Titanium isotope heterogeneity in the Earth's mantle: A case study of the Horoman peridotite massif

Titanium isotopes are a promising novel tracer for magmatic processes. The Ti isotope composition of the Earth’s mantle is key for using this tracer on a planetary scale, however, recent studies point to potential Ti isotope variations in mantle rocks. This study presents stable Ti isotope data for well-characterised peridotites from the Horoman massif (Japan) as well as for primitive, orogenic (Variscan and Alpine belts) and South African ultrapotassic rocks. The peridotites originate from a continuous section across the layered Horoman peridotite body and comprise highly depleted harzburgites to fertile lherzolites as well as metasomatically overprinted peridotites. Unlike most primitive mantle-derived magmas, which display very limited Ti isotope variations, d 49 Ti (deviation of 49 Ti/ 47 Ti from the OL-Ti standard) values of the Horoman peridotites vary from –1.523 ± 0.029 to 0.547 ± 0.015 ‰ (2SD), spanning a total d 49 Ti range of 2.07 ‰ . Heavy, continental crust-like Ti isotope compositions are correlated with Nb/Th and are likely related to metasomatism of the mantle wedge above the Hidaka subduction zone. In such a scenario, iso-topically heavy Ti was inherited from the subducted terrigenous sediments and might have been further enhanced by Ti isotope fractionation during mobilisation from the slab. Signiﬁcant d 49 Ti differences between coexisting clino-and orthopyroxene ( D 49 Ti opx-cpx = 0.16–0.29 ‰ ) and notably light Ti isotope compositions ( d 49 Ti = (cid:1) 1.523 ± 0.029 to (cid:1) 0.677 ± 0.036 ‰ , 2SD) recorded in highly refractory peridotites might be related to kinetic diffusion-driven isotope fractionation during ﬂuid/melt percolation. The analysed ultrapotassic rocks display distinctly high d 49 Ti values (0.114 ± 0.040 to 0.290 ± 0.030 ‰ , 2SD) compared to the majority of primitive mantle-derived magmas. The observed enrichment in heavy Ti isotopes likely originates from recycled continental material characterised by high d 49 Ti, in line with their derivation from strongly metasomatised mantle sources. Our results demonstrate the presence of signiﬁcant small-scale Ti isotope heterogeneity within the upper mantle. (cid:1) 2022 The Authors. Published by Elsevier Ltd. ThisisanopenaccessarticleundertheCCBYlicense(http://

Partial melting and subduction recycling introduce mineralogical and isotope-geochemical heterogeneities within the mantle.However, compared to the comprehensive Ti isotope studies of mantle-derived magmas, peridotites have received little attention so far.While an orogenic lherzolite and subduction zone serpentinites displayed no d 49 Ti deviation from primary mantle-derived melts (Millet et al., 2016), a limited dataset of orogenic lherzolites and dunites revealed more variable Ti isotope compositions (Mandl, 2019).This study aims to constrain the origin of the observed small-scale mantle heterogeneity by determining the Ti isotope compositions of orogenic peridotites.Here, we report Ti isotope data for well-characterised samples from the Horoman Massif (Japan), comprising (i) fertile plagioclase lherzolites, (ii) depleted harzburgites, and (iii) metasomatically overprinted peridotites.To further constrain the impact of metasomatism on the Ti isotope composition of the mantle, we analysed juvenile magmatic derivatives of a metasomatised mantle, i.e. ultrapotassic rocks (lamproite, orangeite, mica-amphibole-rutile-ilmenite-diopside (MARID)xenolith).In addition, we present Ti isotope data for two international geological standard reference materials, UB-N and WPR-1a.
The Horoman peridotites were metasomatised by slab-derived fluids or melts in the mantle wedge of the Hidaka subduction zone.Based on Rb-Sr isotope data, incorporation into the mantle wedge likely took place at < 150 Ma (Ranaweera et al., 2018).Cryptic metasomatism is locally indicated by an enrichment in light rare earth elements (LREE) and large ion lithophile elements (LILE) (i.e.Ba, Rb, Cs) as well as radiogenic Pb isotope signatures (e.g.Malaviarachchi et al., 2008;Takazawa et al., 2000;Yoshikawa and Nakamura, 2000).Some peridotites show evidence for latestage modal metasomatism, documented by the presence of phlogopite.Rubidium-Sr dating of a phlogopite-bearing spinellherzolite yielded a late Miocene age of $ 23 Ma (Yoshikawa et al., 1993).Furthermore, Lai et al. (2015) reported anomalously light Li isotope compositions that may be related to kinetic fractionation during melt infiltration.

