The experiment was carried out at one station in the marine Lake Mrtvo More (English: 'Dead Sea', 42°37'21" N, 18°07'14" E) on the island of Lokrum located in front of the Old City of Dubrovnik, South Croatia (Fig. 1). The island of Lokrum (72 ha) is a nature reserve and NATURA 2000-Ecological network-site (Site of Community Importance, code: HR 4000017). Lake Mrtvo More (surface area 1310 m², perimeter 150 m, max. depth 6 m and average of 2 m) is linked to the open sea by 15-20 m long underwater tunnel. The lake is a favourite swimming spot for many visitors during the summer (On-line Suppl. Fig. 1A).

The region experiences a typical Mediterranean climate: Summers are warm and dry, and winters are mild and rainy. Annually, average air temperature is 16 ºC and precipitation 1308 mm (data from Dubrovnik meteorological station for 1961–2018, Croatian Meteorological and Hydrological Service). Average temperature during the coldest month (January) was 9.1 ºC, and during the warmest (August) 25.2 ºC. The highest rainfall is from October to March. During the dry season (June–August) total rainfall is only 155.4 mm. Average annual wind speed is 3.33 m s^-1, with the dominant southerly winds (SE, SSE) blowing from April to September. Annual potential evapotranspiration is around 1500 mm year^-1 with maximum values in vegetation period (April-September) (Orešić and Čanjevac 2020). In the area, the range of diurnal sea-level oscillations is close to 19 cm (Mihanović et al. 2006). Seawater surrounded the island is under the direct influence of incoming currents from the Ionian Sea (Garić and Batistić 2016).

The experiment was carried out over a period of 25 weeks from April to October 2016. Water samples for analysis physico-chemical parameters were taken weekly (On-line Suppl. Tab. 1) from 19^th April to 12^th October, 2016 at the same place where diatom sampling was carried out, i.e. at the bottom (1 m depth). Temperature (T) and salinity (S) were measured using a WTW Multiline P4 multiparametric sounding lineprobe. Seawater samples for nutrient analyses (Strickland and Parsons 1972, Ivančić and Degobbis 1984) and chlorophyll a concentrations (Chl a, Holm-Hansen et al. 1965) were taken by 5 L Niskin bottles. Measured nutrients included nitrate (NO₃^-), nitrite (NO₂^-), ammonium (NH₄⁺), total inorganic nitrogen (TIN = NO₃^-+ NO₂^-+ NH₄⁺), orthophosphate (PO₄^3-) and orthosilicate (SiO₄^4-).

Samples for NO₃^-, NO₂^-, PO₄^3- and SiO₄^4- were frozen (−22 °C) and analysed in a laboratory (Strickland and Parsons 1972). Subsamples (50 mL) for NH₄⁺ were fixed immediately after collection with 2 mL of 1 mol L^-1 phenol/EtOH, kept at 4 °C and later analysed according to Ivančić and Degobbis (1984). Chl a was determined from 1 L sub-samples filtered through Whatman GF/F glass-fiber filters and stored at −20 °C for a period of less than a month. Filtered samples were homogenized and extracted in 90% acetone for 24 hours at room temperature (Holm-Hansen et al. 1965). Chl a was determined fluorometrically using a Turner TD-700 Laboratory Fluorometer (Sunnyvale, CA) calibrated with pure Chl a (Sigma). Due to the exceptionally high Chl a value of 39 µg L^-1 on 20^th of July, this record was removed from further analysis.

Dissolved oxygen was determined by the Winkler method and oxygen saturation (O₂/O₂′) was calculated from the 100% solubility of oxygen (O₂) in seawater (Weiss 1970, UNESCO 1973). Trophic status was characterized by the TRIX index (Vollenweider et al. 1998), commonly used to classify coastal marine areas in the Mediterranean (see Primpas and Karydis 2011): TRIX = [log₁₀(Chl a×D%O×DIN×TP)+k]/m. Each of the factors represents a variable reflected in the trophic state: Chl a – chlorophyll a concentration (µg L^-1), D%O – dissolved oxygen (absolute deviation from 100% oxygen saturation), dissolved inorganic nitrogen DIN and TP – total phosphorus (µg L^-1). The parameters k = 1.5 and m = 1.2 set the range of the TRIX scale from 0 to 10: 0–4 oligotrophic, 4–5 mesotrophic, 5–6 eutrophic, 6–10 extremely eutrophic.

