communities in large(1,570-L) enclosures and small (2-L) microcosms presented nosignificant differences in plankton dynamics for longer periodsthan those encompassed by our experiment (30). Furthermore,the difference between days was probably also caused by me-dium-term responses of the community to initial environmentalchanges because stabilization of communities in response todisturbances is a continuous process that can extend for longerperiods than those covered by our incubation period (33).Our results also are at odds with those of previous experiments

that found that P. parvum abundances were correlated with nu-trient concentrations, but not with additions of propagules (30).In that experiment, however, initial communities had differingbackground levels of P. parvum cells, and the number of intro-duced cells was low, equivalent to our low propagule pressuretreatment, which, as shown here, did not exert enough propagulepressure to become established in the community. Thus, themain differences between treatments in that experiment werelikely driven by the response of the background levels of P. parvumto the nutrient additions.Studies in macroorganisms have predicted that one of the con-

sequences of invasion is species loss (13), a phenomenon not ob-served in our microcosms, where diversity and richness were notdifferent across propagule pressure treatments on day 15. How-ever, we observed a strong effect of propagule pressure in thecommunity composition. The use of next generation sequencingallowed us to characterize the effects of this community responsewith greater precision and less bias than previous studies, whichmeasured community composition using either microscope iden-tification (34) or quantification of photosynthetic pigments (30, 35).Both of these methods yield a skewed version of community com-position, with a bias toward large, well-characterized taxa clusteredinto broad taxonomic categories (e.g., nanoflagellates). Comparedwith these studies, effects that we observed were subtler, with pri-mary effects in the abundances of specific OTUs rather than inabundances of entire higher level taxonomic groups.Analysis of indicator species showed that a wide diversity of

freshwater microbial taxa responded to invasion. Most of thesechanges were in agreement with previous observations inP. parvum invasions, such as the reduction in abundances of

diatoms, cercozoans, and chlorophytes (30, 34, 36, 37) and theincreased abundances of chrysophytes (37). We also observed ageneral increase in fungal OTU number, which has also beenobserved in P. parvum blooms in Lake Texoma (37). Ciliates canbe both prey and predators of P. parvum (38), and the presence ofdistinctive but closely related ciliate OTUs in both invaded andnoninvaded microcosms suggests that species within this groupcan have opposite responses to the community changes inducedby the invasion of P. parvum.Although there are few previous studies on microbial com-

munities associated with P. parvum blooms, the most abundantindicator species for high propagule pressure communities weremembers of Alphaproteobacteria and Bacteroidetes, groups thatare predominant in the bacterial communities attached to algal cells(39). Non-high propagule pressure communities were enriched withthe order Actinomycetales, a ubiquitous and abundant group oflimnetic bacteria (40), whose abundance also was found to de-crease during P. parvum blooms (37).Overall, even if none of our microcosms reached the high

densities that characterize the dense nearly monospecific bloomsthat are the usual focus of P. parvum studies (25), the observedcommunity changes detected agree with previous observationalresearch in Lake Texoma. There, a year-long study found that aP. parvum bloom was associated with a drastic disruption in thenormal pattern of microbial community seasonal succession andwith community changes in microbial components of the plank-tonic assemblage (27, 37) that mirror those seen in this study.At first glance, our results seem to contradict previous findings

of a primary role for environmental conditions in P. parvumblooms in Lake Texoma, regardless of dispersal of propagules(29). In this reservoir, winter blooms of P. parvum have occurredalmost yearly since 2003, but only in the western arm of the res-ervoir, characterized by relatively high salinities (28, 29). Althoughstill presumably subjected to downstream transport of propagulesfrom these large, sustained blooms, which can reach densities over2 × 105 cells per mL, the rest of the lake has not experienced anyestablishment of P. parvum (defined as recurrent winter blooms).However, it is important to keep in mind that, before propaguleexposure, we manipulated our microcosms’ salinity levels to mimicthose thought to be conducive to P. parvum establishment: i.e.,∼2 ppt (29, 41), effectively creating a niche space for P. parvum.The role of salinity is particularly important, P. parvum being amarine species that is already at the limit of its salinity tolerance inthis reservoir (41) and with most recorded blooms occurring whensalinity exceeds 1.7 practical salinity units (psu) (29).Based on our studies, it is clear that environmental conditions

are primary factors determining P. parvum distributions and bloomformation. However, once these conditions are met, the intensity ofdispersal (measured as propagule pressure) becomes a determiningfactor in establishment success and biogeographic patterns of thisinvasive species. It is important to note that the temperature, nu-trient, and salinity conditions present in our microcosms fall wellwithin the ranges found naturally in Lake Texoma and otherfreshwater systems in the southern United States and that theseparticular conditions are projected to increase in frequency in thecoming decades (42, 43), creating situations that, in tandem withthe already existing high propagule supply, could drive furtherrange expansions and blooms of P. parvum.Our results failed to support several of our initial hypotheses

and seemingly contradict common assumptions about invasionsin natural ecosystems. Increases in nutrient supply and decreasesin community diversity did not facilitate P. parvum invasions and,at least within the ranges used in this experiment, had no influ-ence in invasion success. Instead of resource availability, invasi-bility was determined by propagule pressure, and microcosmswithin the same propagule treatment presented similar outcomesin terms of establishment success. Although the important roleof propagule pressure is in agreement with previous observations

