The high-K calc-alkaline to shoshonitic volcanism of Limnos, Greece: implications for the geodynamic evolution of the northern Aegean

Genetic models for the formation of K-rich magmas in subduction-related settings range from partial melting of subduction-affected mantle sources to melting of crustal rocks depending on the local tectonic framework. The Miocene high-K calc-alkaline to shoshonitic rocks of Limnos Island reflect the magmatic activity in the northern Aegean, which migrated southwards in response to trench retreat and the collision of continental terranes in the Hellenic subduction system. New whole rock and mineral data of basaltic andesites, dacites and monzonites from Limnos indicate that the magmas underwent fractional crystallization of olivine, clinopyroxene, amphibole, apatite, and Fe-Ti oxides at 1100 to 700 °C and 0.5 to 0.1 GPa without significant assimilation of crustal rocks during the magma evolution. The strong enrichment of large ion lithophile elements and light rare-earth elements relative to depleted heavy rare earth and high-field strength elements points towards a mantle source that has been extensively hybridized by subducted sedimentary material. New Sr–Nd-Pb isotope data reveal a distinct isotopic composition of the Limnos rocks with high 207Pb/204Pb at low 206Pb/204Pb and 143Nd/144Nd ratios that is likely related to the subduction of the continental crustal succession of the Apulian block which was subducted prior to the onset of magmatism on Limnos. Partial melting models assuming a hybridized mantle source suggest that the primary melts of Limnos formed by melting of a phlogopite pyroxenite at melting degrees of 5 to 10%. Compositional differences between high-K calc-alkaline and shoshonitic magmas are explained by variable melting degrees and varying amounts of sediment supply to the mantle. The magmatic and tectonic evolution of Limnos island is typical for the Oligocene and Miocene volcanic centres of the migrating western Aegean arc front.


Introduction
Magmas with highly variable compositions form along the Earth's subduction zones reflecting complex processes in the upper mantle of mixing mantle and crustal rocks, of element transport by hydrous material and of partial melting (Miyashiro 1974;McCulloch and Gamble 1991;Hawkesworth et al. 1993). Additionally, the composition of the magmas is frequently altered during the ascent through the lithosphere, particularly by reaction with the rocks of the continental crust (e.g. Davidson et al. 2005). Thus, many volcanic rocks along continental subduction zones are andesitic to rhyolitic yielding little evidence for primary mafic magmas from the mantle in their genesis. There is general consensus that most primary magmas at subduction zones are basaltic and form in the asthenospheric mantle wedge by hydrous melting with variable involvement of decompression melting (Kushiro 2007;Schmidt and Jagoutz 2017). Potassium-rich magmas are frequent in both continental and oceanic subduction zones, and magmas in many arcs range from medium-to high-K calc-alkaline to shoshonitic compositions (Wheller et al. 1987;Robert et al. 1992;Sun and Stern 2001). For example, the K-rich magmas in oceanic island arcs like the Marianas or the Sunda-Banda arcs are generally believed to form from the mantle wedge mainly reflecting the addition 1 3 73 Page 2 of 20 of fluids and partial melts from subducted sediments, continental crust and basaltic crust (e.g. Sun and Stern 2001;Elburg et al. 2004). Other models suggest the formation of K-rich magmas in continental arcs by melting of sediment diapirs or mélanges of sediment with hydrated peridotite in the hot region of the mantle wedge (Behn et al. 2011;Marschall and Schumacher 2012). Primitive magmas in some arcs have high-Mg andesitic compositions which could represent melts from mantle peridotite with high water content or mixtures of peridotite with sedimentary material (Kelemen 1995;Straub et al. 2008). Additionally, many K-rich magmas seem to form after subduction ceased and are thus believed to be 'post-collisional' due to melting of metasomatized lithospheric mantle or lower crustal rocks (e.g. Sekine and Wyllie 1983;Foley 1992;Condamine and Médard 2014). However, lithospheric mantle of the upper plate that is not part of the convecting mantle wedge has temperatures < 1100 °C (Perrin et al. 2018). Partial melting of peridotite with such low temperature requires either massive addition of volatiles or the re-heating by hot mantle material, for example, following extension and rifting of the lithosphere causing an ascent of hot asthenosphere.
The formation of Cenozoic magmas in the Aegean is closely related to the complex tectonic history with alternating periods of ocean subduction and collision/subduction of continental fragments from Eocene to present (Menant et al. 2016;Schaarschmidt et al. 2021a). The occurrence of widespread high-K to shoshonitic magmatism during upper Miocene makes it a suitable area to study the origin of K-rich magmas in subduction-related settings. Here, we provide new petrological and geochemical data on the high-K calc-alkaline to shoshonitic volcanic and plutonic rocks from Limnos island that were interpreted to have formed either during or after subduction. The data are used to assess the influence of fractional crystallization and crustal contamination during the magma evolution. We use trace element and new Sr-Nd-Pb isotope data to determine the source components of the primary melts and to develop a quantitative partial melting model for different mantle lithologies that can be applied to the genesis of K-rich magmas elsewhere.

