SPATIO-TEMPORAL DISTRIBUTION OF MICROBIAL COMMUNITIES IN THE LAURENTIAN GREAT LAKES
Mark Jeremy Rozmarynowycz
A Dissertation
Submitted to the Graduate College of Bowling Green State University in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
December 2014
Committee:
R. Michael L. McKay, Advisor
William H. O'Brien Graduate Faculty Representative
George S. Bullerjahn
Scott O. Rogers
Zhaohui Xu
ii ABSTRACT
R. Michael L. McKay, Advisor
Freshwater microbial communities have received comparatively little attention compared to their marine counterparts, despite the importance of these systems. Using next-generation sequencing (Illumina itags), this study examined the microbial communities of the Laurentian
Great Lakes during both the summer-stratified period and during the winter. Additionally, the winter communities of the Laurentian Great Lakes were compared with the winter community of
Lake Onega, one of the largest freshwater lakes in Europe.
Winter communities were examined from 2010 through 2013. Lake Erie was examined during periods of high- (2010 and 2011) and low ice cover (2012). Lower ice-cover resulted in an 89% decrease in phytoplankton biovolume between years of expansive ice cover and nearly ice-free 2012. Principal coordinate analysis (PCoA) of UniFrac distance matrices revealed a strong separation between high-ice year 2010 and low-ice 2012, indicating a shift in microbial community structure. An examination of winter communities in 2013 from both Lake Erie and the upper Great Lakes revealed phylogenetically different communities for Lake Erie, Lake
Michigan, and the waters of the St Mary’s River. Samples from Lake Michigan and the Straits of Mackinac clustered with Lake Erie samples, which were correlated with concentrations of chloride and sulfate. The communities of the Laurentian Great Lakes were then compared with the communities of Lake Onega, and revealed strong differences in community structure.
Summer communities were examined from 2011 and 2012. A cruise from oligotrophic
Lake Superior to eutrophic Lake Erie revealed differences in community structure of the surface mixed layer. Concentrations of phosphorus and ammonium were correlated with the PCoA iii plots. A comparison of the surface waters of the upper Great Lakes with their oxygenated hypolimnions revealed a unique community in deep waters. This community had high abundances of Planctomycete and Chloroflexi reads, which were stable across spatially and temporally. Resampling in 2012 confirmed the stability of this community, and also examined cyanobacterial communities in both Lake Superior and Lake Erie. The community of Lake
Erie’s hypoxic ‘dead zone’ was also examined. iv
Dedicated to my wife, Clair
v ACKNOWLEDGMENTS
I would like to thank my advisor, Dr. R. Michael L. McKay, and the members of my committee, Dr. George S. Bullerjahn, Dr. Scott O. Rogers, Dr. Zhaohui Xu, and Dr. William H.
O'Brien. I would also like to thank Ben Beall, Ben Oyserman, Tim Davis, Caren Binding, Rick
Bourbonniere, Michelle Palmer, Nigel D’Souza, Steve Wilhelm, Matt Saxton, Euan Reavie, Bob
Sterner, Chip Small, Jacques Finlay, Sandy Brevold, Sue Watson, Michael Twiss, and Derek
Smith, as well as the Technical Operations personnel from Environment Canada. A special thanks to Nikolay Filatov, Sergey Komulaynen, Yuliya Slastina, Andrei Sharov, and everyone else at the Karelian Research Center of the Russian Academy of Sciences. I would also like to thank the officers and crews of CCGS Limnos, CCGS Griffon, USCGC Neah Bay, USCGC
Mackinaw, R/V Blue Heron, and R/V Lake Guardian.
This material is based upon work supported by the National Science Foundation under grant no. OCE- 0927512, 0927277, and 1230735 (RMLM, GSB). Additional support was provided by the Ohio Sea Grant College Program (grant R/ER-081 to RMLM and GSB), New
York Sea Grant (grant R-CE-29 to MRT), the Lake Erie Protection Fund (grant 430-12 to
RMLM) and the U.S. Environmental Protection Agency (grant GL-00E00790-2 to EDR). The work conducted by the U.S. Department of Energy Joint Genome Institute was supported by the
Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 and Community Sequencing Project 723 (RMLM, GSB, RAB).
vi
TABLE OF CONTENTS
Page
INTRODUCTION: HETEROTROPHIC BACTERIA OF FRESHWATER LAKES ...... 1
CHAPTER 1: ICE COVER EXTENT DRIVES MICROBIAL COMMUNITY STRUCTURE IN A LARGE NORTH-TEMPERATE LAKE: IMPLICATIONS FOR A WARMING CLIMATE ...... 8
1. INTRODUCTION ...... 8
1.1 MATERIALS AND METHODS ...... 9
1.1.1 STUDY SITE AND SAMPLING ...... 9
1.1.2 HIGH-THROUGHPUT MICROBIAL COMMUNITY ANALYSIS . 11
1.2 RESULTS AND DISCUSSION ...... 14
1.2.1 LAKE PHYSICO-CHEMICAL PROPERTIES ...... 14
1.2.2 PHYTOPLANKTON- AND MICROBIAL COMMUNITY SHIFTS
ASSOCIATED WITH ICE COVER ...... 15
1.2.3 ECOLOGICAL IMPLICATIONS OF LOW ICE COVER ...... 33
CHAPTER 2: TRANSITIONS IN MICROBIAL COMMUNITIES ALONG A 1,600 KM FRESHWATER TROPHIC GRADIENT ...... 35
2. INTRODUCTION ...... 35
2.1 MATERIALS AND METHODS ...... 37
2.1.1 STUDY SITE AND SAMPLING ...... 37
2.1.2 HIGH-THROUGHPUT MICROBIAL COMMUNITY ANALYSIS . 39
2.1.3 STATISTICAL ANALYSIS ...... 40
2.2 RESULTS AND DISCUSSION ...... 41
2.2.1 PHYSICO-CHEMICAL PROFILES OF SAMPLING LOCATIONS 41
2.2.2 MICROBIAL COMMUNITIES ...... 45
vii
2.2.3 PHYTOPLANKTON- AND MICROBIAL COMMUNITY SHIFTS
ASSOCIATED WITH LAKE TROPHIC STATE ...... 51
2.2.4 COMMUNITIES OF THE OXYGENATED HYPOLIMNION OF THE
UPPER GREAT LAKES ...... 57
2.2.5 RESAMPLING EFFORTS IN 2012 ...... 62
CHAPTER 3: SPATIO-TEMPORAL DYNAMICS OF WINTER MICROBIAL
COMMUNITIES IN LARGE FRESHWATER LAKES ...... 67
3. INTRODUCTION ...... 67
3.1 METHODS AND MATERIALS ...... 68
3.1.1 STUDY SITES AND SAMPLING ...... 68
3.1.2 HIGH-THROUGHPUT MICROBIAL COMMUNITY ANALYSIS . 70
3.1.3 STATISTICAL ANALYSIS ...... 72
3.2 RESULTS AND DISCUSSION ...... 73
3.2.1 ICE CONDITIONS AND SAMPLING ...... 73
3.2.2 ILLUMINA SEQUENCING RESULTS ...... 73
3.2.3 WINTER MICROBIAL COMMUNITIES OF LAKE ERIE ...... 78
3.2.4 MICROBIAL COMMUNITIES OF THE UPPER GREAT LAKES .. 80
3.2.5 UPPER GREAT LAKES VERSES LAKE ERIE ...... 83
3.2.6 LAKE ERIE CACHE SITES AND ICE SAMPLES ...... 90
3.2.7 MICROBIAL COMMUNITIES OF LAKE ONEGA ...... 91
CONCLUSIONS: ANTHROPOGENIC INFLUENCES ON FRESHWATER
MICROBIAL COMMUNITIES ...... 96
REFERENCES ...... 98
viii
APPENDIX A: TOP BLAST HITS FOR OTUS ABOVE 0.5% ABUNDANCE ...... 130
ix
LIST OF FIGURES
Figure Page
1 Mid-winter limnological surveys captured extremes of ice cover on Lake Erie ...... 10
2 Vertical water quality profiles for a representative central basin station (EC 1326)
occupied during winter surveys of Lake Erie...... 15
3 Phosphorus and silicate measurements ...... 16
4 Central basin phytoplankton chl a biomass ...... 17
5 Phytoplankton biomass accumulation during extremes of ice cover ...... 18
6 Rarefaction curves of observed species ...... 21
7 OTU abundance: Lake Erie 2010-12 ...... 23
8 Principal coordinate analysis: Lake Erie 2010-12 (unweighted) ...... 25
9 Principal coordinate analysis: Lake Erie 2010-12 (weighted) ...... 25
10 Maximum-likelihood tree of OTUs identified by supervised learning ...... 28
11 LDA effect size cladogram: Lake Erie 2010 and 2012 ...... 30
12 Sampling locations: July 2011 ...... 38
13 Rarefaction curves for bacteria and total reads: July 2011 ...... 46
14 Principal coordinate analysis (all samples) ...... 48
15 Principal coordinate analysis (surface mixed-layer) ...... 48
16 OTU abundance: July 2011 (surface mixed-layer) ...... 53
17 LDA effect size cladogram: July 2011 (surface mixed-layer) ...... 56
18 OTU abundance: July 2011 (hypolimnion) ...... 58
19 LDA effect size cladogram: July 2011 (hypolimnion) ...... 60
20 OTU abundance (2012 resampling) ...... 64
x
21 Sampling sites in the upper Great Lakes and Lake Erie (2013) ...... 69
22 Read abundance: Winter 2013 ...... 75
23 Rarefaction curves for bacterial OTUs ...... 76
24 OTU abundance: Lake Erie (winter 2013) ...... 79
25 OTU abundance: upper Great Lakes (winter 2013) ...... 81
26 Principal coordinate analysis: Laurentian Great Lakes (winter 2013) ...... 84
27 LDA effect size cladogram (Laurentian Great Lakes) ...... 88
28 OTU abundance: Lake Onega, Russia (winter 2013) ...... 92
29 Principal Coordinate Analysis of the Laurentian Great Lakes and Lake Onega ...... 94
30 LDA effect size cladogram: Laurentian Great Lakes and Lake Onega ...... 95 xi
LIST OF TABLES
Table Page
1 Sampling sites and physico-chemical parameters ...... 43
2 Alpha Diversity: July 2011 ...... 47
3 BEST analysis: July 2011 ...... 50
4 Mantel correlations: July 2011 ...... 50
5 Feature importance scores: surface mixed-layer (July 2011) ...... 56
6 Feature importance scores: hypolimnion (July 2011) ...... 60
7 Ice cover at sampling locations ...... 74
8 Alpha Diversity: Winter 2013 ...... 77
9 Chemical measurements for the Laurentian Great Lakes ...... 85
10 BEST and Mantel tests for the Laurentian Great Lakes samples (2013) ...... 86
11 Feature importance scores (Laurentian Great Lakes) ...... 88
12 Feature importance scores (Laurentian Great Lakes and Onega) ...... 95 1
INTRODUCTION: HETEROTROPHIC BACTERIA OF FRESHWATER LAKES
Bacteria have been recognized as integral components of all ecosystems, where they transform and cycle nutrients. While much attention has focused on marine microbial communities, little has been given to those of freshwater systems. Many freshwater systems suffer from nutrient loading and pollution stemming from anthropogenic inputs such as agricultural runoff. Increased nutrient loads can lead to eutrophication of such systems, the primary threat to ecological integrity (Schindler, 2006), and may lead to shifts in the composition of microbial communities. In freshwater communities, much of the focus has been on the
Cyanobacteria. Here we look at the heterotrophic members of freshwater communities, represented primarily by the Proteobacteria, Actinobacteria, Bacteroidetes, and members of the
PVC superphylum (Zwart et al., 2002; Newton et al., 2011)
The Proteobacteria are a highly diverse lineage, and represent the most studied of the bacterial phyla. This phylum contains 6 classes; the alpha-, beta-, delta-, gamma-, epsilon-, and zetaproteobacteria. Of these classes, the alpha-, beta-, and gammaproteobacteria are most often detected in freshwater and marine systems. Class Alphaproteobacteria, from which mitochondria are believed to have originated (Andersson et al., 1998), are ecologically important and are well known for nitrogen fixation in association with leguminous plants (Kersters et al.,
2006). Alphaproteobacteria are ubiquitous in freshwaters, although less numerous than in the oceans, where the SAR11 clade accounts for ~33% of surface water cells (Morris et al., 2002).
