Invasive dreissenid mussels and benthic algae in Lake Michigan: characterizing effects on sediment bacterial communities

demonstrates ABSTRACT Dreissenid mussels have invaded the Laurentian Great Lakes causing dramatic changes to benthic–pelagic interactions. Despite research on food web impacts, there is limited data on mussel effects on benthic bacterial communities. This study examined effects of dreissenid mussels and benthic algae on sediment bacterial community composition and diversity. Triplicate experimental sediment plus lake water microcosms were used and either mussels, benthic algae or both were added. Changes in water nutrient chemistry and sediment bacterial communities were monitored using 16S rRNA amplicon sequencing, over 21 days. When mussels were present, nitrate and soluble reactive P increased significantly as the dominant N and P forms. Bacterial diversity increased in all microcosms, although bacterial community composition was distinct between treatment. Higher nitrate in mussel microcosms was accompanied by increases in nitrifying taxa ( Nitrospira, Nitrosomonas ), which are important in oxidizing mussel-excreted ammonium. Microcosms with algal additions showed increases in bacterial taxa capable of degrading algal cellulose, and Pelagibacter (SAR11) disappeared from all but control microcosms. This study suggests that bacterial communities in lake sediments respond to mussels and algae. Functional analysis of bacterial communities provides insights into changes in microbially mediated benthic nutrient transformations associated with invasive dreissenid mussels and benthic algae in lake ecosystems.


INTRODUCTION
Bacteria play critical roles in biogeochemical cycling in lakes, especially within lake sediments, which exchange nutrients with the water column. Nearshore sediments are affected by nutrient inputs from terrestrial sources, and by benthic communities of algae and invertebrates. Large-scale ecological changes in lakes are often documented as perturbations to macroscopic organisms, yet the critical biogeochemical nutrient cycling processes are mediated by sediment bacterial communities. A key example is the ecological impacts of invasive species on lake ecosystems. The Laurentian Great Lakes have been invaded by a succession of non-native species via ballast water discharge from transoceanic ships (Ricciardi and MacIsaac 2000). Among the world's most problematic biological invaders are the zebra mussel, Dreissena polymorpha (Pallas) and quagga mussel, D. rostriformis bugensis (Andrusov) (Ricciardi and MacIsaac 2000) (collectively known as dreissenid mussels), which have expanded from their native range in the Ponto-Caspian region and have spread worldwide, including into waterways of North America. Invasion by these species has caused significant ecological changes to Laurentian Great Lakes (Barbiero et al., 2006;Higgins et al., 2008). Many of the perturbations to food web structure, changes to nutrient cycling and promotion of benthic algal growth have been extensively examined (Pillsbury et al., 2002;Higgins and Vander Zanden 2010;Vanderploeg et al., 2012). However, the role of bacteria in these changes and the impacts of mussels on sediment microbial communities have received only scant attention (Frischer et al., 2000;Lohner et al., 2007).
Aquatic ecosystems invaded by dreissenids can show increased benthic nutrient concentrations from mussel waste (e.g. feces and pseudofeces) released in the benthos (Roditi, Strayer and Findlay 1997). Filter feeding by mussels removes particulate P from the nearshore water column, and mussel wastes do not readily suspend far above the substratum, concentrating nutrients in the benthos and sediments, where bacteria must be involved in cycling of these nutrients. In Lake Michigan, this benthic eutrophication promotes nuisance blooms of the filamentous green algae, particularly Cladophora, which grow attached to dreissenid mussels. The alga supports diverse assemblages of epiphytic organisms, particularly diatoms (Lowe, Rosen and Kingston 1982;Dodds 1991;Young, Tucker and Pansch 2010). The dense filamentous benthic algal blooms can also provide diverse niches for bacteria (Zulkifly et al., 2012). The effects of these changes in food web and benthic nutrient accumulation related to mussels on benthic bacterial abundance and community structure are poorly understood.
The few studies of bacteria related to Dreissena showed that mussels can graze larger bacteria (Lavrentyev, Gardner and Yang 2000), and that benthic microbial communities associated with zebra mussels can be distinct from those associated with water and sediments, and suggested that mussels were promoting the enrichment of γ -proteobacteria (Frischer et al., 2000;Lohner et al., 2007). In Lake Erie, Dreissena was suggested to influence bacterial density, community structure and metabolic activity in the benthos (Lohner et al., 2007). Bacterial degradation of organic material in the benthos is essential in nutrient transformations and cycling in lakes (Lohner et al., 2007). This previous research highlighted the need for more multiple factor analyses to better characterize the consequences of dreissenid mussel invasion in lakes for benthic processes, and particularly the role of bacteria in nutrient cycling and aquatic food webs altered by dreissenid invasion.
The objective of this study was to examine the impact of dreissenid mussels and benthic algae on Lake Michigan sediment bacterial communities using an experimental microcosm approach. We hypothesized that additions of algae and/or mussels to microcosms would result in changes to water and sediment nutrient availability compared to control treatments. We also hypothesized that bacterial abundance and diversity would increase and that bacterial community composition would change in response to additions of mussels and/or benthic algae. To examine more detailed bacterial community composition than used in previous studies (Frischer et al., 2000;Lohner et al., 2007), changes in the bacterial diversity and community structure were assessed using 454 pyrosequencing.

