Isotopic evolution of prehistoric magma sources of Mt. Etna, Sicily: Insights from the Valle Del Bove

Mount Etna in NE Sicily occupies an unusual tectonic position in the convergence zone between the African and Eurasian plates, near the Quaternary subduction-related Aeolian arc and above the down-going Ionian oceanic slab. Magmatic evolution broadly involves a transition from an early tholeiitic phase (~ 500 ka) to the current alkaline phase. Most geochemical investigations have focussed on either historic (> 130-years old) or recent (< 130-years old) eruptions of Mt. Etna or on the ancient basal lavas (ca. 500 ka). In this study, we have analysed and modelled the petrogenesis of alkalic lavas from the southern wall of the Valle del Bove, which represent a time span of Mt. Etna’s prehistoric magmatic activity from ~ 85 to ~ 4 ka. They exhibit geochemical variations that distinguish them as six separate lithostratigraphic and volcanic units. Isotopic data (143Nd/144Nd = 0.51283–0.51291; 87Sr/86Sr = 0.70332–0.70363; 176Hf/177Hf = 0.28288–0.28298; 206Pb/204Pb = 19.76–20.03) indicate changes in the magma source during the ~ 80 kyr of activity that do not follow the previously observed temporal trend. The oldest analysed Valle del Bove unit (Salifizio-1) erupted basaltic trachyandesites with variations in 143Nd/144Nd and 87Sr/86Sr ratios indicating a magma source remarkably similar to that of recent Etna eruptions, while four of the five subsequent units have isotopic compositions resembling those of historic Etna magmas. All five magma batches are considered to be derived from melting of a mixture of spinel lherzolite and pyroxenite (± garnet). In contrast, the sixth unit, the main Piano Provenzana formation (~ 42–30 ka), includes the most evolved trachyandesitic lavas (58–62 wt% SiO2) and exhibits notably lower 176Hf/177Hf, 143Nd/144Nd, and 206Pb/204Pb ratios than the other prehistoric Valle del Bove units. This isotopic signature has not yet been observed in any other samples from Mt. Etna and we suggest that the parental melts of the trachyandesites were derived predominantly from ancient pyroxenite in the mantle source of Etna.


Introduction
Mount Etna, Europe's largest active volcano, is located on the Mediterranean island of Sicily ( Fig. 1) and lies in a complex tectonic setting near the convergence of the African and Eurasian plates. Eruptions of submarine and subaerial tholeiitic basalts between ~ 500 and 300 ka were gradually replaced by transitional to alkaline volcanism after ~ 250 ka via a succession of eruptive centres; this activity continues to the present day (McGuire 1982;Chester et al. 1985;Kieffer and Tanguy 1993;Scarth and Tanguy 2001;Patanè et al. 2006;Branca et al. 2008Branca et al. , 2011b. The magmatic evolution of Mt. Etna has been the subject of intense geochemical and isotopic research (e.g., Armienti et al. 1989;Marty et al. 1994;Tonarini et al. 1995Tonarini et al. , 2001D'Orazio et al. 1997;Tanguy et al. 1997;Gasperini et al. 2002;Viccaro and Cristofolini 2008; Communicated by Dante Canil.  Chester et al. (1985). c Map of the southern wall of the Valle del Bove, adapted from the Geological Map of Etna Volcano (Branca et al. 2011b). Widths of sections A, D, E, and C are exaggerated for clarity. All samples analysed in this study were collected from one of the four sections labelled A, D, E and C. d Lithology of the six volcanic units as exposed in the four sections A, D, E and C samples included here are < ~ 15-kyr old (D'Orazio et al. 1997), whereas the oldest historic lavas are from the eighteenth century CE (Viccaro and Cristofolini 2008). Nevertheless, some basal tholeiites (here termed "ancient Etna") and prehistoric (i.e., < ~ 250 ka) alkaline lavas from the eastern flank (Carter and Civetta 1977;Marty et al. 1994;D'Orazio et al. 1997;Tanguy et al. 1997;Tonarini et al. 2001;Corsaro et al. 2002) have been previously analysed, as have some of the older Plio-Pleistocene-age Iblean Plateau lavas ( Fig. 1; Carter and Civetta 1977;Tonarini et al. 1996;Trua et al. 1998;Miller et al. 2017).
Several authors have discussed the isotopic evolution of the magma sources beneath Sicily and Mt. Etna (Marty et al. 1994;Tonarini et al. 2001;Corsaro and Pompilio 2004;Branca et al. 2008;Viccaro and Cristofolini 2008). Strontium, Nd, and Hf isotopic compositions of eruptive products from the Iblean Plateau and Mt. Etna show a broad temporal evolution from DMM-like (depleted MORB [mid-oceanridge basalt] mantle) for the Iblean Plateau toward BSE (bulk silicate Earth) for recent (< 130-year-old) eruptions. This has been interpreted by some authors as evidence for subduction of Ionian oceanic lithosphere beneath Mt. Etna and modification of the sub-Etna mantle by subductionderived fluids (Gvirtzman and Nur 1999). Doglioni et al. (2001) and Gasperini et al. (2002) considered that differential slab rollback between the Sicilian and Ionian segments of the Apennines slab may have opened a slab window that has enabled upwelling and decompression partial melting of underlying asthenospheric mantle. Schiano et al. (2001) also suggested that subduction-derived fluids might be metasomatising the upwelling asthenospheric mantle beneath Mt. Etna. In contrast, Kamenetsky et al. (2007), Viccaro and Cristofolini (2008), Nicotra et al. (2013), Correale et al. (2014), Miller et al. (2017), and Viccaro and Zuccarello (2017) considered Etna's isotopic variation to be the result of melting of a heterogeneous mixture of mantle spinel lherzolites and pyroxenites.
In this paper we investigate prehistoric lavas from Mt. Etna that erupted during a poorly known period in the evolution of the volcano between ~ 85 and ~ 4 ka. The lavas are exposed in the southern wall of the Valle del Bove (latitude 15° 00′ 53″-15° 01′ 42″; longitude 37° 42′ 50″-37° 42′ 32″) and were collected in 1988 from four separate near-vertical sections (Fig. 1). They consist of several moderately evolved, geochemically distinct lava groups that correlate stratigraphically across the sections (Spence and Downes 2011). These lavas were previously considered to be products of a single eruptive centre referred to as "Vavalaci" (Lo Giudice 1970;McGuire 1982;Guest et al. 1984). They are now, however, reinterpreted as derived from several discrete volcanic centres dating from ~ 85 to ~ 4 ka (Calvari et al. 1994;De Beni et al. 2011;Branca et al. 2011a). Thus, they represent an ~ 80 kyr time span between the older magmatism of Mt. Etna and the present day, providing a link between the earlier eruptive phases and historic eruptions.
In this study, we provide new radiogenic isotope data (Sr, Nd, Pb, and Hf) for these prehistoric Valle del Bove units to investigate whether the broad trends in magma source evolution from ancient Etna eruptions to the modern day continued systematically throughout the ~ 80 kyr period that they cover. We also investigate whether variations in isotopic compositions are the result of melting of subducted Ionian oceanic lithosphere or rather a consequence of melting of ambient heterogeneous mantle.

Background and sampling
Mount Etna is located near the convergence of the African and Eurasian tectonic plates (Fig. 1a). North of Mt. Etna, the Quaternary Aeolian volcanic arc formed by northwards subduction of the Ionian oceanic plate. The Iblean Plateau, situated ~ 100 km south of Etna (Fig. 1a), represents the relatively undeformed foreland that was uplifted and underwent discontinuous intraplate volcanism during the Tertiary and Quaternary (Tonarini et al. 1996;Trua et al. 1998).