Members of the LD12 clade, a freshwater branch of the marine SAR11 clade, have since been discovered to be the most numerous of the Alphaproteobacteria in lakes (Logares et al., 2009), where they have adopted an oligotrophic lifestyle (Salcher et al., 2011). Along with their 2 efficient, but slow nutrient uptake, small size may be an adaptation to this lifestyle. It has been observed that increased grazing by size-selective protists resulted in an increase in
Alphaproteobacteria (Šimek et al., 1999; Salcher et al., 2005). Like Alphaproteobacteria,
Gammaproteobacteria are more abundant in marine systems than in freshwaters (Biers et al.,
2009). This class appears to lead a copiotrophic lifestyle, exhibiting fast growth rates in response to nutrient pulses (Šimek et al., 2006). In contrast to the Alpha- and
Gammaproteobacteria, Betaproteobacteria are found in low abundance in the ocean (Rusch et al., 2007). They are, however, often numerically dominant in freshwater (Glöckner et al., 2000;
Zwisler et al., 2003), where they are susceptible to flagellate grazing (Šimek et al., 2001).
Betaproteobacteria lead opportunistic lifestyles, responding rapidly to nutrient pulses (Šimek et al., 2006), and have been associated with extracellular phytoplankton production (Šimek et al.,
2008). Deltaproteobacteria, while not as abundant as other classes of Proteobacteria, are also ecologically important. Among its members are those that prey upon gram-negative bacteria
(Bdellvibrionales), as well as anaerobic sulfur-reducing bacteria. Among the Proteobacteria, phototrophic members utilizing proteorhodopsins can be found in the Alpha- (Rhodobacter),
Beta- (Rhodoferax), and Gammaproteobacteria (Chromatiales)(Kersters et al., 2006).
The Actinobacteria, another major bacterial lineage is found in a variety of habitats ranging from soils to marine and freshwater ecosystems. Previously thought to be exclusively soil bacteria (Goodfellow and Williams, 1983), freshwater and marine strains were discovered to be distinct from soil strains (Rappé et al., 1999; Newton et al., 2011). Moreover, they were often the most numerically abundant organisms in lakes, accounting for up to 63% of biomass attributed to bacterioplankton (Glöckner et al., 2000). Actinobacteria have been observed worldwide, retaining a strong 16S sequence similarity (Hahn and Pöckl, 2005; Zwart et al., 3
1998), suggesting similar clades across the globe, where they adapt to local conditions (Hahn and
Pöckl, 2005). Freshwater Actinobacteria have been divided into nine lineages representing 32 tribes (Newton et al., 2011). Two classes accounted for the vast majority of Actinobacteria in this study; Actinobacteria (acI lineage), and Acidomicrobia (acIV lineage). AcI Actinobacteria are the most abundant lineage in freshwater lakes, having been observed in abundances of >50% of the bacterial community (Allgaier et al., 2007; Warnecke et al., 2005; Glöckner et al., 2000).
Actinobacteria have a number of adaptations to freshwater lakes. Actinobacteria are generally accepted as having high GC content, ranging from 51-70% (Ventura et al., 2007), which has been suggested as a mechanism for UV protection (Warnecke et al., 2005). However, a study by Ghai et al., (2011) indicated that the numerically dominant acI clade, among other lineages, may have GC content as low as 37%. The discovery that Actinobacteria possess light- harvesting proteins similar to proteorhodopsins, dubbed actinorhodopsins, suggests that some members of the Actinobacteria may lead photoheterotrophic lifestyle in surface waters (Sharma et al., 2008). Recently, actinorhodopsins were identified in the genomes of several
Actinobacteria (Ghylin et al., 2014; Hahn et al., 2014). Size and cell wall structure may also be important mechanisms for survival. In culture, Actinobacteria were observed to bloom when in the presence of Ochromonas, a size-selective bacteriovore (Pernthaler et al., 2001; Hahn et al.,
2003). These studies identified cell size as well as cell wall structure, a feature known as the S- layer, as factors reducing grazing (Hahn et al., 2003). Supporting this, removal of the S layer resulted in increased grazing by Ochromonas by up to 5.2% (Tarao et al., 2009). Environmental conditions also play a role in the distribution of Actinobacteria, and have been observed to decrease in abundance with oxygen levels (Allgaier et al., 2006) and in culture with increased eutrophication (Haukka et al., 2006). The decrease in Actinobacteria in response to 4 eutrophication may be the result of a slow growth rate when compared to the growth rates of other freshwater bacteria (Šimek et al., 2006).
The Bacteroidetes are a metabolically diverse phylum found in a variety of habitats. It contains both chemoorganotrophs and phototrophs, the later being Flavobacteria utilizing proteorhodopsins (Gómez-Consarnau et al., 2007; González et al., 2008). Bacteroidetes are often particle associated, where they, along with Proteobacteria, are often numerous (Nold and
Zwart, 1998; Lemarchand et al., 2006). Bacteroidetes are known to degrade high molecular weight dissolved organic matter, such as chitin and cellulose, and as such are thought to have a major role in the environment. While this has been observed in the oceans (Fandino et al.,
2001), Betaproteobacteria have been observed to out-compete Bacteroidetes in some freshwater systems (Schweitzer et al., 2001; Böckelmann et al., 2000). Flavobacteria are found in high abundance during periods of cyanbacterial blooms, where they accounted for nearly 100% of the heterotrophic community during the decline of one bloom (Eiler and Bertilsson, 2007).
The PVC super phylum is an assemblage of bacteria that includes, among others, the
Planctomycetes, Verrucomicrobia, and Chlamydiae, from which the group derived its name.
The PVC super phylum is unique among bacteria for a number of archaea- and eukaryotic-like traits, including compartmentalization of cells (Lindsay et al., 2001; Lee et al., 2009), a lack of division protein FtsZ (Pilhofer et al., 2008), the production of sterol (Desmond and Gribaldo,
2009) and the lack of peptidoglycan in their cell walls. While these traits do not occur uniformly among members of the superphylum, they suggest a common ancestor of the superphylum
(Reynaud and Devos, 2011). Molecular and cellular functions of members of the PVC clade are poorly understood. Of those species which genomes have been sequenced, >50% of their genes are of unknown function (Devos, 2014). In this study, the phyla that occurred in significant 5 abundance were the Planctomycetes and Verrucomicrobia.
The Planctomycetes are another ubiquitous phylum, making up a significant portion of the bacterial community in a variety of environments, including soil (Buckley et al., 2006), fresh water , brackish, and marine habitats (Wang et al., 2002; Janssen, 2006; Gade et al., 2004,
Dedysh et al., 2005; Zeng et al., 2013; Vergin et al., 1998; Shu and Jiao, 2008). They have been found in diverse habitats ranging from epibacterial communities of macroalgae (Bondoso et al.,
2014), to deep-sea hydrothermal vent systems (Kato et al., 2010; Lanzén et al., 2011; Storesund and Øvreås, 2013), to the hyperarid Atacama Desert (Drees et al. 2006). Indeed, Planctomycetes were found in the highest known naturally occurring densities in the hindgut of soil-feeding termites (Köhler et al., 2008).
Planctomycetes are distinctive among bacteria. They display complex compartmentalization of cells, including a membrane-bound nucleoid. Furthermore, many members produce stalks and reproduce through budding. The environmental importance of these organisms is just beginning to be realized. Anaerobic ammonium oxidation (anammox) has been identified only in Planctomycetes (Strous et al, 1999) and it is estimated that they have generated as much as 50% of atmospheric N2 (Jetten, 2008). Interestingly, some possess genes for C1 transfer, typically found in methanogenic archaea and methanotrophic Proteobacteria, suggesting a possible role in the evolution of the methane cycle (Chistoserdova et al., 2004).
Of the samples analyzed in this study, two classes of Planctomycetes were represented in appreciable abundance, the Phycisphaerae and Planctomycetia. Class
Phycisphaerae, formerly group WPS1, was first discovered in PVC contaminated soils (Nogales et al., 2001). Uncultured representatives of this class have been associated with hypoxic conditions and include an anaerobic sulfur-rich spring (Elshahed et al., 2007), the suboxic zone 6 of the Baltic Sea (Fuchsman et al., 2011), and the sulfate-methane transition zone (Harrison et al., 2009). Two species have been brought into culture from this class, both isolated from marine algae. P. mikurensis is a facultatively anaerobic species (Fukunaga et al., 2009), while
Algisphaera agarilytica, is a strict aerobe (Yoon et al., 2013). While most planctomycetes reproduce by budding, both cultured species of this class reproduced by binary fission.
Class Planctomycetia is represented by a number of cultivated freshwater and marine strains. One cultured freshwater representative of this class, isolated from a lake in Michigan, was found in association with Daphnia (Staley, 1973). Members of this class may be important members of the sulfur cycle, and are capable of reducing elemental sulfur to sulfide (Elshahed et al., 2007). Indeed, comparative genomic analysis of 9 Rhodopirellula strains revealed a high diversity of genes annotated as sulfatases (>1,100)(Wegner et al., 2013). No representatives from the anammox clade, the Brocadiales, were detected in this study.