Acosta et al. PNAS | September 15, 2015 | vol. 112 | no. 37 | 11597

ECOLO

GY

in terrestrial and aquatic ecosystems (44), the absence of anyobserved interaction between nutrient availability, biotic resistance,and propagule pressure as proposed in conceptual models (45)was unexpected.These results expand our knowledge of the general processes

behind microbial invasions of planktonic communities and vali-date results gleaned from synthetic ecosystems, such as the im-portant role of propagule pressure for invasion success (18) andthe lack of an intrinsic biotic resistance by more diverse com-munities (14, 15). Our experiment also reveals possible mecha-nisms behind the observed increase in the range of P. parvumpopulations within the United States and may provide a meansof identifying systems with high risk of invasion. P. parvumblooms can maintain densities over 1.5 × 105 cells per mL forprolonged periods of time (27), and many of these blooms areterminated through hydraulic flushing during periods of highinflow (27, 46), potentially exerting significant propagule pres-sure to downstream reservoirs and lakes (29, 47). Nevertheless,propagule supply remains an unknown factor in natural systems,even as efficient tools to measure P. parvum cell densities alonglake and riparian systems have been developed (48). We believethat the implementation of routine sampling programs is a fea-sible goal that not only will serve as an early-warning system forthe formation of P. parvum blooms but also may validate the roleof propagule pressure in ecosystem invasibility and provide in-formation for implementing accurate monitoring and successfulmanagement strategies.

Materials and MethodsExperimental Design. Experiments were conducted with natural freshwatermicrobial communities in microcosms. Samples from surface waters of threesites in Lake Texoma (sampling stations: L2, frequent winter P. parvumblooms; L4, P. parvum present in winter; and L6, P. parvum rarely observed)(28) were retrieved, prefiltered with a 63-μm mesh to remove large zoo-plankton, mixed, and distributed into 21 2.5-L experimental microcosms. Theabsence (here defined as below the detection limit of 200 cells per mL) ofP. parvum was confirmed by counting 10 hemocytometer fields at 200×magnification. Microcosms were incubated under artificial light in a 12-hlight:12-h dark cycle on a bottle roller at 15 °C for 3 d, and then salinity wasmanipulated (day 3) by additions of synthetic sea salt (Instant Ocean;Spectrum Brands; final salinity = 2.3 ppt) to simulate ambient winter lakeconditions, which are conducive to P. parvum blooms (29). Nutrients wereadded to nine microcosms on day 3 to generate high nutrient (0.199 mg/Lphosphorus, 1.25 mg/L nitrogen; n = 9) and low (ambient) nutrient (0.069 mg/Lphosphorus, 0.79 mg/L nitrogen; n = 9) treatments. We also maintained acontrol treatment (n = 3) in which neither nutrients nor propagule pressurewere manipulated. All microcosms were then incubated for 5 d under the samelate winter and early spring temperature and light conditions to allowmicrobialcommunities to respond to nutrients.

On day 8, bottles were divided into low, medium, and high propagulepressure treatments by adding cells from cultures of P. parvum isolated fromLake Texoma (strain UOBS-LP0109). Culture cell densities were estimatedusing the hemocytometer method described above, and samples were di-luted to achieve the desired cell concentrations. Each nutrient × propagulepressure treatment combination was triplicated.

At the beginning of the invasion experiment (day 8), a 200-mL samplewas taken from three low nutrient and three high nutrient microcosmsfor genetic material extraction and sequencing. Afterward, microcosmswere incubated for 7 d, at the end of which (day 15) 250-mL samples weretaken from all microcosms for genetic material extraction and sequencing.Chlorophyll a was measured on days 8 and 15, with a TD 700 bench-topfluorometer.

DNA Extraction and Sequencing. DNA was extracted using a standard phenol/chloroform extraction protocol. Sequences encompassing variable regions ofthe ribosomal SSU gene were used as genetic markers for microbial com-munities. For bacteria, we amplified the SSU v6 region using the 967F and1064R primers (49, 50), and, for eukaryotes, we amplified the SSU v9 regionwith the 1380F and 1510R primers (51). Amplicon sequencing was done in a454 GS FLX Titanium system in the University of Oklahoma Advanced Centerfor Genome Technology (ACGT) (52, 53). Sequences obtained were quality

checked; sequences with divergences from the primers or with an averagequality <25 in a 50-bp sliding window were discarded. Minimum length fortrimmed sequences was >50 bp for bacteria and >120 bp for eukaryotes.Chimeric sequences were detected using the ChimeraSlayer algorithm ofthe QIIME 1.8.0 software (54) and removed from the database. Sequenceswere uploaded to the NCBI Sequence Read Archive, under BioProjectPRJNA271537 and Biosamples SAMN03274828–SAMN03274857.

Community Analyses. Communities were analyzed using QIIME; OTUs weredetermined de novo using a 97% similarity threshold. Taxonomic assignmentof bacterial OTUs was performed using the assign_taxonomy pipelinefrom QIIME, using the Greengenes database trees from August 2012 (55).Unclassified OTUs, singletons, and OTUs classified as archaea, eukary-otes, chloroplasts, or mitochondria were removed from the dataset.Taxonomic assignment of eukaryotic OTUs was performed using the SilvaNGS online platform (56) with default parameters. Because someremaining OTUs had high abundances in our dataset, they were manuallychecked using BLASTN (57) searches and assigned the best consensusclassification based on similarity to available 18S sequences. Unclassified,bacterial, archaeal, and metazoan OTUs and singletons were removedfrom the dataset.

Community data were exported as a BIOM file, imported into R 3.1.0 (58),and analyzed using phyloseq v1.8.1 (59). Before community analyses, all li-braries were normalized to the size of the smallest library using randomsampling without replacement; a random number generator seed was se-lected a priori to ensure reproducibility.

Alpha diversity on days 8 and 15 was measured using the reciprocalSimpson’s index whereas richness was estimated using Chao1. Differencesbetween community composition of each microcosm were visualized usingnonmetric multidimensional scaling (NMDS) with phyloseq. Beta diversitywas calculated with the Bray–Curtis index using vegan v2.0–10 (60). Detec-tion and measurement of P. parvum cells in the microcosms were done atthe end of the experiment, using hemocytometer-based microscopy asdescribed above.