Experimental microcosm design and preparation
Twelve wide-mouth 3.78 L glass jars formed the experimental microcosms testing four treatments in triplicate: (1) sediment with no additions (Control), (2) sediment with benthic algae, (3) sediment with mussels and (4) sediment with benthic algae and mussels. Sediment was collected at 1 m water depth from a site lacking mussel beds, in Lake Michigan near Milwaukee WI, (43.11 • N 87.89 • W) and subsamples were immediately pooled into a single mixed sample. 750 mL wet sediment was placed in each microcosm and then covered with pre-filtered lake water (0.22 μm Millipore, Bedford, MA). Benthic algae and dreissenid mussels were collected via SCUBA in Lake Michigan off Atwater Beach, WI (43.09 • N, 87.87 • W) at 10 m depth in September 2010. Epiphytic material and biofilms attached to mussels were removed by scrubbing with a toothbrush until no visible material could be scraped off the shells with a razor blade. Samples of benthic algae collected were predominately Cladophora sp., but other filamentous algal species and epiphytes were present. Benthic algal wet mass of 9.56 ± 0.44 g was added and/or ∼35 mussels of wet mass 57.24 ± 7.37 g. Mussels were suspended 1-2 cm above sediment in plastic mesh baskets. Microcosms were maintained in an incubation chamber at 11 • C supplied with ∼100 μmol photons m −2 s −1 fluorescent light (Alto, Philips, Eindhoven, Netherlands) on a 16:8 light: dark cycle. All microcosms were aerated with sterile air at a rate of ∼43 mL min −1 .

Mussel feeding
Mussels were fed every two days with microalgae (Pseudokirchneriella subcapitata Canadian Phycological Culture Collection CPCC 37, formerly Selenastrum capricornutum). Algal cells were grown in low P (20 μM K 2 HPO 4 ) DY-V freshwater algal growth medium (Andersen 2005). On feeding days, two new cultures were inoculated to replace used cultures. Each culture grew ∼7 days exhausting available phosphate, resulting in P-depleted cells mimicking natural phytoplankton in Lake Michigan. To feed mussels, two P. subcapitata batch cultures (2 L) were pooled and mixed in a sterile flask and 50 mL subsampled to estimate the starting cell concentration and available P. 300 mL was added to six feeding jars and diluted with 700 mL of filtered lake water (0.22 μm Millipore, Bedford, MA). Then mussel cages were removed from microcosms and suspended in individual feeding jars for 1 h; sterile stir magnets kept cells suspended. After feeding, mussel cages were rinsed in filtered lake water and returned to microcosms. At the conclusion of the experiment (30 days), mussels were examined: closed shells were counted as alive and open shells outside of water were counted as dead.

Water sampling and analysis
Microcosm water samples were collected from the water column and within sediment (pore water) for determination of dissolved and particulate N and P concentrations. Day 0 samples were collected prior to addition of algae or mussels, and then samples were collected on days 1-4, 7, 10, 14, 21 and 30. Water column samples were collected at 14 cm depth and pore water collected 2-3 cm under sediment surface. For dissolved nutrients, water samples were filtered (GF/F, Whatman Inc., Florham Park, NJ). All P analyses were completed immediately and samples for N analyses were stored at −20 • C. Water samples were analyzed for concentrations of soluble reactive phosphorus (SRP/molybdatereactive P), total dissolved phosphorus (TDP) and total phosphorus (TP). SRP was analyzed spectrophotometrically by the ammonium-molybdate method (Parsons, Maita and Lalli 1984). TP and TDP were analyzed on unfiltered and filtered water samples respectively following potassium persulfate digestion (Menzel and Corwin 1965) followed by SRP determination. Dissolved nitrite (NO 3 − ) and nitrate (NO 3 − ) concentrations were determined spectrophotometrically before and following cadmium column reduction (Parsons et al., 1984). Ammonium concentrations were assayed by a phenol-hypochlorite method (Parsons et al., 1984). In each microcosm, pH was measured using an Orion semimicro pH meter (Thermo Fisher Scientific, Waltham MA, USA). Measurements were collected at a depth of 14 cm prior to water and sediment sampling.