The Verrucomicrobia of the PVC superphylum are found both in water and soil habitats, as well as in association with eukaryotes (Sakai et al., 2003; Wang et al., 2005). They have been found in both surface and deep-water habitats, and have been associated with cyanobacterial blooms, where at times they have been observed to be the most abundant heterotrophic bacteria
(Kolmonen et al., 2004; Eiler and Bertilsson, 2004). In mesocosm experiments, their occurance has been correlated with eutrophic conditions (Haukka et al., 2006). Some have been discovered to have the ability to fix nitrogen, while other extremely acidophilic members have been found to oxidize methane (Dunfield et al., 2007; Islam et al., 2008, Pol et al., 2007).
In this study, 5 classes of Verrucomicrobia were present in appreciable numbers,
Pedosphaerae, Spartobacteria, Opitutae, Verrucomicrobiae, and Methylacidiphilae. The
Spartobacteria dominate verrucomicrobial communities in soils (Gaddy et al., 2011), and were 7 also observed as the dominant class in a humic lake (Arnds et al., 2010). In brackish regions of the Baltic Sea, a single OTU, ‘Spartobacteria baltica,’ was the dominant read (Herlemann et al.,
2011; Herlemann et al., 2013).
Other minor phyla may also have important ecological roles in freshwater lakes. A member of the Gemmatimonadetes isolated from a lake in the Gobi Desert was discovered to produce bacteriochlorophyll a (Zeng et al., 2014), and suggests a photoheterotrophic lifestyle for some members of this phylum. Another phylum, the Chloroflexi, may also be important in the ecology of deep, oxygenated hypolimnions. Members of the CL500-11, a subclass of the deep- ocean SAR202 clade, have been found to be the prominent group in the hypolimnions of both ultraoligotrophic Crater Lake, USA (Urbach et al., 2001) and mesotrophic Lake Biwa, Japan
(Okazaki et al., 2013), where it has been speculated to play a role in mineralization.
In this study, the bacterial communities of the Laurentian Great Lakes were explored using Illumina 16s tags. The influence of trophic status on these communities was examined along a gradient beginning in oligotrophic Lake Superior, the headwaters of the Laurentian Great
Lakes, and continuing to Lake Erie’s eutrophic Western Basin. Temporal resolution was provided by winter surveys of Lake Erie along with samples collected from the upper Great
Lakes. To provide insight into communities present in other large lakes in the Northern
Hemisphere, these winter communities were also compared with those of Lake Onega, one of the largest lakes in Europe. Additionally, the effects of global climate change were examined in
Lake Erie, comparing two years of high ice cover with a rare, nearly ice-free winter. 8
CHAPTER 1: ICE COVER EXTENT DRIVES MICROBIAL COMMUNITY
STRUCTURE IN A LARGE NORTH-TEMPERATE LAKE: IMPLICATIONS FOR A
WARMING CLIMATE
1. INTRODUCTION
Lakes and reservoirs serve as rapid responding sentinels of human influence on the natural environment (Adrian et al., 2009; Williamson et al., 2009) rendering them powerful tools to advance our understanding of a changing climate on ecosystem structure and function. The
Laurentian Great Lakes are especially valuable in this respect in that they share characteristics of both oceans and closed basin systems (Rao and Schwab, 2007) such that knowledge gained from their study can be used to gain insights for our coastal oceans. The effects of climate change have been especially pronounced in the Great Lakes where winter ice cover has declined by 71% over the past 4 decades (Wang et al., 2012). The decline is not constant; rather it is driven by high inter-annual variability combined with an increase in the frequency of years with low ice cover (Wang et al., 2012; Fujisaki et al., 2013).
The manifestations of declining ice cover have likely far-reaching effects on biogeochemical cycles and ecosystem functioning in lakes. In Lake Superior where ice cover has declined by 79% since 1973, summer surface water temperatures have increased > 2.5 °C over the same time period, a trend related to the decline in winter ice cover resulting in an earlier start to thermal stratification (Austin and Colman, 2007). Meteorological implications arise as a result of the weakened temperature gradient between air and water with stronger wind speeds possibly altering large-scale circulation patterns in the lake (Desai et al., 2009). Climate warming may also have important implications for eutrophication in the Great Lakes. In the 9 lower Great Lakes, phosphorus concentrations were negatively correlated with the extent of winter ice cover with extremes of 200-300% greater concentrations than normal coinciding with strong El Niño years (Nicholls, 1998). Ice cover provides the time needed for settling and consolidation of suspended particles into sediments thereby reducing the extent of resuspension following ice thaw (Kleeberg et al., 2013).
The ecological integrity of aquatic systems is intimately tied to the activities of microbial consortia. Whereas we are accumulating knowledge of microbial diversity in the Great Lakes
(e.g. Mou et al., 2013; Wilhelm et al., 2014), we know little about how these communities respond to the manifestations of climate change. Mid-winter surveys of Lake Erie conducted between 2010-2012 captured extremes in ice cover (Fig. 1). Expansive ice cover during 2011 coincided with a negative North Atlantic Oscillation (NAO) event together with La Niña whereas the combined effect of La Niña and a positive NAO event resulted in negligible ice cover on the lake in winter 2012 (Fujisaki et al., 2013). Here we show that pronounced changes in annual ice cover are accompanied by equally important shifts in phytoplankton- and bacterial community composition and structure.
1.1 MATERIALS AND METHODS
1.1.1 STUDY SITE AND SAMPLING
Lake Erie was sampled on surveys aboard CCGS Griffon and USCGC Neah Bay during winters 2010-2012 (Fig. 1). Early April monitoring surveys conducted by the U.S.
Environmental Protection Agency (EPA) were on board R/V Lake Guardian. Sampling aboard
CCGS Griffon was conducted during weeklong surveys in mid-February and included routine 10
Figure 1. Mid-winter limnological surveys captured extremes of ice cover on Lake Erie. A. Moderate Resolution Imaging Spectroradiometer (MODIS) image captured on 3 February, 2011 showing expansive ice cover. B. MODIS image captured 15 February showing the mainly ice-free condition of Lake Erie during winter 2012. Sampling locations are shown for surveys onboard CCGS Griffon (mid-February) and USCGC Neah Bay (January – March). MODIS images obtained from Great Lakes CoastWatch Program, NOAA-Great Lakes Environmental Research Lab (http://coastwatch.glerl.noaa.gov/).
hydrographic stations as well as underway sampling (2011-2012 only). Sampling aboard
USCGC Neah Bay comprised underway water collection through the ice-breaking season
(January – March) in a partnership with the U.S. Coast Guard described elsewhere (Oyserman et al., 2012). At each sampling location, the extent- and characterization of ice cover, if present, along with meteorological conditions were recorded. Also included in the analysis were weekly water intake data collected at municipal water treatment plants near the inflow of the Detroit
River into Lake Erie (Amherstburg), at a location in the central basin (Elgin) and near the outflow of the lake to the Niagara River (Rosehill). These data were acquired through the Great
Lakes Intake Program, an initiative of the Ontario Ministry of the Environment and Climate
Change (Nicholls et al., 2001).
Underway winter sampling from CCGS Griffon and USCGC Neah Bay followed the procedure detailed in Binding et al., (2012). Briefly, surface water samples were collected at approximately 1 h intervals using a stainless steel sampling bottle and processed immediately onboard the ship. Sampling at routine hydrographic stations occupied by CCGS Griffon and R/V
Lake Guardian was as described elsewhere (Twiss et al., 2012 and Reavie and Barbiero 2013; 11 respectively). Briefly, water column profiles of temperature, dissolved oxygen and conductivity were recorded during the February CCGS Griffon surveys using a Model 660 Sonde (YSI Inc.,
Yellow Springs, OH, USA) lowered at approximately 0.1 m s-1. Aboard R/V Lake Guardian, depth-resolved sampling was preceded by conductivity, temperature, depth (CTD; SBE 911plus;
Sea-Bird Electronics Inc., Bellevue, WA, USA) casts.
Samples were processed for determination of size-fractionated chlorophyll (chl) a biomass (0.2- and 20-µm polycarbonate filters) and for dissolved- (< 0.2 µm) and particulate nutrients. Chl a biomass was measured by fluorometry following extraction in 90% (v/v) acetone at -20 °C (Twiss et al., 2012). Nutrients were measured by the National Laboratory for
Environmental Testing (Environment Canada, Burlington, ON, Canada), the Laboratory Services
Branch (Ontario Ministry of the Environment and Climate Change, Toronto, ON, Canada) and the National Center for Water Quality Research (Heidelberg University, Tiffin, OH, USA) using standardized techniques (NLET, 1994; Chow et al., 2010; and U.S. EPA, 1979; respectively).
Sample preparation for microscopic analysis conducted as part of spring EPA surveys was as described elsewhere (Reavie and Barbiero 2013). Additional physico-chemical data from EPA annual spring surveys were obtained from the online Great Lakes Environmental Database
(GLENDA; accessed from http://www.epa.gov/cdx/).
1.1.2 HIGH-THROUGHPUT MICROBIAL COMMUNITY ANALYSIS
At three central basin hydrographic stations (EC 341, 880 and 1326) occupied during annual February surveys by CCGS Griffon, a comprehensive taxonomic analysis was completed by Illumina MiSeq targeting the 16S rRNA V4 hypervariable region of bacterial and plastid genomes. Targeting the larger V4 region over V6, which has been typically used in such studies, 12 should better reflect the microbial diversity present (Degnan and Ochman, 2012). Microbial biomass from surface water (and ice melt) was concentrated on Sterivex cartridge filters (0.22
µm; EMD Millipore, Billerica, MA, USA) and immediately frozen in liquid nitrogen. DNA was extracted from the Sterivex cartridges using the PowerWater Sterivex DNA Isolation Kit (MO
BIO Laboratories, Inc, Carlsbad, CA, USA) following manufacturer’s instructions. Short 16S rRNA tag (Itag) sequencing was completed at the Joint Genome Institute (Walnut Creek, CA,
USA) using an Illumina MiSeq benchtop sequencer (2 × 250 bp reads) incorporating a PhiX library control. Primer design for universal amplification of the V4 region of 16S rDNA was based on Caporaso et al., (2011), with the forward primer remaining unchanged and 96 variations of the reverse primer, each having 0 - 3 nucleotides added between the padding and the V4 sequence. PhiX reads and contaminating Illumina adaptor sequences were filtered and unpaired reads were discarded. Sequences were then trimmed to 165 bp and assembled using
FLASH software (Magoč and Salzberg, 2011). Resulting sequences were demultiplexed and filtered for quality: sequences were trimmed using a sliding window of 10 bp, required a mean quality score of 30, and contained less than 5 Ns or 10 bases with a quality score less than 15.
