Structure and Composition of Aggregates in Two Large European Rivers, Based on Confocal Laser Scanning Microscopy and Image and Statistical Analyses

Abstract

Floating riverine aggregates are composed of a complex mixture of inorganic and organic components from their respective aquatic habitats. Their architecture and integrity are supplemented by the presence of extracellular polymeric substances of microbial origin. They are also a habitat for virus-like particles, bacteria, archaea, fungi, algae, and protozoa. In this study we present different confocal laser scanning microscopy strategies to examine aggregates collected from the Danube and Elbe Rivers. In order to collect multiple types of information, various approaches were necessary. Small aggregates were examined directly. To analyze large and dense aggregates, limitations of the technique were overcome by cryo-sectioning and poststaining of the samples. The staining procedure included positive staining (specific glycoconjugates and cellular nucleic acid signals) as well as negative staining (aggregate volume) and multichannel recording. Data sets of cellular nucleic acid signals (CNAS) and the structure of aggregates were visualized and quantified using digital image analysis. The Danube and Elbe Rivers differed in their aggregate composition and in the relative contribution of specific glycoconjugate and CNAS volume to the aggregate volume; these contributions also changed over time. We report different spatial patterns of CNAS inside riverine aggregates, depending on aggregate size and season. The spatial structure of CNAS inside riverine aggregates was more complex in the Elbe River than in the Danube River. Based on our samples, we discuss the strengths and challenges involved in scanning and quantifying riverine aggregates.

In rivers, primary particles are frequently and perhaps characteristically transported as larger flocculated aggregates. They are structurally very stable because they are exposed to a constant shear force, resulting in relatively small aggregates compared to aggregates in lakes and marine systems (41, 49). Abiotic mechanisms such as physical coagulation, collision frequency, and stickiness are involved in particle aggregation (12). These aggregates, which may be regarded as mobile biofilms (e.g., 8, 9), can be very heterogeneous in their composition. Up to 97% of the biofilm matrix is actually water (46). Apart from water, biofilms may consist of dissolved, colloidal, and particulate materials varying in size and composition (12). They are composed of a complex mixture of components including inorganic (minerals), living organic (bacteria, archaea, fungi, algae, protozoa, and viruses), and nonliving organic (extracellular polymeric substances [EPS], allochthonous and autochthonous detritus, lignins, tannins, etc.) material from the respective aquatic habitat and its terrestrial environment (41). Cellular material within a biofilm can vary greatly. Measurements of organic carbon content suggest that cellular material represents 2 to 15% of the biofilm (46). Up to 95% of the biofilm is composed of EPS (13, 23, 42). The actual structure of the biofilm matrix varies greatly depending on the microbial cells present, their physiological status, the nutrients available, and the prevailing physical conditions (46).

Knowledge about the structure and the function of aggregates, both in environmental and engineered systems, is very important (12). In engineered systems such as wastewater treatment plants, understanding flocculation can help in the management of that process. In environmental systems, such structure-function relationships can provide ecologically relevant information about material transfers between particulate and dissolved matter or about spatial distribution of microorganisms, with the related impacts on the aquatic food web. Numerous methods are available to help characterize aggregate properties. Microscopic as well as photographic techniques have been used to analyze aggregate structure. In recent years, confocal laser scanning microscopy (CLSM) data sets have allowed the visualization and quantification of three-dimensional (3-D) structures (18, 31).

In this study, we analyzed aggregates from the Danube and Elbe Rivers by collecting reflection, nucleic acid, glycoconjugate, and negative stain signals using CLSM. In order to receive multifarious information about the aggregates, various approaches were necessary: small aggregates were examined directly, and large and dense aggregates were physically sectioned and poststained. Although most of the detected nucleic acid signals derive from bacteria, we refer to them as cellular nucleic acid signals (CNAS) including potential archaea and virus signals. Nucleic acid signals can potentially also be obtained from fungi, algae, and protozoa. But the detection of extracellular DNA can be excluded due to its type of appearance (7). Data sets of specific glycoconjugates, CNAS, and aggregate structure were visualized and quantified by using digital image analysis. The distribution of CNAS within riverine aggregates was determined by autocorrelation. Based on our samples, we describe the strengths and challenges in scanning and quantifying riverine aggregates using CLSM. Additionally, we discuss structure, function, and potentially important differences in aquatic aggregates from these two large European rivers.

MATERIALS AND METHODS

Study sites and sampling.

The Danube River was sampled at Wildungsmauer, Austria (16°48′0″E, 48°6′0″N), river kilometer 1894, located in the national park area on the southern bank of the river. Samples of the Elbe River were collected on the left downstream bank at river kilometer 318, near Magdeburg, Germany (11°40′0″E, 52°4′0″N). Both locations are in the braided river region and, therefore, can be regarded as representative and comparable areas. We sampled repeatedly during an annual period to accommodate typical seasonal changes in water discharge, temperature, and variable suspended solid loads and quality. Hence, water samples from the Danube River were taken in spring (3 May 2005), summer (20 July 2005), fall (13 October 2005), and winter (8 January 2006). From the Elbe River, water samples were taken in spring (21 March 2005), summer (8 August 2005), fall (3 November 2005), and winter (23 January 2006). Sampling depth was approximately 30 cm.

Riverine aggregates were sampled in 1-liter Plexiglas bottles. The samples were always kept at +4°C until analysis, which was completed within 24 h (10). With an inverted 10-ml glass pipette (14), the aggregates were carefully transferred to Eppendorf tubes, where staining was performed.