Glass slides were used as a substrate for biofilm formation because of their convenience compared to a natural substrate. The dimensions of a standard microscope slide for bacteriological and for algological sampling are the same, measuring about 75 mm by 25 mm and about 1 mm thick. Microhabitats were made of 33 microscope glass slides which were arranged in three rows at a distance of approximately 1 cm and fixed on the upper side of a Plexiglas sheet which was then submerged horizontally with four diving weights at one station in the lake at a depth of approximately 1 m (i.e. on the bottom of the Lake Mrtvo More) about 2 m offshore on 19 April 2016 (On-line Suppl. Fig. 1B). After three weeks, the Plexiglas sheet was hauled up and the first microscopic slide for diatom analysis was removed. Every week another microscope slide was taken out and gently plunged into filtered seawater (Millipore, acetate cellulose 0.22 μm). In total, there were 21 diatom samples (On-line Suppl. Tab. 1). For bacteriological analysis, 12 glass slides were collected in a period from 20th May to 6^th September (On-line Suppl. Tab. 1). All samples were preserved with 4% formaldehyde.

The total number of heterotrophic bacteria was determined by using a direct counting method counting under epifluorescent microscopy (Hobbie et al. 1977). All samples were analyzed within five days, and before processing were stored in the dark in a refrigerator at a temperature of about 5 °C. Glass slides were gently brushed and washed with sterile freshly filtered seawater (Millipore, acetate cellulose 0.22 μm) and the biofilm was dispersed. For bacteria colouring a 0.01% solution of acridine orange was used and the 2 mL subsamples were filtered through Nucleopore filters (pore diameter of 0.2 µm). Bacterial cells were counted using a Jenalumar Zeiss fluorescent microscope under 1500 × magnification. These values are expressed as cells per cm².

A microscopic glass surface of 1 cm² was scraped using a razor blade, and the microalgae were collected in Falcon tubes preserved by adding a known amount (3 mL) of solution (3%) of formaldehyde-filtered seawater (Millipore, acetate cellulose 0.22 μm). Quantitative analysis of homogenized samples was determined with an inverted microscope (Olympus IX 71) equipped with phase contrast. In these samples, taxa were not determined. Results are expressed as total number of diatom cells per cm².

A detailed diatom analysis was performed on permanent slides of processed material (hydrogen peroxide treated before mounting in Naphrax® as reported by Car et al. 2019) with a Nikon E600 microscope at a magnification of 1000×. The species abundances were expressed as percentages of the total number of frustules counted (relative abundances, in %). In total, 400 valves per each sample were counted.

Permanent slides of light microscopy (LM) have been deposited in the diatom collection of the Institute for Marine and Coastal Research, University of Dubrovnik, Dubrovnik, Croatia. Identifications were made following keys and guides reported by Hafner et al. (2018). Nomenclature follows AlgaeBase (Guiry and Guiry 2019).

Cluster analysis was used to determine the similarity level among physico-chemical parameters in samples (Clarke et al. 2008). A hierarchical clustering algorithm based on Euclidean distances on log(x+1)-transformed, normalized data and the average group linkage method were used. The similarity profile routine (SIMPROF, P < 0.05) were used to define the significantly different clusters, and analysis of similarities (ANOSIM) was applied to evaluate a differences among seasons/months (Clarke and Warwick 1994, Clarke et al. 2008).

Nonmetric multidimensional scaling (NMDS) was used for analysis of the community composition variability, i.e. to define the diatom abundance with relation to sampling dates. In order to normalize data, diatom abundances expressed as relative percentages were square root transformed. The Bray Curtis matrix included 285 taxa over 21 samples. In this case, SIMPROF (P < 0.05), SIMPER and ANOSIM randomization were also used: (i) to define the significantly different clusters, (ii) to identify the taxa making the greatest contribution to differences among clusters, (iii) to test differences in diatom community over the sampling period.

