How dry was the Mediterranean during the Messinian Salinity Crisis ?

The Messinian salinity crisis (MSC; 5.97–5.33 Ma) is an enigmatic episode of paleoceanographic change, when kilometers-thick evaporite units were deposited in the Mediterranean basin. It is generally accepted that during the MSC interval there was a dry climate in the Mediterranean region. It is difficult to assess how dry the climate was during the MSC because a modern analogue, in size and duration, is absent. Here we reconstruct hydrological changes in the Mediterranean basin during the three main MSC stages using excellently preserved biomarkers. We used the hydrogen isotopic composition of the long chain n-alkanes (δDn-alkanes) to reconstruct the hydrological changes on the land adjacent to the Mediterranean Sea. Additionally, the δD of long-chain alkenones (δDalkenones) is used to observe changes in the Mediterranean Sea water source. The δDn-alkanes recorded during the deposition of Primary Lower Gypsum (stage 1) in Monte Tondo indicate a δD of the precipitation comparable to the present-day Mediterranean implying a similar hydrologic regime (indicated by experiments modelling the Miocene-Pliocene transition). Elevated δDalkenones values from halite unit (stage 2) of the Realmonte mine are associated with kainite and giant polygons, consistent with presumably high evaporative conditions during halite deposition. The δDn-alkanes recorded during the deposition of Upper Gypsum (stage 3) in Eraclea Minoa indicate a δDprecipitation typical for much drier settings, similar to the Red Sea region. The relative contribution of the different alkenones from Eraclea Minoa is similar to the one observed in present-day marine settings suggesting that, during stage 3, connections to the open Ocean were likely maintained. However, the δDalkenones records during deposition of the evaporites in Eraclea Minoa are similar to those synchronously registered in the Black Sea implying that a similar hydrologic regime, characterized by extended drought, covered large areas of southeastern Europe. Based on the δDalkenones similarity and the Paratethys type of ´Lago Mare´ fauna in the Mediterranean we speculate that the surface water during stage 3 was, at times, derived from the Black Sea.


Introduction
The Messinian salinity crisis (MSC) of the Mediterranean is one of the paleoceanographic events that had a profound impact, regionally and probably globally in the past ten million years. During this episode, between 5.97 and 5.33 Ma (e.g. Hsü et al., 1973;Krijgsman et al, 1999;Manzi et al., 2013), the paleoenvironment and paleogeography of the circum-Mediterranean was completely modified (Fig. 1). A severe deficit in the water budget of the Mediterranean interrupted the already limited water exchange with both the Atlantic Ocean and the Black Sea (e.g. Roveri et al., 2014b) causing the precipitation of an up to 2 km-thick evaporite unit. According to the shallow water-deep basin model (Hsü et al., 1973;Roveri et al., 2014a), evaporite precipitation was associated to a sea level drop in the range of 1500 meters, up to the almost complete desiccation of the Mediterranean. This occurred at the MSC's peak, between 5.60 and 5.55 Ma (Fig. 2), culminating in halite precipitation and marked by the incisions of deep canyons at the Mediterranean margins. Debate is still ongoing regarding the paleoenvironmental conditions of the final 'desiccation' phase. The deep desiccation theory has been recently challenged by new findings suggesting that the depositional facies association and the morphological features, including the canyons, may have been produced without a significant drop of the Mediterranean Sea level (Roveri el al., 2014b(Roveri el al., , 2016. The presence of marine fish during the latest Messinian powered the hypothesis that fully marine environments existed in several Italian areas well before the Pliocene flooding (Carnevale et al., 2006). At the same time, overspilling of Paratethys waters into the Mediterranean has been argued to explain the presence of brackish water fauna (ostracod and mollusk assemblages of Paratethys affinity) in the 'Lago Mare ' (5.55-5.33 Ma) (Suc et al., 1999;Gliozzi et al., 2002;Bertini, 2006;Roveri et al., 2008bRoveri et al., , 2008cStoica et al., 2016;Marzocchi et al., 2016). For the same time interval, palaeosalinity values in the adjacent Paratethys Sea (Fig. 1) were considered to have varied to a much lesser extent.

