Eutrophication, of either natural or anthropogenic origin, is closely connected to hypoxia in coastal seas (Nixon 1990). The cause of deficiency or absence of oxygen in aquatic systems can be of a physical origin. For example, a) sills and barriers affect circulation of water; b) excessive precipitation over evaporation intensifies the stratification that prevents mixing; and c) weak cooling of surface waters hampers water sinking and decreases water column mixing and aeration of deep waters. In addition, if a pronounced pycnocline develops in such a system, diffusion of oxygen through the density gradient can be further prevented. There are large portions of marine systems that are hypoxic, such as subsurface waters of the East China Sea (Chen et al. 2007, Wang et al. 2016) and deep waters of the Baltic Sea (Conley et al. 2009a, Rolff and Elfwing 2015), the Black Sea (Konovalov et al. 1999) and the Gulf of Mexico (Justić et al. 1996, 2005). Layers of the open ocean can be hypoxic due to a significantly larger supply of organic matter than oxygen (Kamykowski and Zentara 1990), but it is hypoxia/anoxia in coastal seas (Diaz 2001) that causes greater concern regarding damage to the biodiversity of ecosystems (Ciglenečki et al. 2015), which many human activities depend on, such as fisheries and tourism.

Lakes behave like natural laboratories because of their comparatively prompt responses to environmental changes (Ciglenečki et al. 2015). Those which contain sea water originated a few thousand years ago by processes related to late postglaciation, i.e. isostasia and elevation of sea level. The marine lakes in Scandinavia were transformed into freshwater lakes due to isostasia, and hence some of them contain the trapped sea water at the bottom (Bøyum 1973). In comparison, the marine lakes of the eastern Adriatic coast are incorporated into the system of karst, containing sea water in the entire water column (Žic et al. 2012, Hrustić et al. 2013, Ciglenečki et al. 2015). The Mljet Lakes (MLs) were formed during the Mesozoic Era as karstic depressions filled with fresh water (Govorčin et al. 2001). The postglacial sea level rose around 4 to 5 thousand years ago and consequently the ingression of sea turned them into a marine environment (Bognar and Curić 1995). Unlike the well-oxygenated Jezero Mir (Lake Mir) (43.8864°N 15.1667°E), which is up to 10 m deep (Žic et al. 2012), and the meromictic lake Zmajevo oko (Dragon’s Eye) (43.5309°N 15.9588°E), which is up to 15 m deep, permanently hypoxic below 13 m and occasionally completely anoxic (Ciglenečki et al. 2015), episodes of hypoxia are expected in deep waters of the MLs (Buljan and Špan 1976) within the system of coastal karst in Croatia.

The strait (depth ~0.8 m, width ~2 m) between SL (~241000 m², ~3349000 m³) and BL (~1450000 m², ~58000000 m³) and the strait (depth ~2.5 m, width ~10 m) between BL and Soline Channel (SC) (Fig. 1) hinder the exchange of deep lake waters with the adjacent coastal sea. BL is up to 49 m deep, with an average depth of 13.6 m, while SL is up to 29 m deep, with an average depth of 7.6 m (Cuculić et al. 2012). The advection from the open sea occurrs strictly in the surface layer, while circulation is dominantly driven by tides <25 cm and tidal phase shifts through the straits acting as filters at semidiurnal frequencies (Peharda and Vilibić 2008). The absence of upwelling, unlike from coastal waters (Batistić et al. 2013), constrains the mixing and supports the development of hypoxia in deep waters of the MLs.

Some instabilities, i.e. internal waves in deep water of BL, were explained by severe cooling and sinking of the surface water to the bottom (Buljan and Špan 1976). Such an event has not been recorded in SL so far, but some water column instability can come from lateral and bottom subaquatic karstic springs (Wunsam et al. 1999). Cuculić et al. (2012) estimated that water renewal times (WRTs) in SL and BL during the mixing period (October-March) were 81 and 70 days and during the period of stratification (April-September) they were 44 and 36 days, respectively. WRT for the stratification period refers only to the upper waters above the depth of 10 m in SL and 17 m in BL. When SL and BL were considered as one system exchanging the water with the open sea, WRTs of the MLs during the mixing and stratifed periods were 82 and 43 days, respectively. These values, being closer to WRT in SL, indicated that the exchange between BL and SL is the limiting factor for the total exchange of water in the system (Cuculić et al. 2012).

