Around 0.1 g of freeze-dried sediments from LC21 and ODP 967 were weighed individually into a porous crucible. HCl (1 mL, 4 mM) was added to remove any inorganic carbon from the sediment. After HCl was drained, samples were neutralised with deionised water and were dried at 65 ∘C. TOC content of each sample was obtained by means of non-dispersive infrared spectrometry using a LECO CS230 analyser. A standard (Chinese stream sediment, NCS DC 73307; LGC, Teddington, UK) was analysed after every 10 samples to check accuracy. TOC content of the 64PE406-E1 sediments was determined by a Thermo Scientific Flash 2000 elemental analyser coupled to a Thermo Scientific Delta V isotope ratio monitoring mass spectrometer (EA-irMS) via a Conflo IV.

Freeze-dried 64PE406-E1 sediments were analysed to determine both bulk δ15N and bulk δ13C values. For carbon isotope analysis, the sediment was first decalcified using a 2N HCl solution for approximately 18 h. The sediment was rinsed three times using double-distilled water and then freeze-dried again. δ15NTOC and δ13CTOC were measured using a Thermo Scientific EA-irMS (see above). The 15NTOC and 13CTOC are expressed relative to air and the Vienna Pee Dee Belemnite (VPDB) standard, respectively, and the isotope analysis precision was 0.2 ‰. For nitrogen isotope analysis, acetanilide, urea, and casein with predetermined isotope values were used as reference material; for carbon analysis benzoic acid and acetanilide were used.

Sapropels spanning four of the most recent five events in the Aegean Sea were sampled from core LC21 from the Aegean Sea and analysed for anammox biomarkers (Fig. 3a). Ladderane fatty acids (i.e. C18–[3]–ladderane fatty acid, and C18–[5]–ladderane fatty acid, C20–[3]–ladderane fatty acid, and C20–[5]–ladderane fatty acid; Fig. 1), the traditional biomarkers for anammox bacteria (Jaeschke et al., 2009; Rush et al., 2012a; Sinninghe Damsté et al., 2002), were found in the most recent sapropel (290–610 ng g-1 TOC; in S1, ∼ 7 ka; Fig. 3a) in abundances comparable to those found in sediments of the Peru Margin and Arabian Sea (Rush et al., 2012a). Conversely, ladderanes were not detected in the sediment sampled directly below this sapropel layer (out S1, Fig. 3a), indicating anammox was an important process during S1 deposition, but likely not before the onset of sapropel deposition. Ladderane concentration progressively decreased with increasing age of the deeper sapropels: 80–170 ng g-1 TOC in S3 (∼ 85 ka); not detected in S4 (∼ 100 ka); and 0–90 ng g-1 TOC in S5 (∼ 125 ka). It is worth noting that two of the three sediments from within S5 did not contain detectable ladderanes. This demonstrates the previously described sensitivity of ladderane lipids to diagenesis (Rush et al., 2012a; Jaeschke et al., 2008) and highlights their potential weakness as a biomarker proxy for past anammox bacteria in ancient sediments. Residues of TLEs were also saponified for ladderane analysis, as these have previously been shown to extend the detection of anammox in older sediments by releasing more matrix-bound ladderanes (Rush et al., 2012a). However, this did not show any difference in the presence of anammox (i.e. there was no detection of ladderanes in residues in which the original TLEs did not contain these biomarkers). The non-detection of ladderanes in most of the S5 samples is particularly surprising as this is the most intense of the Late Quaternary sapropels (Struck et al., 2001), having been described as analogous to the modern-day Black Sea (Menzel et al., 2006). Since anammox is currently present and actively removing N in the cline of a strong redox gradient (redoxcline) of the Black Sea (Jensen et al., 2008; Kuypers et al., 2003), it was expected that anammox behaved similarly in the nitrogen cycle of the eastern Mediterranean during deposition of the S5 sapropel. Given that the oldest detection of ladderanes comes from a slightly older record in the Arabian Sea (Jaeschke et al., 2009), it is unclear why ladderane detection in S5 is sporadic. Perhaps degradation is responsible for the rapid removal of ladderanes from the system during deposition, or the low resolution in the S5 record made these specific sediment depths not ideal targets for anammox activity.

