Earth and Planetary Science Letters Boron isotope insights into the origin of subduction signatures in continent-continent collision zone volcanism

We present the ﬁrst boron abundance and δ 11 B data for young (1.5-0 Ma) volcanic rocks formed in an active continent-continent collision zone. The δ 11 B of post-collisional volcanic rocks ( − 5 to + 2 (cid:2) ) from the Armenian sector of the Arabia-Eurasia collision zone are heavier than mid-ocean ridge basalts (MORB), conﬁrming trace element and isotope evidence for their derivation from a subduction-modiﬁed mantle source. Based on the low B/Nb (0.03-0.25 vs 0.2-90 in arc magmas), as well as low Ba/Th and Pb/Ce, this source records a subduction signature which is presently ﬂuid-mobile element depleted relative to most arc settings. The heavier than MORB δ 11 B of post-collision volcanic rocks argues against derivation of their subduction signature from a stalled slab, which would be expected to produce a component with a lighter than MORB δ 11 B, due to previous ﬂuid depletion. Instead, the similarity of δ 11 B in Plio-Pleistocene post-collision to 41 Ma alkaline igneous rocks also from Armenia (and also presented in this study), suggests that the subduction signature is inherited from Mesozoic-Paleogene subduction of Neotethys oceanic slabs. The slab component is then stored in the mantle lithosphere in amphibole, which is consistent with the low [B] in both Armenian volcanic rocks and metasomatic amphibole in mantle xenoliths. Based on trace element and radiogenic isotope systematics, this slab component is thought to be dominated by sediment melts (or supercritical ﬂuids). Previously published δ 11 B of metasediments suggests a sediment-derived metasomatic agent could produce the B isotope composition observed in Armenian volcanic rocks. The lack of evidence for aqueous ﬂuids preserved over the 40 Myr since initial collision supports observations that this latter component is transitory, while the lifetime of sediment melts/supercritical ﬂuids can be extended to > 40 Myr. © 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Boron (B) and its stable isotopes 10 B and 11 B are a key tracer for the fate of slab-derived components under volcanic arcs (De Hoog and Savov, 2018;Hulett et al., 2016;Ishikawa and Nakamura, 1994;Ishikawa and Tera, 1997;Le Voyer et al., 2008;Morris et al., 1990;Palmer, 1991;Peacock and Hervig, 1999;Rose et al., 2001). This is due to a scarcity of boron in the mantle (<0.2 ppm; Chaussidon and Jambon, 1994;Kamenetsky and Eggins, 2012;Marschall et al., 2017;Ryan et al., 1996) and its strong fluid partitioning, with concomitant isotope fractionation during metamorphic slab dehydration reactions which release aqueous fluids at T < 800 • C Fig. 1. Location maps for Armenia volcanic rocks (a) Location map for the Arabia-Eurasia collision zone, showing the study region in Armenia (country outline in bold on the inset). Plio-Pleistocene active volcanic centres are shown by red triangles (Kaislaniemi et al., 2014). Location of the Pontide-Lesser Caucasus Mesozoic-Paleogene arc is modified from Rolland et al. (2009). (b) Geological map of major geological units within the territory of Armenia (Kharazyan, 2005;Mederer et al., 2013;Neill et al., 2015).
Letters refer to locations in the text: T = Tezhsar alkaline complex; A = Aragats stratovolcano; G = Gegham volcanic highland; V = Vardenis volcanic highland; S = Syunik volcanic highland. Also shown here is the location of the Armenian capital city, Yerevan (population > 1 million people). rocks from Armenia (Sokół et al., 2018;Sugden et al., 2019), in the northern part of the Arabia-Eurasia collision zone (Fig. 1a), representing the first boron isotope data for volcanic rocks from an active continent-continent collision. This collision zone is unique on Earth as a modern continental collision zone associated with widespread mantle-derived magmatism. Magmas from the southern part of the collision zone (e.g. Lake Van, Eastern Anatolia) have OIB-like (ocean island basalt) characteristics, whereas those from the north have arc-like geochemistry (Pearce et al., 1990). The arc-like geochemistry of Armenian volcanic rocks (Fig. 2, see also Table S1 in Supplementary Material) reflects a subductionmodified magma source. The [B] and δ 11 B data presented in this study come from 1.5-0 Ma post-collisional volcanic rocks, as well as ∼41 Ma alkaline igneous rocks, the latter are used to investigate the δ 11 B variations in the subduction-modified mantle since the onset of continental collision. We then use these new [B] and δ 11 B data, alongside previously published trace element and Sr-Nd isotope data in order to investigate the nature of the slab component.