In this study, we intended to investigate the link between natural eutrophication, water column stratification and episodes of hypoxia in deep waters of moderately warm marine lakes. The arrangement of data concerning the mixing and stratification periods and water column division was adjusted to be comparable to the recent report on WRT in the MLs (Cuculić et al. 2012), while offering an extension to parts of the research that revealed an anoxic event in deep water of BL more than a decade ago (Benović et al. 2000). Anoxia in the water of the MLs was not detected after August 1996 and vertical homogenization has been noted to occur regularly since 2003 (Vilibić et al. 2010). Having apparent oxygen utilization (AOU) as an important parameter of our study, including the report on the WRT (Cuculić et al. 2012), we were able to determine oxygen utilization rate (OUR) and organic carbon remineralization rate (OCRR) (Feely et al. 2004) in the MLs for the first time, and to provide information and conclusions that may be valuable for future studies of similar environments worldwide.

The lack of significant differences in any of the parameters related to O₂ between the mixing and stratification periods in SL was most likely related to persistent stratification (no significant difference in Δσ_T between the mixing and stratification period), which constrained the mixing between its upper and deep water. In addition, we found the lowest numbers of parameters with significant (p<0.05) changes between the periods in upper and deep water of SL (Supplementary material Table S12), which are most likely related to the same reason. In comparison, the significant (p<0.05) differences in the upper water of BL (HPF) and G (Cyano and HBa) between the periods (Table S12) are most likely a consequence of changes in the food web, which is a topic beyond the scope of this study. Of note, N_oth in BL and G was significantly (p<0.05) higher in the mixing period, but we cannot offer a plausible explanation for these observations.

The thermocline was crucial for stratification at all explored stations in the stratification period, while a relatively shallow (up to 7 m) and persistent halocline was crucial for stratification of SL in the mixing period, to the point that difference in t in the 2-3 m layer in December, which impacted negatively on the pycnocline by ~30%, was compensated by an 130% positive influence of ΔS on Δσ_T in that layer. High range (>10°C) between low and high t in the upper water of SL during less and more productive periods (Table S9, S6), respectively, played a role in balancing between different levels of O₂ production and solubility. These include an exponentially positive relation between the rate of metabolic processes and O₂ requirements of marine organisms at increasing t (Brown et al. 2004), along with decreasing O₂ solubility in seawater with the rise of t (Garcia and Gordon 1992). Ciglenečki et al. (2015) detected an increase in t of 1.1°C in the surface layer (0-2 m) and 0.64°C at 12 m in Dragon’s Eye for 2001-2012 in comparison with 1996-2000. Therefore, we speculate that the presented trend will have greater influence on SL than on BL concerning the diminished solubility of O₂ with rising t. Decreased O₂ solubility in the upper water of SL in both periods due to higher t is mitigated by lower S in comparison with BL (Table S6, S9). An important difference between the MLs was the notably warmer (2.12 and 2°C) and somewhat saltier (0.32 and 0.38; negative effect on solubility of O₂) deep water of SL during periods of stratification and mixing (Tables S7, S10), respectively. Although phytoplankton blooms produced organic matter and DO throughout the water column of SL in April and July, saltier (on average 0.198 and 0.167) and warmer (on average 2.242°C and 2.933°C) deep water of SL in April and July, respectively, caused reduced solubility of O₂ in comparison with deep water of BL, while having higher O₂ requirements due to organisms at higher t at the same time (Brown et al. 2004). It was predicted by Vaquer-Sunyer and Duarte (2011) that survival times of benthic macrofauna would decrease by a mean of 35.6% under hypoxia with the average 25.5% increase of DO concentration threshold for high mortality in a 4°C warmer ocean, representing the average increase in t expected at the end of the 21^st century under moderate CO₂ emission scenarios. Along these lines, Lojen et al. (2010) have detected 4-5°C higher t (12°C in May and 13°C in October) in the pore water of SL than in the pore water of BL (8°C). Therefore, the difference of ≥2°C between deep waters of SL and BL should be monitored in the future.

Detailed discussion on the impact of p(CO₂) and potential synergistic effects of hypoxia and increased p(CO₂) on benthic invertebrates and fish can be found in Steckbauer et al. (2015), where measurement of calcification rates are stressed as an important additional tool to provide more prediction on the future outcome of marine invertebrates. Influence of hypoxia on benthic communities of the MLs (Kružić 2002, Mikac 2003) has not been investigated. BL is a home for the scleractinian coral Cladocora caespitosa, which forms a reef (650 m²) between depths of 6 and 18 m near the Vrbovačka sampling station (Hrustić et al. 2013). While experimenting on another scleractinian coral reef, Reynaud et al. (2003) found out that calficification was decreased by 50% when t and p(CO₂) were both elevated, with no significant change in response to increased p(CO₂) under normal t. Their findings indicated a need to re-evaluate the projected decrease of marine calcification by the year 2100. Taking into account their findings, McNeil et al. (2004) reported that the annual average coral reef calcification rate will increase with future ocean warming and eventually exceed pre-industrial rates by about 35% by the year 2100.