Bacteriohopanetetrol isomer (BHT isomer; Fig. 1) has recently been proposed to be an alternative biomarker for anammox bacteria in paleo-records (Rush et al., 2014b). Our analysis of non-derivatised BHT isomer was based on the previously published method analysing intact polar lipids via reverse phase liquid chromatography (Wörmer et al., 2013) and achieved better separation of BHT isomer from BHT compared to the acetylated LC–MS (liquid chromatography–mass spectrometry) method (Rush et al., 2014b; Fig. S1).The concentration of BHT isomer in the Aegean Sea sapropels showed a similar trend to ladderanes in the shallow sediment layers (Fig. 3b): the concentration was high in S1 (71–360 µg g-1 TOC) and low in the underlying sediment (12 µg g-1 TOC; out S1), in good agreement with the ladderane data. In contrast, however, BHT isomer was detected in all deeper sapropels at higher concentrations (64–180 µg g-1 TOC in S3; 67–90 µg g-1 TOC in S4; and 68–160 µg g-1 TOC in S5) than the ladderanes. Sediments from outside the sapropel had relatively low, but measurable, BHT isomer concentration (8–17 µg g-1 TOC). As BHT isomer was detected in all sapropels, including the oldest S5 sediments, it appears that the rapid removal of ladderanes from the system is due to degradation during deposition. These results clearly demonstrate the utility of BHT isomer as a biomarker for anammox in paleo-records compared to the more labile ladderane lipids. A hemipelagic, light, non-sapropel sediment sampled between S3 and S4 contained neither ladderanes nor BHT isomer (Fig. 3; out S4), indicating a period where anammox was likely not active in the Mediterranean nitrogen cycle. Furthermore, the detection of BHT isomer in the non-sapropel sediments underlying S1 and S5 and overlying S3 shows that this lipid is a better biomarker than ladderanes for recording trace amounts of anammox throughout the history of the Mediterranean system, especially in sediment deposited under oxic (bottom) water conditions.

To further investigate the occurrence of anammox during sapropel deposition, we analysed in high resolution the well-developed S5 (TOC content up to 12 %; Fig. 4) recovered from the Levantine Basin in the eastern Mediterranean during a cruise of the R/V Pelagia in 2016 (64PE406-E1; Fig. 2). X-ray fluorescence scanning of this core showed no peak in Mn / Ti in the top of the sapropel, indicating this S5 record does not contain the burndown effect of oxygen diffusing downward post-deposition (Dirksen et al., 2019). This was corroborated by the Ba / Ti record, used as a proxy for paleo-productivity, which followed the same trend as organic carbon throughout this sapropel. Thus, it was expected that ladderane fatty acids would be preserved in the high-TOC sediments of this S5 record. However, in line with the earlier results of ladderane analyses for S5 in the Aegean Sea record, the results from the Levantine Basin were inconclusive. Ladderanes were detected in all, except two, of the thirty sapropel samples but were at the detection limit (i.e. peak area of 3× background), preventing interpretation of the ladderane profile in S5. The cause of low ladderane concentration even in sediments with high TOC may be due to unknown degradation in Mediterranean sapropel sediments, and future work should include anoxic degradation experiments with anammox biomass to elucidate potential mechanisms.

The BHT isomer does not appear to have been affected by degradation in the same way as ladderane lipids; it was above detection limit in all S5 sediments (Fig. 4b). The concentration of BHT isomer increased progressively by a factor of 10 from the onset of S5 until the core of the sapropel event (from average pre-sapropel value 2.69×1011 ru g-1 TOC to 2.28×1012 ru g-1 TOC at 33–34 cm core depth; Fig. 4) and then waned until the termination. This indicates that anammox was an important process during the formation of S5, actively removing nitrogen from the marine system. Photic zone euxinia has been observed in cores from the western part of the eastern basin during S5 formation by the identification of isorenieratene (Marino et al., 2007; Rohling et al., 2006). Isorenieratene is a biomarker lipid for the brown strains of the photosynthetic, green sulfur bacteria (Chlorobiaceae). These organisms require the unique conditions of light, albeit at relatively low intensity, and euxinic waters, as they are very sensitive to the presence of molecular oxygen (Overmann et al., 1992). Although anammox bacteria are inhibited by the presence of free sulfide, they likely thrived at the redoxcline during deposition of S5 (Fig. 5a). This is the case, for instance, in the modern Black Sea: at 90 m water depth, where oxygen and sulfide concentrations are both low and nitrite and ammonium are readily available, the presence and activity of anammox have been confirmed via rate measurements and ladderane biomarker observations (Kuypers et al., 2003; Jensen et al., 2008).