Sediment bacteria
Sediment was collected from each microcosm using sterile glass tubing (3 mm diameter) and a pipette bulb. Sediment samples for DNA analysis were transferred to microcentrifuge tubes and frozen at −70 • C. Samples collected for bacterial abundance counts were fixed with NaCl-buffered formaldehyde (3.7% v/v final concentration; Sigma Aldrich, St Louis, MO, USA). Cells were released from sediment samples by sonication according to the method in Velji and Albright (1993), stained with SYBR R green (Life Technologies, Carlsbad, CA, USA) and enumerated by examination under epifluorescence microscopy (Patel et al., 2007).
Bacterial community genomic DNA was extracted from three 1 g sediment subsamples from three replicate samples collected from each replicate microcosm on days 0, 7 and 21, using the FastDNA TM SPIN kit for soil (MP Biomedicals LLC, Solon, OH). For each microcosm replicate and day 7 or 21 time point, DNA extractions from the three sediment samples were pooled. Day 0 sediment samples (four in total) were collected from jars before the addition of treatments. A total of 28 samples were sent for sequencing.

pyrosequencing
Primer and PCR optimizations as well as sequencing were carried out at the Research and Testing Laboratory LLC (Lubbock, Texas) used a Genome Sequencer FLX Titanium System (Roche, Nutley, New Jersey) employing a bacterial tag-encoded FLX amplicon pyrosequencing approach as described previously (Wolcott et al., 2009) using titanium reagents, and a one-step PCR mixture with Hotstar HiFidelity taq polymerase (Qiagen, Valencia, CA, USA) with primers 28F GAGTTTGATCNTGGCTCAG and 519R GTNTTACNGCGGCKGCTG targeting 16S rRNA gene variable regions V1-V3 (Wolcott et al., 2009). Diverse amplicon variants were expected, so a unidirectional setup was utilized to increase the number of reads from a common primer starting point reducing the complexity of sequence processing.

Bacterial diversity analysis
To examine indices of bacterial diversity within the microcosms over time and between treatments, operational taxonomic units (OTUs) generated from bacterial sequence data were analyzed using mothur v. 1.21.1 (Schloss et al., 2009). All data processed in mothur was completed following the standard operating procedure for 454 pyrosequencing of 16S rRNA gene sequences. Amplicon noise was removed from all flow files using the Ampli-conNoise algorithm (Quince et al., 2011). All sequences <200 bp long were removed along with barcodes, primers and sequences with identical bases (homopolymers) longer than 8 bp. To further quality control the data, an alignment was completed using SILVA-compatible alignment database (Quast et al., 2013). With the training set produced during the alignment, chimeras were identified using Chimera slayer (Quince et al., 2011), executed in mothur, and removed prior to preparing analysis inputs. Sequences identified as chloroplast were removed. OTUs for each library were generated by a distance matrix followed by a clustering of sequences using a distance value of 0.03. The proportion of total taxa represented in each sample was calculated using Good's coverage estimate (Good 1953). Data was subsampled down to 800 sequences before generating rarefaction curves, Chao1 richness estimates and invSimpson diversity indices (Simpson 1949;Chao 1984), calculated based on OTUs. Rarefaction curves using OTU data output from mothur were designed using packages 'vegan' and 'mvpart'in the statistical programming language R.

DNA sequence analysis
To examine changes in bacterial taxa and identify which taxa changed significantly between treatments and over time, composition of bacterial genotypes was compared. All sequences were trimmed using CLC Genomics Workbench (CLC Bio, Muehltal, Germany). Classification of sequences to bacterial taxa was completed using the Ribosomal Database Project (RDP) and the RDP classifier (Wang et al., 2007), training set v. 7, based on a 98% identity confidence threshold. All individual sequence counts were normalized using the median sequence depth of the 28 libraries generated from pyrosequencing. In addition to analyzing relative numbers of taxa, absolute abundance was calculated using direct count data from microscopic counts of bacterial abundance. Differences in genotype presence (both relative and absolute abundance) between day 0, 7 and 21 and treatments were compared using the DESeq package within the Bioconductor software (Anders and Huber 2010), with the bacterial dataset after removal of the chloroplast sequences. DESeq estimated the variance across triplicate microcosms and looked for differential change in specific bacterial taxa between time and treatment comparisons. To run these comparisons, libraries were placed into nine 'groups': day 0, control day 7, algae-only day 7, mussel-only day 7, algae plus mussel day 7, control day 21, algae-only day 21, mussel-only day 21 and algae plus mussel day 21. All groups comprised of triplicate libraries (from three microcosms per treatment), except for day 0 which consisted of a total of four libraries. More than 20 different pairwise comparisons were completed using DESeq (based on significance of P < 0.05). Changes in individual taxa were expressed as the fold change compared to initial conditions (e.g. day 0), or to controls.