To investigate the community diversity in the diatom samples, the Shannon-Wiener Biodiversity Index and the Margalef index was computed (Kwandrans 2007). As the diversity index is not completely effective in describing community structure, the evenness of benthic diatom assemblages was also computed using both Pielou’s, and Smith and Wilson's evenness values (Pielou 1966, Smith and Wilson 1996, Beisel et al. 2003).

Canonical analysis of principal coordinates (CAP) was used to summarize the structure of diatom assemblages over the months and to determine which physico-chemical parameters were directly responsible for the variations observed in diatom abundances.

The relationship between the most abundant species and physico-chemical parameters was analysed by Spearman-Rank correlation coefficient. Data were transformed [log(x+1)] to enable the correlation tests among variables (Cassie 1962). The Kolmogorov-Smirnov test was used for testing normality of the data distribution. Only significant values (*P < 0.05, **P < 0.01, ***P < 0.001) were reported.

Statistical analyses were performed using the PRIMER v6 software (Clarke and Gorley 2006) and Statistica 7.0 (StatSoft, Inc. 2004).

Over the study period seawater temperature ranged from 16 °C to 27.3 °C, and salinity ranged from 26.6 to 37.3 (average 35.5) (Fig. 2A, B). TIN ranged from 0.96 to 10.02 µM and mostly follow the distribution of NO₃^-. PO₄^3- varied from 0.066 µM to 0.578 µM and SiO₄^4-from 3.23 to 13.02 µM. The highest value both PO₄^3- and SiO₄^4- was recorded on 20^th of July. During whole study period, average nutrient concentrations were: 3.14 µM NO₃^-, 0.58 μM NO₂^-, 0.97 μM NH₄⁺, 0.24 PO₄^3- and 7.65 µM SiO₄^4-. Oxygen saturation (O₂/O₂′) ranged from 0.57 to 1.39 (average 0.92).

In May-June period the average Chl a was 0.3 µg L^-1. During the whole study period, minimum Chl a (0.12 µg L^-1) was on 31^st May and maximum (39 µg L^-1) on 20^th July. Average Chl a in August and September was 2.5 µg L^-1 and 2 µg L^-1, respectively (Fig. 2D).

The trophic index (TRIX) was lower than 4 (oligotrophic character of the lake) during the initial sampling period (up to 24^th June) and towards the end (from the mid-October). Lake showed mesotrophic character (4.03-4.76) in the period the end of June-mid-July, and again at the end of September-beginning of October. The lake was mostly eutrophic (5.54-6.02) in the period the end of July–mid-September, and under highly eutrophic conditions (6.44) on the 20^th July.

Physico-chemical parameters varied significantly (ANOSIM, P < 0.05) among months, seasons (spring, summer, autumn), and between samples collected before the 18^th June (Group 1) and afterwards, with exception on 7^th June (On-line Suppl. Fig. 3, On-line Suppl. Tabs. 1, 2).

On 20^th May 2016, heterotrophic bacteria reached values of 35,479 cells cm^-2 of the glass slide (Fig. 3A). The average number of bacteria during the study was 42,114 cells cm^-2 with the peak (69,268 cells cm^-2) at the beginning of June. During the second part of study a decline in the number of bacteria was observed.

A peak value of 333,076 cells cm^-2 was observed in August. The average abundance over entire study period was 165,946 cells cm^-2 (Fig. 3B).

A total of 285 diatom taxa belonging to 72 genera were found in samples (Appendix). Genera with the greatest number of taxa were: Mastogloia (36), Nitzschia (29), Navicula (20), Amphora (13), Diploneis (17), Achnanthes (13) and Cocconeis (12). The most abundant taxa were Cocconeis scutellum Ehrenberg var. scutellum and Cocconeis dirupta W.Gregory var. flexella (Janisch and Rabenhorst) Grunow which occurred in all samples with average relative abundance of 30% and 25%, respectively. The maximum abundance of C. scutellum var. scutellum (90%) was recorded on 7^th June, while the maximum abundance of C. dirupta var. flexella (65%) was recorded one month latter (7^th July) (see On-line Suppl. Tab. 4). In total, 48 taxa were found only once (sporadic) in all samples (Appendix).