Statistical Analyses. All statistical analysis were conducted in R. To analyze thedifferences in diversity, richness, and chlorophyll in microcosms on day 8 (thebeginning of the invasion experiment), we used a generalized linear modelwith a Gaussian distribution, using nutrient treatment as the independentvariable. At the end of the experiment (day 15), we analyzed the same di-versity estimators for all microcosms using a linear mixed-effects model withday, nutrient, propagule pressure, and the interaction between nutrients andpropagule pressure as fixed factors, and microcosm identity as a randomfactor; both analyses were done with nlme v3.1–117 (61).

To test for differences in P. parvum cell concentrations at the end of theexperiment, we used a generalized linear model, using a quasi-poisson dis-tribution to account for overdispersion (62). Assumptions of normality andheterogeneity of variance for the residuals were checked using R plottingfunctions. The effects of propagule pressure and nutrient supply on micro-bial community composition were tested with a permutational multivariateanalysis of variance (PERMANOVA) using Bray–Curtis distance matrices withvegan. Post hoc comparisons between high and non-high propagule pres-sure communities were done using pairwise PERMANOVAs with Holm–

Bonferroni-adjusted probability values.For determination of indicator species, low-abundance OTUs with <10

sequences for eukaryotes and <20 sequences for bacteria were removedbecause they were unlikely to contribute to significant differences betweenthe different groups. Using these modified OTU libraries, we calculatedthe indicator value d of each OTU as the product of the relative fre-quency and relative average abundance in groups, using labdsv v.1.6–1 (63),and adjusted P values for multiple comparisons using the false discoveryrate control.

ACKNOWLEDGMENTS. We thank J. D. Easton, A. C. Easton, K. L. Glenn, andJ. E. Beyer (Plankton Ecology and Limnology Laboratory) and Y. Xing,L. Zhou, K. Wang, and H. Lau (Advanced Center for Genome Technology) fortechnical assistance, and A. C. Jones and D. A. Caron for helpful discussionsduring the design and implementation of the experiment. Commentsfrom two anonymous reviewers significantly improved this paper. Finan-cial assistance for this project was provided by grants from the NationalScience Foundation (Grant DEB-1011454 to R.M.Z. and K.D.H.), the OklahomaDepartment of Wildlife Conservation (through the Sport Fish RestorationProgram) (Grant F-61-R to K.D.H.), the Oklahoma Water Resources ResearchInstitute (to K.D.H.), and the University of Oklahoma Office of the Vice Presi-dent for Research (to K.D.H. and B.A.R.).

11598 | www.pnas.org/cgi/doi/10.1073/pnas.1505204112 Acosta et al.

1. Litchman E (2010) Invisible invaders: Non-pathogenic invasive microbes in aquatic andterrestrial ecosystems. Ecol Lett 13(12):1560–1572.

2. de Wit R, Bouvier T (2006) ‘Everything is everywhere, but, the environment selects’;what did Baas Becking and Beijerinck really say? Environ Microbiol 8(4):755–758.

3. Fuhrman JA (2009) Microbial community structure and its functional implications.Nature 459(7244):193–199.

4. Hanson CA, Fuhrman JA, Horner-Devine MC, Martiny JBH (2012) Beyond biogeographicpatterns: Processes shaping the microbial landscape. Nat Rev Microbiol 10(7):497–506.

5. Ruiz GM, Carlton JT, Grosholz ED, Hines AH (1997) Global invasions of marine andestuarine habitats by non-indigenous species: Mechanisms, extent, and consequences.Integr Comp Biol 37:621–632.

6. Fuentes S, et al. (2014) Reset of a critically disturbed microbial ecosystem: Faecaltransplant in recurrent Clostridium difficile infection. ISME J 8(8):1621–1633.

7. Britton RA, Young VB (2014) Role of the intestinal microbiota in resistance to colo-nization by Clostridium difficile. Gastroenterology 146(6):1547–1553.

8. Gillis C-A, Chalifour M (2009) Changes in the macrobenthic community structurefollowing the introduction of the invasive algae Didymosphenia geminata in theMatapedia River (Québec, Canada). Hydrobiologia 647:63–70.

9. Reid PC, et al. (2007) A biological consequence of reducing Arctic ice cover: Arrival ofthe Pacific diatom Neodenticula seminae in the North Atlantic for the first time in800 000 years. Glob Change Biol 13:1910–1921.

10. Pertola S, Kuosa H, Olsonen R (2005) Is the invasion of Prorocentrum minimum (Di-nophyceae) related to the nitrogen enrichment of the Baltic Sea? Harmful Algae4:481–492.

11. Briand J-F, Leboulanger C, Humbert J-F, Bernard C, Dufour P (2004) Cylindrospermopsisraciborskii (Cyanobacteria) invasion at mid-latitudes: Selection, wide physiological toler-ance, or global warming? J Phycol 40:231–238.

12. Sakai AK, et al. (2001) The population biology of invasive species. Annu Rev Ecol Syst32:305–332.

13. Lockwood JL, Hoopes M, Marchetti M (2007) Invasion Ecology (Blackwell, Malden,MA), 1st Ed.

14. Robinson JV, Dickerson JE (1984) Testing the invulnerability of laboratory islandcommunities to invasion. Oecologia 61:169–174.

15. McGrady-Steed J, Harris P, Morin P (1997) Biodiversity regulates ecosystem pre-dictability. Nature 390:162–165.

16. He X, McLean JS, Guo L, Lux R, Shi W (2014) The social structure of microbial com-munity involved in colonization resistance. ISME J 8(3):564–574.

17. Jiang L, Morin PJ (2004) Productivity gradients cause positive diversity-invasibilityrelationships in microbial communities. Ecol Lett 7:1047–1057.

18. De Roy K, et al. (2013) Environmental conditions and community evenness determinethe outcome of biological invasion. Nat Commun 4:1383.

19. Hambright KD (2012) Golden algae & the health of Okla. lakes. Lakeline 2012(Fall):33–38.

20. Baker JW, et al. (2007) Growth and toxicity of Prymnesium parvum (Haptophyta) as afunction of salinity, light, and temperature. J Phycol 43:219–227.