All samples for this study were collected from four stratigraphic sections in the southern wall of the Valle del Bove (Spence and Downes 2011). The sections from east to west are: Section A-Valle del Tripodo; Section D-west of Serra dell'Acqua; Section E-Serra Pirciata; and Section C-Serra Vavalaci (Fig. 1c). The stratigraphy, field relations, and petrography of the lavas can be inferred from the recent geological map of Etna by Branca et al. (2011b). In combination with the 40 Ar/ 39 Ar ages of De Beni et al. (2011) (Table 1), they comprise six discrete lithological units (Fig. 1d). From oldest to youngest, these are: (1) Salifizio-1; (2) Salifizio-2; (3) Cuvigghiuni; (4/5) the Valle del Tripodo member of the Piano Provenzana formation (Tripodo member) and Piano Provenzana; and (6) Mongibello. Note that we have tried to follow the terminology of Branca et al. (2011b), choosing terms from the geological map of Etna that was published after field sampling had been completed. However, we have inadvertently employed a mixture of terms relating to different volcanic centres (e.g., Cuvigghiuni, Mongibello) and lithostratigraphic units (e.g., Piano Provenzana, Tripodo). To improve clarity, Table 1 and the table legend at the bottom  of Fig. 1 explain the equivalence between the terms used in this study and those of the geological map of Branca et al. (2011b).
The stratigraphically lowest lavas in sections A and D (Fig. 1c, d) are basaltic trachyandesites belonging to the Valle degli Zappini formation (Salifizio-1) and are therefore older than 85.6 ka. Basaltic trachyandesite lavas from the Serra del Salifizio formation (Salifizio-2), dated at 85.6 ± 6.8 ka, overlie this unit and are present in all four sections (Table 1). Both formations are considered to belong to the Salifizio volcano (Branca et al. 2011b). The overlying trachybasalt and basaltic trachyandesite lavas in sections D, E, and C are from the Canalone della Montagnola formation, dated at 79-70 ka and defined by Branca et al. (2011b) as eruptions of the Cuvigghiuni volcano. The uppermost lavas in section A (Fig. 1c, d) are less evolved trachybasaltic lavas from the Tripodo member of the Piano Provenzana formation of the Ellittico volcano (Branca et al. 2011b), dated at 42-30 ka. In sections E and C, trachyandesitic lavas form the uppermost member of the Piano Provenzana formation, also dated at 42-30 ka. Precise age relationships between the Tripodo and upper Piano Provenzana members have not been determined (Branca et al. 2011b). We assume from field evidence (Spence and Downes 2011) that the upper member (Piano Provenzana) lavas are younger than those of Tripodo, but our conclusions are not affected by this assumption. The two units are mineralogically and texturally distinguishable and are also compositionally distinct. The basaltic trachyandesitic lavas at the top of sections E and C form the top of the southern wall of the Valle del Bove at the Schiena dell'Asino. These are products of the Mongibello stratovolcano and are younger than 15 ka (Branca et al. 2011b;De Beni et al. 2011).
The sampled lavas are extremely fresh, and most are vesicular to highly vesicular (> 20%). They range from aphyric to strongly porphyritic (25-40% phenocrysts), with plagioclase as the most common phenocryst phase. Some of the larger plagioclase phenocrysts exhibit zoning and/or sieve textures. Clinopyroxene (augite) phenocrysts/ megacrysts range up to 5 mm in size; they typically exhibit oscillatory zoning and ophitic textures, enclosing aligned plagioclase laths and FeTi oxides. Glomeroporphyritic aggregates of plagioclase, augite and Ti-magnetite are also common. Although its modal abundance is relatively low, olivine is present as phenocrysts and in the groundmass of all but the most evolved and aphyric lavas. Kaersutite occurs rarely as small, ragged and broken, partially reabsorbed crystals with reaction rims of titanomagnetite, indicating instability during low-pressure fractionation. Accessory apatite is also present as inclusions in augite and some plagioclase. Further details of sample petrography can be found in Spence and Downes (2011).

Sample preparation and methods
All the sampled Valle del Bove lavas were ground to a powder using a tungsten carbide mortar and analysed by X-Ray Fluorescence (XRF) at Royal Holloway University of London (Spence and Downes 2011). XRF data for ten of the samples were reported in Spence and Downes (2011), others are presented in Table 2. Strontium and Nd isotopic analyses were initially obtained on 11 samples at the University of London radiogenic isotope facility at Royal Holloway in the 1990s (data published in this study) but were unevenly distributed among the six lithostratigraphic units. Therefore, for the present study, 12 additional samples were analysed for Sr, Nd, Pb, and Hf isotopic ratios in other laboratories (see below for details).
Strontium and Nd isotopic ratios on the 11 samples analysed at Royal Holloway University of London (Table 3) were determined following the method described by Thirlwall (1991) and Thirlwall et al. (1997). Strontium and Nd isotopic compositions were determined multi-dynamically using a VG354 5-collector mass spectrometer, following conventional dissolution and chemical separation without prior sample leaching. Data were collected during several analytical sessions (methods in Thirlwall 1991), with reference materials reproducible to ± 0.000020 and ± 0.000008 (2SD) for 87 Sr/ 86 Sr and 143 Nd/ 144 Nd, respectively. For a few samples, small machine bias-corrections were made based on the difference between the means of the measuring period for NIST SRM 987 and a Nd in-house standard. The machine bias corrections were never more than + 0.000009 and − 0.000006, respectively. The long-term laboratory averages for SRM 987 were 87 Sr/ 86 Sr = 0.710248 and 143 Nd/ 144 Nd = 0.511420 for the Nd house standard (equivalent to La Jolla of 0.511857; Thirlwall 1991). Measured Nd and Sr procedural blanks were not significant with respect to the sample concentrations.
The new Sr and Nd isotope analyses (Table 3) were carried out at the Department of Earth and Environmental Sciences, Ludwig-Maximilians-Universität of Munich, following the methods described in Hegner et al. (1995). No sample powder leaching was employed. Strontium and Nd isotopes were measured as metals on single W and double Re-filament configurations, respectively. Total procedural blanks of < 500 pg for Sr and < 100 pg for Nd are not significant for the processed amounts of Sr and Nd. Isotope analyses were performed on an upgraded MAT 261 multi-collector thermal ionisation mass spectrometer. Strontium isotope abundance ratios were measured with a dynamic double-collector routine, while Nd isotope compositions were measured using a dynamic triple-collector routine, with corrections made for interfering 87 Rb and 144 Sm. 87 Sr/ 86 Sr and 143 Nd/ 144 Nd ratios were normalised for instrumental mass bias relative to 86 Sr/ 88 Sr = 0.1194 and 146 Nd/ 144 Nd = 0.7219, using a Rayleigh fractionation law. During the period of this study the NIST SRM 987 Sr reference material yielded 87 Sr/ 86 Sr = 0.710234 ± 0.000006 (2SD of population, n = 8) and the La Jolla Nd reference material 143 Nd/ 144 Nd = 0.511847 ± 0.000008 (2SD of population, n = 10). The long-term external precision for 87 Sr/ 86 Sr and 143 Nd/ 144 Nd is estimated at ca. 1.1 × 10 -5 (2SD).
The Pb and Hf isotope analyses (Table 3) were carried out at the Ecole Normale Supérieure in Lyon (ENSL) by solution chemistry (Hf only; Pb separated at University of New Hampshire; see below) and MC-ICP-MS (both Hf and Pb; Nu Plasma 500 HR) following the methods of  and Blichert-Toft and Albarède (2009). 176 Hf/ 177 Hf ratios were corrected for instrumental mass fractionation relative to 179 Hf/ 177 Hf = 0.7325 using an exponential law. The JMC-475 Hf standard was run every two samples and averaged 0.282163 ± 0.000007 (2σ; n = 14) for 176 Hf/ 177 Hf during the single run session of this work. Since this is identical to the accepted value of 0.282163 ± 0.000009  for JMC-475, no corrections were applied to the data. ε Hf values (Table 3) were calculated using 176 Hf/ 177 Hf = 0.282772 (Blichert-Toft and Albarède 1997).