A C C E P T E D M A N U S C R I P T
5 New data from two locations in the circum-Black Sea , however, indicate strongly enhanced evaporitic conditions in the Black Sea area of Paratethys as well (Fig. 1). The high evaporative conditions in the Black Sea are expressed in very elevated hydrogen isotope values of biomarker molecules . At that time it is commonly thought that the Mediterranean experienced its major sea-level drawdown phase and a physical link to Paratethys is highly likely . However, the magnitude of hydrological changes and exchanges between Paratethys Gypsum. Andersen et al. (2001) estimated that the δD values of the source waters varied by ~100‰ (up to +66‰) over the ~50 kyr studied interval, because of changes in evaporation at low relative humidity. Despite rapid methodological developments in the field of compound specific δD analyses, this method has not been applied further to the Mediterranean MSC.
For the present study we use long chain n-alkanes (C 29 and C 31 ) with a distinct odd over even

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 6 predominance, as they are principally derived from higher plant waxes and thus reflect the terrestrial environment. We use their δD to reconstruct large scale hydrological changes on the continent adjacent to the Mediterranean Sea over the extent of~640 kyrs, covering the whole MSC interval. Additionally, δD values of long-chain alkenones, produced by haptophyte algae are used to reconstruct the δD of Mediterranean surface water during the MSC. Subsequently, these δD values are exploited to reconstruct changes in the mixing between freshwater/evaporation and inflow of sea water from other sources for the Mediterranean basin. The newly obtained data will be compared and integrated with compound specific δD results already available for the MSC interval from the Paratethys realm . Furthermore, these δD results will be discussed along the available, applied already, climate modelling published results focusing on the Miocene-Pliocene transition in the Mediterranean domain.

Stratigraphy of sampled interval
The sedimentary record of the MSC developed over an open marine unit including marls, sapropels and diatomites. The MSC can be separated into three main stages (Roveri et al., 2008a(Roveri et al., , 2008b(Roveri et al., and 2014a

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 7 can be traced up to deeper ones at the base of the RLG unit (Roveri et al., 2008a et al., 1995;Lugli et al., 1999;2010;Van Couvering et al., 2000;Bertini, 2006;Manzi et al., 2009;Manzi et al., 2012;Speranza et al., 2013).

Monte Tondo section -Primary Lower Gypsum unit (MSC stage 1)
The Monte Tondo gypsum quarry, located within the Vena del Gesso basin (along the western Romagna Apennines), contains all the 16 cycles of the PLG, deposited during stage 1 of the MSC (Fig. 2)

Realmonte salt mine -Halite and Resedimented Lower Gypsum unit (MSC stage 2)
The halite unit, underlain and overlain by clastic evaporite deposits, is included in the RLG unit deposited during stage 2 of the MSC (Roveri et al., 2008a and2014a

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 9 laminae and contains minor polyhalite and anhydrite. Unit D, up to 60 m thick, begins with a gray anhydrite-rich mudstone passing to an anhydrite laminate sequence followed by halite millimeter-to centimeter-thick layers intercalated with anhydrite laminae and decimeter-thick halite beds. Lugli et al. (1999) proposed that these lithologies reflect the shallowing and the desiccation of the evaporitic basin resulting from a possible combination of factors: (1) uplift of the basin floor by thrust activity, (2) simple evaporitic drawdown and (3) a basin-wide drop of the Mediterranean sea level. Manzi et al. (2012) suggest that the well-developed cyclicity characterizing the halite unit is related to annual climate variations and the deposition of the Sicilian Messinian halite could have lasted only a few thousands of years (Roveri et al., 2008).