21. Nygaard K, Tobiesen A (1993) Bacterivory in algae: A survival strategy during nutrientlimitation. Limnol Oceanogr 38:273–279.

22. Carvalho WF, Granéli E (2010) Contribution of phagotrophy versus autotrophy toPrymnesium parvum growth under nitrogen and phosphorus sufficiency and de-ficiency. Harmful Algae 9:105–115.

23. Driscoll WW, Espinosa NJ, Eldakar OT, Hackett JD (2013) Allelopathy as an emergent,exploitable public good in the bloom-forming microalga Prymnesium parvum.Evolution 67(6):1582–1590.

24. Remmel EJ, Hambright KD (2012) Toxin-assisted micropredation: Experimental evi-dence shows that contact micropredation rather than exotoxicity is the role ofPrymnesium toxins. Ecol Lett 15(2):126–132.

25. Granéli E, Edvardsen B, Roelke DL, Hagström JA (2012) The ecophysiology and bloomdynamics of Prymnesium spp. Harmful Algae 14:260–270.

26. Roelke DL, et al. (2015) A chronicle of a killer alga in the west: Ecology, assessment,and management of Prymnesium parvum blooms. Hydrobiologia, 10.1007/s10750-015-2273-6.

27. Jones AC, et al. (2013) Seasonality and disturbance: Annual pattern and response ofthe bacterial and microbial eukaryotic assemblages in a freshwater ecosystem.Environ Microbiol 15(9):2557–2572.

28. Hambright KD, et al. (2010) Temporal and spatial variability of an invasive toxigenicprotist in a North American subtropical reservoir. Harmful Algae 9:568–577.

29. Hambright KD, et al. (2015) The niche of an invasive marine microbe in a subtropicalfreshwater impoundment. ISME J 9(1):256–264.

30. Errera RM, et al. (2008) Effect of imbalanced nutrients and immigration on Prymne-sium parvum community dominance and toxicity: Results from in-lake microcosmexperiments. Aquat Microb Ecol 52:33–44.

31. Roelke DL, et al. (2007) Effects of nutrient enrichment on Prymnesium parvum pop-ulation dynamics and toxicity: Results from field experiments, Lake Possum Kingdom,USA. Aquat Microb Ecol 46:125–140.

32. Dufrêne M, Legendre P (1997) Species assemblages and indicator species: The needfor a flexible asymmetrical approach. Ecol Monogr 67:345–366.

33. Shade A, et al. (2012) Lake microbial communities are resilient after a whole-eco-system disturbance. ISME J 6(12):2153–2167.

34. Fistarol G, Legrand C, Granéli E (2003) Allelopathic effect of Prymnesium parvum on anatural plankton community. Mar Ecol Prog Ser 255:115–125.

35. Roelke DL, et al. (2010) Factors influencing Prymnesium parvum population dynamicsduring bloom initiation: Results from in-lake mesocosm experiments. J Am WaterResour Assoc 46:76–91.

36. Michaloudi E, Moustaka-Gouni M, Gkelis S, Pantelidakis K (2008) Plankton communitystructure during an ecosystem disruptive algal bloom of Prymnesium parvum. J PlanktonRes 31:301–309.

37. Jones AC (2012) Annual pattern and response of the bacterial and microbial eu-karyotic communities in an aquatic ecosystem restructured by disturbance. PhD dis-sertation (University of Southern California, Los Angeles, CA).

38. Rosetta CH, McManus GB (2003) Feeding by ciliates on two harmful algal bloomspecies, Prymnesium parvum and Prorocentrum minimum. Harmful Algae 2:109–126.

39. Kodama M, Doucette G, Green D (2006) Relationships between bacteria and harmfulalgae. Ecology of Harmful Algae, eds Granéli E, Turner JT (Springer, Berlin), 1st Ed, pp243–255.

40. Newton RJ, Jones SE, Eiler A, McMahon KD, Bertilsson S (2011) A guide to the naturalhistory of freshwater lake bacteria. Microbiol Mol Biol Rev 75(1):14–49.

41. Hambright KD, et al. (2014) Regulation of growth and toxicity of a mixotrophic mi-crobe: Implications for understanding range expansion in Prymnesium parvum.Freshw Sci 33:745–754.

42. Southard GM, Fries LT, Barkoh A (2010) Prymnesium parvum: The Texas experience.J Am Water Resour Assoc 46:14–23.

43. Roelke DL, et al. (2012) Anticipated human population and climate change effects onalgal blooms of a toxic haptophyte in the south-central USA. Can J Fish Aquat Sci 69:1389–1404.

44. Colautti RI, Grigorovich IA, MacIsaac HJ (2006) Propagule pressure: A null model forbiological invasions. Biol Invasions 8:1023–1037.

45. D’Antonio C, Levine J, Thomsen M (2001) Ecosystem resistance to invasion and therole of propagule supply: A California perspective. J Mediterr Ecol 2:233–245.

46. Roelke DL, et al. (2010) Hydraulic flushing as a Prymnesium parvum bloom-termi-nating mechanism in a subtropical lake. Harmful Algae 9:323–332.

47. Zamor RM, Franssen NR, Porter C, Patton TM, Hambright KD (2014) Rapid recovery ofa fish assemblage following an ecosystem disruptive algal bloom. Freshw Sci 33:390–401.

48. Zamor RM, Glenn KL, Hambright KD (2012) Incorporating molecular tools into routineHAB monitoring programs: Using qPCR to track invasive Prymnesium. Harmful Algae15:1–7.

49. Sogin ML, et al. (2006) Microbial diversity in the deep sea and the underexplored“rare biosphere”. Proc Natl Acad Sci USA 103(32):12115–12120.