Abiotic and biotic parameters.

Chemical analyses of nitrate, nitrite, ammonium, orthophosphate, total soluble reactive phosphorus, and total phosphorus were performed based on German Standard Methods for the Examination of Water, Wastewater and Sludge. Determinations of total suspended solids, particulate organic matter, and chlorophyll-a concentrations are published elsewhere (29). In the Danube River, particle abundance was measured with a Galai CIS100L Laser Analyzer (Ankersmid Ltd., Yokneam, Israel); in the Elbe River an optical measuring instrument, PartmasterL (Aucoteam GmbH Berlin, Berlin, Germany), was used. The measuring range for particle abundance was from 2 to 200 μm.

Staining procedure.

Table 1 displays the probes used for characterizing riverine aggregates. To stain the major parts of the matrix material of the aggregates, lectins from Aleuria aurantia (Vector Laboratories, Burlingame, CA) labeled with the fluorochrome Cy5 (Amersham, Buckinghamshire, United Kingdom) or Phaseolus vulgaris (Sigma Aldrich, St. Louis, MO) labeled with fluorochrome Alexa Fluor 633 (Invitrogen, Eugene, OR) were employed in parallel to stain the lectin-specific EPS (30, 32, 44). The lectins were self-labeled according to the data sheet of the supplier.

TABLE 1.

Probes and fluorochromes with their specificities and spectral characteristics

Staining technique and aggregate characteristic or component	Probe(s)a	Fluorochrome	Specificity/target molecule	Ex/Em (nm)b	Source(s)	Reference(s)
Positive staining
Specific glycoconjugates in the EPS matrix	AAL	Cy5	l-Fucose	649/670	Vector Laboratories (AAL), Amersham (Cy5)	30, 32, 44
	PHA-E	Alexa Fluor 633	Galβ(1,4)GlcNAcβ(1,2) Man, not inhibited by simple sugars	632/647	Sigma Aldrich (PHA-E), Invitrogen (Alexa Fluor 633)
CNASc	SYBR Green I		Double-stranded DNA	497/520	Invitrogen	33
Negative staining
Volume		Rhodamine B		570/590	Invitrogen	24
		Rhodamine 123		507/529
		Rhodamine 6G		528/551
		Fluorescein		492/518	Sigma Aldrich
		Cy5		649/670	Amersham
		Alexa Fluor 568		578/603	Invitrogen
	Dextran conjugates	Fluorescein, lysine fixed, anionic; MW of 2,000,000		494/521	Invitrogen
		TRITC, lysine fixed; MW of 2,000,000		555/580
		Alexa Fluor 568; MW of 500,000		578/603
		Rhodamine B, neutral; MW of 70,000		570/590

^aAAL, A. aurantia; PHA-E, P. vulgaris.

^bEx, peak excitation wavelength; Em, peak emission wavelength. Excitation and emission peaks are according to the data sheet of the supplier.

^cIncluding potential archaea and virus signals.

The EPS glycoconjugates were stained with one of the lectins as described previously (30). Briefly, for lectin staining, the lectins were diluted with deionized water to a final concentration of 0.1 g ml⁻¹ of protein. One hundred microliters of this solution was added to each sample and incubated for 20 min in the dark at room temperature. The aggregates were then carefully washed three times with tap water to remove unbound lectins and were never allowed to dry in the air. After staining the lectin-specific EPS compounds, SYBR Green I (concentration of 1 μl ml⁻¹ of deionized water; Invitrogen) was directly applied to the aggregates and incubated for 5 min in the dark at room temperature to detect aggregate-associated CNAS on/in the fully hydrated aggregates. SYBR Green I is a nucleic acid-specific stain widely used to detect prokaryotes and virus particles in aquatic environments (33).

The dimensions and volumes of the aggregates were determined by a negative staining procedure. Lawrence et al. (24) demonstrated the use of fluorescein and size-fractionated fluor-conjugated dextrans in conjunction with CLSM to directly monitor and determine diffusion coefficients within biofilms. We took advantage of this method to determine aggregate volumes. Different probes and fluorochrome-staining solutions such as dextrans (Invitrogen), rhodamines (Invitrogen), fluorescein (Sigma Aldrich, St. Louis, MO), Cy5 (Amersham), and Alexa Fluor 568 (Invitrogen) were tested. These stains and probes are characterized by their chemical properties and molecular weights (MWs) and were used at different concentrations. Probes and fluorochrome-staining solutions such as dextrans, rhodamines, fluorescein, and Cy5 were not appropriate for this analysis, probably due to their MWs or their binding properties (data not shown). Negative staining with Alexa Fluor 568 allowed us to calculate the aggregate volumes. Hence, to analyze our samples, we stained the water phase—the volume of the scanned box which was not occupied by the aggregate—with the fluorochrome Alexa Fluor 568 (C₃₇H₃₃N₃O₁₃S₂; MW of 791.80). The positively stained samples were carefully transferred into Cover Well imaging chambers (0.5-mm spacer; Invitrogen), covered with 150 μl of Alexa Fluor 568, and immediately examined by CLSM.

CLSM.

Aggregates were analyzed by CLSM using a Leica TCS SP1, controlled by the LCS software (version 2.61; build 1537 174192) (Leica, Heidelberg, Germany) and equipped with an upright microscope. Images were collected with a 63× water lens (0.9 numerical aperture).

