Firstly, the most recent sediments (MoWP; <11 cm depth of MUC) were compared with species that thrive in the modern Baltic Sea (Table 1). The recurring late summer (July–August) cHABs of the Baltic are dominated by the taxa Nodularia spumigena, Aphanizomenon flos-aquae and, to a minor extent, by Anabaena spp. and other species from the order Nostocales, family Nostocaceae (Hajdu et al., 2007; Hällfors, 2004; Kanoshina et al., 2003; Karjalainen et al., 2007; Sivonen et al., 2007; Celepli et al., 2017). While the Nodularia genus is usually prevalent, changes in the composition of the community have been observed from the early to the late stage of the cHAB and from one year to another, resulting in a large variation in its features over time (Finni et al., 2001; Hajdu et al., 2007; Kahru et al., 1994; Wasmund, 1997). A recent extensive meta-omics study revealed that in the Baltic Proper (the predominant area for cHABs), 69 % of the heterocystous cyanobacteria belong to Aphanizomenon, 23 % to Anabaena and 8 % to Nodularia (Celepli et al., 2017).

The HG distribution in the MoWP sediments, with the C26 diol as the dominant HG (Fig. 4a, summarized in Table 1), agrees well with the HG distribution in cultures of Nodularia, Aphanizomenon and Anabaena as well as other members of the Nostocaceae family (Table 1), including those that have been isolated from the Baltic (Bauersachs et al., 2009a, 2017). These cultures generally also synthesized minor amounts of the C26 keto-ol HG, as was seen in the MoWP sediments. The C28 diol, present in trace amounts in the MoWP sediments, was found in varying amounts in the Nodularia, Aphanizomenon and Anabaena cultures. Even between different strains of the same species, the amounts present were highly variable, from a dominant component to not detected (Table 1). The C28 keto-ol, C28 triol and C28 keto-diol HGs were minor components in the MoWP sediments. While not produced consistently across the Nodularia, Aphanizomenon and Anabaena cultures, they were found in certain strains, generally as trace or minor components, in agreement with the distribution in the sediments (Table 1). It is possible, however, that the presence of the C28 triol HG in the MoWP sediments may be linked to the presence of the genus Calothrix (cf. Table 1), which is commonly found in the rocky seabed of the basin (Sivonen et al., 2007).

Overall, the distribution of the HGs observed in the MoWP sediments was in good agreement with the HG distribution of the family Nostocaceae (Table 1), which fits in with the reported dominance of members of this family during the summer cHABs of the Baltic. Furthermore, the HG distribution remained relatively constant throughout the MoWP sediments (Fig. 4a), suggesting that overall the community composition of heterocystous cyanobacteria in the Baltic Sea has remained stable during the last ∼60 years.

The HG distribution in the sediment from the pre-MoWP brackish phase (i.e., from the Ancylus Lake–Littorina Sea (AL–LS) transition to the start of the MoWP) reconstructed in this study was similar to that of the MoWP, although the C26 diol and the C28 diol were present in a greater fractional abundance (Table 1; Fig. 4). The other four HGs were either minor or occurred in traces. Although often absent, a number of Nostocaceae strains have been found to contain the C28 diol (Table 1), and in one Anabaena sp. strain (CCY9402), it was found to be the dominant HG (Bauersachs et al., 2009a). The increased proportion of the C28 diol through the pre-MoWP brackish phase suggests there was a somewhat different cyanobacterial community composition than during the MoWP, although most probably still dominated by cyanobacteria belonging to the family Nostocaceae. The HG distribution remained relatively constant from the establishment of the brackish phase to the MoWP (Fig. 4), which suggests that the cyanobacterial community of the Baltic did not undergo major changes from the AL–LS transition to the MoWP and remained dominated by cyanobacteria belonging to the family Nostocaceae.