Itags were processed using default settings of QIIME 1.8.0 (Caporaso et al., 2010a) unless noted otherwise. Operational taxonomic units (OTU) were picked using uclust at 97%
(Edgar, 2010) and OTUs represented 3 times or fewer were filtered. A representative set of sequences was generated for each site with taxonomy assigned using the RDP classifier (Wang et al., 2007) with a minimum confidence of 80% for taxonomy assignment. Assignment was based on the Greengenes taxonomy (McDonald et al. 2012) and reference database version 12_10
(Werner et al., 2012). For analysis of bacterial populations alone, reads which were assigned to chloroplast and mitochondrial sequences, along with those not identifiable beyond Bacteria were 13 filtered. Reads matching “chloroplast” were used in subsequent analyses to examine the phytoplankton community. The combined bacterial- and eukaryotic reads are referred to as
“total reads”. The representative sequences were aligned to the Greengenes core reference alignment (DeSantis et al., 2006) using PyNAST (Caporaso et al., 2010b) and gaps in the resulting alignment were filtered. A phylogenetic tree was generated from the filtered alignment using FastTree 2.1.3 (Price et al., 2010) following which the samples were rarefied in
QIIME. Alpha diversity, which assesses the diversity within each sample, was calculated from the resulting files and collated. Phylogenetic measures of community beta diversity, a measure of diversity between different environments, were calculated using principle coordinates analysis
(PCoA) on both weighted and unweighted UniFrac matrices (Lozupone et al., 2007) rarified at
10,000 reads. Significance between PCoA clustering was tested using ANOSIM (Clark, 1993) with 999 permutations. Machine-learning approaches were adopted to identify OTUs whose abundances were associated with ice conditions present in 2010 and 2012. To reduce noise due to a large number of OTUs, communities used with machine learning approaches were filtered to contain only OTUs that were present at an abundance of at least 1% in any sample. Samples that clustered together along weighted principal component 1 were used for analysis. Random forest analysis (Knights et al. 2012, Breiman, 1999) was done using QIIME’s supervised learning script. One thousand forests were generated using leave-one-out cross validation for communities rarefied to 20,000 OTUs. Discriminatory taxa between high ice and low ice conditions was also investigated using LEfSe (Segata et al., 2011) which uses Kruskal-Wallis sum rank test (α=0.05) to identify significantly differential taxa and linear discriminant analysis
(LDA) to estimate the effect size of each. 14
1.2 RESULTS AND DISCUSSION
1.2.1 LAKE PHYSICO-CHEMICAL PROPERTIES
Mid-winter surveys of Lake Erie conducted in 2010/2011 and 2012 captured extremes in ice cover (Fig. 1). At 23%, Lake Erie possesses the highest median Total Accumulated Ice
Coverage (TAC) among the Laurentian Great Lakes (Canadian Ice Service, 2010), a distinction consistent with its relative shallow bathymetry. Whereas ice conditions during winter 2010
(20.7% TAC) were comparable to the 33-year historical median, ice conditions in 2011 were more severe than normal (33.6% TAC) with total ice concentration exceeding nine-tenths coverage across much of the lake with the majority of this ice characterized as medium lake ice of 15-30 cm thickness. In sharp contrast, winter 2012 (1.4% TAC) was nearly ice-free. The different ice conditions between years were likewise reflected in water column physico-chemical parameters (Fig. 2). Vertical profiles taken at process stations occupied in mid-February portrayed near isothermal conditions with temperatures markedly depressed in 2011 (-0.2 to 0.4
°C) compared with 2012 (0.3 °C in the shallow western basin to 2.5 °C in the deeper eastern basin). Support for a thorough wind-driven mixing regimen in the absence of ice during winter
2012 came from profiles of turbidity with median values (1 m depth) of 14.2 nephelometric turbidity units (NTU) an order of magnitude higher than measured in association with expansive ice cover during February 2011 (1.5 NTU) (two-tailed unpaired t-test, t = 10.70, DF = 6, P <
0.0001) (Fig. 2). Accordingly, the higher turbidity measured in 2012 caused greater attenuation of photosynthetically active radiation (PAR). The increase in lake turbidity recorded in 2012 is consistent with recent models for Lake Erie that show deeper penetration of the vertical eddy viscosity and accelerated coastal current speed compared to winters with expansive ice cover
(Fujisaki et al., 2013). These increases in vertical mixing and horizontal convection 15
Figure 2. Vertical water quality profiles for a representative central basin station (EC 1326) occupied during winter surveys of Lake Erie. Water column profiles of temperature, dissolved oxygen and turbidity 620 were recorded during the February 2011 and 2012 surveys using a Model 660 Sonde (YSI Inc., Yellow Springs, OH, USA) lowered at approximately 0.1 m s-1. Whereas profiles varied distinctly between years of high (2011) vs. low (2012) ice, they showed only modest variation of physicochemical parameters with depth suggestive of isothermal mixing.
likely also have implications for eutrophication because wind-induced mixing will keep phosphorus and other nutrients in suspension (Nicholls 1998). Indeed, analysis of nutrient chemistry from weekly water intake data demonstrated elevated soluble reactive phosphorus and silicate concentrations during winter 2012 (Fig. 3).
1.2.2 PHYTOPLANKTON- AND MICROBIAL COMMUNITY SHIFTS ASSOCIATED
WITH ICE COVER
Expansive ice cover in winters 2010 and 2011 was associated with under-ice phytoplankton blooms dominated by physiologically robust, filamentous centric diatoms as reported previously (Twiss et al., 2012). Coincident with the mainly ice-free conditions of 2012 was a decline in excess of 70% of average total phytoplankton chl a biomass compared with
2011 that was measured as part of both mid-February and early April surveys (Fig. 4). At first glance, these results appear to be at odds with those of previous studies suggesting that high vernal phytoplankton biomass is linked to a positive NAO index resulting in earlier ice break-up 16
Figure 3. Phosphorus and silicate measurements. Soluble reactive phosphorus (SRP; top panel) and silicate (lower panel) measured near the inflow of the Detroit River into Lake Erie (Amherstburg), at a location in the central basin (Elgin) and near the outflow of the lake to the Niagara River (Rosehill). In each panel, the shaded region shows the interquartile range and the thick shaded bar shows the median. Three low ice years of 2002, 2006 and 2012 are then superimposed against the pooled data.
(or negligible ice cover) during winter (Weyhenmeyer et al., 1999; Gerten and Adrian 2000;
Straile et al., 2003). While recognizing growth of phytoplankton under ice, these studies invoke turbulence as important in promoting the growth of diatoms during spring, a notion predicated on these non-motile taxa requiring resuspension to maintain their position in the photic zone. For many lakes in areas of high snowfall, ice surfaces are likely to be snow- covered thereby restricting development of under-ice blooms due to lack of light penetration through the accumulated snow and ice. In contrast, Lake Erie may present a different winter environment given that an expansive, thick snowpack is uncommon. Rather, snow falling on the ice surface of many large lakes is likely to accumulate in discrete drifts with the resulting snow-free ice 17
Figure 4. Central basin phytoplankton chl a biomass reported from U.S. EPA spring monitoring surveys.
exhibiting high transmittance of PAR (Bolsenga & Vanderploeg, 1992; Jewson et al., 2009). The
PAR penetrating through the snow-free ice can support prolific growth of diatom microplankton biomass, which has been documented in Lake Erie (Twiss et al., 2012) and Russia’s Lake Baikal
(Bondarenko and Evstafyev 2006, Jewson et al., 2008).
Winter surveys conducted in 2011 demonstrated tight coupling between microphytoplankton (> 20 µm) chl a biomass and total (> 0.2 µm) chl a with microphytoplankton contributing a median 81% of total chlorophyll (Fig. 5a), consistent with dominance by filamentous diatoms (Twiss et al., 2012). In contrast, a strong departure from a microphytoplankton-dominated system was observed during the low ice winter of 2012, when pronounced declines in this size-class contributed a median of just 27% of total chl a biomass 18
Figure 5. Phytoplankton biomass accumulation during extremes of ice cover. A. Winter surveys conducted on Lake Erie over two years demonstrated tight coupling (r2 = 0.959) between microphytoplankton Chl a biomass and total Chl a during winters 2010 and 2011, years of expansive ice cover. Coincident with the mainly ice-free conditions of winter 2012 was a decline in total chl a biomass along with a strong departure from a microphytoplankton-dominated system (r2 = 0.758). B. During early spring monitoring surveys conducted immediately following the high ice years of 2010 and 2011, the filamentous diatom A. islandica contributed the majority of total phytoplankton biovolume in the lakes central basin (r2 = 0.982). Following the mild winter of 2012, both total phytoplankton biovolume and the share contributed by A. islandica declined precipitously. 19
(two-tailed unpaired t-test, t = 13.73, DF = 146, P < 0.0001). Results obtained by assay of size- fractionated chl a were reinforced by flow cytometry which showed that cell abundance in the fraction containing large nanophytoplankton (6-30 µm) and smaller microphytoplankton (20-30
µm) declined more than 3-fold. Likewise, results from the April EPA monitoring surveys of 10 central basin stations from 2010-2012 support these trends (Fig. 5b). Aulacoseira islandica that emerges as the dominant species during winter persists into the early spring in Lake Erie
(Barbiero and Tuchman 2001; Reavie et al., 2014) where it contributed 86% of total phytoplankton biovolume following ice-out in 2010 and 2011 (Fig. 5b). A remarkable 89% decrease in total phytoplankton biovolume occurred with the transition from years of expansive ice (2010, 2011) to low ice (2012) with most of this decline attributed to a 95% decrease in A. islandica biovolume (two-tailed unpaired t-test, t = 4.1, DF = 28, P < 0.0005) (Fig. 5b).
Reduced cell size is recognized as a universal ecological response of phytoplankton to warming (Daufresne et al., 2009; Winder et al., 2009) although it usually marks a response to climate warming over multiple years as evidenced from sedimentary diatom assemblages (e.g.
Rühland et al., 2008), not a single season as shown here. Based on the phytoplankton enumeration data, it is likely that the reduction in cell size documented here is related to the reduced dominance of A. islandica during the low ice winter combined with increased contributions to percent algal community biovolume by smaller cryptomonads and dinoflagellates (>15-fold and >4-fold, respectively; E. Reavie, unpublished).
