Phosphorus limits diazotrophic cyanobacteria whereas other phytoplankton are co-limited by nitrogen and phosphorus in the lowland Murray River, Australia

Study sites

The Murray–Darling Basin is Australia’s second largest river catchment and is located in the south-east of the country, spanning five states and territories. Within the basin is Australia’s largest river, the Murray, which runs for ~2530 km and is the terminus for the Basin. Post-colonial regulation of the river in the form of weirs, locks and channels has allowed for the development of towns and agriculture along the river. The development of The Snowy Mountains Scheme, a hydroelectric and irrigation project, has also significantly influenced the flow of the Murray River. In many cases, these riverine alterations, such as dams and weirs, have created ideal habitats for cyanobacteria (Scholz et al. 2003).

Weir pools within the Murray River are a specific area of interest because they are both the major contributor to flow reduction in-stream and sites of recurring cyanobacterial blooms (Walker and Hillman 1982; Scholz et al. 2003; Bowling et al. 2018). These weirs are also often situated alongside major townships on the river and are used for drinking water, recreation and irrigation.

Four weir pools and one riverine site were selected along the Murray River (Fig. 1). The weir pool sites chosen for this study are distributed along ~1300 km of river and are often influenced by irrigation canals, tributaries and anabranches. Sites were monitored monthly as part of a larger study, during the summer months of 2019–2020 and 2020–2021, for phytoplankton, nutrients and physicochemistry (Rogers 2023). Algal bloom reports from the New South Wales State Government’s Murray and Sunraysia Regional Algae Coordinating Committees (RACCs) were also obtained so that experiments would coincide with cyanobacterial blooms (Rogers 2023).

Fig. 1.

Map of the Murray River, Australia, with the five sites chosen for experimental setup indicated.

Details on the locations and features of the sites, and the timing of the nutrient enrichment study are provided in Table 1. Additionally, Fig. 1 illustrates the arrangement of these sites along the Murray River and their relation to other significant tributaries and the Hume Dam. Going forth, site names and year of experimentation will be used for reference to sites for clarity. Thus, Kulnine Weir will be called Kulnine_21, Mildura Weir will be Mildura_20 and Mildura_21, Euston Weir will be Euston_20, Torrumbarry Weir will be Torrumbarry_20 and Corowa will be Corowa_21.

Microcosm enrichment assays

In situ microcosm nutrient assays were undertaken at each site to determine the limiting nutrients of cyanobacterial and other phytoplankton growth and to learn how increased nutrient availability influenced phytoplankton community structure. Each microcosm experiment lasted for 8 days, and experimental setup was similar to that in Facey et al. (2021). Microcosm treatments consisted of a control (C), nitrogen (N), phosphorus (P), and a nitrogen–phosphorus combination (NP), which were all tested in triplicate per site. Amended nutrient addition concentrations were 500 μg N L⁻¹ in the form of KNO₃ (N treatment) and 200 μg P L⁻¹ in the form of KH₂PO₄ (P treatment) for each microcosm treatment. A combination of both N and P were used for the NP treatment and there was no addition to the C treatment. The amended nutrient concentrations were selected so that nutrient availability was in excess, but within a natural range expected in inland Australian river systems (Furst et al. 2019).

The experimental setup consisted of 3-L polyethylene bottles suspended with rope within the river with floats. Bottles were positioned within the euphotic zone near the water surface (~90% surface irradiance), ensuring phytoplankton growth was not limited by light during the experiments. Microcosm treatment bottles were pre-rinsed and then filled from a large plastic tub with ~120 L of surface river water, which was pre-filtered through a 75-μm plankton net to exclude zooplankton, preventing or reducing any phytoplankton change owing to grazing. A fraction of the sample retained in the net was analysed post-experiment for each site to determine whether the pre-filtration process removed any phytoplankton that were absent from the filtrate. Bottles were left with a small air space and nutrient amendments were added using calibrated pipettes. Bottles were lightly mixed through inversion and set randomly along the stringline.

Samples of phytoplankton, chlorophyll-a and dissolved nutrients were collected in triplicate on Day 0 and on Day 8, by using the methods outlined below. Single replicates of all parameters were taken from each microcosm bottle on Day 8, and dissolved nutrients were collected from three surrogate bottles after the nutrient additions on Day 0.