50. Winsley T, van Dorst JM, Brown MV, Ferrari BC (2012) Capturing greater 16S rRNAgene sequence diversity within the domain Bacteria. Appl Environ Microbiol 78(16):5938–5941.

51. Amaral-Zettler LA, McCliment EA, Ducklow HW, Huse SM (2009) A method forstudying protistan diversity using massively parallel sequencing of V9 hypervariableregions of small-subunit ribosomal RNA genes. PLoS One 4(7):e6372.

52. Elshahed MS, et al. (2008) Novelty and uniqueness patterns of rare members of thesoil biosphere. Appl Environ Microbiol 74(17):5422–5428.

53. Wiley G, et al. (2009) Methods for generating shotgun and mixed shotgun/paired-endlibraries for the 454 DNA sequencer. Current Protocols in Human Genetics (Wiley,Hoboken, NJ), Chap 18, Unit 18.1, pp 18.1.1–18.1.21.

54. Caporaso JG, et al. (2010) QIIME allows analysis of high-throughput communitysequencing data. Nat Methods 7(5):335–336.

55. DeSantis TZ, et al. (2006) Greengenes, a chimera-checked 16S rRNA gene databaseand workbench compatible with ARB. Appl Environ Microbiol 72(7):5069–5072.

56. Quast C, et al. (2013) The SILVA ribosomal RNA gene database project: Improved dataprocessing and web-based tools. Nucleic Acids Res 41(Database issue):D590–D596.

57. Altschul SF, et al. (1997) Gapped BLAST and PSI-BLAST: A new generation of proteindatabase search programs. Nucleic Acids Res 25(17):3389–3402.

58. R Core Team (2014) R: A Language and Environment for Statistical Computing(R Foundation for Statistical Computing, Vienna). Available at www.R-project.org/.

59. McMurdie PJ, Holmes S (2013) phyloseq: An R package for reproducible interactiveanalysis and graphics of microbiome census data. PLoS One 8(4):e61217.

60. Oksanen J, et al. (2013) vegan: Community ecology package. Available at CRAN.R-project.org/package=vegan. Accessed April 12, 2015.

61. Pinheiro J, Bates D, DebRoy S, Sarkar D; R Core Team (2014) nlme: Linear and non-linear mixed effects models. Available at CRAN.R-project.org/package=nlme. AccessedApril 12, 2015.

62. Ver Hoef JM, Boveng PL (2007) Quasi-Poisson vs. negative binomial regression: Howshould we model overdispersed count data? Ecology 88(11):2766–2772.

63. Roberts DW (2013) labdsv: Ordination and multivariate analysis for ecology. Availableat CRAN.R-project.org/package=labdsv. Accessed April 12, 2015.

Acosta et al. PNAS | September 15, 2015 | vol. 112 | no. 37 | 11599

ECOLO

GY

Supporting InformationAcosta et al. 10.1073/pnas.1505204112

Fig. S1. Composition of (A) eukaryotic and (B) bacterial communities for microcosms at the time of Prymnesium parvum invasion (day 8) and at the end of theexperiment (day 15), visualized using NMDS ordination. Shape of dots indicates nutrient treatment whereas colors indicate day, as indicated in the legend.NMDS stress values are 0.043 for bacteria and 0.119 for eukaryotes.

0%

20%

40%

Perc

enta

ge o

f OTU

s

A) Eukaryo�c communi�es at Day 15

Non-high PP

High PP

0%

5%

10%

Perc

enta

ge o

f OTU

s

B) Bacterial communi�es at Day 15

Fig. S2. Composition of (A) eukaryotic and (B) bacterial communities for high and non-high propagule pressure (PP) treatments, using 97%-similarity OTUs.Eukaryotes are shown at the level of rank 3 (previously phylum) and bacteria are shown at the level of class. OTUs that constituted less than 0.2% of eachmicrocosm library have been omitted for clarity.

Acosta et al. www.pnas.org/cgi/content/short/1505204112 1 of 7

Table S1. Results for the generalized linear models (GLMs) testing the effects of nutrienttreatments in the richness, alpha diversity, and chlorophyll content of microcosms at day 8,before inoculation of propagule treatments

Treatment df Deviance Resid. df Resid. deviance F-value P

GLM—Effects of nutrients on bacterial richnessTotal model — — 5 665,166 — —

Nutrients 1 109,441 4 555,724 0.788 0.425GLM—Effects of nutrients on bacterial alpha diversity

Total model — — 5 24,646 — —

Nutrients 1 7,432.3 4 17,213 1.727 0.259GLM—Effects of nutrients on eukaryotic richness

Total model — — 5 21,126.4 — —

Nutrients 1 18,057 4 3,069.3 23.532 0.008 ↓GLM—Effects of nutrients on eukaryotic alpha diversity

Total model — — 5 1,589.82 — —

Nutrients 1 1,511.8 4 78.07 77.458 <0.001 ↓GLM—Effects of nutrients on chlorophyll concentration

Total model — — 20 138.427 — —

Nutrients 1 135.52 19 2.907 885.85 <0.001 ↑

Bold font indicates significant results. Arrows indicate whether the measured indicator increased (up arrow)or decreased (down arrow) with the nutrient treatment (high nutrients, compared with low nutrients). Dashesrepresent values that cannot be assigned to that component. Resid. df, Residual df; Resid. deviance, Residualdeviance.

Table S2. Results for the permutational multivariate analysis of variance (PERMANOVA)testing the effects of nutrient treatments in the composition of bacterial and eukaryoticcommunities at day 8, before inoculation of propagule treatments

Effects of nutrients in: df SS MS F-value R2 P

Bacterial communitiesNutrients 1 0.193 0.193 1.439 0.171 <0.001Residuals 7 0.940 0.134 — 0.830 —

Total 8 1.133 — — 1 —

Eukaryotic communitiesNutrients 1 0.185 0.185 5.169 0.425 <0.001Residuals 7 0.251 0.036 — 0.575 —

Total 8 0.436 — — 1 —

Community composition was calculated using 97%-similarity OTUs. Bold font indicates significant results.Dashes represent values that cannot be assigned to that component. MS, Mean of sum of Squares; SS, Sumof Squares. F-values obtained by permutation.