Statistical analysis
Water nutrient concentrations and pH, bacterial abundance, bacterial richness and invSimpson values measured or calculated for triplicate microcosms between the four treatments over time were compared by two-way repeated measures ANOVA (treatment, nutrient concentration, pH, bacterial abundance, richness, diversity and time as factors) using the statistical program SigmaPlot 12.0 (Systat Software Inc., San Jose, CA). Nonmetric multidimensional scaling of bacterial community composition in replicate microcosms for each of the four treatments was carried out in PAST v. 2.17c (Hammer, Harper and Ryan 2001) using a Morisita similarity measure. Table 1. Benthic algae wet mass and total mussel wet mass added to microcosm treatments on day 0, and change in mass of algae and mussels from day 0 to 30 (including dead mussels). Loss of mass is indicated as negative numbers. Dashes (-) represent microcosms that did not have the specific treatment added. For the mussel treatment, the total consumption of P. subcapitata cells by mussels is shown and total particulate P measured and estimates of C and N consumed in those algal cells over multiple feeding days during the 30 days microcosm experiment. Each column displays the mean (standard deviation) value from three microcosms for each treatment.

Changes in mussels and algae and mussel feeding
Algal addition microcosms lost more than 60% of the original algal mass during the 30 days experiment (Table 1). By comparison, mussel-only treatments (no algae added) gained a mean of 0.07 g algal biomass per microcosm, reflecting the growth of green filamentous algae on mussel and sediment surfaces. In the mussel and algae plus mussel treatments, between 3 and 6 of the initial 35 mussels died during the experiment which represented 10-14% of the initial mussel mass (Table 1). Mussels in the microcosms consumed about 6 × 10 9 algal cells during all the feeding periods over 30 days (Table 1). Using average particulate P measurements in the algal cells, this consumption represented a total particulate P input to microcosms of nearly 8 μmoles, and based on an average P. subcapitata cell stoichiometry (C:N:P ∼ 550:55:1; Van Donk et al., 1997), this would also represent >4 mmoles C and >400 μmoles N (Table 1).

Water column and sediment nutrients and pH
Addition of mussels resulted in larger increases in dissolved inorganic N (DIN) than in control and algae-only treatments ( Fig. 1; P < 0.001). The majority of DIN was nitrate (80-90% in day 30 mussel treatments). Patterns of changes in nitrite and nitrate concentrations were similar with nitrite at least an order of magnitude lower (the maximum nitrite concentration measured was 4 μM relative to maximum nitrate of 140 μM). Mussel additions resulted in higher water column and pore water dissolved nitrate (Fig. 1). Water column nitrate concentration decreased in the benthic algae-only microcosms (P < 0.001) and treatments without mussels had no changes in nitrite (P > 0.17). Dissolved ammonium concentrations were lower than and more variable than nitrate, but increased significantly in sediment pore water over 30 days when mussels were present (P < 0.001). The largest increase in dissolved ammonium occurred in algae plus mussel treatment (Fig. 1). Total DIN concentrations were higher in the water column than in the pore water (e.g. mussel treatment water column max. 144 μM compared to mussel treatment pore water max. 45 μM).
Mussel additions also resulted in SRP concentrations increasing from <1 μM at time 0 to >10 μM after 30 days (P < 0.001). In contrast, SRP, DOP, TDP and TP did not change in control treatments over 30 days, remaining <0.01 μM detection limit. 70-80% of the TP was SRP (Fig. 1), although mussel treatments had elevated DOP (1.2-2.8 μM after 30 days). Increases in dissolved P in mussel plus algae treatment lagged behind mussel-only treatment but after 20 days reached similar concentrations (Fig. 1). SRP was higher in water column than pore water (e.g. mussel treatment, water column max. 9.9 μM cf. pore water max.

μM).
Over 30 days, the pH range in microcosms was 6.89-8.90. Addition of mussels and/or algae resulted in a different pH to controls (P < 0.02), with lowest pH in the control treatment (6.89-7.82) and highest with mussels (7.91-8.90).