Serpentinised peridotites
We measured the reference material UB-N from the Association Nationale de la Recherche Technique (ANRT).This is a serpentinised fertile garnet-spinel-peridotite from Col de Baganelles, Vosges mountains (France) (Govindaraju, 1982;Meisel et al., 2003).We also analysed the reference material WPR-1a, provided by the Canada Centre for Mineral and Energy Technology (CANMET), an altered peridotite from the Wellgreen Complex, Yukon Territory (Canada) (Centre for Mineral and Energy Technology, 1994).

Orogenic ultrapotassic rocks
Ultrapotassic magmatism is common in collisional orogens such as the Alpine-Himalayan or the European Variscan orogenic belts (e.g.Conticelli et al., 2009;Soder and Romer, 2018;Zhao et al., 2009) and is assumed to originate from metasomatised domains within the lithospheric mantle.The mantle-derived rocks (high Mg# and high Ni and Cr contents) are strongly enriched in incompatible trace elements and have continental crust-like radiogenic isotope signatures, indicative of high amounts of recycled continental crust in their mantle source (e.g.Conticelli et al., 2009;Prelevic ´et al., 2008;Soder and Romer, 2018).We studied two primitive lamproites from classic localities of the Alpine belt (Murcia-Almeria, Spain and Tuscany, Italy) as well as one primitive ultrapotassic rock (durbachite) from the Variscan orogenic belt (Schwarzwald, Germany) for their Ti isotope composition.Sample localities as well as whole-rock data are given in Table S2.

Sample preparation and ion-exchange chemistry
Sample powders were prepared using an agate disc mill or mortar.Pyroxene separates from two plagioclase-lherzolites (BZ-146L and BZ-254) were handpicked from two subsamples that were crushed separately from the whole rock.In addition, spinelbearing aggregates macroscopically composed of olivine, spinel, ortho-and clinopyroxene were handpicked from subsample BZ-146L-S.While the powders used for bulk rock Ti isotope analysis represent aliquots from 200 to 500 g powdered peridotite (Takazawa et al., 2000), approximately 50-100 g material was processed for mineral separation.Up to 200 mg of rock powder and 80 mg of the mineral separates/aggregates were dissolved in Parr Ò acid digestion bombs following the procedure of Schönbächler et al. (2004).Samples were treated with a mixture of 3 ml concentrated HF and 0.7 ml concentrated HNO 3 and placed in Parr Ò bombs in a ventilated oven at 170 °C for 4.5 days.Subsequently, the samples were dried down, taken up in 5-20 ml 6 M HCl and refluxed overnight to decompose precipitated fluorides.Samples were centrifuged to verify complete dissolution.Chemical purification of Ti followed the protocol of Williams et al. (2021), modified after Schönbächler et al. (2004) and Leya et al. (2007).After determining Ti concentrations using a Thermo Scientific Element inductively coupled plasma mass spectrometer (ICP-MS), sample aliquots containing 1.5-20 lg Ti were mixed 1:1 with a 47 Ti-49 Ti double-spike.To ensure sample-spike equilibrium, spiked aliquots were refluxed overnight, dried down and redissolved in 6 M HCl twice.Samples were processed via a threestep column chemistry procedure using Bio-Rad AG1-X8 anion exchange resin (100-200 and 200-400 mesh, chloride form).The first step employed 100-200 mesh resin and eliminated the majority of matrix elements, concentrating Ti, Hf and Zr.During the second step, Ti was efficiently separated from Zr and Hf.Subsequently, the first column chemistry step was repeated, slightly modified using 200-400 mesh resin, in order to attain sufficient purity level of the final Ti fraction, minimising isobaric (Ca, V, Cr) and polyatomic (S) interferences.For aliquots containing < 2 lg Ti, another passage through this step was required to further reduce Ca and S impurities.The total procedural blank was 1-1.5 ng Ti, accounting for a maximum blank contribution of 0.1 %.