Regional tectonic background
The Aegean subduction zone migrated southwards in the past 30 million years due to slab-rollback, that caused extension in the back-arc and arc region (Jolivet and Brun 2010). The Aegean Sea lies within a complex succession of thrusted terranes that have undergone periods of alternating stress regimes as a result of the ongoing convergence between Africa and Eurasia (Fig. 1a). During the Upper Cretaceous to Eocene, the north-eastward drifting continental blocks of Pelagonia and Apulia collided with Rhodopia and Serbo-Macedonia, which led to the amalgamation with Eurasia (Kydonakis et al. 2015). Its boundaries are marked by two suture zones that formed after the closure of two Neo-Tethyan ocean branches surrounding Pelagonia. The Vardar-Suture-Zone between Pelagonia and the Serbo-Macedonian Massif crosses the Aegean Sea south of Limnos and is most likely linked to the Izmir-Ankara Suture (Çelik et al. 2011). The almost parallel running Pindos suture zone extends southeast of the Dinarides into the South Aegean Sea and separates Apulia to the west from Pelagonia to the east (Fig. 1a). However, the exact locations and correlations of these suture zones are still debated (Marroni et al. 2014;Taymaz et al. 2007).
Widespread extension is believed to have commenced in the Rhodope massif in middle Eocene time and from then on progressed southwestwards (Brun et al. 2016). The extension was caused by the southward retreat of the Hellenic trench as a result of slab-rollback, induced by the subduction and accretion of continental blocks (Faccenna and Brun 2008). Paleomagnetic data revealed that the extension followed a clockwise rotation of the continental blocks opening the Aegean Sea (Brun and Sokoutis 2010;van Hinsbergen et al. 2005). Additionally, post-orogenic collapse and crust-mantle delamination may have played a role in the formation of the extensional structures, such as metamorphic core-complexes and sedimentary basins, that can be observed in the Aegean (Brun et al. 2016;Kydonakis et al. 2015). Slab retreat was accompanied by systematic southward migration of magmatic activity in the western Aegean (e.g. Jolivet et al. 2015) which was suggested to reflect consecutive ascent of mélange diapirs from the slab top (Schaarschmidt et al. 2021a). Alternatively, lithospheric delamination or slab detachment may have been the cause for melt generation causing, for example, calc-alkaline and shoshonitic volcanism in western Anatolia (e.g. Aldanmaz et al. 2000). The early to middle Miocene age of the western Anatolian volcanism correlates with the volcanic activity of Limnos, which could suggest a similar origin. Thus, Pe-Piper et al. (2009) proposed that upwelling of the asthenosphere beneath Limnos was a result of slab breakoff and mantle-derived magmas caused partial melting of the lower crust followed by mixing of the different melts. During the upper Miocene the NE-SW extension was further aided by dextral movements along large-scale strike-slip faults, such as the North Anatolian Fault Zone (NAFZ, Fig. 1a) (Armijo et al. 1999;Brun et al. 2016). The crustal thickness beneath the northern Aegean Sea is estimated at 25 to 30 km, but may have been up to 50 km thick prior to the two-stage back-arc extension in the Oligocene/Miocene and again in the Plio-/Pleistocene (Tirel et al. 2004;Sodoudi et al. 2006;Karabulut et al. 2019). Tirel et al. (2004) point out that the  Innocenti et al. (1994) and Pe-Piper et al. (2009) and locations of faults according to Pavlides et al. (2009) crust of the Aegean Sea shows a flat Moho similar to the Basin and Range probably reflecting lower crustal flow.

Geology of Limnos
The island of Limnos covers an area of ~ 400 km 2 in the northern Aegean Sea, southwest of the island of Samothraki (Fig. 1a). Limnos lies in the Thrace Basin and consists of sedimentary and volcanic rocks forming a rugged morphology in the west, whereas the eastern portion of Limnos is low with a smooth morphology (Fig. 1b). These two parts of the islands are divided by the Kondias-Kotsinas fault zone consisting of a series of NE-SW striking right-lateral strike-slip faults that may have also affected the emplacement of magmatic intrusions (Koukouvelas and Aydin 2002;Pavlides et al. 2009). The sediments are mainly sand-and siltstones with ages from Middle Eocene to Lower Miocene representing deep-marine turbidite deposits with occasional tuff layers overlain by continental conglomerates, marls, and mudrocks (Innocenti et al. 1994;Caracciolo et al. 2011). The continental sediments are overlain and intruded by lava flows and domes, subvolcanic rocks, ignimbrites, and volcaniclastic deposits. The oldest rocks of the basement comprise a succession of Middle Eocene to Early Oligocene flysch deposits representing a submarine fan in the fore-arc basin of the migrating subduction zone (Maravelis et al. 2016). From the Late Oligocene to the Middle Miocene three N-Sdirected compressional tectonic phases occurred during the deposition of the overlying sedimentary units causing folding of the topmost strata and the development of tensional structures (Tranos 2009). This compression affected the subduction dynamics and was probably due to the collision of the Apulian block with the Eurasian Plate which led to uplift, erosion of the folds, and a transition from marine to continental sediments prior to the onset of Early Miocene volcanism on Limnos (Erbil et al. 2021;Innocenti et al. 1994;Tranos 2009). Tectonic reconstructions of the Aegean subduction zone suggest that the Apulian continental block had been accreted and partially subducted by ~ 23 Ma, which was followed by subduction of oceanic lithosphere of the eastern Mediterranean (Ring et al. 2010;Jolivet et al. 2013). Metasedimentary rocks of the lowest tectonic unit of the Cyclades, the Basal Unit, show the peak of high-pressure metamorphism with 0.8 to 1.0 GPa and 350 to 400 °C and ages of 24 to 21 Ma indicating the subduction of thick layers of clastic and carbonate sediments in the Early Miocene (Ring et al. 2001).
Three major volcanic rock units were suggested to occur on Limnos island (Fig. 1b) in addition to a small-scale monzonite intrusion at Cape Fakos; the effusive Katalakon unit, the pyroclastic Romanou unit, and the Myrina unit consisting of the youngest lava domes and occasional flows (Innocenti et al. 1994). Potassium-Ar ages of the Limnos volcanic rocks indicate ages of 21 to 18 Ma (Innocenti et al. 1994;Pe-Piper et al. 2009). The oldest volcanic products on Limnos belong to the southern Katalakon unit (Fig. 1b), which consists of altered andesitic and dacitic lava domes and hosts the monzonitic intrusion at Cape Fakos. Towards the east, around the Gulf of Moudros, the Katalakon unit is overlain by pyroclastic deposits, dacites, and latites of the Romanou unit, which show the highest potassium contents. Based on age dates, Pe-Piper et al. (2009) suggested that the Romanou unit is older than the northern Katalakon unit. The Myrina unit is located in southwest Limnos, and separated from the northern Katalakon unit by the NW-SE striking Kaspakas fault (Pavlides et al. 2009). It comprises the island's youngest volcanic domes and lava flows with mainly dacitic compositions. The northern lavas along the coastline have been assigned to the Agios Ioannis subunit by Pe-Piper et al. (2009) due to their transitional compositions in relation to lavas of the two neighbouring Myrina and Katalakon units (Fig. 1b). Whereas Innocenti et al. (2009) suggested an origin of the magmas in the mantle above a subducting slab, Pe-Piper et al. (2009) proposed that the Limnos lavas represent mixtures of a mantle-derived basanitic melt and a partial melt of amphibolite in the lower crust.

Sampling and sample preparation
For this study, 50 magmatic rock samples were collected and studied at the GeoZentrum Nordbayern (Friedrich-Alexander Universität Erlangen-Nürnberg) with a focus on the geochemical whole rock major and trace element as well as radiogenic isotope composition. The samples mostly consist of lavas from domes and flows as well as from dykes. According to the geological map after Innocenti et al. (1994), the samples represent the three major volcanostratigraphic units and are distributed over the entire island (Fig. 1b). In addition, three monzonite samples from the southern Fakos intrusion were examined.