Geological background
The geology of Armenia is defined by the closure of the Neotethys Ocean which resulted in the Arabia-Eurasia continentcontinent collision. It can be summarised as several accreting terranes overlain by younger volcanic and sedimentary units (Fig. 1b). The accreting terranes are the Mesozoic arc of the Lesser Caucasus and the South Armenian Block (SAB; a Gondwanan microcontinental fragment composed of Proterozoic metamorphic basement and its sedimentary cover; Knipper and Khain, 1980), joined at the Sevan-Akera suture defined by several ophiolite terranes (Rolland et al., 2009). Eventual collision was facilitated by closure of several basins along numerous subduction zones, some of the products of which formed the Mesozoic arc in NE Armenia. Closure of the Northern Neotethys ended with collision of the SAB and the active margin of Eurasia in the form of the arc of the Lesser Caucasus ( Fig. 1) at 70-60 Ma (Rolland et al., 2009). Subduction then jumped to the south until closure of the Southern Neotethys with the collision of Arabia with SAB-Eurasia along the Bitlis-Zagros suture (Fig. 1a) at ∼50-40 Ma (Rolland et al., 2012), although some authors have posited a collision as late as ∼25 Ma (Okay et al., 2010).
The investigated Plio-Pleistocene post-collisional volcanic rocks are from three volcanic highlands of distributed volcanism (Syunik, Vardenis and Gegham) and the large Aragats stratovolcano (Fig. 1b). Volcanic products have basanite to rhyolite compositions, with SiO 2 between 45 and 78 wt% and MgO between 0 and 8 wt%. Most eruptions formed lava flows and/or scoria cones, although the occurrence of some ignimbrite deposits indicates that there have been larger Plinian caldera-forming eruptions in the past. All samples show characteristic negative Nb-Ta and Ti anomalies and positive spikes in Ba, K, Pb and Sr (Fig. 2a). Crustal contamination is known to have not played a role in the petrogenesis of the volcanic rocks (Sugden et al., 2019), necessitating a mantle source modified by subduction. Volcanism is attributed to melting of lithospheric mantle in response to heating due to delamination or relaxation of non-linear geothermal gradients (Sugden et al., 2019). K-Ar and Ar-Ar ages for lava flows, pumice layers and ignimbrites from Syunik, Aragats and Gegham are <1.5 Ma (Connor et al., 2011;Fig. 2. N-MORB (Sun and McDonough, 1989) normalised trace element systematics of (a) Plio-Pleistocene post-collisional (Sugden et al., 2019) and Eocene TAC (b) (Sokół et al., 2018) volcanic rocks from Armenia. Average continental arc basalt is shown for comparison (Kelemen et al., 2003). Note the negative Nb-Ta and Ti anomalies observed in all samples, as well as the positive spikes in large ionlithophile elements and light rare earth elements. In the absence of evidence for crustal contamination in Armenia, such "spiky" trace element profiles are commonly interpreted as derivation of the magmas from a subduction-modified mantle source. The more unusual trace element composition of the TAC samples is likely a result of these samples having intermediate compositions (phonolite), while the post-collisional samples are mafic (trachybasaltic andesite). Lebedev et al., 2013;Ollivier et al., 2010), making these samples definitively post-collisional.
We also present data from the Tezhsar alkaline complex in Armenia (Fig. 1b). This complex contains trachyte-phonolite volcanic rocks and syenite intrusives from a ∼10 km wide shallow plumbing system of an extinct large stratovolcano (Sokół et al., 2018). Although there are some differences in the trace element composition of Tezhsar samples (Fig. 2b) when compared to the post-collision samples, they still have the characteristic (if larger) negative Nb-Ta and Ti anomalies and a general enrichment in large ion lithophile elements and light rare earth elements indicative of a subduction-modified mantle source. An 40 Ar-39 Ar age determination of 41.0 ± 0.5 Ma for the Tezhsar alkaline complex reveals its formation either precedes, or is contemporaneous with, Arabia-Eurasia collision. An earlier 50-40 Ma collision would make Tezhsar syn-to post-collisional, while a 25 Ma Eurasia/SAB-Arabia collision, would mean that Southern Neotethys subduction was still ongoing. Tezhsar magmatism has been attributed to small degrees of decompression melting of subduction-modified lithospheric mantle in response to localised extension (Sokół et al., 2018), rather than the classic flux melting model for subduction zones. Hence, taken together the post-collisional and Tezhsar rocks represent two instances of melting of a previously metasomatised mantle source separated by 40 Myr, allowing us to investigate how the inherited slab component has changed since initial continental collision.
In the same way as volcanic arcs, the slab component must be liberated from a subducting slab and transported by a metasomatic agent. However, because the component is inherited, it must also be stored prior to melting. Here, we use the term "slab component" to refer to both the metasomatic agent and the stored component.