Samples were prepared for Pb isotope analyses at the University of New Hampshire (UNH) geochemical laboratories. Samples were first leached in hot 6 M HCl to remove surficial contaminants. Lead was then isolated from the sample following procedures adapted from Bryce and DePaolo (2004). The Pb isotope data obtained by MC-ICP-MS at the Ecole Normale Supérieure in Lyon were corrected for instrumental mass fractionation by thallium normalization and adjusted for machine drift by sample-standard bracketing (White et al. 2000;Albarède et al. 2004) using the values of Eisele et al. (2003) for NIST SRM 981. Three NIST SRM 981 standard analyses run within the bracketed sample suite yielded results (with external reproducibility) of 208 Pb/ 204 Pb = 36.726(4) 207 Pb/ 204 Pb = 15.498(3), and 206 Pb/ 204 Pb = 16.941(3). Total procedural Hf and Pb blanks were < 20 pg and < 100 pg, respectively, negligible relative to the abundances of these elements processed for study.

Results
Valle del Bove lavas are moderately evolved (50-62 wt% SiO 2 , Fig. 2a) and, based on the ratio of Na 2 O to K 2 O (1.9-2.8), the rocks are mildly sodic basaltic trachyandesites and trachyandesites, including one trachyte sample. Figure 2a also shows fields that encompass the range of compositions for recent (i.e., < 130-years old) and historic (i.e., > 130-years old) Etna lavas, along with the field for prehistoric and ancient Etna magmatic products. The Valle  Clocchiatti et al. (1988). Within the Valle del Bove suite, samples from Salifizio-1 and Piano Provenzana have consistently lower concentrations of K 2 O and total alkalis for a given SiO 2 concentration than those of Salifizio-2, Cuvigghiuni, the Tripodo member, or Mongibello. Although the higher SiO 2 concentrations for prehistoric and ancient Etna samples suggest that these rocks are all more evolved than recent and historic Etna samples, plots of major element oxides versus MgO (Fig. 3) show that the least evolved lavas from Valle del Bove, particularly those from Salifizio-1, have similar MgO concentrations (and Mg numbers; see Online Resource 1, Fig. S1) to recent and historic Etna, suggesting similar degrees of magmatic evolution. Valle del Bove lavas are distinct in having higher SiO 2 for a given MgO concentration ( Fig. 3a). They also have lower CaO, Al 2 O 3 , TiO 2 , and FeO total. concentrations and CaO/Al 2 O 3 ratios. Recent Etna lavas have significantly higher K 2 O for a given MgO concentration (see Online Resource 1, Fig. S1) than either historic or ancient/prehistoric Etna, including the Valle del Bove lavas, consistent with their higher K 2 O shown in Fig. 2b. Only two samples from the Tripodo member, A14 and A15 (Table 2), are as enriched in K 2 O at similar MgO concentration as recent Etna lavas. Other samples from the Tripodo member are more evolved (MgO < 3.3 wt%), yet they have lower K 2 O concentrations, i.e., opposite of what would be expected from fractional crystallisation of the observed phenocryst phases. Samples A14 and A15 occur stratigraphically below the more evolved Tripodo member samples and below a 20 m interval with lack of exposure (Spence and Downes 2011); hence, the relationship between these two groups of Tripodo samples is unclear, but they may not be petrogenetically related.
The Valle del Bove lavas overlap the compositional fields for ancient and prehistoric Etna in Fig. 3, with a few notable exceptions. Salifizio-1 lavas, the oldest rocks in the suite, have SiO 2 concentrations higher than those of recent and historic Etna, but also higher than most other ancient and prehistoric Etna rocks (Fig. 3a). They also have lower Al 2 O 3 and TiO 2 (Fig. 3c, e). Ancient and prehistoric Etna samples, including those from Valle del Bove, extend to more evolved compositions (i.e., lower MgO and higher SiO 2 ) than recent or historic Etna lavas. Figure 4a shows the primitive-mantle-normalised incompatible element patterns for the Valle del Bove samples compared with those of ocean island basalts (OIB) representative of EM (enriched mantle) and HIMU (mantle with high time-integrated U/Pb) sources, as well as average global subducted sediments (GLOSS) and bulk continental crust. Figure 4b shows the data normalised to the composition of average recent Etna lavas. All Valle del Bove samples are enriched in most of the incompatible trace elements relative to recent and historic Etna basalts (Fig. 4b). Two units (Mongibello and Piano Provenzana) exhibit particular enrichment in nearly all incompatible trace elements except Sr and Ti, consistent with their evolved compositions (MgO concentrations < 2.6 wt%, Table 2; see also Figs. 2 and 3). In contrast, the least evolved samples (e.g., from Salifizio-1 and the Tripodo member) have compositions typical of intraplate alkaline magmas, e.g., OIB-like, although with relative enrichments in Ba, Th, and Pb and depletions in Ti and K (Fig. 4a). Of note in this context are the high La/Nb ratios relative to average OIB compositions. The Tripodo member has a trace element pattern that is distinct from most other Valle del Bove units. It is more akin to historic Etna, particularly in terms of its low Th/Nb and Th/Ba ratios (Fig. 4b).
However, the least-evolved Tripodo samples, A14 and A15, have higher concentrations of nearly all incompatible trace elements (Table 2) than the most-evolved samples, further Sr and Nd isotopic analyses by P.Z.
Vroon, analyzed at RHUL. Errors of isotopic ratios are 2 mean and refer to the last digits of the ratios. Nd and Hf calculated using the parameters of Bouvier et al. (2008)   .03) ratios with clear differences between the six units (Fig. 5). Samples from the oldest unit (Salifizio-1) have the highest 87 Sr/ 86 Sr ratios of all the Valle del Bove units (> 0.7035). These rocks contrast with those of similar age from the northern wall of the Valle del Bove, i.e., the Rocca Capra (ca. 81-78 ka) (D'Orazio et al. 1997) and the prehistoric/ancient Etna rocks reported by Marty et al. (1994), all of which have 87 Sr/ 86 Sr ratios < 0.70325 and 143 Nd/ 144 Nd > 0.51291. Instead, Salifizio-1 samples overlap the field of recent high-87 Sr/ 86 Sr Etna lavas (Fig. 5a, b) in a Sr-Nd isotope diagram. They also have overlapping to lower 176 Hf/ 177 Hf and 143 Nd/ 144 Nd ratios relative to recent Etna lavas (Fig. 5c, e). Their Hf isotopic compositions overlap those of most younger Valle del Bove units (except Piano Provenzana), but their 143 Nd/ 144 Nd ratios are lower. Thus, the  Maitre (1984). b K 2 O wt% versus SiO 2 wt% with discriminant boundaries from Peccerillo and Taylor (1976). Data for Valle del Bove from Spence and Downes (2011) and this study. Data for recent Etna samples (< 130-years old) from Armienti et al. (1989), Tonarini et al. (1995), Viccaro and Cristofolini (2008), and Wijbrans (unpublished). Historic Etna samples (> 130-years old) from Viccaro and Cristofolini (2008). Prehistoric/ancient Etna from D' Orazio et al. (1997) isotopic data suggest that Salifizio-1 magmas were derived from a source isotopically similar to that of recent Etna magmas and distinct from that of other ancient and prehistoric Etna rocks. In contrast, the slightly younger (~85-70 ka) Salifizio-2 and Cuvigghiuni lavas have lower 87 Sr/ 86 Sr and higher 176 Hf/ 177 Hf ratios and plot within or near the field of historic Etna lavas (Fig. 5a, c, e), as do lavas from the Tripodo member. However, even these samples have higher 87 Sr/ 86 Sr ratios than most samples of ancient and prehistoric Etna (Marty et al. 1994;D'Orazio et al. 1997). All VdB units, however, have lower 87 Sr/ 86 Sr ratios than the unusual prehistoric picrite, FS (pyroclastic fall deposit), analysed by Correale et al. (2014). The higher Sr isotopic ratio for this sample (0.70391) was attributed to crustal contamination, although the authors did not distinguish between contamination occurring in a shallow crustal magma chamber or via subduction modification of the mantle source.