Nutrient sampling and analysis

Water samples for dissolved nutrients were collected from three surrogate bottles before and after the nutrient additions on Day 0 (ambient and amended concentrations respectively). Each water sample was filtered through a 0.45-μm Sartorius cellulose-acetate syringe filter into a 200-mL polyethylene bottle. All filtered water samples (100–150 mL each) were kept frozen for transport to a laboratory. Analyses of the nutrient species included filterable reactive phosphorus (FRP) and oxidised nitrogen (nitrate and nitrite or NO_X) and were determined photometrically by using a QuikChem 8500 Lachat nutrient analyser, following American Public Health Association (1995) methods. FRP was measured with the molybdite blue method, using ascorbic acid as the reductant (Murphy and Riley 1962). NO_X was analysed with the sulfanilamide method after reduction by a cadmium column (Hosomi and Sudo 1986; American Public Health Association 1995).

Phytoplankton collection and enumeration

Phytoplankton samples (in triplicate, 200 mL each) were collected from the tub on Day 0 and from each microcosm bottle on Day 8 of the experiments, and were preserved in Lugol’s iodine solution. Phytoplankton in the samples were identified using an Olympus BX41 compound light microscope and a Sedgwick–Rafter counting chamber. Enumeration followed methods described by Hötzel and Croome (1999). If required, a 5× concentration was made by settling 50 mL of the sample in measuring cylinders for 24 h. The cylinder was checked for any floating phytoplankton and the upper 40 mL was removed. All taxa were identified to genus level, except for potentially toxic genera that were identified to species level.

Chlorophyll-a analysis

For Days 0 and 8 of the experiments, chlorophyll-a concentrations were measured using the methods described in Müller and Mitrovic (2015), by filtering 200 mL of water sample through a 0.7-μm pore-sized glass fibre filter paper (Whatman) by using a Mitivac vacuum hand pump. The filters were then wrapped in aluminium foil and kept frozen until analysed. Prior to analyses, 10 mL of 90% ethanol was added to the glass fibre filters in clean 10-mL centrifuge tubes and subsequently heated in a 75°C water bath for 10 min. Remnant particles were settled by centrifuging at 2000g for 10 min at 20°C. The supernatant was analysed spectrophotometrically by using a Varian Cary 50 Bio UV Spectrophotometer.

Statistical analysis

Chlorophyll-a concentrations, phytoplankton group biovolume and cyanobacterial species biovolume were analysed using permutational ANOVA (PERMANOVA) with PRIMER + PERMANOVA software (ver. 6, see https://www.primer-e.com/; Anderson 2001). Community analysis was based on similarity percentages (SIMPER) and was performed using Primer software (ver. 6, see https://www.primer-e.com/). Principal component analysis (PCA) was conducted on the dominant phytoplankton communities by using CANOCO software (ver. 5.1, see http://www.canoco.com/), with a separate ordination analysis for each year in which experiments were conducted (Mildura_20, Euston_20 and Torrumbarry_20; and Kulnine_21, Mildura_21 and Corowa_21). Ordinations were conducted separately for each year, because the preliminary analysis indicated that the variation was primarily driven by differences in the initial community composition at the onset of experimentation. Dominant phytoplankton species were identified on the basis of the following two criteria: occurrence in more than three samples and constituting over 5% of the community composition. Hierarchical cluster analysis was performed on the phytoplankton community data to identify commonalities between treatments and sites. The optimal number of clusters was determined using the elbow method (Syakur et al. 2018). PERMANOVA was used to confirm the dissimilarities of the clusters.

Initial conditions

Cyanobacteria were present at all sites at the onset of experimentation, and were the dominant phytoplankton group at Euston_20, Mildura_20, Kulnine_21, Corowa_21 and Mildura_21 experiments (Fig. 2a). Diazotrophic cyanobacterial species were dominant in the system at the start of these experiments, particularly Dolichospermum crassum and D. planctonicum at Euston_20, Mildura_20 and Torrumbarry_20 whereas Chyrsosporum ovalisporum and D. planctonicum were dominant at Mildura_21 and Kulnine_21. Similarly, during the experiments at Mildura_21 and Kulnine_21, there were mixed blooms of D. planctonicum and C. ovalisporum (Fig. 2b).

Fig. 2.