Although notably stratified throughout the year, anoxia did not develop in deep waters of SL. Overturn of the water column under influence of strong cooling of the surface water has not been documented in SL so far, but such events were documented in the meromictic lake Dragon’s Eye a few hundred kilometres farther north, causing the mass mortality of benthic and planktonic organisms in the early autumns of 1997 (Kršinić et al. 2000, Barić et al. 2003) and 2011 (Ciglenečki et al. 2015). These events revealed that the copepod Acartia italica has been well adapted to the extreme conditions of anoxia, while some benthic Pennatae diatoms also survived anoxia in Dragon’s Eye in 1997 (Barić et al. 2003). In addition, the research in the northern Adriatic Sea at 24 m conducted by inducing anoxia revealed that meiobenthic harpacticoid copepods from the family Cletodidae were well adapted to anoxia (Grego et al. 2014).

The absence of upwelling in the MLs indicates that diffusion is the most important mechanism of transport of dissolved matter from deep to upper waters. If upwelling were a normal feature of the MLs, PO₄ released from iron-bound P in oxygen-poor sediments would reach the upper water very fast, alleviating potential P limitation according to DIN:PO₄>16 (Table S1) and acting as a positive feedback to increase hypoxia by stimulating phytoplankton production (Conley et al. 2009b). Deep water of SL was >100% and ~150% richer in total Fe than deep water of BL and G (Buljan and Špan 1976), respectively, while iron-phosphorus still remains unexplored in sediments of this system.

The experiment of fertilization of the MLs was started via addition of 21.5 tonnes of “superphosphate” in eight instalments in BL in 1954 (Buljan and Špan 1976). In total, an average of 1.15 μmol PO₄ l^–1 was introduced in SC and the MLs. The system reacted by multiplying the primary productivity more than six-fold in comparison with 1953 (Buljan and Špan 1976). Hydrogen sulfide (H₂S) was almost persistently present at 25 m in SL between 1951 and 1956 (Buljan and Špan 1976). It was occasionally present at notable concentrations at 20 m (e.g. 4.84 mg l^–1; overall maximum 4.95 mg l^–1 was detected in SL at 25 m on the same day, 11 December 1952). The fertilization experiment caused formation of H₂S (2.93 mg l^–1) at 13 m at Pospile Donje in BL on 20 July 1954 (Buljan and Špan 1976). These authors noted “pink and red water” at 20 m in SL in May and June 1953, which was most likely connected to the presence of anoxygenic phototrophic purple sulfur bacteria, commonly found at the chemocline and in the hypolimnion (Pjevac et al. 2015). Consequently, the MLs were recognized as a highly vulnerable ecosystem and became the protected area of the Mljet National Park (NP) in 1960. It has been shown that the widening and deepening of SC and its entrance to BL in 1960 mitigated anoxic conditions in the near-bottom water of SL, according to historical profiles of the redox-sensitive elements in sediment cores of the MLs (Sondi et al. 2017). WRT in the MLs before 1960 was most likely longer than the WRT estimated recently by Cuculić et al. (2012). Buljan and Špan (1976) did not detect H₂S in deep water of SL in 1961, which is in line with the statement of Sondi et al. (2017), indicating poor ventilation as one of the most important parameters linked to estimated persistent anoxia in the near-bottom water of SL before 1960.