The number of taxa per sample ranged from 9 (25^th May and7^th June 2016) to 52 (11^th August 2016), with an average of 25 (Fig. 3C). The Shannon-Wiener Biodiversity Index varied from 0.74 to 4.51, with an average of 2.93 (Fig. 3D). Pielou’s species evenness ranged from 0.23 to 0.86 (the average 0.63) with the minimum occurring in June and the maximum at the end of September (Fig. 3E). Smith and Wilson species evenness ranged from 0.06 to 0.45 (the average 0.21) with the minimum at the end of May and the maximum in August 2016 (Fig. 3G).

Diatom assemblages differed significantly (NMDS, ANOSIM, P < 0.05) between the samples collected up to the middle of July (Group 1) and afterwards (Group 2). Additionally, sample from the 12^th October (Group 3) differed significantly from all the others (Fig. 4, On-line Suppl. Tab. 3). Cocconeis scutellum var. scutellum, C. dirupta var. flexella, Opephora mutabilis (Grunow) Sabbe et Wyverman, Navicula salinicola Hustedt, Cocconeis costata W.Gregory, Halamphora hyalina (Kützing) Rimet et R.Jahn, Licmophora paradoxa (Lyngbye) Agardh, Licmophora flabellata (Greville) C.Agardh, Halamphora coffeiformis (C.Agardh) Levkov and Psammodictyon rudum (Cholnoky) D.G.Mann contributed the most (cumulatively 70%) to the variance between assemblages from Group 1 (10^th May-13^th July) and 2 (20^th July-3^rd October, SIMPER, Tab. 1). Within Group 1, C. scutellum var. scutellum and C. dirupta var. flexella contributed the most (cumulatively 95%) to the similarity among diatom assemblages from the 10 samples.

Diatom assemblages also differed significantly (ANOSIM, P < 0.05) between the samples collected before the end of June and samples collected afterwards so the first group contained sub-groups A and B with similarity of 40%. Cocconeis scutellum var. scutellum, C. dirupta var. flexella, L. flabellata, P. rudum, C. costata, Navicula flagellifera Hustedt, Nitzschia frustulum (Kützing) Grunow, Cocconeis pseudomarginata W.Gregory and Mastogloia cuneata (Meister) R.Simonsen contributed the most (cumulatively 81%) to the variance between assemblages from these two su-bgroups.

Diatom assemblages varied significantly (ANOSIM, P < 0.05) among months (On-line Suppl. Tabs 3, 4). The pioneer colonization diatom taxa observed after one month (20^th May) of exposure of the glass slides were Cocconeis scutellum var. scutellum and C. dirupta var. flexella which occured with average relative abundance of 73% and 15% respectively. These taxa were recorded in all 21 samples. In May, they contributed the most (cumulatively 90%) to the similarity among diatom assemblages. Cocconeis scutellum var. scutellum had the highest average relative abundances in May (72%) and June (58%), while C. dirupta var. flexella dominated (42%) in July. The abundance of C. dirupta var. flexella was 17% and 14% in August and September, respectively. Halamphora hyalina occurred in 50% of the samples with an average abundance of 6% and maximum of 14% recorded on 24^th August. In 15 samples, L. flabellata and H. coffeiformis were observed with average relative abundance of 3% with the maximum of 13.5% on 7^th July and 2^nd September, respectively.

Spearman-Rank correlation coefficient displays positive correlation between diatom relative abundance and temperature in case of C. dirupta var. flexella and P. rudum (Tab. 2). Psammodictyon rudum also had positive correlation with salinity. In the case of nutrients, the species C. costata, O. mutabilis, R. adriaticum and Seminavis sp. correlated with PO₄^3- and NO₃^-. Six diatom taxa (A. kuwaitensis, C. costata, N. salinicola, Navicula sp., O. mutabilis and Seminavis sp.) correlated negatively with oxygen saturation. Interestingly, Nitzschia frustulum was not influenced by any of these 10 environmental variables. In addition, none of diatom taxa correlated with SiO₄^4- (Tab. 2).