Eraclea Minoa section -Upper Gypsum unit (stage 3)
The Eraclea Minoa section is located on the SW coast of Sicily and hosts the global boundary stratotype section and point for the Zanclean stage, marking the beginning of the Pliocene (Van Couvering et al., 2000). The section provides also one of the most complete sedimentary successions of the Upper Gypsum (stage 3; Upper Evaporites) (Manzi et al., 2009). We collected samples from 8 cycles covering the uppermost part of the Miocene and the lower part of the Pliocene (Fig. 2).The cyclic sedimentation starts with an alternation of marls and sandstones, followed by clastic and primary (cumulate facies) and ends with massive selenite (selenite crust, massive selenite separated by thin marl layers, domed selenite and reworked selenite on top of the domes) (Manzi et al., 2009). The section is laying on the RLG unit (Roveri et al., 2008) here consisting of a gypsum turbidite megabed including blocks of PLG in its basal chaotic part, followed by a 2-m thick cumulate gypsum layer suggested to be time equivalent of the halite in Sicily (Manzi et al., 2009). Cycle 1

A C C E P T E D M A N U S C R I P T
10 consists of banded selenite only. Cycle 2 contains thin marl layers, selenite crust and massive selenite. Cycle 3 consists of marls, followed by selenite crust and massive selenite. Cycles 4, 5, the first part of cycle 6 and 7 show the same basic cyclic pattern. Cycle 6 is divided into three parts, of which the last two consist mainly of marls, thin bedded sandstones and are considered as sub-cycles. The interval above cycle 4 contains typical Lago-Mare faunal assemblages (Melanopsis and ostracods; Manzi et al., 2009). Above the gypsum bed of cycle 7, a terrigenous interval, including marls and sandstone ('Arenazzolo' sandstone beds), is present; it consists of deltaic deposits that are also present in the lower part of the previous Upper Gypsum cycles (Manzi et al. 2009). The strata deposited during the subsequent Pliocene consist of an alternation of marls and sapropelitic shales known as Trubi formation (Van Couvering et al., 2000).

Lipid extraction, separation and analyses
In total 41 rock samples (12 from Monte Tondo, 2 from Realmonte mine and 27 from Eraclea Minoa), weighing between 8 and 60 g, were dried and thoroughly ground. Larger samples (i.e. 15-60 g) were extracted using a Soxhlet apparatus with a dichloromethane (DCM)methanol (7.5:1, v/v) mixture. Smaller samples (up to 15 g) were extracted by accelerated solvent extraction (ASE, Dionex 200) using a DCM-methanol (9:1, v/v) organic solvent mixture at 100°C and 1000 psi. All extracts were rotaryevaporated to near dryness and subsequently further dried under a nitrogen flow. The total lipid extracts were dried over an anhydrous Na 2 SO 4 column. Elemental sulfur was removed using activated copper in DCM.
Copper flakes were activated with 2M HCl and afterwards rinsed with MilliQ ultra-pure water, methanol and DCM, a treatment repeated up to three times when necessary. An aliquot

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 11 of the extract was separated using column chromatography with activated Al 2 O 3 as stationary phase by subsequent elution with n-hexane/DCM (9:1, v:v), n-hexane/DCM (1:1, v:v), and a mixture of DCM/ methanol (1:1, v:v) to obtain the apolar, ketone and polar fractions, respectively. n-Alkanes were isolated from the apolar fraction using urea-adduction. To this end, the apolar fraction was dissolved in 200 μl methanol/urea (~10%, H 2 NCONH 2 , Merck) solution. Subsequently, 200 μl acetone and 200 μl n-hexane were added to the solution, frozen (-20ºC) and dried under N 2 flow. Urea crystals were washed with n-hexane to remove the non-adductable branched and cyclic compounds and subsequently dissolved in a 500 μl methanol and 500 μl MilliQ ultra-pure water mixture. The n-alkanes were extracted from the solution using n-hexane. The urea-adduction procedure was repeated up to three times to eliminate non-adductable compounds as much as possible. Alkenones were obtained from the ketone fraction using urea adduction as well, using the same procedure as for the n-alkanes.
All fractions were measured using Gas Chromatography/Flame Ionization Detector (GC/FID) first. The n-alkanes and alkenones were identified based on mass spectra and their retention times using Gas Chromatography-Mass Spectrometry (GC-MS) on a Thermo-Finnigan Trace DSQ instrument. The fractions (dissolved in n-hexane) were injected oncolumn at 70 ºC (CP-Sil 5CB fused silica column (30 m × 0.31 mm i.d; film thickness 0.1 μm). The oven was set at constant pressure (100 kPa) and then programmed to increase to 130 °C at 20 °C min -1 , and then at 5 °C min -1 to 320 °C at which it was held isothermally for 10 min.