(a) Initial phytoplankton communities and (b) cyanobacteria species composition from Day 0 of each experiment. Error bars are one standard error from the mean, n = 3.

Ambient dissolved NO_X concentrations were low at all sites (Table 2). This was particularly notable for Torrumbarry_20, Euston_20, and both Mildura experiments, with NO_X concentrations of <10 μg L⁻¹. NO_X was more available at Kulnine_21 and Corowa_21. Similarly, FRP concentrations were lowest and below detection limit at Torrumbarry_20, Euston_20, Mildura_20 and Mildura_21 sites on Day 0 of the experiments. FRP was slightly higher Kulnine_21 and to a lesser degree at Corowa_21.

Chlorophyll-a responses to nutrients

Permutational analysis of chlorophyll-a concentrations showed significant differences among treatments at all sites (PERMANOVA: P < 0.05). Pairwise comparisons showed that chlorophyll-a concentrations increased significantly in the P and NP treatments in Mildura_20 (PERMANOVA: C v. P, P = 0.001; C v. NP, P = 0.001), Mildura_21 (PERMANOVA: C v. P, P = 0.001; C v. NP, P = 0.001) and Torrumbarry_20 (PERMANOVA: C v. P, P = 0.003; C v. NP, P = 0.001) experiments when compared with the control (Fig. 3). Furthermore, analysis showed the treatment effect of NP was significantly greater than that of the P treatments at Mildura_21 (PERMANOVA: P v. NP, P = 0.003) and Torrumbarry_20 (PERMANOVA: P v. NP, P = 0.032), with no response to N treatments at both sites when compared with the control (PERMANOVA: Mildura_21 C v. N, P > 0.05; Torrumbarry_20 C v. N, P > 0.05), suggesting serial P limitation. In Mildura_20, chlorophyll-a concentration was significantly greater than the control in all treatments, but the greatest response was in the P treatment. This could indicate independent co-limitation with antagonistic effect of N according to the criteria outlined in Sperfeld et al. (2016). Experiments at Corowa_21, Kulnine_21 and Euston_20 showed a significant increase of chlorophyll-a concentration in the NP treatment when compared with the control (PERMANOVA: Corowa_21 C v. NP, P = 0.001; Kulnine_21 C v. NP, P = 0.002; Euston_20 C v. NP, P = 0.009). Additionally, at these sites there was no significant response in the N treatments (PERMANOVA: P > 0.05) or P treatments (PERMANOVA: P > 0.05) when compared with the control, suggesting Corowa_21 and Kulnine_21 are simultaneously co-limited (Fig. 3). However, at Euston_20, the P and NP treatments were found to not be significantly different from each other (PERMANOVA: P > 0.05), making it hard to identify as either serial P limitation or a type of co-limitation.

Fig. 3.

Chlorophyll-a concentrations from the microcosm assays after incubation. Treatments were C (control), N (nitrogen alone), P (phosphorus alone) and NP (combination of nitrogen and phosphorus). Different letters above the bars indicate statistically significant differences between treatments, determined by pairwise comparisons (PERMANOVA, P ≤ 0.05); where letters are absent, the overall test was insignificant. Error bars are one standard error from the mean, n = 3.

Biovolume of major phytoplankton groups

Chlorophytes consistently had the greatest response to NP treatments and had a significantly greater biovolume than in the control at all weir pool sites following incubation (PERMANOVA: Mildura_20 C v. NP, P = 0.001; Euston_20 C v. NP, P = 0.001; Torrumbarry_20 C v. NP, P = 0.006; Kulnine_21 C v. NP, P = 0.001; Mildura_21 C v. NP, P = 0.018). In Mildura_20, Euston_20 and Torrumbarry_20 incubations, no other treatments responded to nutrient addition, suggesting chlorophytes are simultaneously co-limited at these sites. However, chlorophytes responded significantly in the P treatments at Mildura_21 and Kulnine_21 (PERMANOVA: Mildura_21 C v. P, P = 0.008; Kulnine_21 C v. P, P = 0.023), but not in the N treatment when compared with the control (PERMANOVA: Mildura_21 C v. P, P > 0.05; Kulnine_21 C v. P, P > 0.05), suggesting serial P limitation (Fig. 4). Responses of the cyanobacterial group at Torrumbarry_20, Euston_20 and Mildura_20 showed statistical significance in the P treatment when compared with the control (PERMANOVA: Mildura_20 C v. P, P = 0.014; Euston_20 C v. P, P = 0.049; Torrumbarry_20 C v. P, P = 0.002), suggesting P limitation with an antagonistic effect of N (Fig. 4). Additionally, at Mildura_21, cyanobacteria responded significantly to both P and NP treatments when compared with the control (PERMANOVA: Mildura_21 C v. P, P = 0.008; C v. NP, P = 0.019), but did not differ between each other (PERMANOVA: Mildura_21 P v. NP, P > 0.05), again suggesting single P limitation (Fig. 4). Diatom biovolume was significantly greater in the NP treatment in the Kulnine_21 experiment (PERMANOVA: C v. NP, P = 0.048), but growth was not stimulated in any other sites or treatments. Similarly, euglenoid biovolume was not different from the control in any sites or treatments except in the NP treatment at Kulnine_21 (PERMANOVA: C v. NP, P = 0.017). This suggests that both diatoms and euglenoids are simultaneously co-limited at Kulnine_21 (Fig. 4).