Table S3. Results for the generalized linear model (GLM) testingthe effects of nutrient and propagule pressure in the finalconcentration of P. parvum cells at the end of the experiment(i.e., day 15)

Treatment df Deviance Resid. df Resid. deviance P (>Chi)

Total model — — 17 103,235 —

Propagule pressure 2 101,316 15 1,919 <0.001Nutrients 1 0 14 1,919 1

Bold font indicates significant results. Dashes represent values that cannotbe assigned to that component. Resid. df, Residual df; Resid. deviance, Re-sidual deviance.

Acosta et al. www.pnas.org/cgi/content/short/1505204112 2 of 7

Table S4. Results for the linear mixed-effects models (LMEMs) testing the effects of time,nutrient, and propagule pressure treatments and their interactions in the richness, alphadiversity, and chlorophyll content of microcosms at the end of the experiment (i.e., day 15)

Effects of day, nutrients, andpropagule pressure on: Num. df Den. df F-value P

Eukaryotic richness(Intercept) 1 19 770.196 —

Time 1 2 306.31 0.003 ↓Nutrients 1 19 7.154 0.015 ↓Propagule pressure 3 2 4.292 0.195Nutrients × propagule pressure 3 2 0.731 0.622

Eukaryotic alpha diversity(Intercept) 1 19 939.275 —

Time 1 2 366.806 0.003 ↓Nutrients 1 19 49.113 <0.001 ↓Propagule pressure 3 2 1.934 0.359Nutrients × propagule pressure 3 2 16.837 0.057

Bacterial richness(Intercept) 1 19 3,514.135 —

Time 1 2 55.211 0.018 ↓Nutrients 1 19 0.551 0.467Propagule pressure 3 2 1.585 0.409Nutrients × propagule pressure 3 2 1.342 0.454

Bacterial alpha diversity(Intercept) 1 19 1,441.212 —

Time 1 2 93.929 0.011 ↓Nutrients 1 19 8.456 0.009 ↓Propagule pressure 3 2 1.123 0.503Nutrients × propagule pressure 3 2 1.091 0.511

Bold font indicates significant results. Arrows indicate whether the measured indicator increased (up arrow)or decreased (down arrow) with time (at day 15, compared with day 8) and with the nutrient treatment (highnutrients, compared with low nutrients). Dashes represent probability values that cannot be assigned to thatcomponent. Denom. df, denominator df; Num. df, Numerator df.

Table S5. Results for the permutational multivariate analysis of variance (PERMANOVA)testing the effects of nutrient and propagule pressure treatments in the composition ofbacterial and eukaryotic communities at the end of the experiment (i.e., day 15)

Effects of nutrients in: df SS MS F. model R2 P

Bacterial communitiesPropagule pressure 3 0.468 0.156 1.561 0.205 <0.001Nutrients 1 0.229 0.229 2.289 0.100 <0.001Propagule pressure × nutrients 2 0.191 0.095 0.954 0.083 0.596Residuals 14 1.399 0.100 — 0.612 —

Total 20 2.286 — — 1 —

Eukaryotic communitiesPropagule pressure 3 1.052 0.351 4.573 0.415 <0.001Nutrients 1 0.231 0.231 3.009 0.091 0.013Propagule pressure × nutrients 2 0.182 0.091 1.186 0.072 0.271Residuals 14 1.074 0.077 — 0.423 —

Total 20 2.539 — — 1 —

Community composition was calculated using 97%-similarity OTUs. Bold font indicates significant results.Dashes represent values that cannot be assigned to that component.

Acosta et al. www.pnas.org/cgi/content/short/1505204112 3 of 7

Table S6. Post hoc comparison of the composition of bacterial and eukaryotic communitiesbetween different propagule pressure treatments at the end of the experiment (i.e., day 15),using permutational multivariate analysis of variance (PERMANOVA)

Group 1 Group 2 F-Model R2 P Adjusted P

PERMANOVA—Post hoc comparison of bacterial communitiesControl Low 0.955 0.120 0.595 0.678Control Med 1.111 0.137 0.226 0.678Control High 1.795 0.204 0.012 0.047Low Med 1.043 0.094 0.276 0.678Low High 1.995 0.166 0.002 0.015Med High 1.720 0.147 0.002 0.015

PERMANOVA—Post hoc comparison of eukaryotic communitiesControl Low 1.672 0.193 0.048 0.144Control Med 1.837 0.208 0.071 0.144Control High 4.617 0.397 0.011 0.045Low Med 1.793 0.152 0.063 0.144Low High 8.793 0.468 0.002 0.013Med High 4.754 0.322 0.004 0.020

Community composition was calculated using 97%-similarity OTUs. P values were adjusted formultiple comparisonsusing a Holm–Bonferroni correction. Bold font indicates significant results.

Acosta et al. www.pnas.org/cgi/content/short/1505204112 4 of 7

Table

S7.