What drove the decline in phytoplankton standing stock in winter 2012 is not known, although we suspect it to be related to light-limitation of photosynthesis in a well-mixed, turbid water column since nutrients were abundant (Fig. 3). Measurement of photosynthesis in early spring 2012 supported light-limitation of phytoplankton photosynthesis as Ik, the irradiance at 20 which photosynthesis becomes light-saturated, was > 4-fold higher than the calculated mean water column irradiance during the daylight period. Similar values of Ik were estimated for communities sampled in 2010 and 2011, but the presence of expansive ice cover in those years may have ensured adequate light fields for photosynthesis and growth. Previous studies suggest under-ice communities are subject to only limited mixing by convection, which would restrict the mixing of suspended phytoplankton deeper than Iz (Kelley 1997), or that the communities are physically-associated with ice cover (D’souza et al., 2013)
Additional insights into microbial community dynamics were gleaned from temporally- resolved short 16S rRNA tag (itag) Illumina sequencing of samples collected at three central basin locations. The efficacy of Illumina sequencing has been demonstrated to show that known differences between microbial communities can be readily identified on Illumina platforms, making this approach suitable for high-throughput surveys of microbial communities (Caporaso et al., 2012). While recognizing the uncertainty of relying on a single sampling occasion from which to glean information on microbial diversity representative of the winter season, this approach reflects the myriad logistical challenges faced when conducting research during winter on large ice-covered lakes. Whereas partnership with Coast Guards offered greater temporal resolution for sampling physico-chemical parameters, logistical considerations generally prevented scientists from embedding during icebreaking operations leaving a week-long dedicated science survey during each year of the study as the only opportunity for targeted sample collection.
A total of 623,625 itag sequences were recovered with the output from individual sites ranging between 47,269 to 109,653 total reads per sample (Fig. 6b). For clarity, we 21
Figure 6. Rarefaction curves of observed species (97% OTUs) from 16S amplicons for A) Bacteria and chloroplast sequences. Use of rarefied samples of 10 000 random reads per site was based on the lowest return of bacterial reads. C) Breakdown of bacteria and chloroplast sequences recovered from each site. conducted separate analyses of bacterial- and chloroplast sequences. Use of rarefied samples of
10,000 random reads per site was based on the lowest return of bacterial reads (Fig. 6B).
Rarefaction curves indicated that diversity was not fully captured in most communities (Fig. 6A).
A central finding from our winter surveys was that in 50% of the samples, chloroplast reads representing phytoplankton occurred in comparable (± 1%), or higher numbers than
Bacteria (Fig. 6B). Of the eukaryotic phytoplankton, centric diatoms of class
Coscinodiscophyceae, dominated the winter phytoplankton contributing 95% of all chloroplast reads as well as the majority of total reads during winter regardless of ice cover with an average 22 contribution of 46% of total reads (Fig. 7). This trend was supported by a recent pyrosequencing effort that demonstrated that reads of phototrophs were likewise dominant in 50% of the samples collected during a February 2010 survey of Lake Erie (Wilhelm et al., 2014). These results seemingly contradict other surveys of freshwater lakes using next-generation sequencing (NGS) approaches where reads of heterotrophic bacteria clearly dominate (Oh et al., 2011; Eiler et al.,
2013a; b; Mou et al., 2013; Parfenova et al., 2013), a feature likely related to seasonality because most limnological surveys are completed during the de facto spring-summer-fall field season. In support of this assertion, Wilhelm et al., (2014) showed that bacterial heterotrophs dominated sample reads at all sites surveyed in Lake Erie during summer 2010. Whereas the use of chloroplast small subunit ribosomal sequences can be misleading owing to interspecies differences in DNA extraction efficiencies as well as differences in copy number of the SSU rRNA genes, comparison between microscopic abundance and NGS data demonstrate overall positive correspondence (Eiler et al., 2013). Moreover, some evidence suggests that diatoms can even be underestimated in NGS data due to inefficiencies in extracting DNA compared to other members of the community (Medinger et al., 2010). Further, while sequencing data alone cannot show that diatoms were numerically dominant over bacteria during winter sampling, analysis by flow cytometry showed bacterial abundance in winter to be over 4-fold lower than in summer.
A single OTU of Aulacoseira sp. dominated the diatoms sampled during winter contributing nearly 80% of chloroplast reads and an average 37% of total reads (Fig. 7c). Based on sequencing data alone, the species-level identity of this dominant OTU is inconclusive, which likely reflects the paucity of diatom sequences available in the NCBI database. Our previous 23
Figure 7. OTU abundance: Lake Erie 2010-12 A. OTU abundances per station, colored by phylum, rarefied to 10,000 reads. B. OTU abundances per station, colored by class. C. Chloroplast OTU abundances as a percent of total chloroplast and bacterial reads. 24 surveys (Twiss et al., 2012) along with annual spring surveys of the EPA (Barbiero and
Tuchman 2001; Reavie et al., 2014), however, conclusively demonstrated the winter phytoplankton to be dominated by A. islandica (O. Müller) Simonsen based on microscopic analysis.
UniFrac PCoA of the bacterial community supported separation based on expanse of winter ice cover in both unweighted (qualitative) (Fig. 8) and weighted (quantitative) analyses
(Fig. 9). The weighted analysis considers relative taxon abundance and the phylogenetic distance between observed OTUs from the different samples. By contrast, unweighted Unifrac is sensitive to factors affecting the presence or absence of taxa that may be otherwise obscured by abundance information (Lozupone et al., 2007). The first two principal coordinates for the weighted UniFrac explained 67% and 95% of the variation for bacterial and chloroplast communities respectively (Fig. 5). PCoA of unweighted UniFrac distances revealed that 25% of variation (PC1) could be attributed to the expanse of ice cover. Likewise, 46% of variation
(PC1) could be explained by the expanse of ice cover in the weighted analysis when relative abundance of taxa was considered (Fig. 5a). Interestingly, samples collected during 2011 along with a sample collected from EC 341 during 2012 appear to be intermediates along PC1 and cluster separately along PC2. Satellite imagery revealed that sites sampled in 2011 were open water ~2 weeks prior to sampling, and may indicate a transitioning bacterial community. This clustering pattern is significant and reveals a high level of dissimilarity between clusters present in the weighted and unweighted UniFrac analysis (RANOSIM = 0.97, P = 0.002; unweighted
RANOSIM = 0.78, P = 0.002). Further examination of weighted UniFrac revealed that including the intermediate cluster with either the low ice or high ice cluster resulted in significant dissimilarity from the third group (high-ice RANOSIM = 0.58, P = 0.01; 25
Figure 8. Principal coordinate analysis: Lake Erie 2010-12 (unweighted). PCoA of unweighted UniFrac distance matrices for bacteria and chloroplast communities.
Figure 9. Principal coordinate analysis: Lake Erie 2010-12 (weighted). PCoA of weighted UniFrac distance matrices for bacteria and chloroplast communities. 26
low-ice RANOSIM = 0.86, P = 0.01), supporting a transitional condition. There was no significant difference between the intermediate cluster and a combined high- and low ice condition (RANOSIM
= 0.09, P = 0.25). The consistent ice-extent trend for all sites except EC 341 suggests that the observed changes in the composition of the community related to changes in the selective pressures in the environment. The changes in the relative abundance of Flavobacteria and
Verrucomicrobia (discussed below) were symptomatic of these putative changes in selective processes. For example, Flavobacteria may preferentially use high-molecular weight substrates
(Cottrell and Kirchman, 2003), and this observed decrease in Flavobacteria during the low-ice year may be the indirect result of decreases in organic matter production from diatoms.
Analysis of chloroplast sequences similarly revealed clustering by ice conditions (Figures
8 and 9). The first two principal components explained 35% and 95% of variation for unweighted and weighted UniFrac, respectively. Principal coordinate 1 alone accounted for 87% of the variation in the weighted analysis (Fig. 9). These UniFrac results are significant and reveal dissimilarity between low- and high ice conditions (weighted RANOSIM = 0.41, P = 0.019; unweighted RANOSIM = 0.42, P = 0.036), suggesting that specific taxa are associated with the ice cover. Interestingly, samples collected in 2011, which appear to be intermediates in the bacterial
PCoA, cluster with low-ice samples along weighted PC1, while clustering closer high ice samples along unweighted PC1, and may further point to a transition in the microbial community from open waters to ice. Even though stramenopiles are a dominant component of the microbial community in winter, regardless of ice cover, the strong association between ice extent and the variation in the UniFrac PCoA points to a phylogenetically distinct group of ice-associated organisms (e.g., Aulacoseira icelandica). 27
Complementary machine learning approaches were used to identify discriminatory taxa for low-ice and high-ice conditions associated with years 2010 and 2012 (Knights et al. 2012,
Breiman, 1999). Samples for EC 880, EC 1326, and EC 341 were compared using supervised learning, a random forest algorithm (QIIME). Cross-validation estimates for 2010 and 2012 had an average probability of being guessed correctly 70% and 76% respectively. OTUs determined to be important features were identified using NCBI BLAST (Fig. 10). In total, OTUs that resulted in a decrease in accuracy in prediction accounted for an average of ~54% and ~44% of the bacterial community for 2010 and 2012 respectively. LEfSe LDA (Segata et al., 2011) was used to test for significant differences between low ice and high ice samples (Fig. 11). Both random forest and LDA algorithms identified OTUs within the Actinobacteria, Acidobacteria,
Bacteroidetes, Gemmatimonadetes, Planctomycetes, Proteobacteria, and Verrucomicrobia as important features differentiating low- and high-ice conditions.
The dominant phyla among Bacteria during winter in Lake Erie were similar to previous reports for freshwater lakes (Zwart et al., 2002; Newton et al., 2011). A lack of abundant
Cyanobacteria sequences was likely related to seasonality. Indeed, Cyanobacteria can be abundant components of the bacterioplankton during summer in Lake Erie, accounting for 10% of bacterial sequences in Sandusky Bay (Mou et al., 2013) and recently forming massive surface blooms in the western- and central basins of the lake (Allinger and Reavie 2013; Michalek et al.,
2013).
Among heterotrophic phyla, reads of Proteobacteria were uniformly dominant in winter regardless of ice expanse and contributed approximately 33% of bacterial reads in water samples.