Fig. 4.

Biovolume of different phytoplankton community groups in microcosm experiments. Different letters above the bars indicate statistically significant differences between the treatments, determined by pairwise comparisons (PERMANOVA, P ≤ 0.05); where letters are absent, the overall test was insignificant. Error bars are one standard error from the mean, n = 3.

Phytoplankton community responses

In sites Mildura_20, Euston_20 and Torrumbarry_20, phytoplankton community composition was heavily influenced by increased relative abundance of cyanobacteria in every P treatment across experiments (Fig. 5). SIMPER analysis showed that the diazotrophic cyanobacteria D. crassum and D. planctonicum were the largest drivers of differences at Mildura_20 (~50%), Euston_20 (~41%) and Torrumbarry_20 (~42%). At Mildura_21, D. planctonicum was the primary driver of differences in the P (~33%) and NP (~30%) treatments; however, in the N treatment, C. ovalisporum (~14%) contributed approximately equally with D. planctonicum (~16%). This was reflected in the lack of clear distinction between phytoplankton communities across the treatments (Fig. 5). In the Kulnine_21 experiment, there was a greater proportion of cyanobacteria in the N and C treatments than of chlorophytes, which dominated the P and NP treatments. Small increases in relative cyanobacterial abundance in the N treatment when compared with the control were driven by C. ovalisporum. Chlorophyta were most abundant across NP treatments at all sites except Mildura_21, where cyanobacteria were still most abundant (Fig. 5). Scenedesmus sp. was the largest non-cyanobacterial contributor to differences in phytoplankton community structure across the remaining five experiments.

Fig. 5.

Proportions of the major phytoplankton groups across all six experiments.

Cyanobacteria species responses to amendments

Species-specific responses to nutrient additions followed trends similar to those observed for the total cyanobacterial assemblage during the Mildura_20, Euston_20 and Torrumabrry_20 experiments (Fig. 4, 6); however, several species showed differing responses through the Kulnine_21, Mildura_21 and Corowa_21 experiments. The primary cyanobacterial species displaying P limitation were D. crassum (PERMANOVA: Mildura_20 C v. P, P = 0.001; Euston_20 C v. P, P = 0.004; Torrumbarry_20 C v. P, P = 0.002) and D. planctonicum (PERMANOVA: Mildura_20 C v. P, P = 0.008; Torrumbarry_20 C v. P, P = 0.001) in 2020 experiments, and were the primary phytoplankton species blooming that year (Fig. 6). Similarly, in the Mildura_21 experiments, D. planctonicum displayed clear P limitation of growth (PERMANOVA: C v. P, P = 0.005), whereas C. ovalisporum growth did not appear to be limited by N or P in any of the experiments. D. circinale was present in both Kulnine_21 and Mildura_21 experiments and maintained a low overall biomass in all treatments. At Corowa_21, the diazotrophic cyanobacterium Sphaerospermopsis aphanizomenoides was observed at low biomass, but it did not respond to any of the nutrient additions (Fig. 6).

Fig. 6.

Biovolume responses of dominant cyanobacterial species to nutrient additions. Different letters above the bars indicate statistically significant differences between the treatments, determined by pairwise comparisons (PERMANOVA, P ≤ 0.05); where letters are absent, the overall test was insignificant. Error bars are one standard error from the mean, n = 3.