The aggregate structure was analyzed by CLSM using visible lasers (488 nm, 568 nm, and 633 nm). Emission signals were detected from 480 to 500 nm (inorganic and mineral compounds; detected by reflection signals), 500 to 550 nm (DNA signals; detected with SYBR Green I), 570 to 625 nm (negative staining signals; detected with Alexa Fluor 568), and from 650 to 750 nm (lectin signals; detected with fluorochrome Cy5 or Alexa Fluor 633). Optical sections of aggregates were taken every 1 μm.

Cryo-sections.

Conventional analyses of larger riverine aggregates (traditional staining procedure and scanning with a view from the top) were limited by laser penetration and diffusion of staining solutions. To obtain information on the distribution of lectin-specific glycoconjugates and CNAS inside large riverine aggregates, cryo-sections were performed (19). Aggregates from the Danube and Elbe were embedded in liquid cryostat medium (Neg-50; Richard-Allan Scientific), frozen at −26°C, and physically sectioned (50 μm) with a cryo-microtome (Leica CM3050S). Thereafter, sections were stained on slides with lectins and SYBR Green I (see above) and covered with a coverslip before analysis by CLSM (see above).

Quantification and 3-D visualization.

Quantification of CLSM data sets of the whole aggregate (optically sectioned material) and cryo-sections was done with the freely available software ImageJ (http://rsb.info.nih.gov/ij/), developed in Java (45). To analyze cryo-sections with the software ImageJ, special plug-ins were developed. After thresholds were established, the binary image of each cryo-section was divided into three sectors. Within each sector, one subsample area of rectangular shape extending across the whole dimension of the aggregate's cryo-section was chosen (see Fig. 4). The dimension of the aggregate was given by the lectin-specific glycoconjugate. The software cropped each subsample, and pixels of the CNAS were quantified.

FIG. 4.

Single scans of different cryo-sections showing distribution of CNAS (including potential archaea and virus signals) inside an aggregate (a and d). Insets in panels a and d show glycoconjugate and CNAS. The white rectangle indicates a subsample for CNAS volume calculation. The distribution of the calculated CNAS volumes within the subsamples of the cryo-sections is presented (b and e). Spatial autocorrelation shows the distribution and accumulations of CNAS within the cryo-sections of the aggregates (c and f). Autocorrelations for lags 1 to 1/4 of the total number of slices were calculated. The first local maximum provides information on CNAS layer width. Secondary maxima indicate a regular spatial distribution of CNAS within the cryo-section, pointing to the repeating occurrence of CNAS layers (whose width can be estimated from the first maximum). Black lines indicate the level of significance (twice the standard error). Color allocation: green, nucleic acid; red, glycoconjugates; yellow, CNAS in or in contact with lectin-specific extracellular polymeric substances. Scale bar, 15 μm.

To quantify and calculate structural parameters, the interface description language-based program Confocal Analysis (developed by BioCom for the Helmholtz Centre for Environmental Research, UFZ, Magdeburg, Germany), version 1.31, was also applied. This program can handle multichannel data and colocalization of data.

For each channel and data set, the threshold was set manually. Due to the very heterogeneous composition of the aggregates, automatic batch processing could not be applied.

To visualize 3-D data sets, Imaris, version 4.2 (Bitplane AG, Zurich, Switzerland), and Amira, version 3.0 (TGS, San Diego, CA), were used. For each channel and image, thresholds were set manually. Adobe Photoshop CS2 was used to insert calibration bars into the images.

Calculations and statistics.

For statistical analysis, the software program SPSS, version 12.0, for Windows (SPSS, Chicago, IL) was used. The criteria of normal distribution (Kolmogorov-Smirnov test) and homogeneity of variances (Levene test) were not met, so seasonal differences in the data were determined with a nonparametric test (Friedman test). A Tamhane posthoc test (which does not require homogeneity of variances in the data) was used to determine significant differences between seasons. Differences between the two rivers were tested using a Mann-Whitney U test.

To determine the distribution of CNAS within riverine aggregates, cryo-sections were analyzed. Only those subsamples that included a minimum of 10 slices containing CNAS were included in the analysis. Two approaches were used: the first was used to detect abundance differences between the subsamples of the three sectors, and the second was used to characterize the internal structure of the cryo-sections using a finer spatial resolution.

(i) Approach 1.

The mean abundances of CNAS within each subsample were calculated. Then, differences between the two outer subsamples and the middle subsample were computed to detect potential spatial accumulation of cells. The averages of these two differences were tested against zero using a Wilcoxon test. Mean differences between abundances will only be significantly different from zero if CNAS are more abundant either on the two outer subsamples or in the middle subsample of the cryo-section.

(ii) Approach 2.