Bacterial abundance
Bacteria showed net abundance increases in all treatments between days 0 and 7, and day 0 and 30 but decreased from day 7 to 21, and there were no significant differences among controls and treatment microcosms (P > 0.08, two-way repeated measures ANOVA; Fig. 2). In the mussel-only treatment, bacterial abundance increased from 2.2 to 6.3 × 10 7 cells g −1 between day 0 and 7 but declined to ∼ 5 × 10 7 cells g −1 by day 30 (Fig. 2). After 30 days, the algae plus mussel treatment showed the highest bacterial abundance (6.8 × 10 7 cells g −1 ), while the lowest bacterial abundance was found in the mussel-only treatment.

Bacterial diversity comparisons
Bacterial communities were well represented in the libraries with Good's coverage estimates of 72-80% across all samples (Table 2). There were 213-344 unique OTUs found in samples, with the highest number of OTUs in the day 21 algae plus mussel and algae treatments. By day 21, bacterial richness was higher in treatments with algae plus mussel than controls, plus mussel and plus algae treatments on day 7 (P < 0.05; Table 2). Bacterial diversity increased in all microcosm libraries (P < 0.002; Table 2) but was not affected by treatment over 21 days (P > 0.1). Day 21 mean invSimpson indices for algae-only and algae plus mussel treatments were 43 and 54, respectively, indicating greater diversity than day 21 control microcosms (mean in-vSimpson index 21). Left-hand graphs are water column dissolved nutrients and right-hand column is pore water dissolved nutrients. The four treatments were control (no additions to sediment), plus benthic algae, plus mussels or algae plus mussels. The nitrate and nitrite values are shown together, but nitrate was always at least 90% of these combined values. Points are means of water samples collected from the triplicate microcosms for each treatment. Error bars (where visible) are standard deviation.
At the family level, there were strong similarities between microcosm treatments with dominant families present in similar proportions (Fig. 3). However, the distinct composition of bacterial families on a multidimensional scaling plot between the microcosm treatments was also related to both the elevated nutrients in mussel treatments and presence of algal biomass (Fig. 4), with the bacterial family composition of mussel plus algae treatments influenced by both nutrients and algae. Furthermore, pairwise comparisons of triplicate libraries for each treatment showed that 29 of the families changed significantly with time and/or treatment. By day 21, relative to control microcosms, Sphingobacteriaceae decreased and Pseudomonadaceae and Opitutaceae increased in all treatments (Fig. S1, Supporting Information). Other significant family-level changes included increases in Nitrospiraceae in both mussel treatments but decreases in the algae-only treatment. Nitrosomonadaceae increased in mussel-only microcosms. Some of the changes in bacterial families relative abundance between day 0 and 7 and 21 corresponded to changes in total bacterial abundance, which increased between day 0 and 7 but declined before day 21. For example, in the algae-only treatment, 11 families significantly increased between day 0 and 7, but 3 of these families, Neisseriaceae, Comamonadaceae and Rhodocyclaceae (3-20-fold increases), subsequently declined 2.5-11-fold between day 7 and 21.
Comparing triplicate libraries, 83 of the genera changed significantly with time and/or treatment. In pairwise comparisons between day 21 control and treatment microcosm libraries, there were some common and distinct changes in abundance of genera from the phyla Acidobacteria, Actinobacteria, Bacteriodetes, Nitrospirae, Planctomycetes, Proteobacteria, TM7 and Verrucumicrobia (Fig. 5). Addition of algae and/or mussels resulted in loss of Pelagibacter (SAR11 clade) and Limnobacter which were both present in control microcosms, along with significant decreases in other Proteobacteria genera. Relative to controls, the most changes in genera were in the mussel addition treatment. There were more common changes in algae vs algae plus mussel treatments and in mussel vs algae plus mussel treatments than between algae vs mussel treatments, but there were also distinct changes in each treatment relative to the control (Fig. 5). A number of genera disappeared or appeared relative to the control. N-transforming genera Nitrosomonas and Nitrospira appeared or increased in mussel addition treatments while Nitrospira declined in algae-only microcosms. The largest magnitude changes relative to controls were Acidovorax (43-fold decrease in algae, 7-fold decrease in algae plus mussels), and in all addition treatments, Caulobacter (5.5-13-fold decrease), Pseudomonas (4.2-6.9 increase), and Sphingobium (5-6-fold decrease). In both mussel addition treatments, Microbacterium (5.6-5.7-fold) and Leadbetterella (14-24-fold) increased. In algae plus mussels, GpIIa (21-fold) and Gp18 (9.1-fold) decreased, and in the musselonly treatment, genera that increased were Undibacterium (19.8fold), Luteolibacter (8.5-fold), Malikia (12.2-fold) and Devosia (7.4fold). In algae only, Paludibacter increased (9.2-fold).