The negative correlation (p<0.05) between PO₄ and O₂ in the water columns of the MLs is common in hypoxic waters, e.g. bottom waters of the Baltic Sea (Conley et al. 2002). The periods (Table S5) rather than water layers (Tables S8, S11), except for BL in the period of stratification, differed significantly (p<0.05) in concentration of PO₄ at all stations, having higher values during the mixing period (Tables S3, S4, S5). Because of the relatively high DIN/PO₄ at high productivity in a P-enriched environment, Buljan and Špan (1976) assumed N₂ fixation in the MLs. A similar hypothesis, derived from the excess of NO₃ in SL in early spring, was proposed by Hrustić et al. (2013). N₂ fixation can increase eutrophication, especially if coupled with a flux of PO₄ from deep water or sediments (Jensen et al. 1995) to the photic zone (Conley et al. 2009a). In oligotrophic P-starved regions, N₂ fixers can increase their competitiveness by using excessive N and energy to produce exoenzymes involved in utilization of DOP (Landolfi et al. 2015). The problem of eutrophication in the Baltic Sea was a good reason for the development of a new bioassay to test the distance between N and P limitation in mesotrophic surface waters (Hrustić et al. 2017). The integrated (depth of 0–10 m) samples from the ocean acidification experiment in the Tvärminne field station (Baltic Sea) in August 2012, without a bloom of diazotrophic cyanobacteria, showed the surplus of P, P⁺=0.30±0.10 μmol l^–1, but the effect of ocean acidification was not detected by this method (Hrustić et al. 2017). It would be interesting to use this new bioassay in the future to explore the distance between N and P limitation in the protected mesotrophic MLs, where the food web is controlled by Aurelia sp., and the particulate feeding of Cladocora caespitosa on phytoplankton from the surface currents in BL (Hrustić et al. 2013) may supply missing nutrients in the coral reef (Wyatt et al. 2010).

Anoxia below 39 m in BL in the period 26-28 August 1996 (Benović et al. 2000) did not last for long and wintertime cooling homogenized the entire water column. Quiescent or slowly moving speciemens within a swarm of Aurelia sp. were noted just above the anoxic layer. Aurelia sp. were repeatedly observed in swarms with very large numbers of individuals in BL except in September 1997. The quiescent individuals above the layer of anoxia were most likely affected by hypoxia since decreasing O₂ sat. from the maximum of 101% at 18.5 m to depleted O₂ at 39 m (Benović et al. 2000) probably caused death to many individuals of Aurelia sp., which might have been a significant source of NH₄ in BL (Tinta et al. 2010, 2016). However, deep water of SL had significantly (p<0.01) higher NH₄ than deep water of BL, which is, together with the gradient of NH₄ concentration below the depth of 20 m in SL (Benović et al. 2000), in line with our results (Fig. 5A, B; Tables S7, S10).

These findings point out other reasons why NH₄ in deep water of SL is significantly higher than NH₄ in deep water of BL. Higher t and Chl a are positively related to phytoplankton growth (Agawin et al. 2000). Increased NO₂ in the photic zone might be notably related to excretion during phytoplankton growth (Al-Qutob et al. 2002). This fits with the detection of maximum NO₂ concentrations in the photic zone of BL and G in March, but is less likely to be relevant for the maximum value of NO₂ in deep water of SL in October. According to the highest yearly means of t and Chl a (p<0.05) (Tables S1, S2), a comparatively intense phytoplankton growth was estimated for SL. However, the difference in NH₄ between the MLs was markedly higher during lower production, i.e. in the mixing period when the difference in Chl a was lower (Tables S3, S4). Similar findings can be seen throughout the layers in different periods, epecially in upper waters during the mixing period when BL had higher Chl a but lower NH₄ than SL (Table S9). This leads us to conclude that phytoplankton is probably not responsible for significantly higher NH₄ in SL.

Because of its smaller size, SL receives more terrestrial organic and mineral detritus relative to the amount of authigenic sediment than BL (Lojen et al. 2010). Particulate organic matter (POM) undergoes remineralization through the water column. A negligible portion of POM (~1%) escapes remineralization, being passively buried in oceanic sediments (Delaney 1998), but the percentage of POM received by sediments in the shallower SL is expected to be notably higher. Sondi et al. (2017) determined oxic-anoxic boundaries in the sediment cores of SL, BL and G at depths of 2, 5 and 15 cm, respectively, followed by the profile of dissolved iron. These results led us to conclude that surface sediments of SL have significantly higher oxygen demand than sediments of BL and G, producing notably more NH₄ from the increased amount of POM. That would be an explanation for the negative correlation (p<0.05) between NH₄ and O₂ in SL (Table 3). Wang et al. (2017) reported that re-aeration and sediment oxygen demand were more important than photosynthesis and water column respiration as counterbalanced coupled processes affecting the near-bottom DO concentration in the shallow estuary. Re-aeration of deep waters in the MLs is not expected in summer (Vilibić et al. 2010). Intermittent summer (July and September) hypoxia in the warmer deep water of SL can be attributed to presumably higher organic matter supply to the surface sediments, which have an increased oxygen demand. The plunge of the intense pycnocline might have contributed to hypoxia in the deep water of SL in summer.