Mass spectrometry
Titanium isotopic analyses were performed on a Thermo Scientific Neptune Plus multi-collector ICP-MS housed at ETH Zürich.Samples and standards dissolved in 0.5 M HNO 3 -0.015M HF were introduced via a Cetac Aridus II desolvating nebuliser system.Standard sample and x skimmer cones were used.Measurements were carried out in high mass resolution with a mass resolving power > 11,000 [m/(m 0.95 -m 0.05 )], except for aliquots with low Ti contents < 2 lg, where the instrument was operated at a medium mass resolution of % 7,000.To avoid polyatomic interferences, measurements were performed on the low-mass peak-shoulder.Samples were measured relative to the in-house Ti Alfa Aesar wire standard (Williams et al., 2021) and concentrations for samples and standard were matched within 10 %.Isotope data were acquired in 40 measuring cycles (8.4 s integration time) in static mode, using 10 11 X amplifiers for 46 Ti, 47 Ti, 48 Ti, 49 Ti and 50 Ti and 10 12 X amplifiers for 44 Ca and 90 Zr ++ , which is monitored at half mass 45.Backgrounds were recorded in pure 0.5 M HNO 3 -0.015M HF solution.Background-and Ca-corrections were performed offline.Data was processed offline following the procedure of Siebert et al. (2001) using the signals of 46 Ti, 47 Ti, 48 Ti and 49 Ti.The Ti isotopic composition is expressed in the delta notation relative to the in-house Ti wire standard using standard-sample bracketing: d 49 Ti TiÀAA Wire ð‰Þ ¼ 49 Ti= 47 Ti Sample 49 Ti= 47 Ti Ti Alfa Aesar wire À 1 Â 1000 The Ti isotope data, that is denoted as d 49 Ti, is re-scaled to the OL-Ti standard by applying a systematic offset of 0.224‰ of OL-Ti relative to the Ti Alfa Aesar wire standard (Mandl, 2019;Williams et al., 2021; Table S3).Our average analytical precision was 0.024‰ (2SD), determined on United States Geological Survey (USGS) reference material BHVO-2 (n = 27; Table S3).To confirm data quality, replicate digestions and analyses were carried out (Fig. 3b, Table 1).To assess the influence of processing small quantities of Ti on data quality, aliquots of a peridotite sample (UB-N) with selected amounts of Ti (1.5-20 lg) were processed through the ion exchange procedure.The Ti isotope data of the different aliquots overlap within our analytical precision (Table S4).The obtained Ti isotope compositions for the Durham peridotite standard GP13 and BHVO-2 are within uncertainties identical to published data (e.g.Deng et al., 2019;Millet et al., 2016;Fig. 3a).

Major element-and Ti concentration data
Major element analyses of ultrapotassic rocks were carried out on fused glasses by wavelength-dispersive X-ray fluorescence spectrometry (WD-XRF) using a PANalytical AXIOS instrument at ETH Zürich.Titanium concentration data reported in Tables 1 and  2 are derived by isotope dilution as a by-product of the doublespike analysis.A conservative estimate of uncertainty is 1 percent or less and errors are dominated by the weighing uncertainties.
To further constrain the origin of the Ti isotope variations on a bulk rock level, coexisting clino-and orthopyroxene extracted from two pl-lherzolites were analysed for their Ti isotope composition.Additionally, fine-grained aggregates macroscopically composed of spinel, olivine, ortho-and clinopyroxene were analysed in one sample.Orthopyroxene in both samples displays distinctly higher d 49 Ti values than clinopyroxene (Table 2, Fig. 4b), with D 49 Ti opx- cpx = 0.16-0.29‰.The spinel-bearing aggregates from BZ-146L-S are heavier than the respective pyroxene separates and display a Ti isotope composition that is close to the bulk value (Table 2).We also note a discrepancy between the Ti isotope composition obtained for aliquots from powders studied by Takazawa et al. (2000) and powders prepared from hand specimen used for mineral separation (Fig. 4b), indicating substantial heterogeneities on a centimetre-scale.