DISCUSSION
This is the first study to address the need for more detailed analysis of the consequences of dreissenid mussel invasion in lakes for benthic bacterial community involved in nutrient transformation processes (Frischer et al., 2000;Lohner et al., 2007), by  using experimental manipulations to relate specific changes in freshwater sediment bacterial community to altered habitat and nutrient conditions associated with mussels and benthic algae.

Microcosm nutrients
The marked increases in N and P concentrations in microcosm water associated with mussel microcosms (Fig. 1) support the hypothesis that changes in nutrient concentrations in the water column and sediment pore water result from mussel and/or benthic algae additions, and the most marked effects of mussels were when there was no benthic algae present to take up released nutrients. The predominance of N as nitrate and P as SRP supports previous observations of Dreissena (Nicholls et al., 1999;Makarewicz, Bertram and Lewis 2000). Although mussels excrete NH 4 + (Arnott and Vanni 1996;Conroy et al., 2005), NH 4 + concentrations remained low (<3 μM); high NO − 3 concentrations were likely due to high nitrification rates in the presence of dreissenid mussels (Lavrentyev et al., 2000). Despite the enclosed microcosms, SRP, TP and NH 4 + concentrations accumulated were lower than in western Lake Erie (Conroy et al., 2005) but higher than is typical in nearshore Lake Michigan (Young et al., 2010). The less marked nutrient changes in sediment pore water than in the water column could relate to mussels being suspended slightly above the sediment in cages for feeding, so that excreted nutrients and feces/pseudofeces either diffused into the water column (dissolved N and P) or precipitated to the sediment (particulate N and P) (Schneider 1992). Bubbling of microcosms could also have allowed nutrients to remain dispersed in the water column. However, similar trends in dissolved N and P changes between sediments and the water column suggest that there were no fundamental differences in nutrient transformations between the two. The microcosm sediment depth was ∼3 cm, so may have remained aerobic. In contrast, sediment cores collected in Lake Michigan found oxygen saturation ∼0% at 2-3 cm depth (MacGregor et al., 2001a); such low oxygen will not only influence bacteria present but also the nutrient transformations occurring (Gardner, Nalepa and Malczyk 1987). If microcosm sediments had become anoxic, pore water NO 3 − should have been lower and NH 4 + higher than the water column. NH 4 + was higher in the sediment of mussel plus algae microcosms, supporting this idea. However, the sediment in our microcosms may not have represented the field conditions so well with respect to degree of anoxia which is more likely in deeper lake sediments.
Mussel feeding represented drawdown of particulate nutrients from the water column to the benthos ( Table 1) and release of mussel-excreted SRP. In mussel plus algae treatments, the delayed increase in SRP concentrations relative to the mussel-only treatment suggests algal SRP uptake. Algae were most likely Plimited at the beginning of the experiment (Auer et al., 1982;Young et al., 2010) and readily took up mussel-excreted SRP initially, but after 7 days, internal P saturation possibly suppressed algal SRP uptake (Young et al., 2010). This illustrates how P-limited benthic algae such as Cladophora in Lake Michigan, growing in close association with dreissenid mussels, effectively take up mussel-excreted nutrients, contributing to nearshore benthic nutrient trapping (Hecky et al., 2004).

Bacterial abundance and algal biomass changes
Although bacteria grew in all microcosms over 30 days, bacterial abundance fluctuated. The lack of distinct differences in bacterial abundance between microcosm treatments does not support the hypothesis of higher bacterial abundance associated with mussel additions. Introduction of dreissenid mussels in the eutrophic Hudson River, NY, was associated with increased planktonic bacterial abundance (Findlay, Pace and Fisher 1998). However, in oligotrophic lakes, including Lake Michigan, bacterial growth can be limited by organic carbon availability (Cotner and Wetzel 1992). Analyses of bacterial abundance in field sites with and without mussels and benthic algae would help clarify effects on benthic bacterial abundance.
Algae grew in mussel-only additions, despite initial cleaning of mussel shells. Dreissena can promote benthic algal growth in the Great Lakes (Pillsbury et al., 2002;Tomlinson et al., 2010). In algal addition treatments, there was a decline in algal biomass, possibly related to P limitation stress. However, algal biomass when mussels were also added appeared healthier for a longer time period, suggesting less nutrient stress, but the algae also degraded over the 30 days. Cladophora and other benthic algae are known to be difficult to maintain in laboratory culture (Hoffman and Graham 1984).
This degradation of algal biomass would have contributed nutrients to microcosms, resulting in increases in bacterial abundance coinciding with algal biomass loss; the highest bacterial abundance by day 21 was in the algae plus mussel treatment. A loss of 6 g algal mass (Table 1) represents ∼32 μmoles (9 μM) particulate P, in the algal addition microcosms so the lack of increase in dissolved P in the algae-only treatment (Fig. 1), suggests that P released by algal breakdown either remained as particulate matter or was used up by sediment microbes. Using published C:N:P ratio of 90:5:1 estimated for nutrient replete Cladophora glomerata (Stankovich 2004), this represents ∼2880 μmoles C contributed to microcosms, compared with >4000 μmoles C in microalgae consumed by mussel feeding. This comparison demonstrates that, in addition to particulate C harvested from the water column by efficient mussel feeding (e.g. Higgins and Vander Zanden 2010), release of C originally fixed by benthic algae growing in close association with mussels may also benefit growth of heterotrophic bacteria which can be limited by C supply in Lake Michigan (Cotner and Wetzel 1992).