Positive AOU in the whole water columns of SL and BL in December (not presented), as well as higher AOU at all investigated locations in autumn-winter, were most likely related to the lower photosynthetic activity. The critical depths of dominantly higher consumption than production of O₂, i.e. where AOU>0 μmol kg^–1, involving a notable regeneration of nutrients by organic carbon remineralization, were situated ~10 m below the water columns’ division points of 10 m for SL and 17 m for BL (Fig. 5A, B). There were no significant differences (p>0.05) between MLs concerning O₂, AOU, OUR and OCRR. The larger volume of water in BL was characterized by a net utilization of O₂ (AOU>0 μmol kg^–1) (Fig. 5A, B) at a higher average OUR than in SL. A larger amount of organic carbon was remineralized in deep water of BL in the mixing period, due to a higher OCRR and a larger volume of water with positive values of OCRR than in deep water of SL in the same period (results not presented, derived from AOU). This finding is in line with a presumably higher % of POM that escapes remineralization during sinking in SL. Since we did not have data on WRT for deep waters of the MLs in the stratification period or any data on WRT in G, we performed statistical analyses for OUR and OCRR to a lesser extent.

According to the vertical profiles of SiO₄ and NO₃ in the MLs in April (Fig. 6A, B), we assume that diatoms were a major player in a spring bloom. It is interesting that Chl a had high concentrations only at depths ≥30 m in BL during the whole water column blooms in SL in April and July. Since a major player in a mesozooplankton community of BL, the calanoid copepod Calanus helgolandicus, cannot reproduce in SL (Miloslavić 2012), we speculate that the higher grazing on diatoms in the upper water of BL, at least in April, may explain this result. In the warmer and more oligotrophic July, when C. helgolandicus is expected to reside below the thermocline in BL, it is possible that Aurelia sp. feeding on copepods had a positive effect on autotrophic biomass below the thermocline (Turk et al. 2008). In our study, maximum concentrations of nutrients (Figs 5A, B and 6A, B) in the MLs were higher than those in SL in 1997 and BL in 1997-1998, except for PO₄ in SL and TN in BL, which were similar to the values detected in 1997-1998 (Benović et al. 2000). We could not compare TP as it was not reported by Benović et al. (2000). Average nutrient concentrations in our study seem to be similar to those reported in Benović et al. (2000) and we cannot infer more about any change in the trophic status of the study area. Under the circumstances, we could not calculate the trophic status of the MLs from the data collected two decades ago, but we present TRIX for the period October 2009-September 2010.

Steckbauer et al. (2011) proposed 3.5 mg O₂ l^–1 as a new operational threshold to designate hypoxic waters, representing 88% of the distribution of mean lethal DO concentrations in warm coastal waters, except for those of the 12% most sensitive species. That threshold (~52% DO saturation) is less conservative than 4.6 mg O₂ l^–1 (~70% DO saturation) for warm coastal waters, representing 90% of the distribution of mean lethal DO concentrations (Vaquer-Sunyer and Duarte 2008). Therefore, it is a scientific effort of the highest priority to re-evaluate DO thresholds that define hypoxic waters, taking into account climate change trends. In this context, the MLs seem to be an excellent laboratory for studying differences in benthic communities related to a future potential t increase.

Detailed research on the thermohaline properties of SL in the mixing period should be conducted to review whether the upper water tends to be markedly stratified year after year. Persistent stratification along with the present trends in climate change, e.g. higher precipitation (Justić et al. 2005), have a potential to extend the duration and impact of hypoxia. Since small systems like marine lakes respond fast to environmental changes (Ciglenečki et al. 2015, Miloslavić and Lučić 2015), research on these lakes and similar systems should be encouraged in order to gain more insights into the impacts of driving factors of hypoxia on biodiversity in moderately warm marine systems worldwide.

In the present study, we have demonstrated that two moderately eutrophic Mediterranean marine lakes that communicate and depend on each other, with only a few significant differences, were affected by hypoxia in different frequencies throughout the year. A long-term chemical feature of significantly higher ammonium concentration in a smaller, shallower lake more isolated from the coastal sea has been shown by comparison with data collected from the period more than a decade ago. A notable difference in temperatures of deep waters in these lakes might be one of the key parameters to focus on in future studies, for example, on the impact of driving factors of hypoxia on the benthic communities in the Mljet NP. For the first time, an apparent oxygen utilization and the OUR as well as OCRR have been presented for this study area. Although not significantly (p>0.05) different between the lakes, AOU, OUR and OCRR indicated that a larger volume of deep water in BL in the mixing period was affected by net oxygen utilization and organic carbon remineralization. These findings are in line with the expectation that more POM in the shallower water escapes remineralization, and is then subsequently oxidized in the surface sediments of SL.