Fig. 3. Titanium isotope data of (a) USGS reference material BHVO-2 and peridotite GP13 obtained in this study and from the literature (Deng et al., 2019;Greber et al., 2017a;He et al., 2020;Millet et al., 2016) given relative to the OL-standard.(b) Replicate digestions of five peridotites and a lamproite show good reproducibility.Uncertainties are 2SD (this study; He et al., 2020), else 95 % CI.

Discussion
The Horoman peridotites have highly variable Ti isotope compositions, exhibiting peculiarly light d 49 Ti values as well as a prominent spread towards heavy, continental crust-like Ti isotope compositions (Fig. 4a).The Horoman harzburgites and lherzolites represent melting residues formed beneath a mid-ocean spreading ridge that have been subsequently affected by metasomatism in the mantle wedge and exhumed to crustal levels (e.g.Yoshikawa and Nakamura, 2000).Thus, several processes are superimposed that may have modified the Ti isotope compositions of the samples.Ultrapotassic rocks, which are derived from a mantle source metasomatised by continental crust material, are characterised by a marked enrichment in heavy Ti isotopes (Fig. 7).In the following sections, we first discuss the heavy Ti isotope compositions of the peridotites and ultrapotassic rocks with regard to mantle meta-somatism and crustal recycling.Subsequently, the anomalously light d 49 Ti values are discussed with respect to partial melting and melt percolation.
5.1.Titanium isotope heterogeneity in the mantle -the role of mantle metasomatism and crustal recycling

Horoman peridotites
Most mantle-derived magmas (i.e.basalts of different tectonic settings and komatiites) exhibit a narrow range in d 49 Ti scattering around 0‰ (Fig. 7; Deng et al., 2018a;Greber et al., 2017b;Millet and Dauphas, 2014;Millet et al., 2016).By contrast, peridotites have variable Ti isotope compositions -harzburgites and lherzolites from the Horoman massif span a d 49 Ti range of 2.07‰, including both lighter and heavier values relative to mantle-derived melts.Heavy Ti isotope compositions of 0.129 ± 0.027 to 0.547 ± 0 The Ti isotope data were measured relative to the Ti Alfa Aesar wire standard (d 49 Ti Ti-AA Wire ) and are reported relative to the OL-Ti standard (d 49 Ti) for comparison.n: number of analyses..015‰are akin to the upper continental crust (Fig. 5).The high d 49 Ti values are negatively correlated with Nb/Th (Fig. 6c) and indicate the addition of a crustal component.The Horoman mantle section was affected by at least two distinct metasomatic events -an early event documented by cryptic metasomatism of the peridotites and a late one resulting in phlogopite formation (e.g.Takazawa et al., 1992Takazawa et al., , 2000;;Yoshikawa and Nakamura, 2000).Heterogeneous enrichment in LREE and LILE, as well as phlogopite crystallisation were likely related to subduction of terrigenous sediments and metasomatism of the mantle wedge above the Hidaka subduction zone (Malaviarachchi et al., 2008(Malaviarachchi et al., , 2010;;Yoshikawa et al., 1993, Yoshikawa andNakamura, 2000).(Meta-) pelitic and -psammitic rocks crop out in the Hidaka metamorphic belt (Komatsu et al., 1994;Osanai et al., 1991) and it is likely that sediments of similar composition were recycled during subduction (Yoshikawa and Nakamura, 2000).The observed heavy Ti isotope compositions of the studied peridotites are thus likely related to metasomatism by crust-derived fluids/melts.Solubility of Ti in aqueous fluids is generally low, but may be significantly enhanced in the presence of alkalis, Al and Si (e.g.Antignano and Manning, 2008;Manning et al., 2008;Mysen, 2019), and increasing fluid peralkalinity (Manning et al., 2008).Thus, fluids released by dehydration of felsic lithologies may act as efficient solvents for Ti.However, experimental studies suggest subsolidus aqueous fluids are dilute (e.g.Hermann and Rubatto, 2009;Manning, 2004;Spandler et al., 2007) and, hence, ineffective in transporting trace elements from the slab to the mantle wedge (e.g.Hermann et al., 2006).Instead, hydrous granitic melts produced by fluid-fluxed sediment melting (e.g.Hermann