Analysis of mineral compositions
Polished thin sections of ten samples of both shoshonitic and high-K rocks from variable locations were petrographically examined and the minerals were analysed using the JEOL JXA-8200 Electron Probe Microanalyser at the GeoZentrum Nordbayern (GZN). For the measurement of core and rim compositions of plagioclase, amphibole and clinopyroxene an electron beam of 3 µm diameter with a current of 15 nA and an acceleration voltage of 15 kV was used. Magnetite and ilmenite grains were analysed using a 1 µm beam with a beam current of 20 nA and an acceleration voltage of 20 kV. The counting times and detection limits of the analysed elements are listed in Schaarschmidt et al. (2021b) and the results of all mineral analyses can be found in the electronic supplement of this work (Table S1). The mineral compositions were used to estimate the pressure-temperature conditions and the volatile content of the magmas during crystallization by applying different thermobarometric models (Table 1).

Whole rock geochemistry
Sample blocks of fresh rock material were ultrasonically cleaned in deionized water and dried for 24 h in a drying oven at 60 °C. After crushing of the blocks by a hydraulic press, an agate mill was used to pulverize the samples to grain sizes below 40 µm. The powder was dried for 24 h at 100 °C.
The contents of the major elements (> 0.1 wt.%) were determined by X-ray fluorescence (XRF) using an energy dispersive Spectro Xepos He X-ray fluorescence spectrometer at GZN. Samples were homogenized to melt discs using lithium-tetraborate (Li 2 B 4 O 7 ). Loss on ignition (LOI) was calculated by determining the weight difference of each sample before and after the ignition in a muffle furnace. Accuracy and precision of the measurements were determined by multiple measurements of the international rock standards BR and BE-N for basalts and GA and AC-E for granites (Govindaraju 1994; electronic supplement Table S1). Repeated analyses of the standards BR and BE-N gave an accuracy better than 9% for P 2 O 5 , better than 6% for Na 2 O and better than 3% for the remaining major elements. The precision of the analysis, expressed as the relative standard deviation (RSD, 2 s), was better than 3% for Na 2 O, better than 2% for MgO and P 2 O 5 , and better than 1% for all other major elements. For the granite standards GA and AC-E, the accuracy was better than 7% for Fe 2 O 3 and better than 5% for all other major elements, except for MgO and P 2 O 5 , which have too low absolute MgO and P 2 O 5 contents in the felsic standard material. The relative precision (2 s) was better than 1% for all major elements save MgO (12%) and P 2 O 5 (6%).
Sixteen samples were selected for trace element analyses based on their locations and their major element contents to cover a preferably wide range in SiO 2 and K 2 O. A detailed description of the preparation procedure is given in Schaarschmidt et al. (2021b) and is briefly described here. The trace element concentrations were determined on a Thermo-Fischer X-Series 2 Q-ICP-MS at GZN together with multiple measurements of the basalt standard BHVO-2 (GeoReM; Jochum et al. 2016). All trace elements had an accuracy better than 9% except for Zn (< 17%), Sb (< 46%), Cs (< 28%) and the longterm reproducibility of 2020 (2 s; n = 30) was better than 14% for all trace elements excluding Ni and Cu (< 20%) and Cs (< 40%) (electronic supplement Table S1). Due to the partial dissolution of zircon during digestion of rock powder, the Zr concentration determined by XRF Table 1 Temperature, pressure, water content and oxygen fugacity calculations based on different models for amphibole (amph), hornblende (hbl), clinopyroxene (cpx), magnetite/ilmenite (mt/ilm) and plagioclase (plg)  Andersen and Lindsley (1985), Lepage (2003), Powell and Powell (1977), Spencer and Lindsley (1981)  analysis was considered for all interpretations and diagrams. For Sr, Nd, and Pb isotope analysis 100-150 mg of rock powder was dissolved and processed following the procedure described in Schaarschmidt et al. (2021b). Lead, Sr and Nd was separated from the rock matrix using different ion exchange columns (Schaarschmidt et al. 2021b). Isotope measurements of Sr and Nd were carried out on a Thermo-Fischer Triton TIMS (thermal ionization mass spectrometer) at GZN. Strontium isotope measurements were corrected for instrumental mass fractionation assuming 88 Sr/ 86 Sr = 0.1194 and corrected for the contribution of 87 Rb to mass 87. Neodymium isotope data were corrected for mass fractionation using a 146 Nd/ 144 Nd ratio of 0.7219. Samarium interference on masses 144, 148, 150 was corrected by measuring 147 Sm. The Sr isotopic ratios have a maximum internal uncertainty of ± 0.000008 (2 s) and were normalized to the NBS987 standard with 87 Sr/ 86 Sr = 0.710250 and an average measured value of 0.710276 ± 0.000011 (2 s, n = 3). Likewise, the Nd isotopic ratios (2 s = ± 0.000005) were normalized to the Erlangen Nd standard with 143 Nd/ 144 Nd = 0.511541 and an average value yielding 0.511538 ± 0.000010 (2 s, n = 4). Lead isotope measurements were carried out on a Thermo Fisher Neptune multi-collector ICP-MS at GZN using a 207 Pb/ 204 Pb double spike to correct for instrumental mass fractionation. The Pb fraction was diluted with 2% HNO 3 to a concentration of approximately 20 ppb. One part of this solution was spiked in order to obtain a 208 Pb/ 204 Pb ratio of about 1. The double spike, with a 207 Pb/ 204 Pb ratio of 0.5, was calibrated against a solution of the NBS982 equal atom Pb standard. Spiked and unspiked sample solutions were introduced into the plasma via a Cetac Aridus desolvating nebulizer, and measured in static mode. Interference of 204 Hg on mass 204 was corrected by monitoring 202 Hg. An exponential fractionation correction was applied offline using the iterative method of Compston and Oversby (1969 The initial isotope ratios of Sr, Nd and Pb were calculated for each sample by assuming an emplacement age of 20 Ma. The results of whole rock major element, trace element and Sr-Nd-Pb isotope analyses can be found in the electronic supplement (Table S2).