To test whether CNAS occurred in accumulations anywhere within the cryo-sections of the aggregates and to test for regularity in distributions of CNAS within the cryo-sections, we calculated the spatial autocorrelation (25). Since the total number of slices per subsample was variable, autocorrelations for lags 1 to 1/4 of the total number of slices were calculated. If the autocorrelation values were higher than twice the standard error for positive autocorrelations, they were considered significant. The highest autocorrelations always occurred at lag 1 and then typically decreased uniformly. The first local maximum was defined as the number of lags with significant positive autocorrelations, starting from lag 1 until the significance level was reached or the autocorrelations started to increase again. This first maximum provides information on the width of the CNAS layers. The width of the layers in micrometers can be calculated from the number of lags of the first local maximum +1. Local secondary maxima were counted only if (i) positive autocorrelations were higher than twice the standard error and (ii) the difference between local maximum and presiding minimum was higher than 0.05 autocorrelations. These secondary maxima indicate a regular spatial distribution of CNAS within the cryo-section, pointing to the repeating occurrence of CNAS layers (whose width can be estimated from the first maximum). For example, if a secondary maximum occurs at lag 20, most CNAS layers will be placed approximately 20 μm apart. To characterize the position of secondary maxima, we used the lag numbers at which the respective secondary maxima occurred. For further calculations, we defined the following variables: layer width (calculated from the first local maximum as described above), spatial structure (the number of secondary local maxima), diameter of cryo-section (the total number of slices of each subsample), and percentage of CNAS (the percentage of slices that harbored CNAS within each subsample).

The structural parameters derived from the autocorrelation analysis (layer width and spatial structure) may not be independent from aggregate size and the overall occurrence of CNAS within it. Hence, two multiple linear regressions were carried out to define this relationship, with the diameter of cryo-section and percentage of CNAS as independent variables and with layer width and spatial structure as dependent variables. The residuals from these two regressions gave estimates for layer width and spatial structure without the effect of aggregate size and percentage of CNAS. As mentioned before, each cryo-section was divided into three sectors and, within each sector, one random subsample was chosen. For these three groups of subsamples the regressions described above were calculated separately to (i) determine whether the position within the aggregate had a strong effect on the regressions and (ii) to avoid having strongly dependent samples within one regression analysis. The residuals of these regressions were used as dependent variables in a second set of multiple regressions, where season and river (dummy coded) were defined as the independent variables. The objective was to determine whether seasonal or between-river differences in the spatial structure could be detected when the effects of aggregate size and percentage of CNAS were factored out.

RESULTS

Abiotic and biotic parameters characterizing the Danube and Elbe Rivers.

The Danube River and Elbe River were characterized by an annual mean discharge of 1,818 m³ s⁻¹ and 469 m³ s⁻¹, respectively, during the sampling period. The measured chemical characteristics, particle abundances, seston parameters, and chlorophyll-a values for both rivers during the four seasons are summarized in Table 2. On average, nutrients, seston parameters, particle abundances, and chlorophyll-a values were higher in the Elbe River than in the Danube River.

TABLE 2.

Selection of abiotic and biotic parameters characterizing the Danube and Elbe Rivers during the four seasons of the sampling period

Parametera	Danube River				Elbe River
	Spring	Summer	Fall	Winter	Spring	Summer	Fall	Winter
Discharge (m3 s−1)	2,809	2,656	1,236	935	1,459	330	260	379
Conductivity (μS cm−1)	415	367	408	552	1,262	1,106	1,523	1,380
P-PO4 (μg liter−1)	31	30	34	48	53	78	61	81
P-Psoluble (μg L−1)	36	35	35	49	61	83	70	88
Ptotal (μg liter−1)	74	87	56	63	202	129	352	145
N-NO3 (μg liter−1)	2,487	1,497	1,829	3,048	2,526	3,478	5,627	5,802
N-NO2 (μg liter−1)	49.1	14.9	6.5	13.3	9.7	0.6	20.1	45.1
N-NH4 (μg liter−1)	3	44	35	9	25	13	164	90
No. of particles (107 liter−1)	1.07	2.61	0.36	0.05	25.21	11.38	4.18	3.83
Relative contribution of POM to TSS (%)	11.02	6.98	16.24	30.41	16.15	36.01	16.35	40.36
Chlorophyll-a (μg liter−1)	0.97	0.56	0.75	0.37	0.87	10.37	1.03	0.43

^aP-PO₄, orthophosphate; P_soluble, total soluble reactive phosphorus; P_total, total phosphorus; N-NO₃, nitrate; N-NO₂, nitrite; N-NH₄, ammonium; TSS, total suspended solids; POM, particulate organic matter.

Scanning aggregates under in situ conditions.

Aggregates were examined after a conventional staining procedure and scanned by simple view from the top. The aggregates incorporated a large amount of noncellular matter such as sand, clay, and detritus. This limited diffusion of the staining solution, laser penetration, and detection of emission signals in our samples. This approach enabled us to properly investigate only the outer 20 μm of the aggregates (Fig. 1).

FIG. 1.

3-D reconstruction of a riverine aggregate. Four-channel presentation of mineral compounds, glycoconjugates, and CNAS (including potential archaea and virus signals) as well as negative stain. Polygons indicate border of aggregate dimensions based on negative staining (see Materials and Methods) without specific glycoconjugates and CNAS. Color allocation: reflection mode, gray; nucleic acid, green; glycoconjugates, red.

Distribution of aggregates' components based on the outer 20 μm.

First, the aggregate volume was calculated by subtracting the negative stain volume from the scanned box volume. In some cases the fluorochrome Alexa Fluor 568 penetrated into the aggregate matrix and revealed the existence of channels and pores within aggregates. Therefore, colocalization of the specific glycoconjugate and the negative stain was used to calculate the aggregate volume. Colocalization is a tool for quantifying the degree of association or codistribution of labeled structures between any two channels in an image (34). In the first step of the calculation, we subtracted the volume of colocalized specific glycoconjugates and negative stain from the negative stain; in the second step we subtracted this term from the scanned box volume.

(i) Specific glycoconjugate composition of aggregates.