Compound-specific hydrogen isotope analyses
Compound-specific hydrogen isotope (δD) compositions of individual n-alkanes and alkenones were measured on the adducted fractions using a HP 6890N Gas Chromatograph

A C C E P T E D M A N U S C R I P T
12 (GC) coupled to a Thermo-Finningan Delta Plus XP Isotope Ratio Mass Spectrometer (IRMS). The fractions (dissolved in hexane) were injected on-column at 70 °C, the oven being programmed to increase to 130 °C at 20 °C min -1 , and then at 5 °C min -1 to 320 °C at which it was held isothermal for 10 min. The film thickness of the CP-Sil 5 column was 0.4 μm and a constant flow of He was used at 1.5 ml min −1 . Eluting compounds were pyrolyzed on-line in an empty ceramic tube heated at 1450 °C, which was pre-activated by a 5 min methane flow of 0.5 ml min −1 . H 3 + factors were determined daily on the isotope ratio mass spectrometer and were at any time < 5. H 2 gas with known isotopic composition was used as reference and a mixture of C 16 -C 32 n-alkanes with known isotopic composition (ranging from −42‰ to −256‰ vs. Vienna Standard Mean Ocean Water (V-SMOW)) was used to monitor the performance of the system (Schimmelman Mixture A and B, Biogeochemical Laboratories, Indiana University). A squalane standard was co-injected with every sample with an average value of −171±3‰, which compared favorably with its offline determined value of −168.9‰. Each fraction was measured between two and five times, depending on the amount of material available.

n-Alkane abundance
The apolar fractions contain a series of n-alkanes ranging from n-C 16 to n-C 34 , with the longchain (C 27 -C 31 ) n-alkanes having the highest abundances. These long-chain n-alkanes show a strong odd-over-even carbon number predominance (Table 1, Figs 3a, e and k). At some levels the contribution of the shorter chain n-alkanes is to some degree higher (Figs 3c and e) although the long-chain (C 27 -C 31 ) n-alkanes clearly dominate the distribution.

ACCEPTED MANUSCRIPT
There are marked differences between the n-alkanes contribution form the samples collected in Monte Tondo and Eraclea Minoa. These differences are in the ranges for the degree of oddity (CPI), expressed as following relation: CPI = (((A 25 +A 27 +A 29 +A 31 +A 33 )/(A 24 +A 26 +A 28 +A 30 +A 32 )) + ((A 25 +A 27 +A 29 +A 31 +A 33 )/( A 26 +A 28 +A 30 +A 32 +A 34 )))*0.5 where A represents the area under the chromatogram peak for individual n-alkanes. The CPI values vary from 3 to 7.9 in Monte Tondo samples while for Eraclea Minoa CPI vary from 1.7 to 3.7 (Table 1).
Although all 41 samples yielded enough lipids to identify the organic compounds (e.g. Fig. 3a and g), since the stable isotope analyses requires much more material to be injected, the extracted amounts were enough for δD analyses in 23 samples only.

n-Alkane δD ratios
The δD of the C 29 and C 31 n-alkanes (majorly produced by higher terrestrial plants) from Monte Tondo section range between −175‰ and −138‰ (Table 1) and show a strong correlation (R 2 = 0.92; Fig. 4), with no appreciable offset between the isotopic values between the compounds. In general, the δD values of the C 29 n-alkane are somewhat less negative than those of the C 31 n-alkane ( Fig. 5 and Table 1). From the analyzed Monte Tondo interval, four out of twelve samples contained organically bound sulphur (Sinninghe Damsté et al., 1995) to such a degree that they could not be measured on mass-spectrometers (GC-MS nor GC-IRMS).The extracts from the Realmonte samples contained insufficient nalkanes to allow for δD isotope measurements.
Only six out of the total twenty-seven extracted samples from Eraclea Minoa section contained sufficient long chain n-alkanes that could be analyzed for δD of the C 29 and C 31 n-

ACCEPTED MANUSCRIPT
and correlate (R 2 = 0.75; Fig. 4). As for Monte Tondo, the δD values of the C 29 n-alkane are somewhat less negative than those of the C 31 n-alkane ( Fig. 4 and Table 1).