The Piano Provenzana lavas have the lowest 176 Hf/ 177 Hf ratios of all the analysed prehistoric lavas and almost the lowest 143 Nd/ 144 Nd (the Nd isotope values overlap those  Tonarini et al. (1996) and Trua et al. (1998) of Salifizio-1). The youngest Mongibello lavas return to isotopic ratios that are more typical of historic Etna, with higher 143 Nd/ 144 Nd (Fig. 5b, d, f). The Hf and Sr isotopic compositions thus indicate the existence of at least three different components, two of which are predominant, i.e., the isotopically distinct sources of recent Etna lavas and historic/prehistoric Etna lavas, with five of the six Valle del Bove units overlapping these two fields. The third component, with an unusually low Hf isotopic composition for a given Sr or Nd isotopic composition, is seen only in the Piano Provenzana unit (Fig. 5c, e). This component has not previously been observed in samples from Mt. Etna, regardless of eruption age or major element composition.
Strontium, Nd, and Hf isotopic ratios (Fig. 5) show a change over time from more depleted DMM-like compositions (e.g., ε Nd of ~ + 10) for ancient Iblean tholeiites to less depleted isotopic compositions for recent Etna eruptions. However, this simple trend breaks down during the ~ 80 kyr time span of the Valle del Bove eruptions (Fig. 5). Instead, a wider range of compositions appears to have been available for prehistoric Etna than exists today, in particular with addition of the new data of this study for Salifizio-1 and Piano Provenzana.
Lead isotope data for lavas from the Iblean Plateau and Mt. Etna (Fig. 6) also show a trend from depleted towards more enriched source compositions over time. However, some of the variation in the literature data, particularly in 207 Pb/ 204 Pb, may result, in part, from the greater analytical uncertainty of the older data collected using thermal ionisation mass spectrometry as compared with modern MC-ICP-MS data, which better control instrumental mass discrimination. There may also be some differences in sample preparation, particularly with regards to whether predissolution leaching was applied. Lead isotope compositions of the Valle del Bove samples analysed using MC-ICP-MS with Tl-normalisation and sample-standard bracketing show tight linearity in Pb isotope space over the ~ 80 kyr time span considered (Fig. 6)  ), whereas the Tripodo member is the most radiogenic (i.e., its source shows the highest timeintegrated U/Pb of all samples). The radiogenic Pb isotope composition of the Tripodo member contrasts with its lower 87 Sr/ 86 Sr than in the other samples (Fig. 5a), which suggests a source that is more incompatible element depleted (i.e., time-integrated depletion in Rb relative to Sr) than that of most other Valle del Bove lavas.

Fractionation of Valle del Bove prehistoric magmas
The data points in the TAS diagram (Fig. 2a) suggest that most of the prehistoric Valle del Bove lavas are more evolved than recent and historic Etna eruptions (Tanguy et al. 1997  Differentiation Indices (DI = normative q + or + ab + ne + lc) of 63-77 (Fig. S2a), but they lie along two different differentiation trends (Fig. S2b). Samples from the Tripodo member, Cuvigghiuni, Salifizio-2, and Mongibello become more silica undersaturated as they become more evolved, whereas, those from Salifizio-1 and Piano Provenzana become more silica saturated as differentiation increases.  Armienti et al. (1989), Marty et al. (1994), Tonarini et al. (1995Tonarini et al. ( , 2001, and Viccaro and Cristofolini (2008). Historic Etna data from Marty et al. (1994), Tonarini et al. (2001), and Viccaro and Cristofolini (2008). Ancient/ prehistoric Etna data from D' Orazio et al. (1997) and Marty et al. (1994). Iblean Plateau data from Tonarini et al. (1996) and Trua et al. (1998). Hafnium isotopic ratios for historic and recent Etna from Viccaro et al. To better understand the effects of fractional crystallisation on the Valle del Bove lavas, we have modelled isobaric fractionation using alphaMELTS 1.9 (Ghiorso and Sack 1995;Asimow and Ghiorso 1998;Smith and Asimow 2005). The results are shown in diagrams of MgO versus major element oxides SiO 2 , Al 2 O 3 , CaO, FeO total , as well as Na 2 O + K 2 O versus SiO 2 (Fig. 7); details of the calculations are provided in Online Resource 2. Given the observation that the Valle del Bove suite includes at least two distinct differentiation trends, two different parental compositions have been assumed, one to represent the trend of increasing silica saturation (i.e., Salifizio-1 sample A5; Spence and Downes 2011) and the other to represent the trend of increasing silica undersaturation with increased differentiation (i.e., Tripodo member sample A15; Table 2). Figure 7 shows that, for both starting compositions, fractional crystallisation at low pressure (i.e., 1 kb or ~ 3 km) fails to reproduce the data, in particular the observed increase in Al 2 O 3 and decreasing FeO total with decreasing MgO concentration. It also fails to reproduce the distribution of data on the TAS diagram (Fig. 7i, j), with low-pressure fractionation resulting in excessive enrichment in total alkalis. For these lowpressure calculations, we assumed 'dry' conditions (H 2 O of 0.01 wt%), because of the low loss on ignition of the studied samples (Table 2). However, even if we assume higher water concentrations (e.g., 2 wt%), the model curves at low pressure (not shown) do not reproduce the data, in particular for Al 2 O 3 , FeO total , and total alkalis.
In contrast, at moderate pressure (3.5 to 4 kb or ~ 10-12 km) and H 2 O concentrations in the melt of ~ 3-4 wt%, the model curves reproduce the shape of the observed data distributions well. The difference in the residual melt pathways for low vs. moderate pressure is due to the expansion of the olivine stability field as pressure increases. At low pressures, plagioclase is the liquidus phase, followed by olivine and clinopyroxene in fairly rapid succession (see Online Resources 2 for details). The early stabilisation of plagioclase leads to a rapid depletion in Al 2 O 3 as fractionation proceeds. At higher pressures, the olivine stability field expands, and the crystallisation sequence is typically spinel followed by olivine, followed closely by clinopyroxene, with plagioclase appearing much later. Under these conditions, Al 2 O 3 increases until plagioclase starts to fractionate.
None of the models accurately reproduce the high Al 2 O 3 concentrations of some samples from Cuvigghiuni and the Tripodo member (Fig. 7c, d). The MELTS models also fail to reproduce the variations in TiO 2 versus MgO (not shown). These discrepancies may, in part, be a function of uncertainties in the thermodynamic data used in the MELTS calculations and are associated with an over-stabilisation of spinel, a recognised problem of the MELTS software (Ghiorso et al. 2002). This leads to depletion in Al 2 O 3 , FeO, and TiO 2 in the calculated melts in excess of that in the actual rocks. Nonetheless, some of the Valle del Bove samples are plagioclaserich, in particular the Tripodo lavas (Spence and Downes 2011). Hence, some of the observed compositional variation, particularly the enrichment in Al 2 O 3 , may be due to plagioclase accumulation at low pressures prior to eruption.
In summary, modelling with alphaMELTS 1.9 indicates that fractionation of the Valle del Bove lavas occurred in a magma chamber at moderate depths of ~ 10-12 km, which is consistent with interpretations based on geophysical data for the depth of present-day magma storage Mollo et al. 2015;Miller et al. 2017;Cannata et al. 2018). For the Valle del Bove lavas, fractionation is dominated by olivine and clinopyroxene. This contrasts with models for recent Etna in which the fractionation sequence is typically described as plagioclase, followed by clinopyroxene, olivine, Pb/ 204 Pb ratios for the six Valle del Bove units; prehistoric/ancient, historic, and recent Etna lavas; and tholeiitic and alkali lavas from the Iblean Plateau. Data for historic and recent Etna lavas from Carter and Civetta (1977) and Viccaro and Cristofolini (2008). Prehistoric/ ancient Etna data from Carter and Civetta (1977) and Miller et al. (2017). Data for Iblean tholeiitic and alkali lavas from Carter and Civetta (1977), Tonarini et al. (1996), and Trua et al. (1998). Analytical uncertainties (2SD) for Valle del Bove data are within the size of the symbols Fig. 7 Major element variation diagrams for a SiO 2 , b FeO total , c Al 2 O 3 , d CaO, e K 2 O, and f TiO 2 versus MgO wt% showing fractional crystallisation pathways calculated using alphaMELTS 1.9. Details of calculations in Online Resources 2. a, c, e, g, and i assume a starting composition of Salifizio-1 lava A5. b, d, f, h, and j assume Tripodo member A15 as the starting composition. Three model scenarios are presented for both starting compositions: (1) 1 kb, QFM, 'dry'; (2) 3.5 kb, QFM, 4 wt% H 2 O; (3) 4 kb, NNO, 3 wt% H 2 O. Data sources for recent, historic, and ancient Etna as in Fig. 3 titaniferous magnetite, and, in some cases, amphibole (Viccaro et al. 2011), all of which suggest a stronger imprint from shallower fractional crystallisation processes. Water concentrations of 3-4 wt% are consistent with the compositions of melt inclusions in olivine (Spilliaert et al. 2006), studies of Etna magma ascent rates (Armienti et al. 2013), and thermodynamic modelling (Kahl et al. 2015). Low values of loss on ignition (Table 2) suggest the Valle del Bove lavas were thoroughly degassed before eruption.