The PCA for site experiments conducted in 2020 used 18 dominant species. The first two axes explained 66.52% of community data variability (Axis 1 = 39.58%; Axis 2 = 26.95%) (Fig. 7). The most important species for Axis 1 ordination were D. crassum (1.52) and D. planctonicum (1.13). For Axis 2, the most important groups for its ordination were Scenedesmus (1.69), Synedra (−0.887) and Fragillaria (−0.821). PCA for site experiments performed in 2021 used 20 dominant species. The first two axes explained 54.16% of community data variability (Axis 1 = 35.19%; Axis 2 = 18.97%) (Fig. 7). Species important for Axis 1 ordination included Nitzschia (1.47), Euglena (1.42), Cryptomonas (1.34) and Chrysosporum (−1.41). Important species for Axis 2 included D. planctonicum (2.02), Peridinium (−1.14) and Chrysosporum (−0.951).

Fig. 7.

Principal component analysis (PCA) applied to phytoplankton species data (dominant) from (a) 2020 sites and (b) 2021 sites. Sampling units distinguished by site and treatment. Hierarchical cluster analysis identified six clusters indicated by numbered ellipses.

Hierarchical clustering identified six distinct clusters, which further distinguish the community data displayed in the two PCA ordinations (Fig. 7). All clusters were significantly different from each other (PERMANOVA, P = 0.001). There were notable similarities in the NP treatments from Torrumbarry_20, Euston_20 and Mildura_20 (Cluster 4), which had a high proportion of chlorophyta such as Scenesdesmus sp. This cluster was distinct from the other treatments from these sites. For example, Cluster 3 was composed of samples from C, N and P treatments primarily exhibiting co-limitation. Where the community was of lower average total biomass, but greater cyanobacterial biomass (Fig. 5). Similarly, Cluster 6 comprised samples from Mildura_20 and represents a P-limited, low-biomass community with a high diatom biomass and low chlorophyta biomass. Corowa_21, the only non-weir pool site, comprised the majority of the samples in Cluster 1, along with NP treatments at Kulnine_21. This cluster represents a community of high overall biomass, with a mixed community of chlorophyta, diatoms and euglenoids. Cluster 2 was the lowest total biomass and was made up primarily of samples from Kulnine_21 (C, N, P treatments) and Mildura_21 (C and N) and was notably distinct from NP treatments from these sites. Cluster 5 represents high cyanobacterial biomass (D. planctonicum), which occurred when P limitation was alleviated in the Mildura_21 experiment.

Chlorophyll-a and biovolume responses to nutrient additions

At all sites, there was evidence of phytoplankton nutrient limitation on the basis of chlorophyll-a concentrations, indicating that nutrient concentrations play an important role in regulating overall phytoplankton growth. Phosphorus was the key limiting nutrient at Torrumbarry_20, Mildura_20 and Mildura_21, whereas simultaneous co-limitation was evident at Kulnine_21 and Corowa_21. At Euston_20, it was difficult to determine whether the system exhibited serial P limitation or a type of co-limitation (Fig. 3). Total phytoplankton biovolume reflected these trends and was highest in either P or NP treatments at all sites (Supplementary Fig. S1). This indicates that P, and sometimes P with N, are key nutrients limiting total phytoplankton growth in the Murray River, as has been observed globally in other freshwater environments (Schindler 1977; Elser et al. 2007; Harpole et al. 2011; Müller and Mitrovic 2015).

At the beginning of the experiment, ambient concentrations of FRP were low across all sites, with the exception of Kulnine_21, which is influenced by Darling River inflows. NO_X concentrations were at levels conducive to nitrogen limitation at all sites (Dolman et al. 2016; Shatwell and Köhler 2019), and were particularly low at Torrumbarry_20, Euston_20 Mildura_20 and Mildura_21 (≤10 μg L⁻¹). At sites with lower NO_X availability, the experimental response of chlorophyll-a concentrations were greatest in the NP treatment and were significantly higher than the control, which initially indicates co-limitation of the phytoplankton community (Fig. 3). This finding is supported by Harpole et al. (2011) who proposed that under conditions where concentrations of both N and P are low, phytoplankton growth will typically be limited by both nutrients. Similar findings of co-limitation in nutrient amendment bioassays were documented by Müller and Mitrovic (2015) and Bratt et al. (2020). However, we also found significant increases in chlorophyll-a concentration relative to the control in the P treatments, at Torrumbarry_20 and Mildura_21, although to a lesser extent than in the NP treatments, and with no response in the N treatments. This is indicative of serial limitation, where P is the primary limiting nutrient and N is a secondary limiting nutrient, which takes effect after P limitation is alleviated, suggesting that the community is not truly co-limited at all sites (Harpole et al. 2011).