Petrography
The Limnos lavas and the monzonite are characterized by a fine-to medium-grained porphyritic texture with a holoto hypocrystalline groundmass surrounding phenocrysts of plagioclase, quartz, amphibole, biotite and clinopyroxene. The colour of the groundmass varies between greenish light grey to dark brown. Some samples exhibit orientated phenocrysts . Plagioclase is the dominant phase in all samples and mostly has a euhedral to subhedral shape and frequently shows a zonation. Quartz is more abundant in samples of the Katalakon unit, whereas clinopyroxene phenocrysts only occur in rocks of the Romanou unit and the Fakos monzonite. The clinopyroxenes are of light brown to green colour and rarely show growth zoning. The samples contain accessory apatite, zircon, magnetite, ilmenite and titanite. The Fakos monzonite (LIM-14) further contains euhedral orientated K-feldspar phenocrysts. Lavas from Katalakon south and Romanou host enclaves of several cm size with a similar mineralogy, but smaller phenocryst size. Most samples went through minor degrees of hydrothermal alteration resulting in partial chloritization of pyroxene and sericitization of feldspar. Vesicles occur in several samples and are filled with calcite.

Mineral composition
Plagioclase is the most abundant mineral in all rocks from Limnos. The An content ranges from a maximum value of 54 (LIM-03) to the lowest value of 21 (LIM-14) with the majority of crystals with An contents between 30 and 50 (andesine). Plagioclase of one shoshonitic sample (LIM-42) is considerably more potassic (Or content 4 to 9) than crystals of the other samples. Within single plagioclase crystals, the An content decreases from the cores toward the rims in all samples and profiles across individual grains show normal zoning patterns (LIM-03, LIM-37, LIM-42) but also few irregularly zoned grains with more sodic cores in LIM-14. The An contents and the whole rock compositions of each sample were used for the plagioclase-liquid hygrometer by Waters and Lange (2015) to calculate the water content of the melt, ranging from 4.0 to 4.8 wt.% in shoshonitic samples and from 4.3 to 6.7 wt.% in high-K samples (Table 1).
Amphibole in samples LIM-14 and LIM-37 largely classifies as Mg-hornblende besides some crystals of tschermakitic compositions with higher Mg and Si contents in Page 7 of 20 73 sample LIM-37 (Leake et al. 1997). The thermobarometric model by Ridolfi et al. (2010) was applied to calculate the temperature, pressure, water content and oxygen fugacity of the samples (Table 1). The amphiboles yield crystallization conditions of 0.15 to 0.40 GPa at 900 to 960 °C for the high-K lavas and ~ 0.04 GPa at 760 °C for the monzonite (Fig. 2, Table 1). The calculated water contents differ slightly compared to the results of the plagioclase hygrometer, but also record the lowest H 2 O contents for the monzonite (Table 1). The calculated oxygen fugacity log fO 2 ranges from -12 (LIM-14) to -10 (LIM-32), which corresponds to a ΔFMQ of + 2 (relative to the fayalitemagnetite-quartz buffer). The geobarometer of Mutch et al. (2016) for granitic rocks depends only on the Al content of the amphiboles and yields slightly higher pressures for all three samples (Table 1).
Clinopyroxene occurring in two analysed samples (LIM-14 and LIM-42) has diopside-augite compositions with the monzonite (LIM-14) clinopyroxenes having slightly higher Ca contents. In the lack of glass analysis, the whole rock compositions of the samples were used for the clinopyroxene-liquid thermobarometer by Putirka et al. (2003). Clinopyroxene records crystallization temperatures around 1052 °C and 1130 °C and pressures of 0.24 GPa and 0.52 GPa for the monzonite (LIM-14) and the shoshonitic lava (LIM-42), respectively (Fig. 2, Table 1).
Magnetite-ilmenite pairs were used to estimate the crystallization temperature and the oxygen fugacity of the magmas. Based on their Fe content and V/Ti ratio only magmatic and no re-equilibrated magnetites were taken into account (Wen et al. 2017). The remaining mineral pairs provide approximations of the temperature and oxygen fugacity, which correspond to FMQ + 2 at 730 to 890 °C for the shoshonitic and high-K lavas. The values overlap with previous data from Methana (Schönhofen et al. 2020) and Nisyros (Seymour and Lalonde 1991) at the SAVA, but are higher than the values observed at Santorini (FMQ to FMQ + 1; Fabbro et al. 2013).

Major element composition
The SiO 2 contents of the Limnos samples range between 55 and 65 wt.%, i.e. between basaltic andesites and dacites, overlapping with data from previous studies (Innocenti et al. 1994;Pe-Piper et al. 2009). About half of the samples can be classified as high-K rocks with the other half belonging to the shoshonitic series (Fig. 3a, Peccerillo and Taylor 1976). All high-K samples follow the calc-alkaline trend of decreasing FeO T and MgO contents and increasing alkali contents with increasing SiO 2 (Fig. 3). The samples have LOI values between 1 and 7 wt.% indicating variable amounts of hydrous minerals and possibly carbonate alteration. For example, the dike sample LIM-07 is propylitized suggesting late hydrothermal alteration. However, most of the samples contain hydrous magmatic minerals biotite and amphibole that can probably account for up to 3 wt.% LOI. The lavas generally follow continuous major element trends of increasing K 2 O and decreasing CaO contents with increasing silica (Fig. 3) whereas alteration should cause less systematic variation. Thus, we conclude that the major element composition of samples with LOI < 3 wt.% is not significantly altered, with the possible exception of some samples with anomalous K 2 O, CaO, Al 2 O 3 or Na 2 O contents (LIM-07, LIM-10, . However, even samples with high LOI do not show anomalous Sr-Nd-Pb isotopic ratios supporting the reliability of the data. The rocks of the southern Katalakon unit, the adjacent Fakos monzonite and the Romanou unit mainly belong to the shoshonitic series, whereas the Katalakon north and Myrina units follow the high-K trend. Based on the published age record, this suggests a general evolution from shoshonitic to high-K calc-alkaline magmatism with time (Innocenti et al. 1994  Al-in-hornblende barometer after Mutch et al. (2016). Amph: amphibole thermobarometer after Ridolfi et al. (2010). Cpx: clinopyroxene thermobarometer after Putirka et al. (2003). Mgt-ilm: thermometer for magnetite-ilmenite pairs after Andersen and Lindsley (1985), Lepage (2003), Powell and Powell (1977) and Spencer and Lindsley (1981) 2013; Innocenti et al. 1981;Kirchenbaur et al. 2012;Nicholls 1971;Pe-Piper and Piper 1994). Overall, our results are consistent with the data of previous studies, but cover a larger range of SiO 2 especially at low SiO 2 contents. Magmas from Lesbos and Samothraki have a larger range of SiO 2 (50 to 72 wt.%) than rocks from Limnos, but follow the same trends. The shoshonitic rocks and monzonites exhibit slightly lower Al 2 O 3 and higher P 2 O 5 contents than the high-K calc-alkaline samples, while all other major element contents show no major differences between the two series.  Bailey et al. 2009;Büttner et al. 2005;Elburg et al. 2014Elburg et al. , 2018Elburg and Smet 2020;Fabbro et al. 2013;Innocenti et al. 1981;Kirchenbaur et al. 2012;Nicholls 1971;Pe-Piper and Piper 1994)