Source components of Etna magmatism: evidence from major and trace elements
Fractional crystallisation has clearly affected the compositions of the Valle del Bove lavas (Figs. 2, 3, 7). Nonetheless, some of the major and trace element variations observed may be an indication of differences in mantle source composition or variations in percentage of melting, rather than differences in extent of fractionation.
We test this hypothesis using Th, a highly incompatible trace element, as a proxy for alkaline magma fractionation   (Fig. 8). Thorium correlates positively with Differentiation Index in the Valle del Bove suite (R 2 = 0.75), further justifying this trace element approach (see Online Resources 1, Fig. S3). In the absence of significant crustal contamination, the ratio of two highly incompatible elements (i.e., D ≪ 1) should remain largely unchanged during fractional crystallisation. Therefore, plots of Th versus other highly incompatible trace elements should produce linear arrays that project back to the ratio of the two elements in the parental magma, providing an indication of source composition and possible heterogeneity (e.g., Allegre et al. 1977;Joron and Treuil 1989). Given that the Valle del Bove lavas are characterised by an anhydrous phenocryst assemblage consisting of clinopyroxene + plagioclase + magnetite ± olivine (Spence and Downes 2011), elements such as Nb, Zr, and Ba should have similar incompatibility during fractional crystallisation in this system (see GERM database, https:// earth ref. org/ KDD/). Figure 8a shows that most Valle del Bove lavas overlap the field of prehistoric/ ancient Etna and form arrays with Nb/Th ratios similar to recent and historic Etna. Some lavas from Salifizio-1, however, appear to have evolved from a parental magma that was more enriched in Th relative to Nb. Moreover, if the Piano Provenzana lavas evolved from a transitional/silicasaturated parent similar to Salifizio-1, as suggested by Fig. 2 and S2, these lavas are significantly more enriched in Nb (or depleted in Th) than expected by fractional crystallisation alone. In the plots of Ba and Zr vs Th (Fig. 8b, c), the Valle del Bove samples again largely overlap the field for prehistoric/ancient Etna, but the slopes of the data at the highest degrees of fractionation are shallower than expected within individual units; that is, at the highest Th concentrations (i.e., highest fractionation), Ba and Zr concentrations are lower than expected. Instead, the data are better described by second-order polynomials (R 2 > 0.86), as shown in Fig. 8b, c. At lower Th concentrations (lower fractionation), the curves indicate similar Ba/Th and Zr/Th ratios for recent and historic Etna, but as magma differentiation increases, they deviate to lower Ba/Th and Zr/Th. This could be due to  Fig. 3 fractionation of trace phases with higher D values for these elements that appear later in the fractionation sequence, such as amphibole, biotite, or K-feldspar (for Ba), and zircon (for Zr). Nonetheless, as for Nb vs. Th, lavas from Salifizio-1 appear to have evolved from a parent that was enriched in Th relative to Ba and Zr, and the Piano Provenzana lavas have higher concentrations of Ba and Zr relative to Th than would be expected if they were derived from a silica-saturated parent similar to Salifizio-1. In addition, the two least evolved Tripodo member samples, A14 and A15 (MgO > 4.75 wt%; samples circled in Fig. 8a-c), clearly have higher concentrations of incompatible trace elements than the more evolved lavas from this unit (MgO < 3.3 wt%), opposite to what would be expected if the rocks were related by fractional crystallisation.
The data, therefore, demonstrate that at least some of the trace element variations observed are unrelated to fractional crystallisation (Fig. 8). These findings further suggest that crustal contamination is unlikely to be responsible for the trace element variations, since average continental crust (Rudnick and Gao 2003) has concentrations of Nb (8 ppm), Zr (132 ppm), Ba (456 ppm), and Th (6 ppm) that are lower than those in the least evolved Valle del Bove samples. This conclusion is consistent with similar trace element arguments against significant crustal assimilation put forward for recent Etna by Corsaro et al. (2007). In addition, D'Orazio et al. (1997) observed Sr and Nd isotope equilibrium between clinopyroxene and host lavas from the north wall of the Valle del Bove, which indicates that if crustal contamination occurred, it could not have taken place after clinopyroxene began to crystallise. Similar evidence for phenocryst-melt isotopic equilibrium was also reported by Miller et al. (2017) for the older Timpe lavas. As shown by our alphaMELTS 1.9 modelling, crystallisation of clinopyroxene occurs very early in the fractionation sequence at moderate pressures, thus limiting the potential for impact from shallow-level crustal contamination.
Despite the similarities in incompatible element ratios and, hence, potential source regions for recent to ancient Etna as suggested by Fig. 8, Fig. 4b shows that even the most primitive of the Valle del Bove samples, i.e., some samples from Salifizio-1 and the Tripodo member, are enriched in Ba, Th, Nb, Pb, and Zr relative to recent and historic Etna magmas. Furthermore, these two units are not only distinct from recent and historic Etna, but also different from one another, suggesting differences in source composition or petrogenesis. The Tripodo member has a trace element pattern similar to average historic Etna lavas, just at higher overall concentrations, consistent with the lavas being slightly more evolved (Fig. 3). In contrast, Salifizio-1 lavas have higher Rb and Th, but lower light rare earth elements (LREE) and Sr. Strong relative depletions in Sr indicate a role for plagioclase fractionation, and the trough at Ti reflects magnetite fractionation (Fig. 4b). The low K concentration would be consistent with fractionation of a K-bearing phase, such as biotite or K-feldspar, but neither of these phases is observed in the rocks. Instead, the relative depletion in K is more likely an indication of derivation from a source in which a K-bearing phase, such as phlogopite or amphibole, was residual during partial melting, as suggested by Viccaro and Cristofolini (2008) for recent Etna lavas.
This interpretation is consistent with our own modelling of K and Rb systematics for the Valle del Bove lavas (Fig. 9). In a plot of K/Rb ratio versus Rb concentration, the Valle del Bove samples overlap the fields for historic and recent Etna but extend to higher concentrations of Rb for a given K than seen in the younger Etna rocks, particularly for the highly evolved lavas from Piano Provenzana and Mongibello. The effects of clinopyroxene and amphibole fractionation are shown as the solid and dashed lines, respectively. Fractionation in the absence of a hydrous phase would increase Rb Fig. 9 K/Rb ratios versus Rb concentrations showing calculated trends for partial melting of a source containing 1 modal % phlogopite (dotted line) and 1 modal % amphibole (dot-dash line). Variations due to fractional crystallisation of amphibole from starting compositions A5 (Salifizio-1) and A15 (Tripodo member) shown as dashed lines. Variations due to fractional crystallisation of clinopyroxene from starting compositions A5 and A15 shown as solid lines. Data sources for ancient to recent Etna samples as in Fig. 3. Fractional crystallisation modelled using the equation C L = C o *F (D−1) , where C L is the calculated liquid composition, C o is the initial concentration, D is the bulk distribution coefficient and F is the fraction of liquid remaining. Partial melting calculations assume simple batch melting, i.e., C L = C o /(D + F − DF), where C L is the calculated liquid composition, C o is the initial concentration, D is the bulk distribution coefficient and F is the degree of partial melting. Mineral modes are based on anhydrous lherzolite xenoliths from the Iblean Plateau (ol = 72%; opx = 15%; cpx = 12%, Correale et al. 2014) modified to include 1% amphibole or 1% phlogopite. Partition coefficients from GERM (https:// earth ref. org/ KDD/); Rb partition coefficients as follows: ol = 0.0003; opx = 0.001; cpx = 0.01; amph = 0.023; phlog = 1.5; K partition coefficients: ol = 0.0001; opx = 0.0003; cpx = 0.004; amph = 1.36; phlog = 3 concentrations but have little effect on the K/Rb ratio (solid lines). Because K is more compatible in amphibole than Rb, fractionation of this phase would decrease the K/Rb ratio while increasing the Rb concentration (dashed lines). While some of the Valle del Bove data are consistent with a small amount of amphibole fractionation, in most cases, our modelling suggests that fractionation is dominated by anhydrous phases, in support of our MELTS modelling (Fig. 7).