Analytical methods
As one of the most fluid-mobile elements, B and its isotopes are extremely susceptible to surface alteration. For detailed information on sample selection, the extent of alteration in Armenian igneous rocks, and how alteration was avoided, refer to the Supplementary Material.
Most (19) of the samples were prepared for boron isotope analysis at the IGG-CNR Pisa, Italy; ∼0.2 g of sample powder was fused with K 2 CO 3 in platinum crucibles with a 4:1 flux to sample ratio. Boron was then extracted from the fusion cakes by repeated crushing and centrifuging of the cakes in high pH B-free water. It was further purified by passing the solution through anion and cation exchange columns. Anion columns were packed with Amberlite IRA-743 boron-specific anion exchange resin, while cationexchange columns were packed with AG 50W-X8 resin. The procedure used an anion column step, followed by a cation column step and then a final (repeat) anion column step to produce the final purified boron solution, as described by Tonarini et al. (1997). Of those samples prepared at IGG-CNR Pisa, the B isotope composition of the mafic-intermediate Plio-Pleistocene post-collisional samples with the exception of sample 9.31B.04 were measured on a Thermo Scientific TM Neptune series multi-collector (MC)-ICP-MS in Pisa, specially tuned for 11 B/ 10 B analysis (following Foster, 2008). Samples were diluted to contain ∼20 ppb B and were then bracketed with NBS 951 boric acid standard solution of the same concentration, to correct for machine induced mass fractionation. Within run errors are between 0.08 and 0.22 (2σ for this and subsequent errors). Several samples were re-prepared and re-analysed, reproducing the original value to within ±0.5 or better. The accuracy of the measurement was monitored as follows: 28 replicate analyses of NBS 951 gave an average δ 11 B of +0.01 ± 0.41 , 7 replicate analyses of the IAEA standard B1 (seawater) gave an average δ 11 B of +39.38 ± 0.27 (accepted value ∼ +38.6 ± 1.7 ; , and 3 replicate analyses of the JB2 (basalt) gave an average δ 11 B of +7.25 ± 0.57 (accepted value +7.33 ± 0.37 ; Tonarini et al., 2003). All replicate analyses of these standards were performed after they had been processed through the full B separation procedure.
The boron concentrations of these purified boron solutions were also measured during MC-ICP-MS analysis. B solutions at variable concentrations (10, 50 and 100 ppb) were used for calibration, and analysed repeatedly to correct for instrumental drift. These concentrations were converted to sample contents using accurately measured sample weights (error < 0.1%) and reagent volumes (error < 1%). Sample loss during flux fusions and column chromatography is minimal (<2%). For both B concentration and 11 B/ 10 B measurements, analytical blanks were subtracted. Blanks were generally ∼0.2 ppb (sample/blank > 100). Two analyses allow for the accuracy of these concentration measurements to be verified. The standard BCR-2 had a measured concentration of 3.7 ppm (accepted values 4.1 to 4.7 ppm; Menard et al., 2013). A veined gabbro which had previously been measured for [B] at the University of South Florida (sample AM20; [B] = 6.9 ppm; Mc-Caig et al., 2018), had a measured [B] of 6.6 ppm. It is therefore reasonable to assume that concentrations are accurate to within 1 ppm.
The Tezhsar samples, as well as the one mafic sample from the Gegham volcanic highland (9.31B.04, Fig. S1) were prepared in the Table 1 [B] and δ 11 B of Armenian post-collisional and Tezhsar volcanic rocks presented in this study. Sr-Nd isotope compositions including precision and analytical methods can be found in Sokół et al. (2018) and Sugden et al. (2019). same way as the Plio-Pleistocene samples. However, isotopic analyses were by thermal ionisation mass spectrometry (TIMS) using a VG Isomass 54E mass spectrometer at IGG-CNR Pisa following the methods outlined in Tonarini et al. (2001). The accuracy of these measurements was monitored by analysis of the SRM-951 boric acid standard, which had been processed through full column chemistry alongside the samples. Uncertainties on measurements are 0.4 to 0.6 . Boron isotope analyses of Plio-Pleistocene obsidians from the Gegham volcanic highland (Fig. S1) were made using multiple multiplier laser ablation inductively-coupled plasma mass spectrometry (MM-LA-ICP-MS) at the Department of Terrestrial Magnetism (DTM) of the Carnegie Institution of Washington. For the full method description see Savov et al. (2009) and references therein. The accuracy of these measurements was monitored by repeated analyses of NBS 610, 612 and 614 glasses, as well as B-5 (Mt. Etna volcano basalt) and B-6 (Lipari obsidian) homogeneous glass standards (produced at DTM, P = 4 GPa), which yielded an average δ 11 B (±1 ) of −3.59 and −0.95 respectively. This can be compared to respective accepted values of −3.8 ± 2 and −1.6 ± 1.4 , suggesting reproducibility is better than 1 , while within run uncertainties were <1 .