There are two considerable peaks in BHT isomer that fall outside of the S5 trend (Fig. 4b), occurring at the onset (2.43×1012 ru g-1 TOC; 46–47 cm core depth) and termination (1.12×1012 ru g-1 TOC; 16–17 cm core depth) of the sapropel. Sea level rise and gradual freshening of the Mediterranean are believed to have caused a stepwise removal of oxygen and subsequent slow build-up of anoxia ca. 3 kyr before the (massive) freshwater discharge from the African continent instigated the real onset of S5 (Schmiedl et al., 2003). The intense anammox peak pre-sapropel formation could be a response to this marine redox shift (Fig. 5a). Anammox would have thrived, consuming the residual low levels of ammonium and nitrite in an anoxic Mediterranean water column. Then, once monsoonal discharge brought in the initial pulse of nutrients from the Nile, the slow-growing anammox bacterial population would have been rapidly outcompeted by heterotrophic denitrifiers consuming sinking organic carbon being produced in the overlying oxic waters. As S5 progressed and N supply became scarcer, anammox would have repopulated the niche of redoxcline N remover at core sapropel conditions. The peak of BHT isomer observed at S5 termination (Fig. 4) shows that the conditions were again favourable for anammox to thrive. However, this may have occurred at the anoxic sediment–water interface rather than in the water column, where low concentrations of nitrite and ammonium could have persisted from the degradation of organic matter settling on the seafloor after the reoxidation of the water column. The BHT isomer ratio (BHT isomer/total BHT; Sáenz et al., 2011) normalises the contribution of the anammox biomarker to other potential sources of BHT. The ratio in the S5 record (Fig. 4c) showed the same trend as BHT isomer concentration in the sapropel (e.g. the ratio was highest during the core sapropel, 0.58 at 30–32 cm, and showed distinct peaks at its onset and termination). The slight decrease in BHT isomer ratio before and after the sapropel event is likely due to an increased production of BHT by other bacterial sources rather than a change in the BHT isomer producer.

Short-chain (SC) ladderane fatty acids (i.e. C14–[3]–ladderane fatty acid and C14–[5]–ladderane fatty acid; Fig. 1) are oxic biodegradation products of ladderane fatty acids (Rush et al., 2011) and are used to infer exposure of ladderane lipids to oxic conditions either pre- or post-deposition. SC ladderane fatty acids were only detected in three of the S5 sediments (Fig. 4b), specifically at sapropel onset (46–47 cm core depth) and termination (15–16 and 16–17 cm core depth). This implies that, during sapropel maximum, anammox was thriving at the Mediterranean redoxcline. Anammox detritus would then have sunk through an anoxic (euxinic) Black Sea-type water column, unexposed to oxygen and the effects of β oxidation that produces SC ladderane fatty acids (Rush et al., 2011). This has been seen in the modern Cariaco Basin, where ladderanes are observed, but SC ladderanes are absent (Rush et al., 2012a). The presence of SC ladderanes at the onset and termination, yet absence in the core S5 record, could also corroborate the concept of “split-anoxia” (as proposed for S1 by Bianchi et al., 2006), which hypothesises for the first 100 to 1000+ years of sapropel formation euxinia was present as a mid-depth oxygen minimum zone rather than a continuation from the seafloor. During these periods where the water column was not fully euxinic, ladderanes would have been oxidised to SC ladderanes in the underlying waters, which would have contained a certain amount of available oxygen. Alternatively, as productivity waned, sedimentation rates would have decreased in the Levantine Basin. Lower sedimentation rates at the onset and termination of S5 would suggest a longer residence time of ladderanes in sediment that would periodically be exposed to (sub)oxic bottom water conditions. Oxic water inflow of pore waters would have stimulated the β oxidation responsible for SC ladderane formation (Rush et al., 2011). It is worth noting that, in the low-resolution Aegean Sea sample set (LC21), all samples from S1 and S3 that contained ladderanes also contained a high concentration of SC-ladderane fatty acids, whereas the singular S5 sediment did not contain SC ladderanes. This would appear to indicate that the Aegean water column during S1 and S3 deposition was not fully euxinic and that S5 in the Aegean mirrored the euxinic Levantine Basin.