Trace element composition
The analysed samples from Limnos are compared to shoshonitic rocks from Lesbos (Pe-Piper et al. 2014) and Samothraki (Vlahou et al. 2006) in a MORB-normalized incompatible element diagram (Sun and McDonough 1989;Fig. 4). The magmas from Limnos, Lesbos and Samothraki all show similar element patterns characterized by strong enrichments of Pb and the large ion lithophile elements (LILE) Cs, Rb and Ba compared to a depletion in the high field strength elements (HFSE) Nb, Ta, Zr and Ti. Yet, the shoshonitic samples are slightly more enriched in both element groups compared to high-K samples (Fig. 4). Uranium and Th are greatly enriched in all rocks, but the shoshonitic rocks show a slightly lower Th/U ratio than the high-K calc-alkaline samples, whereas the monzonites exhibit the highest Th/U ratios. All samples are enriched in the light rare earth elements (LREE) and gradually decrease in their contents towards the heavy rare earth elements (HREE), resulting in flat HREE patterns (Fig. 4). The samples further show a small negative Eu anomaly. One shoshonitic sample (LIM-45) has a stronger enrichment in middle and heavy REE. The Limnos magmas show the same correlation of increasing La/Sm and decreasing Ba/Th with SiO 2 as observed throughout the Aegean (Fig. 5a,b). However, the magmas of Limnos, Lesbos and Samothraki reach much higher La/Sm ratios compared to the SAVA. Within the samples of Limnos, the monzonites and many of the shoshonitic lavas have higher intermediate REE (Sm to Tb) contents, lower Ba/Th and higher Th/Nd than the high-K samples, whereas Nb/Zr shows no consistent variation among the magma series (Fig. 5b-d).

Radiogenic isotope composition
The age-corrected Sr, Nd and Pb isotope ratios of the Limnos magmas exhibit some of the highest 87 Sr/ 86 Sr and lowest 143 Nd/ 144 Nd and 206 Pb/ 204 Pb ratios (Fig. 6a,b) observed in Aegean magmatic rocks and resemble the magmas from Lesbos (Pe-Piper et al. 2014) and Samothraki (Vlahou et al. 2006). In contrast to the younger volcanic rocks of the SAVA (Bailey et al. 2009;Büttner et al. 2005;Elburg et al. 2018;Elburg and Smet 2020;Pe-Piper and Piper 1994), the Limnos magmas have higher 207 Pb/ 204 Pb and 208 Pb/ 204 Pb at lower 206 Pb/ 204 Pb (Fig. 6c,d). The majority of the Limnos magmas shows a relatively narrow range of Sr, Nd and Pb isotope ratios, that does not show any variation with increasing SiO 2 content (Fig. 5e,f). However, three samples show a distinctly different isotope composition. One high-K andesite from the Romanou unit has lower Pb isotope ratios, but higher 143 Nd/ 144 Nd, while two dacites from the Myrina unit have lower 87 Sr/ 86 Sr and 207 Pb/ 204 Pb, but higher 143 Nd/ 144 Nd and 206 Pb/ 204 Pb ratios (Fig. 6).

Fractional crystallization processes affecting the Limnos magmas
The compositional trends of the magmatic rocks of Limnos resemble those of the SAVA lavas (Fig. 3) and probably reflect fractional crystallization, which has been suggested previously (Innocenti et al. 1994;Pe-Piper et al. 2009). The andesitic and dacitic rocks have similar radiogenic isotope and Nb/Zr ratios implying that they have closely Fig. 4 Multi-element diagram normalized to MORB (Sun and McDonough 1989 (Fig. 5d-f). Variable partial melting of crustal rocks and mixing with mantle-derived magmas would have caused significant changes in composition between andesites and dacites (e.g. Deschamps et al., 2017). The geothermometric calculations indicate that the samples record a large temperature range between 1130° and 700 °C (Fig. 2, Table 1) in agreement with conditions for extensive fractionation. The decrease in CaO and MgO with SiO 2 can be attributed to the fractionation of clinopyroxene, spinel and olivine at temperatures above 1000 °C and pressures of at least 0.5 GPa. Normal zoning patterns of plagioclase ranging from An54 to An20 in andesitic to dacitic lavas also reflect decreasing CaO contents during fractional crystallization. However, some of the variation of the bulk rock and mineral compositions may be due to magma mixing of closely related melts, which can also explain the abundance of subhedral plagioclase. The constant Al 2 O 3 , Sr and Eu contents with increasing SiO 2 indicate that plagioclase  is not continuously fractionating from the melts probably due to high water contents which suppresses the crystallization of plagioclase during the early stage of fractionation (Almeev et al. 2012). Amphibole fractionation is indicated by decreasing FeO, MgO, Sc and Y concentrations. Ratios of light REE and middle/heavy REE also correlate positively with the SiO 2 content (Fig. 5a), which can be explained by fractional crystallization of amphibole and/or clinopyroxene. Fe-Ti oxide fractionation causes continuously decreasing FeO and TiO 2 contents which is in accordance with the relatively oxidizing conditions (FMQ + 2) calculated for magnetite-ilmenite pairs and amphiboles at ~ 900 to 730 °C. Apatite fractionation is reflected by decreasing P 2 O 5 contents with increasing silica. Thus, the variation of major elements of the Limnos magmatic rocks is due to the crystallization and fractionation of olivine, clinopyroxene, amphibole, biotite, apatite and magnetite/ilmenite, whereas plagioclase fractionation is not observed during the early magma evolution. Mixing between related magmas within the two different magmatic series may have also contributed to the observed variation. Innocenti et al. (1994) suggested that the variations in K 2 O and other incompatible elements between high-K and shoshonitic lavas at Limnos reflect crystal fractionation processes at varying crustal depths from one parental magma. According to this model, fractionation of olivine, pyroxene, plagioclase, and magnetite at high pressures led to the formation of shoshonitic magmas, whereas fractionation of plagioclase, clinopyroxene, and magnetite at low pressures generated high-K melts.  2009, Büttner et al. 2005, Elburg et al. 2018, Elburg and Smet 2020, Pe-Piper and Piper 1994 and Nile fan sediments (Klaver et al. 2015) are shown. Initial isotope ratios for basement samples were calculated for an age of 20 Ma. Mixing lines are shown between depleted MORB mantle (DMM: Workman and Hart 2005) and Ios gneiss (Büttner et al. 2005) for clinopyroxene of the shoshonitic sample LIM-42 gave crystallization pressures of ~ 0.5 GPa which is similar to the pressures of the high-K samples (0.2 to 0.5 GPa for amphibole) and corresponds to a maximum crystallization depth of 18 km. The shoshonitic monzonite sample LIM-14 yields lower pressures of ~ 0.2 GPa for clinopyroxene and < 0.1 GPa for amphibole, which corresponds to a crystallization depth of ~ 5 km and an intrusion depth of ~ 2 km (Cas and Simmons 2018). Hence, these results do not support a model of systematically different pressures of crystallization for shoshonitic and high-K magmas. Rather, the lavas show variable pressures probably reflecting different magma stagnation levels and the monzonite represents a shoshonitic magma batch that crystallized in the upper crust. Additionally, the monzonites and many of the shoshonitic rocks have lower Ba/Th and higher Th/Nd and Nb/Yb compared to the high-K lavas (Fig. 5b,c) which implies different parental magmas for the magma series. High Th/Nd and Th/La ratios in magmatic rocks are often attributed to sediment subduction (Plank 1993(Plank , 2005, but the increasing Th/Nd and (La/ Sm) N probably reflect the fractionation of amphibole which has very low Th/Nd < 0.1 and low (La/Sm) N < 1 (Tiepolo et al. 2000;Schönhofen et al. 2020). An increasing Th/Nd by fractional crystallization is also in agreement with the constant 143 Nd/ 144 Nd (Fig. 5c), whereas assimilation would lead to changing Nd isotope ratios. We conclude that the shoshonitic and the high-K calc-alkaline rocks on Limnos have different parental magmas causing different crystal fractionation processes and liquid lines of descent.