Boron concentrations for Tezhsar and Gegham rhyolite samples were measured on a Perkin Elmer Optima 2000 DV inductively coupled Plasma-Optical Emission Spectrometer (ICP-OES) at the School of Geosciences of the University of South Florida. Samples were fluxed with Na 2 CO 3 in platinum crucibles with lids using a furnace at 1400 • C in a boron-free clean lab environment. Sample preparation techniques followed the methods outlined in Snyder et al. (2005). The analytical blank was measured as 1.5 ppm. The blank-corrected concentration of the JB-3 (basalt) external standard was correct to within 1 ppm (18.8 ppm, vs. accepted value 18 ppm). The similar [B] for the veined gabbro sample (AM20) measured in both South Florida (6.9 ppm) and Pisa (6.6 ppm), confirms that the concentration measurements for post-collisional mafic samples can be compared directly with those of the Tezhsar alkaline complex and post-collision rhyolite samples.
All measurements of boron and its isotopes were made using the same sample powders as were used to measure the major and trace element concentrations, and Sr-Nd isotope ratios (methods in Sugden et al., 2019).

Results
All [B] and δ 11 B values are shown in Table 1; δ 11 B ranges from −5 to +2 (Fig. 3) for post-collision samples, consistently heavier than mid-ocean ridge basalts (MORB; −7.1 ± 0.9 ; Marschall et al., 2017). There is no consistent variation with geographic position in the post-collision samples: samples from Aragats (−5 to −3 ), Gegham (−4 to +2 ), Vardenis (−3 to 0 ) and Syunik (−4 to +1 ) all show a similar δ 11 B range (Table 1), confirming previous observations that despite changes in lithospheric thickness, the slab contribution is uniform across Armenia (Sugden et al., 2019). The B/Nb in mafic samples varies from 0.03 to 0.25, i.e. lower than in any modern volcanic arc (Fig. 3;De Hoog and Savov, 2018), and in fact overlapping with the range of MORB (0.15-1.05; Marschall et al., 2017). This suggests a fluid-mobile element depleted source when compared to the sources of arc volcanism. The δ 11 B of the studied rocks are similar to those from hot arcs such as the Cascades (circled red in Fig. 3; −21.3 to −0.4 ; Leeman et al., 2004;Rose et al., 2001;Walowski et al., 2016), or the intraplate volcanoes of the Oregon-Snake River Plain-Yellowstone region (−8.9 to −0.8 ; Savov et al., 2009), both shown to represent melting of fluid-starved sources. B/Nb ratios are higher in the rhyolite samples by an order of magnitude (0.75-1.5; Fig. 3), but the δ 11 B of the rhyolites is comparable (−4 to +2 vs −5 to +1 for mafic samples; Fig. 3).  (Table 1). Tezhsar samples have higher B concentrations (4-20 ppm) and B/Nb ratios (0.2-0.8) than the post-collision samples, although B/Nb is still at the lower end for arcs generally (Fig. 3). Surprisingly, magma sources tapped at 41 Ma and <1.5 Ma have comparable Sr-Nd-B isotopic characteristics (Fig. 4) -with overlapping 143 Nd/ 144 Nd (0.51275-0.51286), and only slightly lower 87 Sr/ 86 Sr in Tezhsar samples (initial values 0.7040-0.7044 vs 0.7042-0.7049). The lower B/Nb and 143 Nd/ 144 Nd; and higher 87 Sr/ 86 Sr of these collision-related magmas when compared to arc rocks, means they define a geochemical reservoir distinct to volcanic rocks from both arcs and oceanic ridges/islands.

Origin of the subduction signature
Prior to this study, Western Anatolia was the only region in the world where young (17-0 Ma) volcanic rocks, which erupted after subduction ceased had been studied for [B] and δ 11 B. Western Anatolia has experienced rapid geodynamic changes over the past 25 Myr, from subduction to an extensional setting , which although not a continental collision, does provide an analogue of what happens when subduction ends. Here, 23-17 Ma calc-alkaline rocks have δ 11 B (−7.1 to −0.1 ) and B/Nb which extend from MORB to more arc-like values (Fig. 3), reflecting their formation during subduction. However, 17-14 Ma ultrapotassic rocks have very light δ 11 B (−15.1 to −11.2 ), interpreted to reflect the progressive dehydration of a stalled slab (Agostini et al., 2008), which is also shown by the gradual reduction in B/Nb (Fig. 3). . The dehydration preferentially removes 11 B, giving the residual slab an increasingly light δ 11 B (Ishikawa and Tera, 1997). These samples provide a useful comparison for the likely effects of a stalled slab on δ 11 B and B/Nb with the Armenian samples presented in this study.