and Rubatto, 2009;Skora and Blundy, 2010) or supercritical fluids released at higher depths and pressures > 5 GPa (e.g.Mann and Schmidt, 2015 and references therein) are capable of dissolving high amounts of solutes.Titanium isotope fractionation is driven by differences in the bonding environment (bond stiffness, coordination number [CN]).At thermodynamic equilibrium, heavy isotopes will be preferentially incorporated into sites with stronger bonds, i.e. sites with a lower CN (e.g.Schauble, 2004).Whether the isotopic composition is directly transferred to the fluid/melt phase during sediment dehydration or partial melting is thus dependent on Ti coordination in the coexisting phases (minerals vs fluid/melt).The Ti budget in sediments subjected to eclogitefacies metamorphism during subduction will be mainly controlled by rutile and subordinate by phengite, garnet and clinopyroxene (e.g.Hermann and Spandler, 2008), where Ti is 6-fold coordinated at relevant P-T conditions (Ackerson et al., 2017;Waychunas, 1987).By contrast, various studies on Ti solubility and complexation in the systems NaAlSi 3 O 8 -H 2 O, Na 2 O-SiO 2 -H 2 O-TiO 2 , CaO-SiO 2 -H 2 O-TiO 2, and Na 2 O-Al 2 O 3 -SiO 2 -H 2 O-TiO 2 suggest the stability of 4-and 5-fold coordinated Ti in fluids, melts, and supercritical fluids (Antignano and Manning, 2008;Mysen, 2012Mysen, , 2019)).Predominantly 4-and 5-fold coordinated Ti has been further detected in natural rhyolitic melts (Farges and Brown, 1997).The coordination difference between residual minerals (i.e.rutile) and fluid/ melt phase could lead to Ti isotope fractionation, favouring isotopically heavy Ti in the liquid phase.However, so far little research has been done on the influence of pressure > 2 GPa on the bonding  2019), Millet and Dauphas (2014) and Millet et al. (2016).Average composition of the upper continental crust (UCC, 0.184 ± 0.069‰, 1SD, n = 88) derived from shale data (Deng et al., 2019;Greber et al., 2017a).Uncertainties are smaller than symbol size.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)Fig. 6.Titanium isotope data of the Horoman peridotites versus major and trace element contents and trace element ratios.Uncertainties (2SD) are smaller than symbol size, unless indicated otherwise.Refractory peridotites with particularly low TiO 2 contents show extremely light Ti isotope compositions (a-b).Excluding these distinctly light samples, a moderate negative correlation between d 49 Ti and Nb/Th is evident (c).environment of Ti.Paris et al. (1994) investigated the pressure dependence of Ti coordination in potassic Ti-silicate glasses in the range of 0.5-3 GPa.With increasing pressure, the average CN of Ti progressively shifted from $ 5 to $ 6 (Paris et al., 1994).These experiments are limited to a simple chemical system and no data is available for Ti coordination in HP melts or supercritical liquids of natural compositions.Element transport in subduction zones occurs at pressures of 2.5-5 GPa (e.g.Hermann and Spandler, 2008).It is, therefore, conceivable that Ti in the released (U)HP melts/supercritical fluids has a higher average CN, close to that of minerals in the residue (CN of 6).Experiments suggest the presence of residual rutile up to high pressures and temperatures (4.5 GPa, 1050 °C) and high degrees of sediment partial melting (Hermann and Rubatto, 2009;Hermann and Spandler, 2008).Assuming an increase in Ti coordination in melt/supercritical fluids at (U)HP conditions, the released liquids are expected to primarily reflect the Ti isotope composition of the subducted sediments.However, we cannot rule out isotopic fractionation of Ti during mobilisation from the slab, resulting in further enrichment in heavy Ti isotopes.This latter mechanism has been invoked to explain heavy Mo isotope compositions in arc basalts (Freymuth et al., 2015;König et al., 2016;Willbold and Elliott, 2017).The extent of isotope fractionation would ultimately depend on (i) coordination of Ti in the (U)HP liquids, (ii) subduction zone geotherm and (iii) degree of partial melting of the subducted material (i.e. the amount of residual rutile).