Potential crustal assimilation in the Limnos magmas
The formation of shoshonitic magmas was suggested to be due to assimilation processes within the crust (e.g. Feeley and Cosca 2003), or to re-melting of lower crustal metabasalts by ascending basanitic magmas (Pe-Piper et al. 2009). We do not observe metamorphic xenoliths that would record assimilation of crustal rocks during the magma ascent. Metabasaltic rocks of the Cyclades, e.g. Samos, Paros and Naxos (Stouraiti et al. 2017), have too low Sr and Pb and too high Nd isotope ratios in order to be possible contaminants of the Limnos magmas. The metamorphic units of the Serbo-Macedonian massif have too high 143 Nd/ 144 Nd and 206 Pb/ 204 Pb (Bonev et al. 2012;Frei 1995) to represent a possible mixing end-member for the Limnos lavas (Fig. 6b). However, a sedimentary component is required to achieve the high incompatible element concentrations, the high Th/ Nd and (La/Sm) N , and high Pb-isotope ratios in the Limnos lavas (Spandler et al. 2007;Campbell et al. 2014). The constant Sr, Nd and Pb isotope ratios of the majority of Limnos rocks, including those with LOI > 5%, with increasing SiO 2 contents (Fig. 5e,f) do not indicate assimilation of continental crustal material which typically has 87 Sr/ 86 Sr > 0.710 (e.g. Büttner et al. 2005;Stouraiti et al. 2010;Bonev et al. 2012). Additionally, the Th/Nd and (La/Sm) N of the Limnos samples are also constant with variable SiO 2 (Fig. 5c,d), but much higher than MORB (e.g. Th/Nd > 0.4 compared to 0.02 in N-MORB, Hofmann 1988), suggesting the contribution of partial melts from sediments to the primitive magmas. We conclude that the Limnos magmas are affected by a partial melt from a sedimentary end-member rather than by one from an amphibolitic source, but that this sediment component caused enrichment of the Limnos parental melts in the mantle rather than in the crust.