The δ 11 B of post-collisional volcanic rocks from Armenia (−5.2 to +1.7 ) is not consistent with a model in which a stalled slab progressively dehydrates during collision-related slab break-off. Instead, in Armenia δ 11 B has exhibited consistently heavier than MORB values over the past 41 Myr, on the basis of the postcollisional (−5.2 to +1.7 ) and Tezhsar alkaline complex samples (−8.7 to −3 ). Similarly, mixing with an intraplate mantle source (OIB ∼ −10 ; Walowski et al., 2019), which had previously been underneath (and therefore unaffected by) a subducting slab would produce a lighter δ 11 B in the post-collisional volcanic rocks compared to the older Tezhsar samples. In fact, post-collision samples have slightly heavier δ 11 B (Fig. 3). Both mixing with intraplate magmas and dehydration of a stalled slab would also result in more variable trace element and Sr-Nd isotope compositions, which is not observed.
While slab break-off is not considered a prominent process under Armenia, it could well be an important process elsewhere in the collision zone. In particular the Lake Van region of Eastern Anatolia, where alkaline magma compositions are observed, suggestive of OIB-type mantle from below the slab contributing to the magma source (Keskin, 2003;Pearce et al., 1990).
Volcanism in Western Anatolia continued after the eruption of the 17-14 Ma ultra-potassic rocks, forming two magmatic series: Na-alkaline and K-alkaline (Figs. 3 and 4). Despite the similar δ 11 B of the Na-alkaline volcanic rocks (−2.7 to −0.8 ) to Armenian post-collisional samples (Fig. 3), their higher 87 Sr/ 86 Sr and lower 143 Nd/ 144 Nd (Fig. 4), along with their OIB-like trace element geochemistry , suggests a distinct petrogenesis. They have been interpreted as melts derived from an un-modified asthenospheric mantle . However, the δ 11 B The low δ 11 B of the older K-alkaline sample, might suggest that at this time a stalled slab is still influencing magma δ 11 B (Agostini et al., 2008;Tonarini et al., 2005). The rebound to heavier values in the younger samples, with lower B/Nb (Fig. 3) is explained by the end of the influence of any subducting slab , such that B comes to reflect the composition of the mantle which remains. In the case of these younger K-alkaline rocks, the subduction component in their mantle source must have been inherited. The similar geochemistry of the Armenian volcanic rocks to these K-alkaline rocks suggest that they too have an inherited subduction component.
It seems most likely that the subduction signature observed in the Armenian post-collisional and Tezhsar volcanic rocks, has been stored for at least 41 Myr, to be inherited during later melting events. The slab component is mostly likely to have been stored in the lithospheric mantle, where cooler temperatures (due to a conductive geotherm) stabilise metasomatic minerals able to store the slab component long after subduction has ceased (e.g. Mandler and Grove, 2016). Amphibole, rather than phlogopite, is the most likely such mineral, based on the high Ba/Rb (20-40) and low Rb/Sr (0.01-0.04) ratios of the post-collisional volcanic rocks (Sugden et al., 2019). This is because Rb is an order of magnitude more compatible in phlogopite compared to amphibole (LaTourrette et al., 1995). Temperatures of <1100 • C would be required to stabilise  Neill et al. (2015) and Sugden et al. (2019). Also shown is the MORB field with higher Dy/Dy* and Ti/Ti* (after Davidson et al., 2013 and references therein). Dy/Dy* is a measure of the curvature of a rare-earth element profile, whereas Ti/Ti* is a measure of the size of the Ti anomaly on a mantle-normalised trace element pattern (Davidson et al., 2013). pargasitic amphibole (Mandler and Grove, 2016), showing that the slab component must be stored in the mantle lithosphere.
A residual amphibole phase is supported by the positive correlation between Dy/Dy* and Ti/Ti* ( Fig. 5; Davidson et al., 2013). Only amphibole and clinopyroxene preferentially partition the middle rare earth elements, decreasing the Dy/Dy* ratio of any melt equilibrating with these phases. The positive correlation with Ti/Ti* confirms that amphibole must be a residual phase in the source of the post-collisional magmas because Ti is an order of magnitude more compatible in amphibole than in pyroxene (Davidson et al., 2013). Thus, amphibole is able to store slab-derived boron for >41 Myr, soaking up metasomatic components to be released during later melting episodes caused by partial amphibole breakdown (Sugden et al., 2019).