During the four seasons on average 59.95% (standard deviation [SD], 18.67; n = 124) in the Elbe River and 36.49% (SD, 23.86; n = 88) in the Danube River of the aggregate volume were labeled with lectins (Fig. 2a). In a comparison of the relative contribution of the specific glycoconjugate volume to the aggregate volume from both rivers during the four seasons, significant differences occurred in spring, summer, and fall (Mann-Whitney U test, P values of <0.001, 0.034, <0.001 for 53, 49, and 57 samples for spring, summer, and fall, respectively). No significant differences were found in winter (Mann-Whitney U test, P = 0.318; n = 51).

FIG. 2.

Averages of the specific glycoconjugate volume (a), the CNAS (including potential archaea and virus signals) volume (b), and the remaining nonvisualized components (c) to the aggregate volume (in percent) from the Danube River (spring, n = 18; summer, n = 20; fall, n = 30; winter, n = 20) and Elbe River (spring, n = 34; summer, n = 29; fall, n = 27; winter, n = 34) during the four seasons. Error bars indicate SDs of all inspected aggregates. Note different scales of y-axes.

A comparison of the relative contribution of the specific glycoconjugate volume to the aggregate volume of the Danube (Friedman test, P < 0.001; n = 14) and the Elbe (Friedman test, P = 0.003; n = 14) on a seasonal basis revealed significant differences in both rivers. In the Danube, spring values were significantly lower than in summer (Tamhane, P < 0.001), fall (Tamhane, P < 0.001), and winter (Tamhane, P < 0.001). Furthermore, the relative contribution of the specific glycoconjugate volume in summer was also significantly higher than in fall (Tamhane, P < 0.001). During fall in the Elbe River, this relative contribution was significantly higher than in spring (Tamhane, P = 0.008), summer (Tamhane, P < 0.001), and winter (Tamhane, P < 0.001).

(ii) CNAS associated with aggregates.

On a yearly average, the relative contribution of CNAS volume to aggregate volume was higher in the Elbe (mean, 1.53%; SD, 2.51; n = 124) than in the Danube (mean, 1.00%; SD, 3.35; n = 88). This was not always the case when the values from both rivers were compared for the four seasons individually. Significant differences occurred in spring, fall, and winter (Mann-Whitney U test, P values of <0.001, 0.007, and 0.006 for 52, 57, and 51 samples for spring, fall, and winter, respectively), but no significant differences occurred in summer (Mann-Whitney U test, P = 0.20; n = 49) (Fig. 2b).

When we tested for seasonality in this relative contribution within the rivers, both the Danube (Friedman test, P < 0.001; n = 13) and the Elbe (Friedman test, P < 0.001; n = 26) exhibited significant differences. During spring in the Danube, the relative contribution of the CNAS volume was significantly higher than in fall (Tamhane, P < 0.001). In the Elbe, values were significantly lower in spring versus summer (Tamhane, P = 0.02), fall (Tamhane, P = 0.02), and winter (Tamhane, P < 0.001). Furthermore, the relative contribution was also significantly lower in summer than in winter (Tamhane, P = 0.001).

(iii) Remaining, nonvisualized components.

The variability of the remaining material, which was not stained by SYBR Green I and the specific lectin, was also assessed (Fig. 1). The nonvisualized components are based on the volume calculated by negative staining. On average, the relative contribution of these volume components to the aggregate volume was higher in the Danube (mean, 62.84%; SD, 24.22; n = 88) than in the Elbe (mean, 38.48%; SD, 18.70; n = 124) during the four seasons.

When the values from both rivers were compared for the four seasons individually (Fig. 2c), significant differences occurred in spring, summer, and fall (Mann-Whitney U test, P values of <0.001, 0.033, and <0.001 for 51, 49, and 57 samples for spring, fall, and winter, respectively), but no significant differences occurred in winter (Mann-Whitney U test, P = 0.084; n = 50).

When we tested for seasonality in the relative contribution of nonvisualized components within the rivers, both the Danube (Friedman test, P < 0.001; n = 16) and the Elbe (Friedman test, P < 0.001; n = 23) exhibited significant differences. During spring in the Danube, the relative contribution of the nonvisualized component volume was significantly higher than in summer (Tamhane, P < 0.001), fall (Tamhane, P < 0.001), and winter (Tamhane, P < 0.001). Furthermore, in summer the relative contribution of this volume was significantly lower than in fall (Tamhane, P < 0.001). In the Elbe, values were significantly lower in fall than in spring (Tamhane, P = 0.005), summer (Tamhane, P < 0.001), and winter (Tamhane, P < 0.001).

Cryo-sections.

Cryo-sections in combination with poststaining provide information on the presence and distribution of CNAS and glycoconjugates inside the aggregates (Fig. 3).

FIG. 3.

(a) Single scan of an unsectioned aggregate. CNAS (including potential archaea and virus signals) and glycoconjugates deeper inside cannot be investigated due to scattering properties (black) of the aggregates themselves and the limited depth of laser penetration. (b) Single scan of a cryo-section showing distribution of CNAS and glycoconjugates inside an aggregate. Color allocation in both panels: green, nucleic acid; red, glycoconjugates; yellow, CNAS in or in contact with lectin-specific extracellular polymeric substances; blue, negative stain (Alexa Fluor 568). Scale bar, 10 μm.