Nitrogen isotope ratio (δ15N) values of bulk nitrogen in S5 sediment show a strong shift towards low values within the sapropel (Fig. 4a), a feature seen in most sapropels (Calvert et al., 1992; Sachs and Repeta, 1999; Struck et al., 2001; Higgins et al., 2010; Möbius et al., 2010). This could potentially be explained by either enhanced diazotrophic N2 fixation because N was limited in the system (Möbius et al., 2010) or the preferential uptake and burial of 14N when nitrate is present in excess and primary producers have the opportunity to fractionate maximally (Calvert et al., 1992). As a biomarker for N removal from the system was not available, previous work has only been able to approach this conundrum with evidence for N fixation processes. Using isotopic evidence of diazotrophic phytoplankton, Sachs and Repeta (1999) and Higgins et al. (2010) argue that Mediterranean surface water was nitrogen-limited during sapropel events. Here, for the first time, we present evidence of N loss in a Mediterranean sapropel using BHT isomer as an anammox biomarker. The fact that BHT isomer concentration increases towards the core of S5 appears to suggest that N species were not limited and rather that freshwater run-off could be resupplying these nutrients to microorganisms in the water column and enhancing the pool of N. However, anammox thrive at the redoxclines of modern oxygen minimum zones (Pitcher et al., 2011; Rush et al., 2012b) and euxinic basins (Wakeham et al., 2012; Kuypers et al., 2003), where pulses of fresh N species do not necessarily reach. At the S5 Black Sea-type redoxcline, anammox did not need a riverine supply of N but could have instead been sustained by the advection of N from deeper waters (Rohling et al., 2006) or by N remineralised from the sinking pool of (diazotrophic) organic matter from above. Thus, BHT isomer results show that N removal by anammox was at its highest flux during core S5 sapropel conditions and that the anammox process appears to play an integral role in N cycling during sapropel events.

To confirm that anaerobic ammonium oxidation has occurred throughout the history of anoxia in the Mediterranean Basin, not only in the most recent Quaternary sapropels, BHT isomer concentration was analysed across a high-resolution Pliocene sapropel (ODP Leg 160, Site 967; Fig. 2). The Ba record of this sapropel shows the same trend with depth as TOC, indicating no significant burndown of organic matter after its deposition (Grant et al., 2017). BHT isomer is present throughout this older record (Fig. 6b), and as the BHT isomer ratio (BHT isomer / total BHT) is consistently elevated (average 0.48; Fig. 6c), anammox is the likely source in the entirety of the record. Much like the trend seen in the S5 Levantine sapropel, sapropel S65 showed two distinct peaks in BHT isomer concentration at its onset (110–240 µg g-1 TOC; 69–73 cm core depth) and termination (640– 1100 µg g-1 TOC; 54–59 cm core depth). However, BHT isomer concentration displayed a distribution different to that of the S5 record during the core Pliocene sapropel event (Fig. 6b). BHT isomer concentration was low, likely representing unfavourable conditions for anammox during this sapropel. Isorenieratene has been detected in the Pliocene record of Site 967, albeit in a different sapropel event (Menzel et al., 2002). It is possible that euxinia shoaled further into the photic zone during this Pliocene sapropel, forcing anammox at the redoxcline to compete for N with phytoplankton (Fig. 5b). Anammox would have therefore only thrived during the build-up and termination periods when photic zone euxinia would have been deeper/less intense. Nevertheless, this hypothesis should be confirmed through future analysis of photic zone euxinia biomarkers (e.g. isorenieratene). There was a spike in BHT isomer concentration mid-sapropel that coincided shortly after with a decrease in TOC (65–67 cm core depth; Fig. 6a). Mid-sapropel breaks have been reported elsewhere, as repopulation events of benthic fauna (e.g. Rohling et al., 1993), and could be due to inflow of freshly ventilated deep water. Re-ventilation would have directly stimulated anammox bacteria that were inhibited by euxinia, whereas there may have been a slight delay on the effect of decreasing TOC (Fig. 5b). The concentration of BHT isomer was still high after sapropel deposition (∼ 250 µg g-1 TOC; < 40 cm core depth), relative to that pre-sapropel. This may indicate that the anammox process remained an important N process in the Mediterranean after bottom water anoxia waned.

Combined, the high-resolution results from the S5 and Pliocene sapropels indicate that the functioning of anammox is not always the same during periods of Mediterranean anoxia. This demonstrates that the response of the N cycle to anoxic conditions can vary drastically from one sapropel event to the next.