The Ancylus Lake phase displayed a distinct HG distribution from the brackish phase (Fig. 4b; summarized in Table 1). The C28 diol was often dominant and both the C26 and C28 keto-ol were present in a higher proportion than during the brackish phase. This is most evident for the Ancylus Lake phase I and the middle section (ca. 230–210 cm) of the Ancylus Lake phase II. Yet, at the first (ca. 250–230 cm) and last part (ca. 210–193 cm) of the Ancylus Lake transitional phase II, the HG distribution is more similar to the one observed in the brackish phase (Fig. 4b). This is also evident from the PCA analysis with more negative values for PC1 and PC2 at those depths (Figs. 4b and 5b). The AL–LS transition did not happen instantly (Borgendahl and Westman, 2007; Emeis et al., 1998; Gustafsson and Westman, 2002; Hyvarinen, 1984), and the sediment intervals showing a brackish-like distribution of the HGs probably correspond to weak pulses of marine water that might have occasionally entered the basin already during the Ancylus Lake transitional phase II and consequently influenced the overall distribution of the HGs (Fig. 4b). This final stage of this transition is also evident from the lithology and TOC profile (Fig. 3c).

When the Baltic evolved from a freshwater lake into a brackish semi-enclosed basin, it experienced an increase in salinity from fresh to values of 10–15 ‰ (Gustafsson and Westman, 2002). The observed changes in the HG distribution over the AL–LS transition suggest that this change from freshwater to brackish resulted in a different cyanobacterial species composition and hence a different HG distribution. Indeed, several freshwater species have been found to contain a HG distribution dominated by the C28 diol (Table 1), including Cyanospira rippkae (Soriente et al., 1993), Tolypothrix tenuis (Gambacorta et al., 1998) and Aphanizomenon aphanizomenoides (Wörmer et al., 2012), although we emphasize that we did not analyze HGs with C28+ alkyl chains for this stage and, therefore, cannot exclude the contributions of cyanobacteria producing such extended HGs. Alternatively, an increased influx of soil organic matter during the Ancylus Lake phase could be responsible for the distributional HG changes. However, since HG lipids contain an attached sugar moiety, we feel it is unlikely that HGs produced in soil will make it to the sediments of the Baltic Sea since they would be exposed extensively to oxygen during transport and only relatively stable components such as lignin, wax lipids, and branched glycerol dialkyl glycerol tetraethers (GDGTs) will likely survive this transport to the middle of the Baltic Sea where our core was taken.

For Nodularia spumigena, the most abundant heterocystous cyanobacterium in the present Baltic, its basic physiological features, such as growth, production of the toxin nodularin and differentiation of heterocysts are substantially affected at extreme salinities (Mazur-Marzec et al., 2005; Moisander et al., 2002). This is thought to be the predominant reason why Nodularia blooms only occur within a certain salinity range (i.e., 7–18 ‰) in nitrogen-deficient waters (Mazur-Marzec et al., 2005). This would imply that during the Ancylus Lake phase, the low salinity was limiting the growth of Nodularia sp. Other heterocystous cyanobacteria such as Anabaena and Aphanizomenon may be better adapted to freshwater conditions.

A high variability in the HG distribution is also observed for the Yoldia Sea sediments (Figs. 4b and 5b). The most distinct feature is the relatively high fractional abundance of the C28 diol HG, which sometimes reaches 50 %, the highest value recorded for all sediments. The Yoldia Sea phase was a relatively short period when a connection with the sea was established and waters may have become brackish. Nevertheless, the HG distribution is not at all similar to that of the brackish phase.