However, fractionation of the observed phenocryst phases cannot account for the lower Rb and higher K/Rb values of some Tripodo or Mongibello samples, assuming parental magma compositions like those of the most primitive lavas in the suite. Indeed, the most primitive lavas, e.g., A5 from Salifizio-1 and A15 from the Tripodo member, overlap the field of recent Etna, whereas the more evolved samples from the Tripodo member overlap the field for historic Etna, and evolved rocks from Mongibello have both higher K/Rb ratios and Rb concentrations (Fig. 9). These differences suggest either different source compositions or differences in the degree of melting that gave rise to the parental magma, or both.
We evaluate these two alternatives via a simple batch partial melting model of a spinel lherzolite source containing either 1% amphibole or 1% phlogopite in the mode (Fig. 9). The two curves have similar shapes and show similar trajectories of increasing K/Rb ratios with decreasing Rb concentration as degree of partial melting increases, similar to the distribution shown by the Etna data. The phlogopite-bearing source, however, produces lower concentrations of Rb and higher K/Rb ratios for the same degree of partial melting relative to the amphibole-bearing source. More sophisticated melting models, such as non-modal batch melting or fractional melting, tend to produce higher Rb concentrations and K/Rb ratios for smaller degrees of partial melting. In contrast, partial melting of an anhydrous peridotite source would produce a nearly horizontal array on this diagram given the similar partition coefficients for the relevant mineral phases.
Collectively, the data suggest that the Etna lavas were derived by partial melting of a source containing hydrous phases, such as amphibole or phlogopite. The trend of increasing K/Rb ratios with decreasing Rb concentration as a function of degree of partial melting is more consistent with the observed data distribution than melting of an anhydrous source. The offset of the Tripodo member samples and historic Etna to lower Rb and higher K/Rb relative to recent Etna suggests these magmas were either derived by larger degrees of partial melting or from a different source.
The modelling results in Fig. 9 do not allow unambiguous distinction between these two options, because details of source mineralogy and elemental concentrations can only be assumed. However, several lines of evidence lead us to conclude that the Tripodo member lavas were derived by smaller degrees of partial melting than the Salifizio-1 lavas, e.g., their more silica undersaturated compositions ( Fig. 2 and S1), higher La/Y, and lower Rb, Th, and K relative to Ba, Nb, and La (Fig. 4). Therefore, source heterogeneity has likely played a significant role in generating the compositional variations observed.

Source components of Etna magmatism: evidence from Sr:Nd:Pb:Hf isotopes
Many authors have interpreted the compositional variations in recent Etna lavas as evidence for source heterogeneity (e.g., Carter and Civetta 1977;Tanguy et al. 1997;Tonarini et al. 2001;Armienti et al. 2004;Viccaro and Cristofolini 2008;Viccaro et al. 2011;Correale et al. 2014 and references therein). Models to explain this source heterogeneity are broadly of two types. One calls for modification of the mantle source via subduction-derived melts and fluids, a model consistent with the complex geodynamic setting in which Etna is located (e.g., Armienti et al. 2004;Tonarini et al. 2001;Viccaro et al. 2011). The other attributes source heterogeneity to lithologic variations in which a predominantly lherzolitic matrix is crosscut by pyroxenite (± garnet) veins (Correale et al. 2014;Miller et al. 2017;Viccaro and Zuccarello 2017). Correale et al. (2014) further argue that the pyroxenitic component contributes about 10% to the melts that give rise to recent Etnean lavas.
Constraining the proportion of pyroxenite to peridotite in the mantle source is challenging because (a) pyroxenite compositions can vary widely Lambart et al. 2016) and (b) phase relations, and hence melt compositions, are pressure and temperature sensitive. Nonetheless, in broad terms, partial melts involving high proportions of pyroxenite in the source tend to be higher in CaO/Al 2 O 3 and TiO 2 but lower in SiO 2 for a given MgO concentration than those derived from peridotite melting (Kogiso et al. 2003;Condamine and Medard 2016). Thus, the higher SiO 2 , lower CaO/Al 2 O 3 , FeO, and TiO 2 of prehistoric Etna magmas relative to recent and historic Etna (Fig. 3) are consistent with an increase in the proportion of pyroxenite in the source of Etna magmas over time.
Variations in isotopic composition also allow us to examine source heterogeneity and to assess several different processes that may be involved in the petrogenesis of the Valle del Bove lavas: (1) melting of isotopically distinct asthenospheric mantle sources; (2) variable degrees of partial melting of metasomatised mantle domains; (3) addition of a subduction-related component to the mantle source; and (4) crustal contamination of magmas during fractionation. In the following sections we consider the relative contributions of these various processes.
Recent and historic Etnean lavas are generally considered to have been derived from an asthenospheric mantle source (Wilson and Downes 2006), known variously as the low-velocity composition (LVC) of Hoernle et al. (1995), the European asthenospheric reservoir (EAR) of Cebrià and Wilson (1995), the common component ('C') of Hanan and Graham (1996), or the FOcal ZOne (FOZO) of Hart et al. (1992). This mantle component may itself be a mixture, with contributions from DMM (low 87 Sr/ 86 Sr and radiogenic (high) Nd, Hf, and Pb isotopic compositions) and possibly EM1 sources (enriched mantle-1) (Stracke et al. 2005;Viccaro et al. 2011). The isotopic compositions of the Iblean Plateau lavas (Figs. 5, 6) indicate an origin from a more depleted mantle source than that feeding present-day Mt. Etna (Tonarini et al. 1996;Trua et al. 1998). Hence, modern Etna is no longer tapping the depleted mantle source of the older Iblean tholeiitic magmatism. Instead, modern Etna is being fed by an isotopically heterogeneous mantle source more typical of intra-plate ocean islands (Armienti et al. 1989;Tonarini et al. 2001;Viccaro and Cristofolini 2008). Surprisingly, our study shows that Salifizio-1 lavas, the oldest lavas in our suite, have isotopic compositions similar to those of recently erupted Etna lavas, whereas, four slightly younger units (Salifizio-2, Cuvigghiuni, Tripodo, and Mongibello) have tapped a source very similar to that of historic Etna lavas (Fig. 5). In contrast, Piano Provenzana lavas show a unique isotopic signature that has not previously been recognised in the data for Mt. Etna. Although major and trace element data provide no evidence for significant involvement of crustal contamination in the petrogenesis of Valle del Bove lavas, we still need to consider this process with regard to the isotope data before we can confidently interpret this new isotopic signature as an indication of source heterogeneity.

Crustal contamination versus a subducted sediment component: evidence from Sr:Nd:Pb:Hf isotopes
The composition of the lower crust beneath NE Sicily is not well constrained, since the only crustal xenoliths available are derived from the upper crust and have experienced extensive thermal metamorphism (Michaud 1995). The outcropping basement NE of Mt. Etna is composed of a variety of metasedimentary rocks intruded by Hercynian granitoids (Fiannacca et al. 2008) and thus closely resembles the crust of the Calabrian arc and other areas of Hercynian crust throughout southern Europe, but there are few studies of the geochemistry and isotopic compositions of these rocks. We have therefore used Sr, Nd, and Pb isotopic data for metaigneous and metasedimentary lower crustal xenoliths from the Hercynian French Massif Central ) as proxies in concert with the limited information available on Calabrian basement . The average Hf isotope composition of the same Hercynian lower crustal xenoliths (Vervoort et al. 2000) has been used here as a proxy for the crust underneath Mt. Etna.