The low [B] of Armenian post-collisional volcanic rocks (1-5 ppm) relative to arc volcanic rocks is consistent with the low [B] observed in vein amphiboles in sub-arc mantle xenoliths from Kamchatka (0.2-3 ppm; Tomanikova et al., 2019). An amphibole source for the subduction signature would be expected to produce magmas low in B given low experimental partition coefficients for boron in amphibole (Brenan et al., 1998), explaining its low capacity for concentrating boron in its structure during the storage of This stored slab component is reminiscent of the "amphibole sponge" concept of Davidson et al. (2007), where it was argued that "cryptic" amphibole fractionation in the lower crust provides a hydrous filter on ascending arc magmas. It was also argued that subsequent breakdown of amphibole in these lower crustal cumulates can lead to the production of intra-crustal melts of intermediate-felsic composition (Davidson et al., 2007). However, with SiO 2 contents as low as 45 wt% and Sr-Nd isotope compositions on the mantle array, the Armenian post-collisional magmas are clearly mantle-derived (Sugden et al., 2019). Magma generation occurs above the lithosphere-asthenosphere boundary (∼120 km; Priestley et al., 2012). This is on the basis of evidence from rare earth elements for both garnet and spinel in the melt residue, and major element geothermobarometry revealing melting at 50-80 km depth and at temperatures of ∼1100 • C, consistent with melting in the lithosphere rather than the asthenosphere (Sugden et al., 2019). Such temperatures are also consistent with amphibole breakdown melting (Mandler and Grove, 2016), which is argued to occur at both the base and interior of the mantle lithosphere (Sugden et al., 2019).
Hence the amphibole sponge of Armenian post-collisional magmatism is in the mantle lithosphere rather than in the lower crust. Rather than arc melts, this sponge filters a slab-derived component. Upon subsequent amphibole breakdown, melting of this sponge creates post-collisional mafic magmas rather than felsic intra-crustal melts.

Origin of the subduction signature in high boron post-collisional rhyolites and Tezhsar volcanic rocks
The high B/Nb of two Tezhsar alkaline complex samples (Fig. 3), and the lower δ 11 B of one Tezhsar sample (−8.7 ; Fig. 3) are similar to the volcanic rocks from "hot" subduction zones ( Fig. 3; Leeman et al., 2004;Savov et al., 2009). This could suggest the subduction signatures in the Tezhsar samples are a mixture of the component observed in the post-collision samples, and a component derived from contemporaneous addition of material from a partially dehydrated slab. The Tezhsar volcanic rocks formed much closer to the time of Mesozoic-Paleogene active slab subduction (Mederer et al., 2013;Sokół et al., 2018), so the observation of a second slab component only in the older Tezhsar samples is reasonable. Tezhsar is analogous to the oldest of the K-alkaline lavas from Western Anatolia (δ 11 B = −12.9 ) when some of the depleted stalled slab material was still present (Agostini et al., 2008). Despite the higher B/Nb of the rhyolite post-collision samples, the matching Sr-Nd-B isotope characteristics of both mafic and felsic samples illustrates that the magma source of the parental magma to these rhyolites is the same long-lived subductionmodified mantle source which supplied the mafic magmas. These isotope characteristics also confirm that crustal contamination did not play a role in the petrogenesis of the rhyolites, meaning that B/Nb varies with the extent of crystal fractionation.
[Nb] is invariant in the remaining melt during fractional crystallisation, whereas [B] increases by an order of magnitude (Table 1). Nb must partition into a crystallising phase, while B remains in the liquid. Here, a reduction in Nb/Ta with SiO 2 content in these same magmas (22-27 in basalts vs. 6-19 in rhyolites) offers a clue. Fractionation of magmatic rutile in the lower crust produces high Nb/Ta cumulates (Tang et al., 2019). A small amount of rutile fractionation may be enough for Nb to be buffered in the remaining melt, while Ta remains an incompatible element.