Proteobacteria are arguably the best-studied phylum of environmentally relevant bacteria 28
Figure 10. Maximum-likelihood tree of OTUs identified by supervised learning. Important features for distinguishing between a high-ice year (2010) and a low-ice year (2012), identified using random forest analysis, along with associated BLAST hits. Percent change reflects changes in the community for low ice. 29 reflecting their ubiquity and abundance as well as their biochemical- and physiological diversity
(Newton et al., 2011). Comparing abundance of proteobacterial classes during winter surveys, the Betaproteobacteria were dominant comprising 57% of all proteobacterial reads followed by
Alphaproteobacteria sequences which contributed 28% (Fig. 4b). These trends in abundance follow the general sense of the literature whereby Betaproteobacteria can be numerically dominant among bacterioplankton in lakes (Glöckner et al., 1999; 2000; Eiler et al., 2013b), even during periods of ice cover (Pernthaler et al., 1998). While ubiquitous in freshwaters
(Newton et al., 2011), Alphaproteobacteria more frequently prevail in marine environments where they are dominated by the SAR11-cluster (Morris et al., 2002). Members of the freshwater
LD12 lineage (now classified as SAR11 subclade IIIb; Grote et al. 2012) dominate the
Alphaproteobacteria in most lakes (Zwart et al., 2002; Newton et al., 2011). Likewise in the present study, a sequence that clusters with members of the LD12 clade dominated the
Alphaproteobacteria with a single OTU that accounted for 66% of all alphaproteobacterial reads and over 6% of all bacterial reads. Whereas LD12 bacteria are ubiquitous in lakes, studies of seasonal bloom dynamics generally show a reduction in their abundance during winter
(Pernthaler et al., 1998; Salcher et al., 2011; Heinrich et al., 2013). LDA indicated a significant decrease in Gammaproteobacteria under low-ice conditions. Supervised learning identified
OTUs from the Alpha-, Beta-, and Gamma- lineages as important features for distinguishing the low- and high ice years (Fig. 10). Shifts in discriminatory Proteobacteria during 2012 resulted in an average net decrease of ~7% of the bacterial community, of which a decrease of ~5% was attributed to an OTU with a high identity to the genus Pelagibacter, a member of the SAR11 clade. 30
Figure 11. LDA effect size cladogram: Lake Erie 2010 and 2012. LDA cladogram comparing taxa of high-ice 2010 with those of low-ice 2012. Significantly discriminant nodes are colored, and branches are shaded by highest-ranking taxon.
Following the Proteobacteria in abundance were the phenotypically- and metabolically- diverse Bacteriodetes that contributed 27% of heterotroph reads under conditions of expansive ice cover (Fig. 4b). Negligible ice cover encountered in February 2012 was accompanied by a
23% decline in Bacteriodetes reads which could be attributed in part to a 48% decrease in reads of Flavobacteria between years of high- and low ice (two-tailed unpaired t-test, t = 3.46, DF = 6,
P < 0.05). In previous studies of lakes (Eiler and Bertilsson 2007; Zeder et al., 2009),
Flavobacteria responded positively to pulses of phytoplankton production, so it is likely that the large decline in chl a biomass measured between high- and low ice years in the present study contributed to the decline of this bacterial class. Supervised learning identified members of the
Sphingobacteria, Saprospireae, Cytophagia, and Flavobacteria as important features for 31 discriminating between high- and low ice years. Abundance of Bacteroidete OTUs identified by supervised learning as important features exhibited an average net decrease of ~8% of the bacterial community, of which decreases in OTUs belonging to the Sphingobacteria accounting for ~2% while decreases in Saprospireae resulted in a decrease of ~3% (Supp tree). LDA revealed significant decreases for Bacteroidetes, and indicated that Sphingobacteria and
Cytophagia, had significantly decreased under low-ice conditions.
In contrast to the Bacteriodetes, reads of Verrucomicrobia increased by 30% between years of high- and low ice with the increase attributed to increases in the reads of class
Verrucomicrobiae (Fig. 4b). A single OTU belonging to the Spartobacteria contributed 50% of all Verrucomicrobia reads in water samples and > 8% of total bacterial reads in water samples regardless of ice cover making it the single most abundant bacterial OTU in our surveys. This
OTU showed close identity (89%) to a putative ‘Spartobacteria baltica’ (Herlemann et al., 2013) identified as the most abundant bacterial read in brackish (salinities 5-10) regions of the Baltic
Sea (Herlemann et al., 2011). Random forest analysis identified this OTU as an important feature for distinguishing between high-ice and low-ice conditions, while LDA indicated a significant increase in the abundance of Spartobacteria during low ice conditions.
Verrucomicrobia are important representatives of soil communities, accounting for nearly 25% of all soil bacteria reads in a recent survey (Bergmann et al., 2011). Likewise,
Verrucomicrobia are ubiquitous members of aquatic environments (Zwart et al., 2003;
Herlemann et al., 2011; Freitas et al., 2012) where they may be active in the hydrolysis of diverse polysaccharides (Martinez-Garcia et al., 2012). While typically present as a minor phylum in aquatic systems accounting for < 2% of bacteria in diverse marine environments
(Freitas et al., 2012) and generally < 6% in freshwater lakes (Newton et al., 2011; Parfenova et 32 al., 2013; Poretsky et al., 2013), higher abundances of Verrucomicrobia have been reported.
Using CARD-FISH, Arnds et al. (2010) showed that Verrucomicrobia contributed up to 19% to the microbial community in a dystrophic lake in Germany. The characterization in the present study of the Verrucomicrobia as abundant members of the Lake Erie winter community, accounting for 18% of total bacterial reads during February 2012 and possessing a single OTU identified as the most abundant sequence in our surveys further underscores the need to elucidate the roles assumed by this globally distributed, but under-recognized group.
The final major phylum represented in our study, the Actinobacteria, collectively represented greater than 17% of all bacterial reads regardless of ice cover (Fig. 4b). Whereas representatives from class Actinobacteria dominated in all water samples, sequences of subclass
Acidomicrobidae consistently accounted for ~25% of the reads within the phylum. Whereas
Actinobacteria can be numerically dominant in lakes (Glöckner et al., 2000; Eiler et al., 2013b), historically they have been associated with the terrestrial environment and their presence in lakes attributed primarily to runoff and aeolian deposition. Distinct aquatic clades have been recognized (Warnecke et al., 2004) with the acI lineage (class Actinobacteria) accounting for >
90% of all Actinobacteria reads in some lakes (Warnecke et al., 2005). Consistent with the results of previous studies (Newton et al., 2011), the dominant Actinobacteria OTUs in the present study cluster with the acI- and acIV (predominantly Acidomicrobidae) lineages which typically dominate freshwater environments. Whereas a lack of cultured representatives leaves questions regarding the role of Actinobacteria in the environment, culture-independent approaches support the existence of both photo- and heterotrophic life styles with genes coding for energy-generating actinorhodopsins identified associated with members of the acI lineage sampled from the photic zone of Lake Erie during summer (Sharma et al., 2009). The machine 33 learning algorithms both indicated shifts in Actinobacteria between high- and low ice conditions.
Supervised learning indicated that two OTUs, both increasing under low-ice conditions, as discriminatory features, one of which shared close identity with Rhodococcus. Likewise, LDA indicated significant increases in Actinobacterial acI and acIV lineages during low ice conditions.
Shifts in minor taxa were also apparent between high- and low ice conditions. Both LDA and supervised learning pointed to members of the Acidobacteria, with LDA indicating a significant increase in Solibacterales during low ice conditions. LDA also indicated significant decrease in Planctomycetes and an increase in Gemmatimonadetes.
1.2.3 ECOLOGICAL IMPLICATIONS OF LOW ICE COVER
Winter ice cover on the Great Lakes has declined by 71% over the past 4 decades (Wang et al., 2012). While ice decline on Lake Erie has been lower (50%) than the system-wide average, ice cover models incorporating 2 × CO2 scenarios predict a future with markedly reduced, or even no ice on the lake (Assel 1991). Thus, the condition of negligible ice cover encountered during winter 2012 provided a window to a future “low ice” state of Lake Erie. Our surveys demonstrated tangible and potentially important shifts in phytoplankton- and microbial community structure between winters of high- and low ice, the most striking of which was the precipitous decline in standing stock of filamentous diatom microplankton in winter 2012. The shift away from a phytoplankton community dominated by filamentous diatoms is likely to have far reaching ecosystem effects including food web disruptions where we predict that the sharp decline in Aulacoseira islandica will disrupt trophic transfer of carbon to zooplankton ultimately having a negative impact on fish recruitment. Suppression of winter diatom growth may also 34 have important biogeochemical implications for events during summer in Lake Erie. Abundant diatom growth combined with low measured rates of bacterial decomposition results in export of algal biomass to the benthos (Wilhelm et al., 2014). As the hypolimnion warms during summer, bacterial remineralization of the exported diatom biomass accelerates, which depletes the hypolimnion of oxygen and results in formation of the Lake Erie “dead zone”, the expanse of which can exceed 10 000 km2 (Hawley et al., 2006). However, a prediction that lower phytoplankton biomass accumulation during low ice winters would lessen the extent and magnitude of central basin hypoxia is not supported by available data. Rather, late summer hypoxia followed the low ice years of 2002 and 2012. Whereas winter production is an important driver of late summer hypoxia (Wilhelm et al., 2014), other factors contribute to the formation of hypoxia including the length of the stratified period and hypolimnetic volume and temperature. 35
CHAPTER 2: TRANSITIONS IN MICROBIAL COMMUNITIES ALONG A 1,600 KM
FRESHWATER TROPHIC GRADIENT
2. INTRODUCTION
Collectively, the Laurentian Great Lakes system holds 21% of global surficial and liquid freshwater making it arguably the most important natural resource on the North American continent. Historically, the lakes were exploited for their productive fisheries and as a transportation route, supporting the manufacturing activities that brought prosperity to America’s
“North Coast”. Today, they support a multi-billion dollar tourist and recreation industry, supply
40 million people with drinking water, and provide habitat for wildlife and dozens of species of fish. Unfortunately, these resources are at risk. Eutrophication is regarded as one of the primary threats to the ecological integrity of our freshwater resources (Schindler, 2006) and the
Laurentian Great Lakes have historically followed the pattern of eutrophication documented globally (Beeton, 1965). However, these transitions in trophic state are fluid, reflecting ecosystem responses not only to directed management efforts (Bunnell et al., 2014) but also to biological invasions (e.g. Hecky et al., 2004) and climate warming (Chapter 2). Indeed, recent evidence has accumulated describing both the oligotrophication (Barbiero et al., 2012) and re- eutrophication (Kane et al., 2014; Scavia et al., 2014) of lakes in the Great Lakes system.
Among the Great Lakes, Lakes Superior and Erie serve as end members in terms of trophic state (Chapra and Dobson, 1981; Sterner 2011). Lake Superior as the headwater of this system is an oligotrophic lake characterized by cold temperatures, deep mixing and a high stoichiometric imbalance between nitrogen and phosphorus (Sterner et al. 2007, Sterner 2011). It has continuous, near-saturation oxygen levels throughout the water column with oxygen 36 extending deep into surface sediments (Li et al. 2012). Its near pristine condition is supported by long term trends demonstrating stable profiles of total dissolved solids and major ions which serve as indicators of anthropogenic impacts on the system (Chapra et al., 2012). Lake Erie, in contrast, has a human-dominated watershed which is home to 13 million people, whose industrial and agricultural activities have contributed to the eutrophication of the western and central basins of the lake. Whereas the combined effects of directed management (Makarewicz and Bertram 1991) and the unintended consequences of invasive dreissenid mussels (Nicholls and Hopkins 1993) resulted in increased transparency of the lake in the 1980’s and 1990’s, the lake has since entered into a period of re-eutophication (Kane et al., 2014; Scavia et al., 2014) characterized by blooms of pelagic- (Twiss et al. 2012; Michalek et al. 2013) and benthic algae
(Higgins et al., 2005) and intensification of seasonal hypoxia (Hawley et al., 2006; Scavia et al.,
2014).