Forty-six cryo-sections were analyzed to obtain quantitative information on the CNAS distribution within riverine aggregates. First, we tested whether the subsamples of the three sectors (see Materials and Methods) differed in their CNAS distributions. No differences were found either in the Danube (Wilcoxon test, P = 0.529; n = 66) or in the Elbe (Wilcoxon test, P = 0.429; n = 56) using data from all seasons.

Second, we tested whether CNAS exhibited a homogeneous distribution or whether they occurred in accumulations within the aggregates. Autocorrelations describe the spatial distribution of CNAS within riverine aggregates. Two examples are given in Fig. 4. Figure 4a to c depict a broad layer width of CNAS (45 μm) and no further significant spatial structure. In comparison, Fig. 4d to f depict a relatively narrow layer width (12 μm) and a clear spatial structure. In the Danube, this width averaged 10.38 μm; in the Elbe it averaged 9.95 μm (Fig. 5a). On average, the spatial distribution of CNAS was more complex in the Elbe River (Fig. 5b). In the Danube, 42% of the examined subsamples contained at least one secondary maximum; the respective value for the Elbe was 52%. For the Elbe the secondary maxima were positioned over a wider range of lags (highest values were 87 for the Elbe and 75 for the Danube), with most values occurring either between 10 and 15 lags or 20 and 25 lags. In the Danube most values occurred either between 10 and 15 lags or between 25 and 30 lags.

FIG. 5.

Comparison of the Danube and Elbe Rivers based on the layer width which harbors CNAS (including potential archaea and virus signals) (a) and the occurrence of secondary maxima (b). Cryo-sections of all four seasons are considered.

Furthermore, two sets of multiple linear regressions were carried out with cryo-section diameter and percentage of CNAS as independent variables and with layer width and spatial structure as dependent variables (Table 3). The three separate regressions for the subsamples gave similar, but not identical, results: slopes for each parameter had the same direction and were within the same order of magnitude within one set of regressions. Ten out of the 12 slope values showed a P value of <0.05. Generally, layer width as well as spatial structure could be significantly predicted by the cryo-section diameter and the CNAS percentage. Increasing diameters of the cryo-sections led to broader CNAS layers and more spatial structure. Layer width also increased with an increasing CNAS percentage, but spatial structure was reduced as this percentage increased. Thus, differences in layer width and spatial structure also reflect differences in aggregate size and colonization.

TABLE 3.

Effect of aggregate size and colonization on layer width of CNAS (cellular nucleic acid signals including potential archea and virus signals) and spatial structure within the three subsamples^a

Variable and sector	Overall multiple linear regression		Intercept		Slope of diameter		Slope of % CNAS
	r2	P	Regression coefficient	P	Regression coefficient	P	Regression coefficient	P
Layer width of CNAS
Sector 1	0.268	0.001	−4.229	0.244	0.033	0.010	0.160	0.000
Sector 2	0.174	0.010	−2.942	0.547	0.043	0.004	0.083	0.131
Sector 3	0.245	0.002	0.096	0.971	0.025	0.006	0.098	0.006
Spatial structure
Sector 1	0.155	0.035	1.575	0.016	0.001	0.790	−0.017	0.023
Sector 2	0.321	0.000	0.768	0.128	0.004	0.004	−0.016	0.007
Sector 3	0.402	0.000	0.053	0.900	0.006	0.000	−0.012	0.036

^aMultiple linear regression analysis (method: include). Independent variables are aggregate diameter (total number of slices in the respective subsample) and % CNAS (percentage of slices in the subsample containing CNAS). Dependent variables are layer width and spatial structure (number of secondary maxima found in the respective subsample [see text]). Sector numbers refer to the position of the subsample within the cyro-section. Boldface values indicate significant parameters (P < 0.05 after Bonferroni corrections). r², coefficient of determination; P, probability value.

To make these parameters comparable between aggregates of different diameters, the residuals of these regressions were used for further analysis, specifically, as dependent variables in a second set of multiple regressions in which season and river were defined as the independent variables (Table 4). In summer the layer widths of CNAS were significantly narrower than predicted by diameter and percentage of CNAS. This effect did not depend on the river as the identity of the river was not selected as an independent variable. When the differences in aggregate size were factored out, the spatial structure was more complex in the Elbe River than in the Danube River, and spatial complexity was also greater in summer and spring than in the other seasons, as indicated by the slopes of the regressions.

TABLE 4.

Differences in layer width of CNAS and spatial structure (corrected for aggregate size) between seasons and the two rivers^a

Variable and parameter	Overall multiple linear regression		Value for the variable
	r2	P	Regression coefficient	P
Residual layer width	0.040	0.015
Intercept			0.778	0.203
Slope of river
Slope of spring
Slope of summer			−2.877	0.015
Residual spatial structure	0.053	0.023
Intercept			−0.306	0.015
Slope of river			0.299	0.047
Slope of spring			0.458	0.052
Slope of summer			0.399	0.017

^aMultiple linear regression analysis (method: backward selection). Independent variables are dummy coded: river (sample from 1 = Elbe; 0 = Danube), spring (sample from spring = 1; other seasons = 0). Coding for summer, fall, and winter were carried out analogously to spring; dependent variables are residual layer width (residuals from regressions given in Table 3) and residual spatial structure (residuals from regressions given in Table 3). Missing slope values, including those for fall and winter, indicate that the respective independent variable was not selected by the linear regression analysis. CNAS include potential archaea and virus signals. Boldface values indicate significant parameters (P < 0.05 after Bonferroni corrections). r², coefficient of determination; P, probability value.