Origin of the Limnos parental melts
All magmatic rocks of Limnos are characterized by high Cs, Rb and Ba, while Nb and Ta show negative anomalies compared to N-MORB (Fig. 4), which is typical for a magma source that has been enriched by subduction components,  (Table 1) suggest a hydrous mantle origin of the parental melts (e.g. Arnaud et al. 1992). The enrichment of the mantle beneath Limnos causing high ratios of (La/Sm) N (> 4.5), Ba/Th (< 100), Th/Nb (~ 2.4), and 87 Sr/ 86 Sr require an influence of subducted sediments (Global Subducting Sediment: e.g. La/Sm ~ 5.0, Plank and Langmuir 1998) rather than fluids released from altered oceanic crust (Ba/Th ~ 600 to 2300 (< 1000 °C), Carter et al. 2015;Elliott 2003). Due to the complex history of the Aegean subduction system, it is challenging to identify the subducted units that affected the Limnos magmas at their time of formation. The magmatic activity on Limnos at 21 to 18 Ma (Pe-Piper et al. 2009) directly follows the partial accretion and subduction of the Apulian continental block after the closure of the Pindos ocean, as recorded by high-P metamorphic ages of 25 to 20 Ma for metamorphosed units of the Tripolitza and Ionian Blocks (Ring and Layer 2003;Ring et al. 2010 (Fig. 6c, dark arrow). We suggest that this variation of the Nd and Pb isotopes reflects the change from subduction of the Pindos oceanic crust overlain by pelagic or flysch sediments to subduction of the Apulian continental crust with Variscan basement overlain by Mesozoic-Cenozoic limestones, evaporites, pelagic sediments and flysch sediments (Sotiropoulos et al. 2003; (Büttner et al. 2005).
Since the Ios gneiss underlies the Cycladic Blueschist Unit (Thomson et al. 2009) we speculate that it might resemble the subducted sediments of Apulia or the Variscan basement of which they were derived. Calculated mixing lines of average depleted mantle (i.e. DMM: depleted MORB mantle; Workman and Hart, 2005) with bulk Ios gneiss can roughly reproduce the Sr, Nd, and maximum Pb isotope ratios of the Limnos magmas (Fig. 6). The Pb isotopes require mixing of 1 to 2% Ios gneiss (IOS 3, Fig. 6c,d), whereas the Nd isotopes require mixing of 4 to 6% Ios gneiss (IOS 2, Büttner et al. 2005; Fig. 6b). However, the curvature of the mixing lines is also strongly affected by the Pb, Sr and Nd fractionation during the slab-to-mantle transfer, e.g. during sediment melting (e.g. Elliott 2003;Spandler et al. 2007). Thus, we suggest that the Ios gneiss represents a suitable end-member for the metasediments of the Apulian block that were subducted and caused the partial melting of the Limnos magmas in the mantle. The Pb isotopes of Limnos rocks as well as those from Lesbos and Maronia-Leptokaria trend towards low 206 Pb/ 204 Pb ratios (Fig. 6c, small arrows). Decreasing 206 Pb/ 204 Pb ratios correlate with decreasing K/La, Rb/Ba, Nb/La (Fig. 7a-c) and P/Nd ratios within the Limnos samples. Especially the younger high-K lavas of the Myrina and northern Katalakon units show lower Rb and Nb concentrations and low 206 Pb/ 204 Pb ratios. The decreasing Nb/La ratios (Fig. 7c) and Nb concentrations at decreasing 206 Pb/ 204 Pb ratios exclude the influence of an enriched mantle component. It rather points towards a crustal component with low Rb, Th, K, Zr, and Nb concentrations (Fig. 7) that affected the source of the high-K Limnos magmas. Unradiogenic Pb and Hf isotopes have been observed for coarse-grained sediments at the African shelf (Klaver et al. 2015) or in the forearc of the Lesser Antilles (Carpentier et al. 2009) due to higher portions of detrital zircon. The Quaternary Nile fan sediments from the African shelf (Klaver et al. 2015) have lower K/La ratios (~ 130), LILE and LREE contents than the Limnos magmas (K/La ~ 500), which could explain some but not all of the decreasing element concentrations (Fig. 7) Fig. 5), Cycladic basement rocks (Büttner et al. 2005;Stouraiti et al. 2017) and Nile fan sediments (Klaver et al. 2015) are shown for comparison Therefore the K/La variation in the Limnos magmas may be explained by a mixture of clastic sediments, such as arkoses and pelites (high K/La, low 143 Nd/ 144 Nd), and more limestone-rich sediments (low K/La, higher 143 Nd/ 144 Nd) derived from the subducted Apulian slab. We conclude that subduction of the Apulian sedimentary sequence and its interaction with the mantle wedge controlled the incompatible element composition and the Sr, Nd, and Pb isotope composition of the parental Limnos magmas. Interaction of mantle with slab-derived sediments or partial melts from the sediments generated a heterogenous source that gave rise to the high-K calc-alkaline and shoshonitic magmas.

Melting model for the Limnos parental melts
Experiments of Sekine and Wyllie (1983) simulated the reaction of granitic melt derived from sediments with mantle peridotite under P-T conditions reached in subduction zones. Their results suggest that hybrid reaction zones form in the mantle wedge that are characterized by orthopyroxene-and clinopyroxene-rich veins in the absence of olivine. Garnet, quartz and phlogopite might form as well, the latter being an important supplier of potassium to partial melts (Sekine and Wyllie 1983). More recent reaction experiments of sediment melt and dunite confirmed the formation of phlogopite-pyroxenites and K-rich magmas under P-T conditions of a subduction zone (1 GPa, < 1000 °C; Förster et al. 2019). Geochemical characteristics of western Anatolian volcanics resemble those of Limnos and were interpreted to be generated by melting of phlogopite-rich veins in subduction-metasomatized lithospheric mantle (Dilek and Altunkaynak 2010). Condamine and Médard (2014) also emphasized the importance of phlogopite for the generation of potassic magmas. The presence of residual phlogopite in the mantle is supported by low Ba/Rb, K/La, and high Rb/Sr ratios (Furman and Graham 1999) which we observe in the Limnos lavas (Fig. 7). Low CaO/Al 2 O 3 (~ 0.29) and low contents of CaO (~ 4.5 wt.%) and Sc (~ 14 µg/g) in the Limnos samples can also be related to a phlogopite-bearing mantle source (Miller et al. 1999). Based on these findings we suggest that the source rock of the Limnos primary magmas was dominated by orthopyroxene, clinopyroxene and phlogopite. Melt contributions from surrounding peridotite can be considered relatively small, because of the much lower solidus temperature of pyroxenite at pressures < 1 GPa and with little involvement of released volatiles (Lambart et al. 2016). We use REE modelling to test a potential melting regime of the Limnos magmas by comparing variable mixing endmembers and source mineralogies (Fig. 8). The composition of the hybridized mantle source for the melt modelling was determined by mixing DMM (Workman and Hart 2005) with up to 20% of sedimentary material (Fig. 8, solid lines). As sedimentary mixing endmember we use either bulk Ios gneiss (Büttner et al. 2005) or a sediment melt derived from melting experiments of a metapelitic starting material (Hermann and Rubatto 2009; run C-1578 at 900 °C, 3.5 GPa and 6% H 2 O added). The resulting REE compositions were used for modelling non-modal batch melting at variable degrees of melting (Fig. 8, dashed lines). The partition coefficients used for the melting model are presented in Table 2. Comparing the REE ratios of the intermediate Limnos magmas (< 60 wt.% SiO 2 ) to the modelled compositions allows us to estimate the degree of partial melting and the influence of different source rock mineralogies (Fig. 8). Condamine and Médard (2014) suggested the following melt reaction for a phlogopite-lherzolite above 1075 °C: 0.49phl + 0.56 opx + 0.47 cpx + 0.05 sp = 0.58 ol + 1.00 melt (phl = phlogopite; opx = orthopyroxene; cpx = clinopyroxene; sp = spinel; ol = olivine).
Since olivine is consumed during the reaction between sediment melt and peridotite, the modified melt reaction for an olivine-free phlogopite-pyroxenite would be: 0.31phl + 0.36 opx + 0.30 cpx + 0.03 sp = 1.00 melt. Small amounts of spinel (~ 1%) are negligible for the modelling due to their low distribution coefficients but have been included nevertheless as spinel is likely to occur as a minor or accessory mineral in mantle rocks at moderate pressures (~ 1 GPa). The REE concentrations of the Limnos magmas can be modelled by assuming the following composition for the hybridized mantle rock: 5% phl, 20% cpx, 74% opx, 1% sp (Fig. 8a). According to this model, the samples of Limnos require about 5 to 10% partial melting after mixing of DMM with either 10 to 20% of bulk Ios gneiss (Fig. 8a) or 2 to 5% of experimental sediment melt (Fig. 8b). The melting behaviour of this phlogopitepyroxenite is compared to other typical mantle lithologies in Fig. 8b-d. Melting of spinel peridotite (Fig. 8c) shows similar melting curves to the phlogopite pyroxenite ( Fig. 8a) but requires pretty low melting degrees < 4%. A garnet peridotite source may provide suitable Sm/Yb ratios even at low amounts of sedimentary material in the mixture (< 5%) but fails to reproduce the high La/Sm ratios of the Limnos magmas (Fig. 8d). We conclude that the REE contents of Limnos can be explained by 5 to 10% partial melting of a pyroxenite containing 5% phlogopite.
Most of the intermediate shoshonitic magmas of Limnos (< 60 wt.% SiO 2 ) differ from the high-K samples in having slightly higher middle REE concentrations and Sm/Yb ratios. A higher amount of sedimentary material could lead to a higher amount of phlogopite involved during partial melting, which at the same time increased the K content in the melt. Higher Nb/Yb, Th/Nd, Sm/Yb ratios and higher concentrations of Cs, Rb, Zr and U in most of the shoshonitic compared to the high-K magmas (Fig. 5) indicate that both, a variable input of sedimentary material and variable melting degrees controlled the formation of shoshonitic vs. high-K magmas. This is in accordance with models from the Maronia-Leptokaria high-K calc-alkaline to shoshonitic rocks where it is believed that variable source enrichment by subducted sediments caused lower water contents and lower degrees of melting at higher temperatures for shoshonitic relative to high-K calc-alkaline magmas (Perkins et al. 2018;Schaarschmidt et al. 2021b). This agrees with lower water contents estimated in the Limnos shoshonitic compared to the high-K lavas (Table 1) (Castillo 2012). This new model is in agreement with the massive subduction of sedimentary material at the time of magmatic activity on Limnos.