The activities of microbial populations and consortia have a direct and profound influence on the ecological integrity of aquatic systems. Microbes are primary and secondary producers. They facilitate nutrient cycling, remediate and detoxify contaminants, produce novel metabolites, and serve as keystone species in interactions with many other organisms. Yet our knowledge of microbial diversity in the Great Lakes is limited both spatially and seasonally.
With a few exceptions (summarized in Reed and Hicks, 2011), most of what we know about planktonic microbial diversity in the Great Lakes comes from recent pyrosequencing studies focused on Lake Erie (e.g. Mou et al., 2013a; Wilhelm et al., 2014), the lake most impacted by eutrophication. Here we describe planktonic microbial diversity captured as part of a 3,200 km round-trip limnological survey extending from western Lake Superior to eastern Lake Erie.
Conducted over a period of 2 weeks during the summer-stratified period, the survey provided a 37 unique opportunity to gain insights into how microbial communities reflect changes in the trophic state of the largest lake system on Earth.
2.1 METHODS AND MATERIALS
2.1.1 STUDY SITES AND SAMPLING
Samples were collected between July 14-29, 2011 aboard R/V Blue Heron during a 3,200 km round-trip survey through Lakes Superior, Huron, and Erie (Fig. 12). At each hydrographic station, sampling was preceded by a conductivity–temperature–depth (CTD) cast. Water was sampled from discrete depths using Niskin bottles attached to the CTD rosette. Samples were processed for determination of total chlorophyll (chl) a biomass (>0.2 µm) and for dissolved- (<
0.2 µm) and particulate nutrients. Sampling also included determination of elemental stoichiometry (C, N, P) for size-fractionated (<80 µm [total] and <2 µm [picoplankton]) planktonic seston.
Chl a biomass was measured by fluorometry following extraction in 90% (v/v) acetone at
-20 °C (Welschmeyer 1994). Subsamples for nutrient determination were frozen in polyethylene bottles on board ship for subsequent laboratory analysis. Nitrate was measured using a Lachat
QuickChem© 8500 Flow Injection Analysis System (Hach Co., Loveland, CO, USA) and a cadmium reduction column whereas measurement of total dissolved phosphate followed the ascorbic acid-molybdate method (Parsons et al. 1984). A fluorometric method was used to
+ measure NH4 as described previously (Kumar et al. 2007; Small et al. 2013). For particulate organic C, N, and P measurements, seston was collected onto precombusted 25 mm GF/F filters
(for P, filters were acid-rinsed prior to sample application), frozen, and then dried at 60 °C as 38
Figure 12. Sampling locations: July 2011. Sampling locations along a transect leaving from station CD-1 in Lake Superior and extending to station EC 879 in Lake Erie. described elsewhere (Sterner 2011). Particulate organic carbon and nitrogen were determined with a 2400 Series II CHN analyzer (Perkin Elmer, Waltham, MA, USA) using acetanilide as a standard. Samples for particulate organic phosphorus were digested in a 5% potassium persulfate solution and autoclaved for 30 min. Liberated soluble reactive phosphorus was analyzed with the ascorbic acid-molybdate method (Parsons et al., 1984).
Microbial biomass from water samples was concentrated on Sterivex cartridge filters
(0.22 µm; EMD Millipore, Billerica, MA, USA) and immediately frozen in liquid nitrogen. DNA was extracted from the Sterivex cartridges using the PowerWater Sterivex DNA Isolation Kit
(MO BIO Laboratories, Inc, Carlsbad, CA, USA) following manufacturer’s instructions. 39
2.1.2 HIGH-THROUGHPUT MICROBIAL COMMUNITY ANALYSIS
A comprehensive taxonomic analysis was completed by Illumina MiSeq targeting the
16S rRNA V4 hypervariable region of bacterial and plastid genomes. Short 16S rRNA tag (itag) sequencing was completed at the Joint Genome Institute (Walnut Creek, CA, USA) using an
Illumina MiSeq benchtop sequencer (2 × 250 bp reads) incorporating a PhiX library control.
Primer design for universal amplification of the V4 region of 16S rDNA was based on (Caporaso et al., 2011), with the forward primer remaining unchanged and 96 variations of the reverse primer, each having 0 - 3 nucleotides added between the padding and the V4 sequence. PhiX reads and contaminating Illumina adaptor sequences were filtered and unpaired reads were discarded. Sequences were then trimmed to 165 bp and assembled using FLASH software
(Magoč and Salzberg, 2011). Resulting sequences were demultiplexed and filtered for quality: sequences were trimmed using a sliding window of 10 bp, required a mean quality score of 30, and contained less than 5 Ns or 10 bases with a quality score less than 15.
Itags were processed in QIIME 1.8.0 (Caporaso et al., 2010a) using default settings unless noted otherwise. OTUs were picked using uclust at 97% identity (Edgar, 2010) and
OTUs represented 3 or fewer times in the data set were filtered along with those shorter than 200 bp. A representative set of sequences was generated, where the most abundant sequence represented its respective OTU. Taxonomy was assigned to each representative sequence using the RDP classifier (Wang et al., 2007) with a minimum confidence of 80% for taxonomy assignment. Assignment was based on the Greengenes taxonomy (Mcdonald et al., 2012) and reference database version 12_10 (Werner et al., 2012). For analysis of bacterial populations, reads that were assigned to chloroplast and mitochondrial sequences, those that were unclassified or unassignable, and those not identifiable beyond bacteria were filtered. Those reads matching 40
‘chloroplast’ were then used in subsequent analyses to examine the photosynthetic eukaryotic community. The combined total bacterial and chloroplast reads are referred to as ‘total reads’.
The representative sequences were aligned to the Greengenes core reference alignment (Desantis et al., 2006) using PyNAST (Caporaso et al., 2010b) and gaps in the resulting alignment were filtered. A phylogenetic tree for all bacterial OTUs was generated from the filtered alignment using FastTree 2.1.3 (Price et al., 2010). The samples were then rarefied in QIIME.
2.1.3 STATISTICAL ANALYSIS
Alpha and beta diversity were implemented through QIIME, and principal coordinates analysis (PCoA) on both weighted and unweighted UniFrac matrices (Lozupone and Knight,
2005) was done at a rarified depth of 20,000 reads. Significance of PCoA clustering was tested using ANOSIM with 999 permutations, each cluster being tested against the remainder of samples.
The importance of environmental features on PCoA clustering was explored using
QIIME’s BEST analysis, an implementation of BIOENV (Clarke & Ainsworth, 1993). Distance matrices were generated for each physiochemical parameter (QIIME), and correlations of physiochemical measurement matrices and UniFrac plots was examined using the Mantel test with 999 permutations. Machine-learning approaches were adopted to identify OTUs whose abundances were associated with UniFrac clustering patterns. To reduce noise due to a large number of OTUs, communities used were filtered to contain only OTUs that were present at an abundance of at least 1% in any sample. Shifts in community composition were investigated using LEfSe LDA (Segata et al., 2011), which uses a Kruskal-Wallis sum rank test (α=0.05) to identify significantly differential taxa and linear discriminant analysis (LDA) to estimate the 41 effect size of each. LEfSe LDA was used to identify significant differences in taxa between the communities of the open waters of the UGL, and those of Lake Erie as well as for identifying differences in community structure between surface and deep waters of the UGL. Communities of the surface waters of Michipicoton and CD-1 (sample B) (Lake Superior), along with EC 54,
EC 29, and EC 17 (Lake Huron) were compared against surface and metalimnion samples from
Lake Erie (91M, EC 879) as observed in UniFrac clustering of mixed layer samples. Samples from CD-1 (sample B), Michipicoton, EC 54, EC 17, and EC 29 were compared with their hypolimnion counterparts with the exception of EC 17, which did not cluster closely with other deep-water samples. Additionally, deep-water samples from EL-7, EL-0, and Sleeping Giant were used in the analysis. Random forest analysis (Knights et al. 2012, Breiman, 1999) was implemented using QIIME’s supervised learning script, and generated one thousand forests using leave-one-out cross validation for communities rarefied to 20,000 OTUs.
2.2 RESULTS AND DISCUSSION
2.2.1 PHYSICO-CHEMICAL PROFILES OF SAMPLING LOCATIONS
Thermal stratification was evident at most sites during the July multi-lake survey with surface mixed-layer temperatures following expected latitudinal trends ranging from 8 - 17 °C
(median: 13.4 °C) in Lake Superior, 18 – 22 °C in Lake Huron and 24 - 26 °C in Lake Erie. Of the physico-chemical parameters monitored, surface mixed-layer waters of the upper Great
Lakes (Superior and Huron) differed from deep waters primarily in temperature although there
+ - was evidence of modest surface water depletion of both NH4 and NO3 (Table 1) consistent with seasonal biological uptake (Sterner et al., 2007). In all cases, dissolved oxygen was near 42 saturation throughout the water column. Characteristic of the summer-stratified period in the upper Great Lakes (Barbiero and Tuchman, 2001), a prominent chlorophyll maximum existed at depth, located below the thermocline. In contrast to the upper Great Lakes, physico-chemical parameters in Lake Erie varied dramatically by basin consistent with the trophic gradient existing within the lake (Ludsin et al., 2001). Lake Erie’s shallow western basin was thoroughly mixed with elevated nitrogen compared to mixed layer water chemistry in the oligotrophic eastern basin. Whereas oxygen remained saturated throughout the water column in both the western- and eastern basins, our survey coincided with the onset of hypoxia in Erie’s central basin with dissolved oxygen declining to ~50% saturation in the hypolimnion.
Phytoplankton- and microbial communities sampled from the surface mixed layer in the upper Great Lakes were depleted in phosphorus compared to samples from depth, a pattern described previously during the summer-stratified period (Sterner 2011). For both total- (>0.2
µm) and picoplankton (0.2 – 2 µm) size fractions, molar ratios of N:P and C:P were ~2-fold higher in the epilimnion compared to the hypolimnion (two-tailed unpaired t-test, P < 0.0001).