Canonical analysis of principle coordinates (CAP) showed that the samples collected in May and June are more related with abundance of adnate diatoms, particularly C. dirupta var. flexella and C. scutellum var. scutellum, while motile forms were better related in the samples from July, August and September (Fig. 5). Taxa presented in the samples collected from July to September were associated with higher seawater temperature and higher nutrient concentrations (e.g. P. rudum, R. adriaticum, T. coarctata). Erect diatoms (e. g. L. paradoxa, L. flabellata, Fig. 6) appeared in October, when salinity was low.

Strong relationships between environmental variables and diatom assemblages were found in Lake Mrtvo More and shifts in dominance at the species level were recognized. In the first stage of the experiment, when generally the nutrient concentrations were low, the lowest number of diatom taxa was recorded and adnate diatoms appeared. In general, adnate taxa adhere strongly horizontally to the substrate by means of their raphe valve and may easily benefit from a nutrient exchange with the substrate due to their adhering mode through the valve face (Round 1981, Sullivan 1984, Romagnoli et al. 2014). Diatom species richness of Lake Mrtvo More was strongly correlated with TIN, constraints during which C. scutellum var. scutellum remained a common species in the diatom community. This taxon was the dominant in the assemblages during the first months of experiment but its relative abundance declined when seasonal anthropogenic disturbance started (July). It seems that the changes in nutrient concentrations induce changes in species diversity. This is in agreement with the results of Marcus (1980), who found differences in diversity between sites with varying levels of nitrogen concentration during investigation of periphytic communities using glass slide substrates when recording a greater algal growth downstream of a dam, which was attributed to nitrogen discharges from the reservoir. Marcus (1980) suggested that while Cocconies became dominant at the three downstream sites because of its greater efficiency in obtaining or incorporating limited nitrogen resources, species other than Cocconies dominated the diatom communities in which nitrogen concentrations were enriched apparently because of higher potential growth rates which could be realized with the elevated nutrient conditions. Cocconeis taxa clearly differ in their response to nutrient supply, leading to an altered community composition, which may be detected only if the species level is considered. As C. dirupta var. flexella was associated with higher temperature values, C. dirupta var. flexella remained a common species and characterized the benthic diatom assemblage of Mrtvo More during the warmer period of the year.

In this study Nitzschia frustulum was not influenced by any of these 10 physico-chemical parameters. This was showed in previous studies in which N. frustulum has been described as a highly tolerant diatom taxa which is resistant to organic pollution and is associated with areas affected by intensive agricultural and industrial activities (Tornés et al. 2007, and references therein). In addition, the genus Navicula has a high adaptability to all trophic status of ecosystem and appears to be tolerant of pollution (Agatz et al. 1999, Cunningham et al. 2005, Cibic and Blasutto 2011). Our findings are consistent with this observation, in particular for N. salinicola. Pollution tolerant genera like Nitzchia and Navicula occurred in abundance through the summer season. Generally, abundance of diatom taxa of Nitzchia and Navicula was low through the oligotrophic state of the lake.

The abundance of opportunistic species provided the possibility of distinguishing possible anthropogenic pressures on the ecosystem. Although the variability of the physico-chemical variables in Lake Mrtvo More suggests the presence of two distinct environmental contexts that enhance the proliferation of different benthic communities, it cannot be clearly connected with anthropogenic impact by visitors (swimmers) during the summer. Apart from an increase in nutrient concentrations during summer, sea temperature and salinity also rise. The increase in salinity is caused by the interaction of several factors, such as low precipitation and higher air temperature and evaporation. Moreover, along with salinity, seawater temperatures also seem to be very important for some species (e.g. P. rudum) as they were associated with higher temperature values. SiO₄^4- was measured and compared with benthic diatom abundance but no significant correlation was recorded. No relationships between SiO₄^4- in the water column and benthic diatom abundance have already previously been described (Facca and Sfriso 2007).