As a consequence of the retreat of the ice sheet and the entry of sea water through the Danish straits, there was an increase in water temperature during the AL–LS transition (Björck, 1995). It is possible that this increase in water temperature could have been responsible for the changes in the HG distribution, as growth temperature has been reported to affect the distribution of the HGs in cyanobacteria belonging to the order Nostocales (Bauersachs et al., 2009a, 2014b, 2015). Specifically, increasing temperature positively correlated with increasing relative proportions of HG diols over HG keto-ols. In our record, the ratio of diols to keto-ols increased from the Ancylus Lake towards the brackish phase (Fig. 4b), which would be in agreement with the higher SWTs during the brackish phase. However, when the HG proxies are used to estimate SWT based on the proxy calibrations from a lake (Eqs. 1–4), the predicted temperatures are somewhat unrealistic. For the brackish phase, the HDI26 and HDI28 values vary between 0.96–1.00 and 0.95–1.00, translating into an average SWT of ca. 24 and 23 ∘C, respectively. This is too high, even for summer temperatures when the cHABs occur (Kanoshina et al., 2003). TEX86-derived summer temperatures (Kabel et al., 2012; Warden et al., 2017) do not exceed 17.5 ∘C (Fig. 3d). The application of the HG-based calibrations in this setting assumes that salinity has no impact since they have been established for a freshwater lake (Bauersachs et al., 2015). For the Ancylus Lake and Yoldia Sea phases, the HDI26 and HDI28 values are highly variable and range between 0.52–1.00 and 0.00–0.99, translating into average SWTs of ca. 20 and 17 ∘C, respectively. This is lower than observed for the brackish phase but also seems too high. Apparently, cyanobacterial species composition exerts an important control on the HG distribution in such a way that the HGs are not able to predict accurate temperatures in the brackish–freshwater system of the Baltic. Cultivation experiments with HG-producing strains isolated from the Baltic Sea (see Table 1) at varying temperatures may improve the HG paleo-thermometry of Baltic Sea sediments.

In the Baltic the occurrence of summer cHABs has intensified since the 1950s (Kabel et al., 2012; Poutanen and Nikkilä, 2001). Yet, due to the spatial patchiness and interannual variability, it has proven difficult to recognize a clear trend in the cHABs on the scale of the entire Baltic (Finni et al., 2001; Kahru and Elmgren, 2014; Pitarch et al., 2016; Wasmund and Uhlig, 2003). However, the general interest in these events has led to intensified research (see Finni et al., 2001; Kahru and Elmgren, 2014; Kutser et al., 2006 among others) and also led to the establishment of the Baltic Marine Environment Protection Commission (HELCOM) in 1992 to monitor this phenomenon. Disparate indices and parameters have been employed to describe and quantify cHABs over time and were applied in the different areas of the Baltic which are biogeochemically heterogeneous and display distinct seasonal dynamics (Kahru, 1997; Kahru et al., 2007; Kahru and Elmgren, 2014; Kononen, 1992; Kutser et al., 2006; Pitarch et al., 2016; Wasmund and Uhlig, 2003). The methods employed and the frequency of the sampling campaigns have improved in the recent past, reducing the inaccuracy associated with previous sampling methods and measurements (Hansson and Öberg, 2016; Kahru, 1997; Kahru and Elmgren, 2014; Wasmund and Uhlig, 2003). However, intrinsic limitations of the techniques in use may still cause difficulties when comparing measurements from different years, even within the same time series (Finni et al., 2001; Kahru, 1997; Kahru and Elmgren, 2014).

Here, the HG abundance over the past ∼30 years (i.e., 2012–1979 of the MoWP), recorded within the first ∼7 cm of the MUC, is discussed in comparison with a time series of the cHABs episodes relative to the Eastern Gotland Basin (Fig. 6), whose intensity is expressed as the frequency of cyanobacteria accumulation (FCA) (Kahru and Elmgren, 2014). FCA is determined by ocean color satellite data and expresses the frequency of the occurrence of cHABs in July–August using 1 km2 pixels (Kahru et al., 2007). Kahru and Elmgren (2014) reported prominent cHABs in the early 1980s, in the period 1990–1996 and again from 1999 until 2008, with the interval 2005–2008 recording the highest FCA percentages with relevant interannual changes of the areal extent (Kahru, 1997; Kahru et al., 1994, 2007; Kahru and Elmgren, 2014). The HG lipid biomarker abundance profile from our sampling site was overall in reasonable agreement with the FCA measurements (Fig. 6). However, it failed to record the intense cHABs of the early 1980s, and there is a mismatch of 1 or 2 years in recording the start of the strong cHABs recorded at the end of the same decade (Kahru and Elmgren, 2014). Furthermore, this comparison is complicated by a certain degree of uncertainty in the age model of the sedimentary record. Moreover, the intrinsic temporal and spatial variability in the cHABs in the modern Baltic Sea, together with the difficulties encountered in an attempt of creating a consistent long time series that combines FCA data from multiple satellite sensors, may provide an explanation for the discrepancies observed (Kahru and Elmgren, 2014; Wasmund and Uhlig, 2003).