Bacterial diversity and richness
This is the first study to present detailed genetic analysis of bacterial community composition in sediments associated with dreissenid mussels and benthic algae based on >280 000 identified sequences from 454 pyrosequencing. Bacterial diversity did not significantly change between treatments; therefore, the hypothesis that net bacterial diversity will be influenced by mussels and/or benthic algae was not supported ( Table 2). The starting sediment for the microcosms was low in organic matter, from a location lacking mussels and benthic algae, so it was not expected to support high bacterial abundance or diversity but diversity also increased in controls. Sampling over a longer period (40-50 days) may have resolved differences in bacterial abundance and diversity between treatments, but microcosm enclosure artifacts would also become more acute with prolonged incubation.
Increases in diversity in treatments with algae also resulted in higher bacterial richness (more unique OTUs) for both algae treatments by day 21 (Table 2). Previous work suggesting increases in bacterial richness with dreissenid mussels, assessed by DGGE, postulated that mussels created conditions with increased benthic substrate concentration and additional niches (Lohner et al., 2007). This study shows that benthic algae in both nutrient-poor (algae only) and nutrient-rich (mussels plus algae) environments can be associated with changes in bacterial community composition and higher benthic bacterial community richness, suggesting that the filamentous algae may be important in providing more niches for benthic bacteria. The filamentous benthic alga Cladophora has been shown to profoundly influence lake benthic communities by altering hydrodynamics and providing complex habitat for diverse epiphytic algae and bacteria (Besemer et al., 2007;Young et al., 2010;Zulkifly et al., 2013); sequences identified to eukaryotic chloroplasts indicated the presence of abundant diatoms, green algae and Cryptomonads in the microcosms.

Bacterial community change
Comparisons of bacterial communities across triplicate microcosms identified distinct community composition, related to algae and nutrients, as well as numerous benthic bacterial taxa that changed in abundance in response to mussel and/or benthic algae additions, supporting the hypothesis that bacterial community structure in the sediment changes with microcosm additions. Changes in bacterial community composition Figure 5. Changes in bacterial genera in day 21 microcosm libraries relative to day 21 control microcosm libraries, with common and distinct changes for the three addition treatments indicated. Genera which changed significantly (DESeq analysis, P < 0.05) are grouped as increasing or decreasing relative to control and genera in bold appeared or disappeared relative to controls. GIS-Genera incertae sedis.
Among the genera identified in microcosms, bacteria known to be pathogenic to humans, such as Enterococcus, Shigella and Campylobacter, were not found. The filamentous benthic alga Cladophora can harbor high densities of enterococci in polluted areas (Byappanahalli et al., 2003;Olapade et al., 2006;Englebert, McDermott and Kleinheinz 2008), but were not identified on healthy Cladophora (Zulkifly et al., 2012). Algae plus mussel treatments did support a potential fish pathogen, Cytophaga, which was also reported on Cladophora (Zulkifly et al., 2012).