Nature of the slab component and its metasomatic agent
The slab signature in arc magmas is generally imparted by aqueous fluids, melts or supercritical fluids liberated from a subducting slab and overlying sediments. Here, melts and supercritical fluids (MSF) will be treated as interchangeable, given the similar trace element partitioning behaviour of the two (Kessel et al., 2005).   (Prelević et al., 2008). Post-collisional samples are all those samples with >4 wt% MgO and <54 wt% SiO 2 from Syunik and Vardenis volcanic highland (Sugden et al., 2019), Gegham volcanic highland (Savov, unpublished), Aragats volcano (Connor et al., 2011), and Northern Armenia (Neill et al., 2015). Also shown are Mesozoic-Paleogene arc samples from the Pontides of Eastern Turkey (Aydinçakir and Sen, 2013) and the Kapan Zone of South Armenia (Mederer et al., 2013). These samples overlap with the post-collision samples in Th/Rb but extend to lower values likely to reflect greater fluid input to the mantle source. Six arc samples which have very high Th/Rb of up to 2.5 are not shown on this figure, because they are interpreted to reflect post-emplacement alteration and leaching of Rb (Aydinçakir and Sen, 2013). fractionation model for the composition of fluid released from an increasingly dehydrated slab is constructed (Figs. 3 and 4). Coupled thermodynamic and thermomechanical modelling of subducting lithosphere can constrain the temperature of the subducting slab, as well the phases which store B: likely to be phengite, chlorite and amphibole (Konrad-Schmolke and Halama, 2014). On this basis, a 10 fluid-residue fractionation factor is assumed (Konrad-Schmolke and Halama, 2014;Wunder et al., 2005). As the slab dehydrates, the fluids released have δ 11 B and B/Nb which diminish with depth (Fig. 3). The δ 11 B and B/Nb of arc rocks (but not the Armenian volcanic rocks), can be explained by mixing between depleted (MORB) mantle and fluids from a variably dehydrated slab (Fig. 3). 87 Sr/ 86 Sr and 143 Nd/ 144 Nd would not vary during dehydration (horizontal trajectories in Fig. 4). An aqueous fluid derived from a previously dehydrated slab could explain the Sr-B isotope composition of the post-collisional magmas (Fig. 4a), but not the lower 143 Nd/ 144 Nd ratios (Fig. 4b).
The metasomatic agent must instead be an MSF derived from subducted sediments or oceanic crust, in order to mobilise Nd and modify the 143 Nd/ 144 Nd of the mantle source. Adakites are thought to be geochemically characterised by an MSF derived from oceanic crust. They have MORB-like Sr-Nd isotope ratios, rather than the high 87 Sr/ 86 Sr and low 143 Nd/ 144 Nd of the post-collisional rocks (Fig. 4). Moreover, although Armenian post-collisional magmas possess the high Sr/Y ratios (15-130) characteristic of adakites, they lack the high primary SiO 2 contents, and have higher Y and Yb abundances (Castillo, 2012).
This leaves a sediment MSF as the carrier of the slab component in the post-collisional rocks, supported by their low 143 Nd/ 144 Nd. A sediment MSF is also supported by the similar Th/Rb (Fig. 6) in the volcanic rocks (average for mafic samples ∼ 0.16 ± 0.06 (2 SD)) and Tethyan flysch sediments (avg. 0.13 ± 0.2; Prelević et al., 2008). For Th/Rb to reflect local subducted sediments, the metasomatic agent must be: a) derived from the sediment, and b) an MSF for Th and Rb to partition similarly between the mobile and the residual phases, given that Th is fluid-immobile (Johnson and Plank, 2000). This is confirmed by fluid-eclogite partitioning experiments, which show that even at high salinities, Rb is typically enriched in the fluid by around 2 orders of magnitude more than Th (Rustioni et al., 2019). In addition, given the relatively low Th content of the oceanic crust (Kelley et al., 2003), the high Th contents of post collisional samples (comparable to the Tethyan sediments) suggest derivation of the MSF phase from the slab sediments (Fig. 6).
The δ 11 B of this sedimentary component is likely to be in the range of −5 to +2 (Fig. 4) based on the δ 11 B of post-collision samples. This can be compared to the δ 11 B of metasediments.
Metamorphosed terrigenous sediments (Catalina schist, California and Lago di Cignana, Italy) have much lighter δ 11 B compared to the Armenian volcanic rocks (−7 to −15 in tourmalines; Bebout and Nakamura, 2003). However mixed marine-terrigenous sediments from Syros, Greece, have comparable δ 11 B to the Armenian samples (−1.6 to +0.9 in tourmaline prograde mantles; Marschall et al., 2008). It is probably the case that there is no δ 11 B fractionation during separation of a melt phase from the residual metasediment, given the tetrahedral co-ordination of B in both magma and most silicate minerals (Maner and London, 2018). A melt of the mixed marine-terrigenous metasediment could produce the slab signature observed in Armenian post-collisional magmas. Release of a supercritical fluid could involve none or only limited isotope fractionation given the likely high temperatures (Konrad-Schmolke and Halama, 2014). If instead separation of a supercritical fluid did involve isotope fractionation, then the isotopically lighter, dominantly terrigenous-derived sediment could be a more viable source for the slab component's metasomatic agent.