Interestingly, the cyanobacterial biovolume often had a greater response to additions of P alone relative to the control, such as in Euston_20, Torrumbarry_20 and Mildura_20, than to the combined NP treatment. Whereas chlorophyta were consistently co-limited in weirpool sites. These two community groups regularly made up the majority of the phytoplankton biomass in the experiments, but their different responses to nutrient additions highlight unique nutrient acquisition strategies. The proliferation of cyanobacteria in the P treatments is likely to be due to the prevalence of diazotrophic taxa at the onset of the experiments (Fig. 2a). Diazotrophic cyanobacteria, such as heterocystous taxa of the order Nostocales, can acquire their N demands by using fixation of N₂ (Issa et al. 2014). The low ambient concentrations of NO_X at these sites may provide beneficial conditions for diazotrophic taxa to outcompete other phytoplankton species (Bradburn et al. 2012; Dolman et al. 2016; Facey et al. 2022). Alleviation of N-limitation by the NP treatment appears to shift the community in favour of other phytoplankton groups.

Growth of D. crassum, a common diazotrophic cyanobacterial species in the Murray River, was highest in the P treatment at Mildura_20, Torrumbarry_20 and Euston_20. The response in the P treatment was notably higher than that in the NP treatment, likely because N demands were met by N₂-fixation, providing diazotrophic species with an ecological advantage over other phytoplankton groups. In contrast, when N availability was increased (NP treatment), D. crassum was no longer competitive. Instead, a proliferation of a non-diazotrophic genera such as Scenedesmus occurred (Fig. 7). Similarly, the response of D. planctonicum, another diazotrophic cyanobacteria, followed patterns similar to those of D. crassum in the Mildura_20 and Torrumbarry_20 experiments, with the greatest growth response occurring in the P treatment. The affinity of D. crassum and D. planctonicum to available P was confirmed by the PCA, which showed a strong relationship with the two Dolichospermum species in the Torrumbarry_20, Euston_20, Mildura_20 and Mildura_21 experiments. This suggests that Dolichospermum spp. will not be competitive when there is high dissolved inorganic-N availability, but can utilise P pulses under N-limiting conditions. This is consistent with previous studies that have found that Dolichospermum spp. improve fitness under N-limited conditions (Dolman et al. 2012; Moisander et al. 2012; González-Madina et al. 2019), even compared with other N₂-fixing taxa (De Nobel et al. 1998; Yema et al. 2018; Grover et al. 2020), whereas other diazotrophic taxa are stronger competitors for dissolved forms of N (Ferber et al. 2004).

At the Corowa_21 and Kulnine_21 sites, the conditions at the onset of the experiments were characterised by higher availability of dissolved inorganic N, albeit at concentrations still likely to illicit some growth limitations (Kolzau et al. 2018). This resulted in notable differences in the community responses compared with the other sites. At Corowa_21, nutrients released from Hume Dam were likely to drive the elevated ambient nutrient concentrations (Rogers 2023), perhaps in combination with less denitrification occurring in the upstream, riverine environment where conditions were likely to be well mixed and oxic. This, in combination with increased flow velocity, may explain the low Nostocales biovolume on Day 0 of the experiment. As a result of these factors, total cyanobacterial biomass as well as individual species did not respond to any of the nutrient treatments. However, the overall phytoplankton biomass (as total biovolume and chlorophyll-a) was greater in the NP treatment than in the control and was driven predominantly by chlorophyta. Similarly, at Kulnine_21, cyanobacteria did not increase in response to any nutrient treatments. However, chlorophyta and diatoms were strongly co-limited and made up a large proportion of the community in the NP treatment.