Long chain ketones
The ketone fractions show the presence of long-chain unsaturated ethyl and methyl ketones (alkenones) (C 37 -C 39 ) in the Eraclea Minoa section and the Realmonte samples (Figs 3b, d and h). The alkenone distribution shows a remarkable dominance of the C 37 ketone, followed closely by the C 38 ketone and lower but nonetheless appreciable concentrations of the C 39 ketone. Both C 37 and C 38 ketones are dominated by the di-unsaturated components.
Alkenones were not detected in Monte Tondo ketone fractions.

Alkenones δD values
The stable hydrogen isotopic composition of the C 37 and C 38 alkenones (δD alkenone ) varies markedly between −203‰ to −125‰ (‰ V-SMOW) ( Table 2) and show a strong correlation between each other (R 2 = 0.81; Fig. 4). Throughout the record, the δD values of the C 37 and C 38 alkenones closely track each other, with the exception of one level in the basal part of Eraclea Minoa, at stratigraphic level of 4.5 m ( Table 2).
In the composite record of Monte Tondo and Eraclea Minoa the amplitude is 50‰, evolving from lower δD n-alkanes values for the PLG from the Northern Apennines increasing to significantly higher δD n-alkanes during the deposition of Upper Gypsum from Eraclea Minoa

Comparison with the n-alkanes δD values from present-day records
In the present-day situation, the measured δD C29n-alkanes values from existing lakes located closest to the Sicily section (Lago Grande and Lago Piccolo di Monticchio; Basilicata Region) record values between −169‰ and −180‰ (Sachse et al, 2004). These values are

A C C E P T E D M A N U S C R I P T
16 40‰ to 56‰ more depleted than the δD C29n-alkanes registered in the Upper Gypsum of Eraclea Minoa (−124‰ and −140‰) (Fig. 5). Seemingly, there are no differences between samples of evaporites and marls. Nevertheless, the data set needs to be significantly larger to have attributed a statistical meaning. However, the large offset between the present-day δD C29nalkanes values and the ones recorded during the deposition of the stage 3 evaporites indicates that conditions were significantly different from today.
To estimate the δD precip we applied a constant biosynthetic fractionation between source water and n-alkane of 157‰ (Sachse et al., 2006;Sessions et al., 1999). Additionally, a different evapo-transpiration enrichment effect was applied for δD precipitation calculations (  (Table 1). These values largely overlap with the δD precipitation of around -41‰ to -25‰ recorded today around Italy ( Fig. 1; IAEA, 2001).
The δD n-alkanes (i.e. δD precipitation ) values for the Upper Gypsum records persistently more δD enriched isotopic values while the values for the δD n-alkanes (i.e. δD precipitation ) of the PLG fluctuate on a significantly larger range (Fig. 5, Table 1

Alkenones and their δD isotopic compositions
Alkenones are long-chain ketones synthesized by unicellular haptophyte prymnesiophyte algae, common in the photic zone of the modern ocean (Marlowe et al., 1984;Volkman et al., 1980). Alkenones have been also reported from brackish and freshwater lakes from around