DISCUSSION

We used visible laser light of various wavelengths in combination with fluorescently labeled probes to describe EPS as well as CNAS abundance and the structure of riverine aggregates by CLSM. Due to their size and fragility, aggregates were quite difficult to examine under in situ conditions. Nevertheless, CLSM permits the analysis of properties of fully hydrated aggregates without fixation and dehydration.

Handling, staining, and analyzing aggregates.

Within 30 min after sampling, aggregates settle in the sampling bottles, and coaggregation of smaller aggregates into larger and loosely associated aggregates may occur (12). Droppo et al. (11) developed a nondestructive technique to stabilize aggregates in low-melting-temperature agarose. However, lectin-binding analyses after aggregate stabilization in agarose did not work due to interference of the lectin with the agarose (B. Luef, personal observation).

Bura et al. (10) demonstrated that the loss of EPS constituents in activated sludge flocs occurred after 24 h at 4°C. DNA and acidic polysaccharides were the most labile components. The analysis of the EPS composition therefore requires fresh samples. All of our samples were analyzed within 24 h.

Although aggregates are exposed to variable physical stress in their natural environments, care must be taken to maintain the structural integrity of riverine aggregates. While mounting and staining probably do not alter aggregates significantly, fixation, freezing, and dehydration may damage their native structure (12). Thus far, transmission electron microscopy has been the method of choice to investigate the microstructures of aquatic aggregates (26). This approach offers high resolution but usually requires fixation and/or dehydration in order to examine the 3-D structure. Recently, CLSM has become a key instrument for investigating fully hydrated biofilms and aquatic aggregates (4, 30). For in situ analysis of hydrated aggregates, a range of fluorescence stains—specific for polysaccharides, nucleic acids, proteins, and lipids—are available. Lectins are a useful probe to examine the 3-D distribution of glycoconjugates in fully hydrated biofilms (31). It has been shown that binding of lectins to river snow proves the presence of specific target monosaccharides in the aggregates' EPS matrix (B. Luef, T. R. Neu, and P. Peduzzi, submitted for publication). Accordingly, lectin binding analysis may be useful to follow the production of lectin-specific glycoconjugates over time. Knowledge about the internal distribution of the specific glycoconjugates, CNAS, and other components such as minerals, proteins, potentially unstained glycoconjugates, humic material, and structural plant components (Fig. 2) will clearly further an understanding of the relationship of structure, function, and spatial dynamics in aquatic aggregates.

Quantification of CLSM data sets.

After CLSM is applied, the data can be visualized and quantified. Such quantification, however, remains difficult because the composition of environmental samples may be highly complex. Ample software is available for quantifying CLSM data sets (16, 47) although such complex images remain difficult to interpret, and their meaning is often inconclusive (6). The capabilities of such software have two aspects: on the one hand, they can dramatically improve image quality; on the other hand, they can also generate artifacts. Image quality is critical and depends on the quality of the input data.

In this study we applied the algorithm colocalization, which is a very powerful tool for determining the degree of associated labeled structures. Obtaining reliable data requires the following considerations: (i) that the signal intensity between the two channels has to be relatively balanced; (ii) that the fluorochromes used should have a wide separation between their emission spectra; (iii) that background noise, blur, autofluorescence, nonspecific labeling, reflections, and bleed-through have a major impact on colocalization quantification; (iv) that setting thresholds to eliminate noise and background in the two channels is a critical step; and (v) that optical misalignment of the instrument also impacts the result (34).

Often a deconvolution algorithm is applied to 3-D sets, which eliminates blur and noise and increases contrast and axial resolution to restore image clarity (34). We did not apply deconvolution to our 3-D data sets although small objects such as bacteria may become more distorted than bigger objects (Fig. 1). Reasonable and reliable solutions using deconvolution for biological samples require nonlinear methods, which are more complex and time-consuming to compute and usually require multiple computation rounds, and thus uncertainties remain.

Determination of the aggregates' size/volume.

Aggregate size is a widely used parameter for characterizing aggregates. Generally, aggregate size is observed in two dimensions, and there is no simple way to estimate exact size and shape. In the literature, the equivalent spherical diameter is often used to estimate aggregate size (28, 29). However, aggregates are 3-D, are irregularly shaped, and have pores (12). Many methods such as using a Coulter counter, photographic techniques, or laser-based sizing instruments have been developed for size measurements (17, 22).

Staudt et al. (45) showed that digital image analysis of confocal stacks allows the spatial and temporal binding of different lectins to be quantitatively evaluated. Our volume determinations may have limitations due to some CLSM drawbacks such as scattering properties of aggregates, depth of laser penetration, and diffusion properties of staining solutions (Fig. 1 and 3a). In addition, using only one lectin to label the specific glycoconjugate of the aggregate matrix may underestimate the actual volume. Lectins might also bind to non-EPS targets or bind nonspecifically to the EPS matrix (20). Nevertheless, the use of CLSM data sets (including cryo-sectioning and the mathematical model of colocalization) yields superior estimates of aggregate volumes.

Ecological aspects of riverine aggregates.

Microbial aggregates in marine systems (2), freshwater lakes (15), and lotic habitats (8, 9, 30) represent hot spots of high nutrient concentrations and microbial activity.