Tectonic framework of the Limnos magmatic system
The formation of the Miocene Limnos magmas has been related on one hand to the subduction at the southwards    (Innocenti et al. 1994), and on the other hand to post-orogenic slab detachment and asthenospheric upwelling (Pe-Piper et al. 2009). Thus, the tectonic situation of the Limnos region at the time of magmatic activity needs to be considered to understand the cause of partial melting. Limnos largely consists of Eocene to Oligocene clastic flysch sediments representing deposits of an accretionary wedge in the fore-arc region of the Oligocene magmatic arc situated in the Rhodopes (Görür and Okay 1996;Maravelis et al. 2016). The Limnos volcanic rocks largely rest on marine clastic sediments of the Late Oligocene to Early Miocene Ifestia Unit that show an erosional top and are overlain by the continental Therma Unit implying rapid uplift (Innocenti et al. 2009). An interpretation of the fault systems on Limnos indicates that this uplift is probably due to Middle Oligocene to Early Miocene collision between Apulia and Eurasia (Innocenti et al. 1994;Tranos 2009). Thus, the setting of the Limnos magmatic activity in the Oligocene closely resembled that of the presently active SAVA with rapid vertical movements due to changes in the dynamics of the fore-arc and arc region of the migrating subduction zone. Tectonic reconstructions also suggest that the active volcanic arc was close to the Limnos region 23 million years ago (Faccenna et al. 2014). Seismic data show an unusually thick lithosphere of possibly 150 km beneath the northern Aegean (Sodoudi et al. 2006), i.e. there is no indication of asthenospheric upwelling and lithospheric thinning beneath Limnos. Furthermore, the isotopic and trace element composition of the Limnos magmas reveals melting of a heterogeneous, hydrous mantle source that was affected by slab-derived crustal components. We conclude that the magmatism of Limnos reflects partial melting above the subducting slab and its southward-directed rollback that caused migration of the magmatic activity (Innocenti et al. 1994;Schaarschmidt et al. 2021a).
The tectonic models of the Aegean subduction zone suggest accretion and partial subduction of continental blocks belonging to the Apulian block during Oligocene and a change to subduction of the oceanic lithosphere of the eastern Mediterranean ~ 23 Ma ago (Ring et al. 2010;Jolivet et al. 2013). Thus, at the time of magma formation beneath Limnos between 22 and 18 Ma, significant volumes of continental crustal material of the Apulia block were already subducted and have likely affected the mantle wedge causing the extreme isotopic compositions of Upper Miocene Aegean magmas. The processes at the Oligocene-Miocene North Aegean arc may have resembled those occurring at the Banda Arc where parts of the Australian continental lithosphere are subducted causing extreme element enrichments in arc magmas (e.g. Elburg et al. 2005), whereas other lithospheric parts are accreted (Harris 2011). Additionally, the subduction of the continental lithospheric blocks causes uplift in the Banda Arc (Harris 2011) which is consistent with the uplift observed in the sediment succession of Limnos prior to and during the magmatic activity (Innocenti et al. 1994). We conclude that the magmatic and tectonic evolution of Limnos can be explained by the subduction of continental material of the Apulian block and we present the first comprehensive model for the geologic processes in this part of the northern Aegean.

Conclusions
The high-K calc-alkaline to shoshonitic trachyandesites and the Fakos monzonite are genetically related and underwent fractional crystallization at temperatures of 1100 to 700 °C and pressures around 0.5 to 0.1 GPa with no detectable crustal contamination. Mixing of different related magma batches can have also affected the composition of the lavas. The radiogenic isotope and incompatible element compositions of the Limnos magmatic rocks indicate the input of a continental sediment component to the depleted mantle wedge, and this component changed from that previously observed in older rocks from the northern Aegean. We suggest that Mesozoic-Cenozoic sediments of the Apulian block contributed to the source and controlled the trace element and isotope signature of the parental magmas, in accordance with upper Miocene ages of the high-pressure metamorphism of the respective units. This highlights the change of sedimentary input into the Aegean subduction zone between 30 and 20 million years ago. The subducted sedimentary component interacted with the mantle rock to form enriched phlogopite-pyroxenite veins at the expense of olivine. Partial melting of this phlogopite-pyroxenite source at degrees of 5 to 10% and varying sediment input into the mantle cause variations between high-K and shoshonitic rocks on Limnos. The magmatic rocks of Limnos indicate that both the melting processes in the mantle wedge and the complex variation of subducted oceanic and continental units in the migrating Aegean arc caused considerable changes in magma composition.