A C C E P T E D M A N U S C R I P T
19 the world (e.g. Volkman et al., 1980;Kristen et al., 2010). Changes in the relative abundance of alkenones with a different degree of unsaturation, expressed in the U k' 37 index, are commonly used to deduce past sea surface water temperature (Brassell et al., 1986;Prahl & Wakeham, 1987).
The occurrence of alkenones in the studied sections was unexpected because the sedimentary successions were deposited under exceptionally highly saline conditions.
Moreover, in the Realmonte and Eraclea Minoa successions, relative abundances of the C 37 , C 38 and C 39 alkenones (Fig. 3b, d and (Thiel et al., 1997) and both in DSDP core 380A from the central Black Sea and the Taman section from the Black Sea coast . Although high alkalinity lakes show a dominant C 37:4 alkenone, this compound is absent from the Mediterranean MSC record. The relative abundances of C 37 , C 38 and C 39 are constant throughout the record (Fig. 3b, d and h). Because of the unknown alkenone producer(s) we refrain from calculating temperatures based on the U k' 37 index.

Alkenones δD values
The δD composition of the alkenones reflects predominantly the δD of the water they live in (Engelbracht and Sachse, 2006;Schouten et al., 2005;Paul, 2002), although values are also influenced by salinity , growth rate  The results from Eraclea Minoa indicate that the hydrogen isotopic composition of the C 37 and C 38 alkenones had a much higher range, showing a variation of more than 75‰ (Table 2 and  and 3 (Upper Gypsum). Still, similarly to alkenone based SST reconstructions (Prahl and Wakeham, 1987), there are limitations when calculating δD water from δD alkenone , since e.g. the species on which the calibration is based on did not exist during the late Miocene (Emiliania huxleyi and Gephyrocapsa oceanica). Therefore, we here consider only relative changes in δD water (Fig. 5).

Comparison and integration of the δD data into the existing models
Evaluation of the role of the Mediterranean's freshwater fluxes in controlling both its environmental evolution and exchange through its gateways is in its early stages being hampered by inadequate rainfall datasets as well as by model-data mismatch on temporal as well as spatial scales (Flecker et al, 2015). Most model data cover the entire MSC event, without distinction between the three specific stages of the MSC. Even more, more complex models like isotope-enabled general circulation models (GCM) are absent for the MSC events of the Mediterranean. Therefore we discuss our δD data in relation to the existing models of hydroclimate analysis available for the Late Miocene of the Mediterranean.
Results from simulations on an atmosphere-only GCM indicate an increase in net precipitation during the Late Miocene, causing increased river runoff around three times greater than today as a consequence of increased input from North African rivers feeding the Eastern Mediterranean (Gladstone et al., 2007). Many more rivers were thought to have transported water from the south (Griffin, 1999) as a result of a stronger African summer monsoon (Marzocchi et al., 2015;Gladstone et al., 2007). However, the same model predicts a smaller net hydrologic budget (river discharge plus precipitation minus evaporation) than for present day. At precession scale, wetter periods in the Mediterranean region may also have resulted from enhanced wintertime storm track activity in the Atlantic and associated increased precipitation (Kutzbach et al., 2014). Our δD data from the stage 3 of the MSC detailed data acquisition, to be used as data-check in further modelling exercises of complex models like isotope-enabled general circulation models.

δD results on biomarkers from the Mediterranean MSC
The δD n-alkanes and δD alkenones results are in line with the presence of vast accumulations of evaporites, indicators for dry environmental conditions (Fig. 5, Tables 1 and 2). However, the δD composition of the individual biomarkers indicates that more extreme conditions were

A C C E P T E D M A N U S C R I P T
24 characterizing the environment during stage 3 in comparison to stage 1. Regardless the applied evapo-transpiration enrichment effect (humid of dry environment-), the δD n-alkanes results indicate that much dryer conditions than today were affecting the Mediterranean between 5.55 and 5.32 Ma (Fig, 5). Dryer conditions were also inferred by Bertini (2006), However, there are also levels δD n-alkanes values, typical for drier conditions (Fig. 5). Intense evaporation at low relative humidity was also suggested by Andersen et al., (2001). These authors calculated deuterium enriched isotope compositions of the Mediterranean waters of, sometimes reaching, +66‰ (typical for highly evaporative settings).
Recurrent connection to the Atlantic was frequently used to explain the quasi-marine 87 Sr/ 86 Sr values recorded during deposition of the PLG (e.g. Müller and Mueller, 1991;Flecker and Ellam, 1999;Lugli et al., 2010;Roveri et al., 2014a). Natalicchio et al. (2014) found that in the northernmost offshoot of the Mediterranean Basin, gypsum did not form just