Indicativeeu

karyoticOTU

s,as

determined

usingtheDufren

e–Le

gen

dre

indicatorsp

eciesan

alysis

OTU

Ran

k1

Ran

k2

Ran

k3

Ran

k4

Ran

k5

Ran

k6

Ran

k7

Indicatorva

lue

Adjusted

Pva

lue

Non-highpropag

ule

pressure

trea

tmen

tsDen

ovo

2228

Archae

plastida

Chloroplastida

Chlorophyta

Chlorophycea

eCarteria

——

0.93

8271

60.00

9Den

ovo

167

Archae

plastida

Chloroplastida

Chlorophyta

Chlorophycea

eCarteria

——

0.76

7848

70.04

3Den

ovo

196

Archae

plastida

Chloroplastida

Chlorophyta

Treb

ouxiophycea

e-

——

0.79

7411

0.04

2Den

ovo

2920

Archae

plastida

Chloroplastida

Chlorophyta

Treb

ouxiophycea

ePa

rach

lorella

——

0.75

0957

90.02

7Den

ovo

925

Cen

trohelida

M1-18

D08

——

——

—0.96

5197

20.00

9Den

ovo

919

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Olig

ohym

enophorea

Peritrichia

0.89

3778

50.01

4Den

ovo

3439

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Olig

ohym

enophorea

Peritrichia

0.80

0266

50.04

2Den

ovo

1651

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Prostomatea

Coleps

0.74

4318

20.03

8Den

ovo

3009

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taSp

irotrichea

Choreotrichia

Tintinnopsis

0.8

0.03

8Den

ovo

326

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taSp

irotrichea

Hyp

otrichia

Halteria

0.86

9808

0.02

4Den

ovo

1650

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taSp

irotrichea

Hyp

otrichia

Halteria

0.85

7233

0.02

7Den

ovo

1369

SAR

Alveo

lata

Dinoflag

ellata

Dinophycea

ePe

ridiniphycidae

Gonya

ulacales

Ceratium

0.93

3333

30.00

9Den

ovo

3588

SAR

Alveo

lata

Dinoflag

ellata

Dinophycea

ePe

ridiniphycidae

Gonya

ulacales

Ceratium

0.93

3333

30.00

9Den

ovo

1335

SAR

Alveo

lata

Dinoflag

ellata

Dinophycea

ePe

ridiniphycidae

Gonya

ulacales

Ceratium

0.86

6666

70.01

4Den

ovo

3366

SAR

Alveo

lata

Dinoflag

ellata

Dinophycea

ePe

ridiniphycidae

Gonya

ulacales

Ceratium

0.8

0.02

7Den

ovo

806

SAR

Rhizaria

Cerco

zoa

——

——

0.98

6124

0.00

9Den

ovo

1201

SAR

Rhizaria

Cerco

zoa

Vam

pyrellid

aeVam

pyrella

——

0.90

2550

60.04

1Den

ovo

2148

SAR

Stramen

opile

sChrysophycea

eLG

01-09

——

—0.78

3464

60.03

8Den

ovo

354

SAR

Stramen

opile

sDiatomea

——

——

0.62

8119

40.01

9Den

ovo

432

SAR

Stramen

opile

sDiatomea

——

——

0.82

1684

0.04

2Den

ovo

1498

SAR

Stramen

opile

sDiatomea

Bacillariophytina

Med

iophycea

e—

—0.90

8377

90.01

6Den

ovo

343

SAR

Stramen

opile

sDiatomea

Bacillariophytina

Med

iophycea

eSk

eletonem

a—

0.86

6666

70.01

4Den

ovo

1271

SAR

Stramen

opile

sDiatomea

Bacillariophytina

Med

iophycea

eTh

alassiosira

—0.68

7861

30.01

0Den

ovo

1216

SAR

Stramen

opile

sDiatomea

Coscinodisco

phytina

Frag

ilariales

Syned

ra—

0.77

3869

30.04

2Den

ovo

3801

SAR

Stramen

opile

sDiatomea

Coscinodisco

phytina

Frag

ilariales

Syned

ra—

0.81

3675

20.04

3Den

ovo

673

Unkn

owneu

karyote

——

——

——

0.89

2367

90.00

9Highpropag

ule

pressure

trea

tmen

tsDen

ovo

1076

SAR

Alveo

lata

Apicomplexa

Conoidasida

Coccidia

Adeleo

rina

—0.91

2698

40.01

4Den

ovo

2321

SAR

Alveo

lata

Apicomplexa

Conoidasida

Cryptosporida

Cryptosporidium

—0.88

2352

90.02

9Den

ovo

1859

SAR

Alveo

lata

Ciliophora

——

——

0.86

1213

50.03

8Den

ovo

2968

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Olig

ohym

enophorea

Scutico

cilia

tia

0.80

1282

10.01

0Den

ovo

1373

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Olig

ohym

enophorea

Scutico

cilia

tia

0.90

1639

30.01

4Den

ovo

2923

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Olig

ohym

enophorea

Scutico

cilia

tia

0.79

3650

80.02

4Den

ovo

2150

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Olig

ohym

enophorea

Scutico

cilia

tia

0.77

6836

20.02

5Den

ovo

2657

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Olig

ohym

enophorea

CV1-2A

-17

0.63

8297

90.04

3Den

ovo

2394

SAR

Alveo

lata

Ciliophora

Intram

acronuclea

taConthreep

Prostomatea

Cryptocaryon

0.80

1282

10.01

4Den

ovo

3365

SAR

Alveo

lata

Ciliophora

Postcilio

desmatophora

Heterotrichea

——

0.64

3274

90.04

2Den

ovo

677

SAR

Alveo

lata

Ciliophora

Postcilio

desmatophora

Heterotrichea

Spirostomum

—0.79

7872

30.01

4Den

ovo

2602

SAR

Alveo

lata

Ciliophora

Postcilio

desmatophora

Heterotrichea

Spirostomum

—0.85

9213

30.02

4Den

ovo

2544

SAR

Alveo

lata

Ciliophora

Postcilio

desmatophora

Heterotrichea

Spirostomum

—0.88

3758

0.04

1Den

ovo

2863

SAR

Stramen

opile

sChrysophycea

e—

——

—0.73

8396

60.04

3Den

ovo

3633

SAR

Stramen

opile

sChrysophycea

eOch

romonad

ales

Och

romonas

Och

romonas

—0.66

6666

70.03

0Den

ovo

152

SAR

Stramen

opile

sChrysophycea

eOch

romonad

ales

Paraphysomonas

——

0.84

0.01

4Den

ovo

3735

SAR

Stramen

opile

sChrysophycea

eOch

romonad

ales

Paraphysomonas

——

0.86

3228

70.01

4

Acosta et al. www.pnas.org/cgi/content/short/1505204112 5 of 7

Table

S7.