In rivers, particulate organic matter is an important energy source for heterotrophic biota. The composition of aggregates varies depending on their origin (soil, rock, riparian vegetation, biofilms, macrophytes, autochthonous algae, etc.) and therefore influences the associated microorganisms (reviewed in references 41 and 49). For example, hydrology apparently influences particle quality, which in turn also affects the abundance and activity of the associated bacteria and viruses (3, 29, 35). Aggregates dominated by mineral particles are less densely colonized by bacteria than aggregates consisting of mostly organic compounds (29, 48). Experimental evidence for the significance of suspended-particle quality on bacterial and viral parameters in Danube water has also been reported (21a). The Danube and Elbe Rivers appear to harbor different particle qualities. The respective aggregates differ in their glycoconjugate compositions, which also change over time. In our study, over the annual cycle the relative contribution of the specific glycoconjugates and associated CNAS to the aggregate volume was on average higher in the Elbe (Fig. 2a and 2b). Moreover, when data were averaged for the four seasons, more chlorophyll-a, a higher proportion of particulate organic matter, and more ambient nutrients were found in the Elbe than in the Danube (Table 2). This supports the observed differences in aggregate composition and structure.

Besemer et al. (5) showed for the Danube that the compositional dynamics of the particle-associated bacterial communities were related to hydrological and seasonal changes, such as inflow of terrestrial material during flood events, leaf fall, or algal blooms. Bacteria that are restricted to the microenvironment offered by aggregates might be more dependent on the available organic matter entrained in a river system than bacteria, which can live on particles as well as freely suspended. In the Elbe River, the structure and composition of the microbial river snow community showed seasonal dynamics (8). The river snow community was characterized by a great bacterial diversity in spring. In summer the typical aggregate features were large amounts of green algae, diatoms, and cyanobacteria. In autumn and winter, algae were absent, and bacterial abundance increased.

We found different spatial patterns and amounts of CNAS inside riverine aggregates, depending on aggregate size and season (Fig. 4 and 5; Tables 3 and 4). Aggregation of small particles to large aggregates may benefit microorganisms: changes in water column conditions may affect microorganisms associated with aggregates less than their free-living counterparts because the former maintain their own microenvironments. Moreover, the aggregate matrix may protect the bacteria from protozoan predation pressure. Concentration gradients of gases (39), nutrients (1), and microbial activity (36) can exist in aggregates. Some studies have focused on anoxic microniches inside marine snow aggregates. Reducing microzones with sulfide production were found immediately adjacent to oxic zones (40). Experimental results, however, have shown that aggregates are seldom anoxic unless they occur in oxygen-minimum zones (38). The oxygen microenvironment of aggregates decreased dramatically when they were sitting on a solid surface compared to when they were sinking (37). Thus, the composition and structure of aggregates, chemical gradients, and the assemblage of microorganisms create distinct microhabitats inside aggregates.

The ecological relevance of our approach is to reveal environmental heterogeneity on a finer scale. Besides information on the spatial distribution of the components forming particles and aggregates, quantitative assessment of the involved constituents can also be of relevance. There is evidence for a higher microbial abundance on particles with a higher organic content (29, 36). Differences between particles such as river snow and fresh and aged leaf litter were also demonstrated in experimental studies (21a). For example, our new approach allows us to distinguish between the EPS structure (an important part of the nonliving organic matrix) and the probably living (or bioactive) fraction (CNAS; bacteria, archaea, and viruses). Prokaryotes play a pivotal role in the formation, transformation, and degradation of floating aggregates and, thus, particle-mediated carbon flow (43). An abundant, attached bioactive fraction (living microorganisms), e.g., produces more extracellular enzymes than free-living forms, with considerable impact on the aggregates themselves and on the surrounding water body (21, 43). Further, since the sinking and horizontal transport of particles (marine, lake, or river snow) are means of carbon transport, relating this specific information to total particle volume can provide important insight to system productivity and organic carbon flux (M. G. Weinbauer, Y. Bettarel, R. Cattaneo, B. Luef, C. Maier, C. Motegi, P. Peduzzi, and X. Mari, submitted for publication). Moreover, since microbial processes may be different inside particles than they are at the interface between the particulate and ambient phases (Weinbauer et al., submitted), spatial resolution of various compartments can provide further key insights into the role of floating particles. Since aggregates are considered as fractal particles (27), the abundance of microorganisms does not directly scale with the surface or the volume of the particle but, rather, with its fractal dimension. Thus, an increase in aggregate size rather implies a decrease in microbial abundance together with an increased porosity of the aggregate as microorganisms are attached to the solid fraction (Weinbauer et al., submitted).

The Danube and Elbe Rivers differed in their aggregate composition. Over the annual cycle, the relative contribution of the specific glycoconjugates and associated CNAS to the aggregate volume changed over time. Different spatial patterns of CNAS inside riverine aggregates, depending on aggregate size and season, were found. The spatial structure of CNAS inside riverine aggregates was more complex in the Elbe than in the Danube. This information underlines the substantial value of visualizing and quantifying CNAS and glycoconjugate distribution in aggregates for a better understanding of the ecology of floating aquatic aggregates. Although analyzing riverine aggregates by CLSM poses a broad range of challenges, these techniques and applications yield better insight into the relationship of the EPS matrix and microbial organisms in riverine aggregates.

So far, we have not yet comprehensively demonstrated substantial differences between the two investigated rivers regarding the ecology of their particle loads. However, we hope that our first results—suggesting pronounced variability in aggregate structure and composition among rivers and across the seasons—will stimulate further and more detailed investigations on suspended matter in aquatic systems.