Comparison between δD results on biomarkers from the Mediterranean and the
Paratethys during MSC

Mediterranean -Paratethys correlation
Overspilling of the Paratethys into the Mediterranean during the latest Messinian (Cita et al., 1978;Orszag-Sperber, 2006;Roveri et al., 2008) is acknowledged by the presence of a Paratethys type of biota found in the Lago Mare facies deposits (Suc et al., 1999;Gliozzi et al., 2002;Bertini, 2006;Roveri et al., 2008b;2008c;Grossi et al., 2008). However, the exact relationship between the Mediterranean and Paratethys during the MSC was for a long time

A C C E P T E D M A N U S C R I P T
26 hampered by the inadequate stratigraphic correlations and insufficiently robust age control on the Paratethys deposits. Integrated magnetobiostratigraphic studies performed in the past 10 years on the Paratethys Messinian sedimentary successions revealed that the entire MSC interval in the Paratethys is included in the Pontian (Vasiliev et al., 2004(Vasiliev et al., , 2011Krijgsman et al., 2010).

Changes marking the pre-onset of the MSC
The

Changes during stage 1
The δD n-alkanes results from Monte Tondo are largely suggesting a precipitation pathway similar, at times, to today. There are also levels where the δD C29n-alkanes values were up to −135‰ more enriched, probably induced by drier conditions (Fig. 5). These values are recognized in the branching and banded selenite facies that have been considered to mark the  (Fig. 6).

Changes during stage 3
The

Conclusions
The δD results on specific biomarkers from the composite record covering the entire MSC interval in the Mediterranean Sea are converging towards large environmental changes. These δD data in relation to those existing in adjacent Paratethys realm (Fig. 6) are indicating: 1) The δD n-alkanes recorded during MSC stage 1 indicate a δD precipitation similar to the presentday Mediterranean hydrologic regime ( Fig. 6 b and c). Only at some levels, the δD n-alkanes (i.e. δD precipitation ) are more enriched suggesting more arid/warm conditions or a proximity of the vapor source. However, in conjunction with the existing δD C22n-alkanes , significantly drier conditions were to be expected during stage 1. Proximity of the vapor source as the cause for the more enriched values cannot be excluded although, during stage 1, the Mediterranean is considered to have experienced on-off connections to the Atlantic, modulated by precession.
2) The absence of the alkenones for the entire PLG unit is most likely related to sulphurisation of the organic matter. Noteworthy is that, at the moment, there is no rock record from the deepest Mediterranean basin for the stage 1, a time when PLG was deposited in the marginal settings. Therefore, because of absence of offshore data, we  (Manzi et al., 2009;Roveri et al., 2014b). ii) Alternatively, since the relative

A C C E P T E D M A N U S C R I P T
30 contribution of the C 37 , C 38 and C 39 alkenones appears similar to the one of the alkenones extracted from samples originating from the Black Sea (part of the Paratethys), we could also speculate that the alkenone producers were common for the Mediterranean and  1 and 3). The δD precip is calculated for a 'wet' climate using Sachse et al. (2006) and for a 'dry' climate using Feakins and Sessions, (2010).
Average, standard deviation (STDEV) and standard error of the means (SEM) are listed; n.d denotes levels where the δ 13 C or δD could not determine.   were calculated by selecting yearly means in which isotope content have been measured at least in 75% of the precipitation for that year and at least over eight months (IAEA, 2001).

A C C E P T E D M A N U S C R I P T 43
The locations of Taman peninsula (TM) with Zheleznyi Rog section used in Vasiliev et al., (2013) and Deep Sea Drilling Project 42B (Hole 380A) used in Vasiliev et al., (2015) are represented by yellow triangles.

ACCEPTED MANUSCRIPT
A C C E P T E D M A N U S C R I P T 53