Bacterial roles in nutrient transformations
Bacteria play critical roles in biogeochemical cycling of nutrients and organic matter in aquatic ecosystems and many nutrient transformations occur on or in sediments (Cross et al., 2005). Changes in bacterial community composition could respond to and effect changes in ecosystem nutrient cycling and thus functioning, particularly if changes occur within specific bacterial functional groups.
In terms of carbon cycling, changes in bacterial genera with algal additions included increases in the genera, Cellvibrio and Opitutus, known to degrade cellulose. Previous studies of bacteria associated with mussels but not benthic algae did not report these taxa (Frischer et al., 2000;Lohner et al., 2007;Winters et al., 2011), but a microbiome of Cladophora identified both genera (Zulkifly et al., 2012). Other potentially cellulose-degrading taxa, Sorangium and Byssovorax (Polyangiaceae) were present as <0.1% total genera. The benthic filamentous green alga Cladophora is a rich source of cellulose (Zulkifly et al., 2013), and filament breakdown in microcosms may have been facilitated by these bacteria. The increase in Paludibacter in algae-only treatments could relate to fermentation of algal C breakdown products; however, such anaerobic process in Paludibacter were unlikely in the aerobic microcosm conditions. Bacterial composition changes also suggested that mussel excretion influenced microbial nitrogen transformations and cycling. For example, the appearance of the ammonium oxidizing Nitrosomonas in day 21 mussel addition microcosm and the increase in Nitrospira in both mussel addition treatments would have been important in oxidizing ammonium outputs to nitrite and nitrate. Nitrifying bacteria, Nitrosospira and Nitrobacter were present in microcosms, helping convert ammonium to the more abundant nitrate. Nitrospira is commonly found in freshwater sediments (Altman et al., 2003) and Nitrospirales were associated with mussel gut tissue (Winters et al., 2011). Future examination of microbial nitrogen transformation in mussel-influenced sediments could target expression of genes encoding critical enzymes, for example, ammonium oxidation (e.g. amo genes).
Other taxa critical for N transformations were denitrifying/nitrate reducing genera, Anaeromyxobacter, Blastobacter, Comamonas, Flavobacterium, Hyphomicrobium, Methylobacter, Opitutus, Pseudomonas, Rhodobacter, Rubrivivax, Sphingopyxis and Sphingomonas; increases in Pseudomonas were common to all microcosms with mussels and /or algae but not in controls. The decrease in Caulobacter in treatments relative to control microcosms could also relate to very low nutrient conditions in the lakes. The species Caulobacter crescentus is known to be associated with oligotrophic conditions (Entcheva-Dimitrov and Spormann 2004) so increasing nutrients from mussels and algae may have changed competition to disadvantage Caulobacter in the microcosms with additions. The algae-only microcosm maintained low water column inorganic N but some increases in sediment ammonium concentrations, possibly related to increases in diazotrophic Dechloromonas in algae-only but decreases in the mussel-only microcosms. There is evidence for N 2 fixation in Lake Michigan, despite areas of elevated nitrate (MacGregor et al., 2001b). Sediment bacteria identified could also be critical in increases in denitrification related to mussel excreta release in the benthos of Lake Michigan (Rowe, Kreis and Dolan 2014).
The disappearance of Pelagibacter (SAR11) from all microcosms with additions of algae and/or mussels may relate to preference for oligotrophic conditions (Salcher et al., 2011). While SAR11 is abundant in marine systems, it was until recently thought to be rare in freshwaters. Pelagibacter in this study probably aligns with the freshwater LD12 lineage (Salcher et al., 2011), and in nutrient-enriched microcosms may have been outcompeted by taxa with faster nutrient uptake and growth rates.
Microbial transformation of other nutrients could also be inferred from known functions of other taxa that were observed in the microcosms; reduction of iron and other metals with the genera Anaeromyxobacter, Rhodoferax, Geobacter, Dechloromonas and Ferribacterium, along with the genus Pedomicrobium which was only present in algae-addition microcosms. Sulfate-reducing taxa Desulfobulbus, Desulforhopalus and Desulfovibrio were probably responsible for S cycling in the microcosms but the thiosulfate oxidizer Limnobacter associated with benthic algae (Zulkifly et al., 2012) disappeared when mussels or algae were added. Loss or decrease of the methanotroph Methylibium and the methylamine utilizer Methylotenera from algal addition microcosms, and increase in methane-oxidizing Methylococcus in mussel-only microcosms suggests methane transformation in mussel-associated bacteria.

CONCLUSIONS
This study demonstrates that mussels and algae can affect bacterial communities in lake sediments and suggests that many of the taxa are responding to habitat changes, including elevated benthic nutrient concentrations, and presence of algal biomass typical of benthic regions of Laurentian Great Lakes which have been invaded by dreissenid mussels and support benthic algal blooms. The bacteria observed in the microcosms represent bacterial communities which are critical to nutrient regeneration and cycling in the lake habitat, particularly since much of the allochthonous nutrients are trapped and transformed in the nearshore benthic sediments (Hecky et al., 2004). This study indicates that changes in aquatic bacterial community composition associated with dreissenid mussel invasion can affect nutrient cycling, but also suggests that the benthic algae are important as bacterial habitat and may supply carbon substrates to support bacterial growth. Further detailed study of lake bacterial communities will facilitate incorporation of specific bacterial nutrient transformation processes into biogeochemical nutrient cycling models.