It is possible to construct a mixing model in an attempt to quantify the contributions of this sediment MSF and the un- (2) X SC is the total mass fraction of the slab component in the magma source. Based on mixing models for Sr-Nd isotope data from postcollisional volcanic rocks in Armenia and neighbouring north-west Iran, 1% is a reasonable estimate (Allen et al., 2013;Sugden et al., 2019).
[B] UM is the concentration of boron in the previously unmodified mantle. Given there is no evidence for a deep mantle plume in this region, the depleted mantle is a reasonable approximation here (0.077 ppm; Marschall et al., 2017). This gives [B] SC of between 9.3 and 30.5 ppm. This range is below the bulk-rock value for the Syros metasediment sample used to constrain δ 11 B SC (94 ppm; Marschall et al., 2008). It is possible that these metasediments owe their elevated B to the influx of B-rich fluids at retrograde conditions (Marschall et al., 2008). When compared with un-metamorphosed subducting sediments, [B] SC is at the lower end of the range (5 to 130 ppm; Plank, 2014). This is to be expected, given that subducting sediments will release much of their B during dehydration in the fore-arc (Snyder et al., 2005), although any melting event would give a melt with a higher [B] than the  Fig. 6. MORB values after Sun and McDonough (1989). source, given the incompatibility of B (D solid/melt ∼ 0.1; Kessel et al., 2005).
It is now possible to estimate [B] of the magma source: Syros could react with the mantle lithosphere to produce the postcollisional magma source. Ratios such as Ba/Th, Pb/Ce and U/Nb (fluid-mobile element/immobile element of similar compatibility during melting) are higher and more variable in Mesozoic-Paleogene arc rocks from Armenia and Eastern Anatolia (Fig. 7; Aydinçakir and Sen, 2013;Mederer et al., 2013) compared to the relatively young post-collision rocks investigated in this study. This indicates that a fluid component was present during subduction, but was not inherited by magmas formed after the continental collision. This points to the sediment MSF being much more long-lasting in the mantle, whereas the aqueous fluid component is transitory. Indeed, studies of U-Th series isotopes in Kamchatka suggest the aqueous fluid component can travel from slab to surface in 1-300 kyr, whereas a sediment component can take 350 kyr -4 Myr to make the same journey (Turner et al., 2000). It could be the case that once the slab stalls the aqueous fluids are quickly removed, such that the subsequent slab component is dominated by sediment MSF.

Concluding remarks
High Ba/Th, Pb/Ce and U/Nb in Mesozoic-Paleogene Tethyan arc magmas indicate the mantle source of these magmas was modified by aqueous fluids. However, the δ 11 B of post-collisional volcanic rocks indicates derivation from a mantle source modified by a fluid-starved slab component. There is also no evidence for continued dehydration of a stalled slab following collision, which would be expected to produce the even lighter δ 11 B, as observed in the 16-17 Ma samples from Western Anatolia. Instead the similar δ 11 B and Sr-Nd isotope ratios in post-collisional and the Tezhsar alkaline complex volcanism indicate that the slab signature was inherited from preceding subduction. The lack of the heavy δ 11 B signature of aqueous fluids being a result of them being transitory and not surviving to impart their signature on post-collisional Armenian mantle sources. For this subduction component to be inherited, it needs to be stored since at least the 41 Ma volcanism recorded at Tezhsar. Amphibole is likely to be the storage repository for the slab component based on low Rb/Sr and co-variation in Dy/Dy* vs. Ti/Ti* in post-collisional volcanic rocks. This is in agreement with the low [B] in post-collisional volcanic rocks, because amphibole has a modest storage capacity for B. Higher B/Nb in Tezhsar compared to post-collisional volcanic rocks (similar to B/Nb in "hot" arcs) and slightly lower δ 11 B may suggest that an inherited slab component was mixed with a component derived from a contemporaneously subducting slab. The more clustered Sr-Nd-B isotope compositions of the post-collisional samples suggests that by the Plio-Pleistocene only the inherited slab component remains. The sediment MSF origin of the slab component suggested on the basis of trace elements and Sr-Nd isotopes may be consistent with δ 11 B in post-collisional volcanic rocks, given similar values for some dehydrated metasediments.
The lower 143 Nd/ 144 Nd in arc rocks compared to fore-arc serpentinites (Fig. 4) is perhaps a "smoking gun" for the presence of this sediment MSF component in all arc rocks, but their heavy δ 11 B is testament to it being obscured by a dominant aqueous fluid component. As a setting in which only sediment MSF metasomatises the mantle source, volcanism in continent-continent collision zones is an ideal setting to separate the effects of slab-derived sediment MSF and aqueous fluids in subduction zones.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.