We observed multiple peaks of the HG abundance in the MoWP section of the MUC core (≤11 cm depth), which reached ∼50–150 r.u.gTOC-1. Below this in the LIA section, the HG abundance declined sharply to <10 r.u.gTOC-1 (Fig. 3a). This decline may be expected given that the MoWP is characterized by higher summer surface temperature (Fig. 3d), increased organic matter deposition and more frequent anoxic events than the LIA phase (Kabel et al., 2012), all being conditions that lead to increased cHABs. Furthermore, the cooler LIA experienced more oxygenated bottom water, which may have affected HG preservation (see also below). However, a substantially increased HG abundance was not observed below the LIA in the MWP section of the MUC core (Fig. 3a). Similar to the MoWP period, the MWP was characterized by higher summer temperatures (Fig. 3d) and increased stratification of the water column that would favor bottom anoxia and, presumably, the occurrence of cHABs. The top of the GC also records the LIA–MWP transition (Fig. 3). Here, the HG abundance reached ∼10–18 r.u.gTOC-1 at <30 cm depth, which is up to 4 times higher than the HG abundance observed in the MUC for the same period. This discrepancy between the HGs records in the two related cores is puzzling. After the MWP, HG abundance declined to ≤ 5 r.u.gTOC-1 during the remaining part of the brackish phase, as recorded in the GC (Fig. 3a), with only a minor increase in the HG abundance during the periods when summer temperature was higher and the Baltic Sea was stratified, resulting in bottom water anoxia (Fig. 3; e.g., during the Holocene Thermal Maximum).

Based on these data from the Baltic Sea, it is not possible to confidently couple the HG abundance record directly with cHAB occurrences and anoxic events in the past. Several factors are thought to affect this relationship. Firstly, it is possible that the occurrence of cHABs varied over time. In the shallow part of both sediment cores, HG abundance was generally high, but it started declining with increasing depth, independently of other factors (Fig. 3). This might suggest that cHABs were less common and intense in the past brackish Baltic Sea, even at times of warmer and more stratified conditions. Secondly, the succession of oxic and anoxic bottom water conditions may impact the preservation efficiency of HGs. Such successions took place in the Baltic Sea during the entire Holocene as is evident from the alternation of dark–laminated with light–homogeneous sections in the sedimentary record (Kabel et al., 2012). In the shallow part of both sediment cores, the high abundance of HGs coincided with dark–laminated sediment phases; low HGs, by contrast, co-occurred with light–homogeneous phases. In contrast, in the deeper part of the section this correspondence was lost. Finally, the generally declining trend in the HG absolute abundance in the shallow sediments might also be due to the anaerobic breakdown of the HGs. A decline in lipid biomarkers with depth has been documented before in anoxic Black Sea surface sediments (Sun and Wakeham, 1994). This process would be seemingly in contrast with previous indications of a high preservation potential of the HGs in ancient marine and lacustrine anoxic sediments (Bauersachs et al., 2010), but it should be realized that even in the older Baltic Sea sediments, HGs are still detected. Apparently, even if diagenesis occurs, it does not result in the complete destruction of HGs.