Cont.

OTU

Ran

k1

Ran

k2

Ran

k3

Ran

k4

Ran

k5

Ran

k6

Ran

k7

Indicatorva

lue

Adjusted

Pva

lue

Den

ovo

3289

SAR

Stramen

opile

sChrysophycea

eOch

romonad

ales

Paraphysomonas

——

0.61

7284

0.04

4Den

ovo

1095

SAR

Stramen

opile

sChrysophycea

eOch

romonad

ales

Paraphysomonas

——

0.66

6666

70.02

7Den

ovo

3070

SAR

Stramen

opile

sPirsonida

Pirsonia

——

—0.78

0141

80.04

2

Theindicatorva

lueofea

chOTU

isdetermined

bytheproduct

oftherelative

freq

uen

cyan

drelative

averag

eab

undan

cein

each

propag

ule

pressure

trea

tmen

t.Th

ePva

lueisthefalsedisco

very

rate

(FDR)-

adjusted

probab

ility

ofobtainingas

highan

indicatorva

lueas

observed

ove

r1,00

0iterationsofthean

alysis.L

ow-abundan

ceOTU

s(those

hav

ingfewer

than

10sequen

cesam

ongallm

icroco

sms)

wereremove

dfrom

thean

alysis.Dashes

representtaxo

nomic

ranks

wherenoiden

tificationco

nsensusco

uld

bereached

fortheOTU

.

Acosta et al. www.pnas.org/cgi/content/short/1505204112 6 of 7

Table S8. Indicative bacterial OTUs, as determined using the Dufrene–Legendre indicator species analysis

OTU Phylum Class Order Family GenusIndicatorvalue

Adj.P value

Non-high propagule pressure treatmentsDenovo32742 Actinobacteria Acidimicrobiia Acidimicrobiales C111 — 0.8580645 0.024Denovo33315 Actinobacteria Actinobacteria Actinomycetales — — 0.8200717 0.034Denovo4218 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.9112426 0.006Denovo11127 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.8666667 0.010Denovo20884 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.9135802 0.016Denovo14048 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.8148148 0.024Denovo19018 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.9107143 0.024Denovo23164 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.8657005 0.024Denovo28240 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.8514851 0.024Denovo1320 Actinobacteria Actinobacteria Actinomycetales ACK-M1 — 0.8307692 0.034Denovo21463 Chloroflexi Anaerolineae Caldilineales Caldilineaceae — 0.6564885 0.043Denovo32663 Cyanobacteria Oscillatoriophycideae Oscillatoriales Phormidiaceae Planktothrix 0.6548507 0.015Denovo795 Cyanobacteria Synechococcophycideae Pseudanabaenales Pseudanabaenaceae — 0.6902655 0.016Denovo33080 Planctomycetes Planctomycetia Planctomycetales Planctomycetaceae Planctomyces 0.9107143 0.010

High propagule pressure treatmentsDenovo33043 Bacteroidetes Flavobacteriia Flavobacteriales Flavobacteriaceae — 1 0.006Denovo9881 Bacteroidetes Flavobacteriia Flavobacteriales Flavobacteriaceae — 0.8333333 0.014Denovo29919 Bacteroidetes Flavobacteriia Flavobacteriales Flavobacteriaceae — 0.8259912 0.015Denovo2352 Bacteroidetes Flavobacteriia Flavobacteriales Flavobacteriaceae — 0.8078231 0.016Denovo5439 Bacteroidetes Flavobacteriia Flavobacteriales Flavobacteriaceae Winogradskyella 0.6666667 0.044Denovo219 Proteobacteria Alphaproteobacteria Caulobacterales Caulobacteraceae — 0.9677419 0.006Denovo32145 Proteobacteria Alphaproteobacteria Caulobacterales Caulobacteraceae — 0.9668508 0.006Denovo6254 Proteobacteria Alphaproteobacteria Caulobacterales Caulobacteraceae — 0.9574468 0.006Denovo27327 Proteobacteria Alphaproteobacteria Rhizobiales — — 0.9863946 0.006Denovo9454 Proteobacteria Alphaproteobacteria Rhizobiales — — 0.9817352 0.006Denovo32791 Proteobacteria Alphaproteobacteria Rhodobacterales Rhodobacteraceae — 1 0.006Denovo11300 Proteobacteria Alphaproteobacteria Rhodobacterales Rhodobacteraceae — 0.7838284 0.034Denovo8546 Proteobacteria Alphaproteobacteria Rickettsiales Rickettsiaceae — 0.9922179 0.006Denovo15444 Proteobacteria Alphaproteobacteria Rickettsiales Rickettsiaceae — 0.8248731 0.015Denovo4158 Proteobacteria Alphaproteobacteria Rickettsiales Rickettsiaceae — 0.8333333 0.024Denovo11138 Proteobacteria Deltaproteobacteria Bdellovibrionales Bacteriovoracaceae — 0.8961474 0.006Denovo32148 Proteobacteria Deltaproteobacteria Bdellovibrionales Bacteriovoracaceae — 0.9375 0.006Denovo27723 Proteobacteria Deltaproteobacteria Bdellovibrionales Bacteriovoracaceae — 0.8995816 0.024

The indicator value of each OTU is determined by the product of the relative frequency and relative average abundance in each propagule pressuretreatment. The P value is the false discovery rate (FDR)-adjusted probability of obtaining as high an indicator value as observed over 1,000 iterations of theanalysis. Low-abundance OTUs (those having fewer than 20 sequences among all microcosms) were removed from the analysis. Dashes represent taxonomicranks where no identification consensus could be reached for the OTU.

Acosta et al. www.pnas.org/cgi/content/short/1505204112 7 of 7