Lead isotope data for the Valle del Bove lavas show a tightly constrained linear array in comparison to circum-Tyrrhenian volcanic rocks (Fig. 10). Most samples plot within or near the FOZO/EAR/LVC/C field with high 206 Pb/ 204 Pb ratios, while the Piano Provenzana lavas have lower 206 Pb/ 204 Pb and trend towards the general fields of sediments or continental crust. The solid black arrows in Fig. 10a, b indicate possible mixing trajectories that could explain the Valle del Bove Pb isotope arrays. The data are inconsistent with assimilation of Hercynian-type upper continental crust, but could be explained by incorporation of subducted sediment like that seen in the eastern Mediterranean (Klaver et al. 2015) or Ionian Sea ) into a FOZO/EAR-type source. Assimilation of Calabrian-type basement is also consistent with the Pb isotope data, and we cannot distinguish these alternatives using Pb isotope data alone.
However, Calabrian basement has an average 87 Sr/ 86 Sr value of ca. 0.717 , so assimilation-fractional crystallisation processes involving such a contaminant would likely have also affected 87 Sr/ 86 Sr, which is not observed (Fig. 5). Furthermore, all analysed Etna lavas have > 700 ppm Sr (compared to < 250 ppm for Calabrian basement), so it is unlikely that assimilation of continental crust would strongly affect their Sr isotope compositions. Nevertheless, the Piano Provenzana lavas clearly provide evidence for the presence of a component with a lower ε Hf and ε Nd composition beneath Etna (Fig. 5d, f).
The uniqueness of this component is particularly apparent in comparisons of the Valle del Bove units in 176 Hf/ 177 Hf vs 206 Pb/ 204 Pb in the context of the wider regional data set (Fig. 11a). Most of the Valle del Bove lavas again show a well-constrained array with one end-member in the FOZO/ EAR field; the data partially overlap the field for recent and historic Etna, but are clearly distinct from the sources of the Iblean Plateau and Aeolian arc (Alicudi and Filicudi), the compositions of which indicate greater time-integrated incompatible element depletion. Paired Hf-Pb isotopic data for other prehistoric Etna rocks are limited, but four analyses from the Timpe Santa Caterina phase of activity, i.e., 220-100 ka (Miller et al. 2017), have high Hf isotope values that overlap the composition of the Iblean Plateau. The Pb isotope compositions of these rocks are, however, more HIMU-like than those from the Iblean Plateau and, instead, are similar to the most radiogenic lavas from Valle del Bove. Thus, based on currently available data, the low Hf isotope values of the Piano Provenzana lavas are unlike any known Etna magmatic products.

Piano Provenzana: a new component in Etna magmatism?
Figures 5 and 11 indicate that the lavas from Piano Provenzana are isotopically unique among Etna magmatic products. One interpretation is that a brief episode of melting of an unusual sediment component in the slab beneath Sicily may have given rise to the source of the Piano Provenzana lavas. Comparisons of 176 Hf/ 177 Hf and Nd/Zr (Fig. 12a) show that all the prehistoric lavas analysed in this study have lower Nd/ Zr than recent or historic Etna, while only the Piano Provenzana lavas have Hf isotope compositions lower than recent and historic Etna. Thus, the lower Hf isotope signature in the Piano Provenzana unit might be related to zircon-rich sediments introduced into the mantle wedge prior to mixing with an upwelling melt generated from a more depleted mantle end-member, i.e., the "zircon effect" (Patchett et al. 1984;White et al. 1986;Blichert-Toft et al. 1999;Carpentier et al. 2009). Mixing calculations (Fig. 11b) show that isotopic signatures of most Valle del Bove samples can be  Fig. 6 with additional data from Ayuso et al. Fig. 11 a 176 Hf/ 177 Hf versus 206 Pb/ 204 Pb ratios for Valle del Bove samples in the context of the wider data set for Italian/Tyrrhenian volcanic rocks and the field for LVC/EAR/FOZO/C field. Data sources as for Figs. 5 and 10 with additional Hf isotopic data for Italy and the Tyrrhenian Sea from Barnekow (2000), Conticelli and Peccerillo (1992), Prelević et al. (2010), andMiller et al. (2017). Average Hf isotopic composition for Hercynian lower and upper continental crust from Vervoort et al. (1999Vervoort et al. ( , 2000. Mediterranean sediment data from Klaver et al. (2015) and Kempton et al. (2018). b Five representative mixing scenarios between EAR-type mantle and possible subducted sediment calculated using IgPet (Carr and Gazel 2017). EAR mantle modelled using two compositions: EAR a is representative of prehistoric Etna from Timpe Santa Catalina (Miller et al. 2017 explained by incorporation into a FOZO/EAR-type mantle source of < 2 wt% sediment similar in composition to that of the Ionian Sea . The Piano Provenzana samples, however, require a sediment contaminant with a lower Hf isotope composition, such as the Zr-rich sediments of the eastern Mediterranean (Klaver et al. 2015). A similar model was used recently to explain the Hf-Nd isotope systematics of Miocene subduction-related rocks from Sardinia .
However, if a subduction component is involved in the mantle source of the Piano Provenzana lavas, this addition must have occurred prior to 42-30 ka, the age of the Piano Provenzana lavas according to De Beni et al. (2011). Moreover, 87 Sr/ 86 Sr appears to have been largely unaffected by this process relative to other lavas of Ellittico. The Tripodo member, for example, comprises low-SiO 2 lavas with an age range similar to that of the upper Piano Provenzana formation (Branca et al. 2011a;De Beni et al. 2011), and these lavas exhibit low 87 Sr/ 86 Sr and high 176 Hf/ 177 Hf (Fig. 5). Furthermore, the degree of isotopic offset of the Piano Provenzana lavas from other prehistoric Etna lavas (Figs. 5,6,12) decreases in the order: Hf > Nd > Pb > Sr. In other words, the fluid-immobile elements Hf and Nd show a greater offset than the fluid-mobile elements Pb and Sr.
Thus, the lower 176 Hf/ 177 Hf values (relative to all other Etna lavas) could imply melting of an additional (ancient?) component in the source of the Piano Provenzana lavas, rather than recent fluid transfer during subduction. The unique isotope composition is consistent with small-scale heterogeneity in the form of veins or metasomatic minerals in the mantle source, i.e., mantle metasomes. Such a process has been described from numerous localities worldwide in which low-degree melts, enriched in highly incompatible elements, ascend and undergo fractional crystallisation, producing a spectrum of mineral assemblages ranging from anhydrous veins (chiefly pyroxenites ± garnet) to hydrous cumulates that contain amphibole and phlogopite (e.g., Frey and Prinz 1978;Irving 1980;Kempton 1987;Bodinier et al. 1990;Harte et al. 1993;. Viccaro and Zuccarello (2017) recently showed that partial melting of a mixed peridotite-pyroxenite mantle source could produce alkaline compositions similar to those of recent Etna magmas. In their model the peridotite component is a spinel lherzolite that contains hydrous metasomatic phases, whereas, the pyroxenite component is a garnet pyroxenite formed by crystallisation of silicate melts. They concluded that the pyroxenite to lherzolite ratio required to explain the compositions of recent Etna lavas ranges from 10 to 35%.