For the total seston fraction, mean mixed layer ratios of N:P and C:P of 35 and 295, respectively, suggested phosphorus-depleted surface communities and were consistent with the characterization of the upper Great Lakes as phosphorus deficient, oligotrophic environments
(Sterner 2011; Barbiero et al., 2012; Bunnell et al., 2014). Whereas both total- and picoplankton size fractions sampled from the surface mixed layer were depleted of phosphorus compared to deep water samples, C:N:P stoichiometry of the picoplankton fraction, which comprised most of 43
Table 1. Sampling sites and physico-chemical parameters Upper Great Lakes + Station Date Sample Depth (m) Site Depth (m) Temp (°C) DO (mg/L) Chl a (µg/L) DIC (mg/L) NH4 (µM) NO3 (µM) P (µM) Latitude Longitude CD-1 07/14/2011 5 250 8.5 13.1 0.46 9.4 0.2 25.8 0.05 47.063 -91.432 CD-1 07/29/2011 2 250 16.4 10.7 0.43 9.3 0.4 24.2 0.04 47.063 -91.432 CD-1 07/29/2011 150 250 3.8 14.0 0.05 9.3 0.6 26.6 0.03 47.063 -91.432 EC 17 07/24/2011 5 80 20.8 9.4 0.14 15.6 0.4 18.8 0.02 45.245 -80.864 EC 17 07/24/2011 73 80 4.1 14.5 0.09 16.1 0.5 20.6 0.03 45.245 -80.864 EC 29 07/24/2011 5 125 21.2 9.3 0.11 17.1 0.4 23.5 0.02 44.367 -81.833 EC 29 07/24/2011 119 125 3.8 14.3 0.02 17.2 0.6 23.2 0.02 44.367 -81.833 EC 54 07/17/2011 5 125 17.8 10.2 0.26 16.3 0.3 20.4 0.04 45.517 -83.417 EC 54 07/17/2011 100 125 3.9 14.5 0.06 17.1 0.5 22.7 0.03 45.517 -83.417 EC 9 07/17/2011 48 50 4.8 14.3 0.58 17.2 0.5 17.6 0.04 43.633 -82.378 EL-0 07/15/2011 145 165 3.8 14.4 0.09 9.2 0.4 26.5 0.05 47.75 -87.5 EL-7 07/26/2011 132 124 3.8 14.3 0.14 9.3 0.5 26.9 0.04 47.167 -85.015 Michipicoten 07/26/2011 5 80 16.7 10.5 0.46 9.0 0.3 23.4 0.04 47.908 -85.007 Michipicoten 07/26/2011 78 80 4.0 14.2 0.06 9.4 0.4 26.2 0.04 47.908 -85.007 Sleeping Giant 07/29/2011 240 272 3.6 14.1 0.11 9.2 0.4 26.5 0.03 48.223 -88.906
Upper Bays + Station Date Sample Depth (m) Site Depth (m) Temp (°C) DO (mg/L) Chl a (µg/L) DIC (mg/L) NH4 (µM) NO3 (µM) P (µM) Latitude Longitude Black Bay 2 07/28/2011 2 8 18.7 9.5 1.71 11.8 0.2 15.6 0.16 48.692 -88.397 Black Bay 1 07/28/2011 2 22 16.5 10.1 0.28 10.5 0.3 21.1 0.10 48.5 -88.608 Nipigon Bay 1 07/27/2011 5 27 17.8 9.8 0.58 11.0 0.3 16.1 0.08 48.933 -88.001
Lake Erie + Station Date Sample Depth (m) Site Depth (m) Temp (°C) DO (mg/L) Chl a (µg/L) DIC (mg/L) NH4 (µM) NO3 (µM) P (µM) Latitude Longitude 91M 07/18/2011 8 10 25.8 8.2 0.47 19.0 1.9 38.4 0.21 41.841 -82.917 EC 879 07/22/2011 5 60 23.5 9.1 0.52 20.5 0.6 6.7 0.11 42.507 -79.9 EC 879 07/22/2011 25 60 5.9 13.6 0.91 20.5 2.8 11.0 0.11 42.507 -79.9 EC 879 07/22/2011 60 60 4.6 12.3 0.02 21.0 2.8 12.1 0.02 42.507 -79.9 44 the microbial community analyzed in this study, was not strongly suggestive of phosphorus deficiency. While mean picoplankton N:P (19) and C:P (163) molar ratios were nominally above
Redfield stoichiometry (C:N:P = 106:16:1 by moles) identified for balanced growth, these values were diagnostic of only modest phosphorus deficiency for the Great Lakes (Guildford and
Hecky, 2000; Sterner 2011). Further, biomolecular modifications such as phospholipid substitution adopted by the microbial community in phosphorus-depleted environments such as
Lake Superior (Bellinger et al., 2014) may help to facilitate balanced growth despite non-
Redfield elemental stoichiometry.
Contrasting the situation in the upper Great Lakes, Lake Erie showed little in the way of depth- or seston size-related trends in N:P and C:P seston stoichiometry. Further, regardless of depth, samples showed no evidence of phosphorus deficiency with median elemental ratios lower than diagnostic thresholds (sensu Healy and Hendzel, 1980) for phosphorus limitation.
Water column profiles of seston C:N in the upper Great Lakes did not vary with depth for either total- (two-tailed unpaired t-test, t = 0.198, DF = 40, P = 0.844) or picoplankton (two- tailed unpaired t-test, t = 1.001, DF = 40, P = 0.323) size fractions. Further, mixed-layer seston
C:N was not different between total- and picoplankton size fractions (two-tailed unpaired t-test, t
= 0.562, DF = 47, P = 0.577). Likewise in Lake Erie, there was no discernible pattern of C:N with depth (two-tailed unpaired t-test, t = 0.741, DF = 17, P = 0.469) nor did mixed layer seston
C:N vary between total- and picoplankton size fractions (two-tailed unpaired t-test, t = 1.092, DF
= 17, P = 0.290). For all lakes, while median mixed layer C:N molar ratios (8.3 and 15 for the upper Great Lakes and Lake Erie, respectively) were higher than Redfield stoichiometry, the metric was diagnostic for nitrogen deficiency only for Lake Erie. Whereas Lake Erie has been historically characterized as a phosphorus-limited system (Lean et al. 1983; Guildford et al. 45
2005), evidence for seasonal nitrogen deficiency has accumulated in recent years (North et al.,
2007; Chaffin et al., 2013).
2.2.2 MICROBIAL COMMUNITIES
Insights into microbial community dynamics were gleaned from short 16S rRNA tag
(itag) Illumina sequencing targeting the V4 hypervariable region of bacterial and plastid 16S rRNA. Targeting the larger V4 region over V6, which has been typically used in such studies, should better reflect the microbial diversity present (Degnan and Ochman, 2012). Reflecting the utility of this approach, the Earth Microbiome Project (Gilbert et al., 2010) has adopted it for rRNA amplicon sequencing.
Consistent with previous studies of freshwater lakes surveyed during summer using next generation sequencing approaches, reads of Bacteria dominated total sample reads (Fig. 13B)
(Oh et al., 2011; Eiler et al., 2013a; b; Mou et al., 2013; Parfenova et al., 2013; Wilhelm et al.,
2014). Chloroplast reads accounted for 8.1% of total reads by average and of the eukaryotic phytoplankton detected through plastid 16S sequences, diatoms were the most abundant chloroplast OTU at most sites in the upper Great Lakes consistent with previous characterization of summer phytoplankton communities in Lakes Superior and Huron (Barbiero and Tuchman,
2001). By contrast, communities in Lake Erie contained fewer diatoms and higher abundances of cryptomonad OTUs. This is in stark contrast to the winter and spring in Lake Erie where diatoms comprise a nearly monoalgal community and diatom sequences frequently surpass those of
Bacteria (Chapters 1 and 3).
Analysis of the bacterial community during the 3,200 km round-trip survey was based on a total of 1,411,554 16S rRNA tag sequences from 18 samples with the output from individual 46
A B Bacteria 200000 chloroplast 3000 bacteria
150000
2000 ed species v
total reads 100000
obse r 1000
50000
0 25000 50000 75000 100000 depth (reads) 0 EL-0 EL-7 Chloroplast EC 9 EC 29 EC 54 EC 91 EC 17 EC 879 CD-1 B CD-1 A CD-1 Black Bay 2 Black Bay 1 EC 879 DCL EC 29 Deep EC 54 Deep 500 EC 17 Deep EC 879 Deep CD-1 B Deep Michipicoton Nipigon Bay 1 Sleeping Giant Michipicoton Deep
400
300 ed species v
200 obse r
100
5000 10000 depth (reads)
Figure 13. Rarefaction curves and total reads: July 2011 A.Rarefaction curves for bacterial and chloroplast OTUs. B. Total reads obtained for each sampling location, with abundance for both chloroplast and bacterial reads. sites ranging between 27,426 to 142,364 reads per sample (Fig. 13B). Rarefaction curves indicated that diversity was not fully captured in most communities (Fig. 13A). This was likewise reflected by the Chao1 species richness estimator, which showed that 47% of all OTUs were identified in rarefied samples for Bacteria (Table 2). Shannon’s diversity index which combines species richness and abundance into a measure of evenness, did not vary for bacterioplankton communities by lake region or depth (one-way ANOVA; P = 0.36). Likewise, observed species counts did not vary (one-way ANOVA; P = 0.25) with surface mixed layer and deep samples across all sites averaging 1351 OTUs. The upper bays had the highest observed 47
Table 2. Alpha Diversity: July 2011
Upper Great Lakes Observed Species Shannon Simpson Chao1 PD surface waters 1347.1 6.1 0.94 2750 119.2 deep waters 1320 6.4 0.96 2919 139.9 EC 17 Deep 1194 6.2 0.96 2723 122.4 EC 9 1378 6.4 0.96 3256 135.1
Upper Bays Observed Species Shannon Simpson Chao1 PD Black Bay 1411 6.3 0.95 3056 140.8 Nipigon Bay 1573 6.4 0.96 3255 148.6
Lake Erie Observed Species Shannon Simpson Chao1 PD Western Basin 1099 5.9 0.93 2028 104.7 Eastern Basin Surface 1350.4 6.6 0.97 2972 123.7 Eastern Basin DCM 1230.3 6.1 0.95 2735 118.7 Eastern Basin Deep 1384 6.6 0.96 2992 126.5
*depth = 20,000 species counts, with 1,411 and 1,573 OTUs for Black Bay and Nipigon Bay respectively.
Likewise, Faith’s phylogenetic diversity was highest in Superior’s upper bays, while phylogenetic diversity was lowest in Erie’s western basin.
Principal Coordinate Analysis (PCoA) was used to examine community similarities between sites by comparing phylogenetic differences measured by UniFrac. Visualization of unweighted UniFrac revealed significant clustering of samples collected from the epilimnion of the upper Great Lakes (RANOSIM=0.46, P=0.001), the deep hypolimnetic waters of the upper
Great Lakes (RANOSIM=0.67, P=0.002), and Lake Erie (RANOSIM=0.48, P=0.003)(Fig. 14).
Weighted UniFrac also revealed a similar pattern of significant clustering of the hypolimnetic waters of the upper Great Lakes from all other samples (RANOSIM=0.87, P=0.001)(Fig. 14).
Hypolimnion samples from the upper Great Lakes clustered along PC1 in both unweighted and weighted analyses, which explained 18% and 46% of variation between samples respectively. 48
Unweighted Weighted