The HG results seem to partly contrast an earlier study that, based on fossil pigment records, suggested that cHABs have been recurring simultaneously with the mid-Holocene anoxic events (Funkey et al., 2014). However, this study used carotenoids (i.e., zeaxanthin and echinenone) that are not entirely specific to cyanobacteria and are certainly not limited to nitrogen-fixing cyanobacteria, as opposed to the highly specific HGs that were used here. For example, zeaxanthin is also produced by Synechococcus, the dominating unicellular cyanobacterial species in the Baltic Sea (Celepli et al., 2017). Furthermore, in this environment of highly variable sediment redox conditions, the effect of diagenesis should be considered. Carotenoids are amongst the most unstable organic biomarkers because of their very labile conjugated system of double bonds. Changes in redox conditions of bottom and sediment pore waters will thus have a major effect on the concentration of carotenoids, and this may explain the enhanced concentration of carotenoids in the mid-Holocene TOC-enriched sections (Funkey et al., 2014).

The general down-core decrease in the HG abundance throughout the brackish phase is continued into the Ancylus Lake and Yoldia Sea phases, where the HG abundance is at least an order of magnitude lower than in the first part of the brackish phase (Fig. 3a). The lower HG abundance in the Ancylus Lake and Yoldia Sea phases, relative to the brackish phase, could indicate that N2-fixing cyanobacteria were much less abundant during this freshwater phase. Indeed, further evidence for a lower abundance of diazotrophic phytoplankton during the Ancylus Lake and Yoldia Sea phases comes from the record of δ15N values (Fig. 3b). During these phases the δ15N values are 4–6 ‰, indicating that most of the phytoplankton community was relying on ammonium or nitrate as nitrogen sources rather than atmospheric nitrogen (Bauersachs et al., 2009b; Emerson and Hedges, 2008). When other forms of nitrogen are abundant the energetically expensive N2 fixation becomes disadvantageous (Arrigo, 2005; Capone et al., 2005; Karl et al., 1997). At the start of the LS phase, δ15N values drop to 1–3 ‰ (Fig. 3b), a range expected when N2-fixing cyanobacteria contribute substantially to primary production (Bauersachs et al., 2009b; Rejmánková et al., 2004; Zakrisson et al., 2014), and remained in this range.

As discussed above, the salinity change from a freshwater lake to a brackish sea may have had a significant effect on the heterocystous cyanobacterial composition in the Baltic. This environmental change may have also been a cause of the increased abundance of heterocystous cyanobacteria. Another environmental factor change that could have promoted increased blooming of heterocystous cyanobacteria is the increase in water temperature over the AL–LS transition (Björck, 1995). Temperature is a crucial factor influencing the growth rate and other metabolic features of free-living heterocystous cyanobacteria (Bauersachs et al., 2014b; Kabel et al., 2012; Mazur-Marzec et al., 2005; Staal et al., 2003). In the modern Baltic Sea, a minimum temperature of 16 ∘C is considered essential to initiate cHABs during summer when other crucial factors like a low DIN/DIP ratio, calm winds and high irradiance occur simultaneously (Kanoshina et al., 2003; Kononen, 1992; Kononen et al., 1996; Paerl, 2008; Wasmund, 1997).

It should also be noted, however, that the homogeneous appearance of the sediments and the much reduced TOC content (Fig. 3c) reveal that the water column was generally well mixed and oxygenated during the Ancylus Lake and Yoldia Sea phases, resulting in a higher degradation of organic matter (including HGs) in settling particles and surface sediments. To compensate for this effect, all HG concentrations were normalized to TOC content (Fig. 3a). However, it is known that oxic conditions in the sediment result in a decreased preservation of biomarkers relative to TOC (see Sinninghe Damsté et al., 2002). This may also explain in part the lower HG abundance in the Ancylus Lake and Yoldia Sea than in the brackish phase. However, it is noteworthy that no substantial change in the concentration of HGs is observed during the brackish phase when bottom water conditions changed from oxic to anoxic (Fig. 3). This suggest that the normalization to TOC content is an effective way to compensate for changing redox conditions of bottom and pore waters. The effect of oxic degradation is probably also not responsible for substantial changes in the distribution of the HGs since they are structurally similar and all contain a relatively labile glycosidic bond, so there is no reason to assume that one HG will degrade faster than another.