This model is consistent with the Valle del Bove data as shown in Fig. 12b, a plot of 87 Sr/ 86 Sr vs. Zr/Nb. Zr and Nb are relatively immobile in fluids mobilised during subduction, and since both are moderately incompatible, variation in the Zr/Nb ratio as a function of partial melting of a homogeneous source should be relatively small. Isotope ratios are obviously not fractionated during mantle melting, so any variation observed is a function of source heterogeneity. The data show a broad correlation between 87 Sr/ 86 Sr and Zr/Nb for ancient Etna and most Valle del Bove lavas. Historic and recent Etna lavas plot at the high-Zr/Nb end of the array, although recent Etna lavas have consistently higher 87 Sr/ 86 Sr for a given Zr/Nb relative to older lavas. If modification of the mantle source via subduction-derived fluids was the primary control on isotopic and trace element variations, there should be no obvious correlation between Zr/Nb and 87 Sr/ 86 Sr.  Nicotra et al. (2013). b 87 Sr/ 86 Sr versus Zr/Nb ratios for the six Valle del Bove units compared with the data fields for recent, historic, and ancient/prehistoric Etna; prehistoric Etna sample labelled FS is the unusual picrite reported by Correale et al. (2014). Solid black arrows indicate direction of compositional variation from slab-derived fluids and/or recycled sediment melts. Dashed arrow indicates compositional variation for recent Etna as a result of partial melting as proposed by Viccaro and Cristofolini (2008); dotted arrow indicates variation as a result of fractional crystallisation for most evolved VdB compositions where there is potential to fractionate Zr/Nb through crystallisation of phases like titanomagnetite or sphene. EAR compositional range (double-headed arrow) from Cebrià and Wilson (1995). Data sources as in Fig. 5 It could be argued that the trend is due to a fortuitous combination of subduction enrichment, which produces a vertical trend on the diagram through addition of radiogenic Sr, plus variation in Zr/Nb as a function of degree of partial melting induced by that fluid addition. This scenario implies that the degree of subduction modification correlates systematically with the degree of partial melting of the metasomatised mantle. However, the difference in degree of partial melting required to create the observed range of Zr/Nb ratios is at least a factor of 10, which would have a significant impact on the major element compositions of the magmas produced. In such a scenario, the lowest Zr/Nb should be significantly more enriched in other incompatible trace elements and probably more silica undersaturated than those at the high Zr/Nb end of the array, and such systematic variation is not observed for the data set as a whole (see Online Resources 1, Fig. S4). While we cannot entirely rule out this scenario, the impact from garnet pyroxenite in the source, which has a Zr/Nb of > 25 based on data for Iblean ultramafic xenoliths (Correale et al. 2012), is likely to outweigh the impact of varying degrees of partial melting of a homogeneous lherzolitic source, which typically has a Zr/ Nb ratio < 4. As such, the broad positive correlation is more likely due to a heterogeneous source consisting of variable proportions of lherzolite and pyroxenite.
This suggests there is a compositional, i.e., source heterogeneity, control involved in the genesis of Etna magmas, particularly ancient and prehistoric Etna. Melts derived from a pyroxenite-rich source should have higher Zr/Nb than those from a lherzolitic source. The positive correlation between Sr isotopes and Zr/Nb, therefore, suggests a role for ancient veins in the mantle rather than recent subduction influences. On the other hand, the offset of recent Etna lavas to higher 87 Sr/ 86 Sr may reflect a more significant role for metasomatism of the mantle source via subduction-derived fluids for these rocks.
We therefore agree with the interpretations of Viccaro and Cristofolini (2008), Correale et al. (2014), Miller et al. (2017, and Viccaro and Zuccarello (2017) that the mantle beneath Etna is heterogeneous, consisting of a lherzolite matrix crosscut by streaks or veins of 'ancient' pyroxenite of variable composition. More recent modification by subduction-derived fluids and/or melts may contribute to the petrogenesis of historic and recent Etna magmas.
The isotopic variation seen in the prehistoric lavas from Mt. Etna may be related to a decreasing degree of mantle melting relative to that producing Iblean tholeiites, so that less depleted low-T melting components of the heterogeneous mantle are preferentially tapped Pilet et al. 2011;Correale et al. 2014). In this scenario, the LREE and large-ion lithophile element (LILE) component of these rocks is likely to be related to decompression melting of a fusible part of the fluid-metasomatised upper mantle enriched with pyroxenites and/or hydrated minerals such as phlogopite or amphibole, as found in mantle xenoliths from the Iblean Plateau (Tonarini et al. 1996;Scribano et al. 2009;Bianchini et al. 2010). The Piano Provenzana parental melt may have formed in a discrete metasomatic zone in the mantle with an enriched crust-like signature , i.e., veins with a lower Hf isotopic composition than the surrounding mantle but similar Sr isotopic composition, thus affecting the isotope composition of Hf more than that of Nd and Pb, and with little effect on 87 Sr/ 86 Sr.

Summary and conclusions
1. Analysed Valle del Bove samples represent six lithostratigraphic units of prehistoric Etna lavas ranging in age from ~ 85 to ~ 4 ka. They show differences in element and radiogenic isotopic compositions that vary between the signatures of recent and historic Etna samples, suggesting melting of two main mantle components in the asthenospheric mantle source over the past ~ 100 kyr. 2. Although, in broad terms, the source of magmatism in the region has changed over time from more DMM-like (for Iblean Plateau lavas) to a more enriched (FOZO/ EAR/LVC/C-like) end-member for Etnean magmatism, variations in Sr:Nd:Pb:Hf isotopic compositions do not support a smooth transition. Instead, some of the oldest rocks for which data are available, e.g., Rocca Capra (~80 ka; D' Orazio et al. 1997) and Salifizio 1 (> 86 ka), encompass almost the entire range in Sr and Nd isotope values for Etna (0.70317 and 0.51293 for Rocca Capra, 0.70363 and 0.51283 for Salifizio-1). 3. Most of the prehistoric Etna eruptions tapped the FOZO/ EAR/LVC/C asthenospheric mantle component. In contrast, evolved trachyandesites from the Piano Provenzana unit were derived from a source that is unique amongst analysed lavas from Mt. Etna, having a less radiogenic Hf isotope signature. 4. The low 176 Hf/ 177 Hf ratios of the Piano Provenzana lavas suggest they formed by melting of an unusual piece of metasomatised lithospheric mantle. Indeed, the data for Valle del Bove as a whole support the existence of a heterogeneous marble-cake-style mantle source variably metasomatized by hydrous phases (amphibole and/or phlogopite) and pyroxenite veins as suggested by several previous studies (Viccaro and Cristofolini 2008;Correalle et al. 2014;Miller et al. 2017;Viccaro and Zuccarello 2017). The proportion of pyroxenite appears to be greater in historic to recent Etna than in the Valle del Bove. 5. The degree of magmatic differentiation of Valle del Bove lavas is similar to that of recent and historic Etna, but parental magmas were different and those for recent Etna were more silica undersaturated. Differences may be a function of degree of partial melting but are more likely due to different source compositions and, particularly, different proportions of pyroxenite in the source. 6. MELTS modelling, together with isotopic and trace element data, suggest that the Valle del Bove parental magmas contained ~ 3-4 wt% H 2 O and fractionated in a magma chamber at moderate depths of ~ 10-12 km.
Acknowledgements JBT acknowledges financial support from the French Agence Nationale de la Recherche (grant ANR-10-BLAN-0603 M&Ms-Mantle Melting-Measurements, Models, Mechanisms). HD is grateful to the University of London for support for fieldwork, and to Dr Giz Marriner for assistance with the XRF analyses at RHUL. EH acknowledges financing of laboratory expenses by the Department of Earth & Environmental Sciences at LMU. JB is grateful to ENSL for a visiting professorship that supported her contributions. We are grateful to Philippe Telouk for support with the ENSL Nu plasma. JBT's ANR-10-BLAN-0603 M&Ms grant covered the analyses. We thank John Morrison and Richard Spence for their work in production of Fig. 1. Two anonymous reviewers are thanked for their courteous and constructive comments that improved the clarity of the paper.
Author contributions HD conceived and designed the study. Material preparation was performed by AS, with analytical preparations and data produced by JBT, JB, EH, and PZV. The first draft of the manuscript was written by AS and all authors contributed with data analysis, interpretation, and writing. PDK performed the modelling and wrote the second draft of the manuscript, which all authors again contributed to.
Availability of data and material (data transparency) All new data are provided in tables presented in this paper.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.