It is surprising then that cyanobacteria as a group did not respond to the P treatments, particularly given the large initial inoculum of heterocyst bearing C. ovalisporum and D. planctonicum at Kulnine_21 and Mildura_21 (Fig. 2b). Despite the lack of response from other co-limited groups such as chlorophyta and diatoms, neither C. ovalisporum nor D. planctonicum responded to the P pulse. In experiments at other sites, D. planctonicum appeared to utilise N-limited, P-replete conditions to gain a competitive advantage over non-N fixing taxa. The lack of response to similar conditions at Kulnine_21 may have been due to the Kulnine_21 experiment taking place later in the year than the experiments in which D. planctonicum was successful, resulting in a seasonal or temperature effect, relating to the high light and energetic requirements of heterocyst production (De Nobel et al. 1998; Fu and Bell 2003). Whereas the negative response of C. ovalisporum to P additions, and the greater biomass in C and N treatments than in P, may indicate that C. ovalisporum occupies a different niche or utilises different strategies in response to nutrient limitation or may have different thresholds at which N₂-fixation is promoted. The optimum phytoplankton N:P varies among genera, becaus nutrient requirements are based on physiological conditions, the habitats in which the taxa evolved and specific acquisition strategies (De Nobel et al. 1998; Osburn et al. 2021; Wagner et al. 2023).

Hierarchical cluster analysis identified six distinct clusters, providing insights into common trends among treatments and sites (Fig. 7). The differentiation of these clusters and the associated conditions and dominant taxa provide a clear illustration of the effect of nutrient concentration on community structure. Cluster 4 comprised only samples from NP treatments at sites with particularly low ambient nutrient concentrations (Torrumbarry_20, Euston_20, Mildura_20). This cluster represents the ‘first responders’, which can rapidly respond to a nutrient pulse such as the chlorophyta Scenesdesmus sp. and picocyanobacteria such as Aphanocapsa sp. (Ma et al. 2015; Müller and Mitrovic 2015; Facey et al. 2022). This cluster was distinct from the other treatments at these sites. For example, Cluster 3 contained samples from C, N and P treatments in which chlorophyta were co-limited. The community was of lower average total biomass, but greater cyanobacterial biomass than was Cluster 4. This further demonstrated that N-fixing cyanobacteria are more effective competitors in co-limited systems and typically reduce in biomass following addition of N and P. Reinforcing this, Cluster 1 is representative of high N and P availability. Corowa_21 had the highest ambient NO_X concentrations and comprised the majority of the samples in Cluster 1 along with Kulnine_21 NP. These conditions are associated with a high biomass, diverse community of chlorophyta, diatoms and euglenoids and, notably, the lowest concentration of cyanobacteria. This diversity includes species that typically dominate in nutrient-enriched or riverine systems (Padisák et al. 2009). Using hierarchical cluster analysis to examine commonalities among treatments and sites under differing nutrient conditions represents a reasonably novel and underutilised approach in riverine settings.

Management of nutrients in the Murray River

Chlorophyta and cyanobacteria were the two most dominant phytoplankton groups in the Murray River during this study. Chlorophyta were consistently co-limited by N and P at all sites, indicating that management of either N or P inputs could reduce overall phytoplankton biomass. However, cyanobacteria were commonly limited by P alone. Owing to the positive relationship between cyanobacterial growth and P concentrations, coupled with low ambient dissolved inorganic N and the high ambient abundance of diazotrophic cyanobacteria, the data from this study suggest that P reductions would be the most effective strategy for reducing the occurrence of the regular blooms of diazotrophic cyanobacteria in the Murray River, while also reducing other dominant phytoplankton groups. These results are consistent with findings of Schindler et al. (2008) and Higgins et al. (2018) who observed that reducing N inputs favoured diazotrophic cyanobacterial growth. However, although they have proposed that N₂-fixation can entirely compensate for external N reductions, many proponents of managing both N and P argue that the high energetic cost of N₂-fixation and its dependence on the sufficient availability of additional parameters such as light and P, will reduce the magnitude of blooms when growth is reliant on N₂-fixation (De Nobel et al. 1997; Agawin et al. 2007; Paerl et al. 2011; Shatwell and Köhler 2019; Grover et al. 2022). Therefore, we cannot exclude the possibility that N reductions would also effectively reduce cyanobacterial biomass to some extent, but would likely exacerbate the competitive advantage of diazotrophic cyanobacteria if not undertaken in combination with P.