In summary, there are several lines of evidence supporting that metasomatism in the mantle wedge gave rise to the observed heavy Ti isotope compositions of the Horoman peridotites (Fig. 5), with or without fractionation of Ti isotopes during release of fluids/melts from the slab.We like to note, however, that d 49 Ti of the peridotites do not correlate with LREE enrichment (Fig. S1) and no phlogopite has been observed in thin section (Takazawa et al., 2000).This likely indicates the percolation of several compositionally distinct metasomatic agents through the Horoman mantle section.Alternatively, the heavy Ti isotope compositions might be related to late-stage infiltration of fluids/melts derived from the metapelitic and metapsammitic country rocks during emplace-ment of the Horoman complex into the lower crust.Dehydration and partial melting of the ambient crust would likely give rise to significant fractionation towards heavy Ti isotope compositions.Thus, both metasomatism in the mantle wedge, or during emplacement of the complex would result in a heavy Ti isotope signature, impeding discrimination.High-temperature (HT), granulite-facies rocks occur in the Hidaka metamorphic belt, however, it is unclear, whether the ascending Horoman mantle fragment could have provided the heat source for HT-metamorphism (e.g.Niida, 1984;Ozawa and Takahashi, 1995 and references therein).Moreover, our samples are very fresh (i.e.little to no serpentinisation of olivine) and show little petrographic evidence for late-stage fluid/ melt-rock interaction.We, therefore, favour metasomatism in the mantle as the process responsible for the observed heavy Ti isotope shifts.
essentially relevant for extraterrestrial redox-environments, the calculations predict no significant fractionation between clinoand orthopyroxene even at low temperatures of 700 °C.The observed large Ti isotope fractionation between ortho-and clinopyroxene (D 49 Ti opx-cpx = 0.16-0.29‰) in the two pllherzolites is, therefore, best explained in the context of the metasomatic history of the Horoman complex and likely reflects Ti isotope disequilibrium.Such deviations from empirically and theoretically determined equilibrium fractionation factors are common for metasomatised peridotites and were reported for other elements such as Mg and Cr (e.g.Chen et al., 2018;Shen et al., 2018;Xiao et al., 2013).Clinopyroxene in Horoman peridotites is characterised by a rimward enrichment in Ti (Takazawa et al., 1996).Since light isotopes diffuse faster than heavy ones (Richter et al., 2009), preferential diffusion of light Ti isotopes during fluid/melt-rock interaction and the small diffusivity of Ti in pyroxene (Cherniak and Liang, 2012) may account for the lower d 49 Ti values of clinopyroxene compared to orthopyroxene.Alternatively, the Ti isotope disequilibrium between clino-and orthopyroxene may be related to subsolidus redistribution of Ti and incomplete re-equilibration of the peridotites during transition from garnet-to plagioclase-facies peridotite during uplift.
We like to further note that the whole-rock Ti isotope composition of sample BZ-146L-S is heavier than the respective pyroxene separates (Fig. 4b, Table 2) and close to the composition of the spinel-bearing aggregates.These aggregates were interpreted as pseudomorphs after pyrope-rich garnet (e.g.Morishita and Arai, 2003;Takahashi and Arai, 1989;Takazawa et al., 1996) and may display elevated incompatible trace element concentrations, possibly related to metasomatism during or after symplectite formation (Morishita and Arai, 2003 and references therein).The spinelbearing aggregates, however, only account for $ 10 % of the whole-rock Ti budget (Table S5).Therefore, mass-balance considerations imply that they are not the main carrier of heavy Ti in the samples.This hints at an unidentified phase enriched in heavy Ti isotopes that significantly contributes to the bulk rock composition.Plagioclase-lherzolites from the Bozu section contain accessory pargasite with TiO 2 -concentrations of up to 2-3.5 wt% (Takazawa, 1996).Heavy Ti isotope compositions are, however, also recorded in spinel-lherzolites (e.g.BZ-120), where amphibole is absent.Its impact on the Ti isotope signature of the peridotites is, thus, unclear.Moreover, it is evident that peridotites sampled from the same location within the Bozu stratigraphic section (146L and 146L-S, Fig. 4b) display distinct d 49 Ti values.This further indicates, that fluid/melt-peridotite interaction was locally highly variable, whereby Ti-bearing phases may have precipitated along grain boundaries (e.g.small flakes of interstitial phlogopite), and significantly influenced the Ti isotope compositions of the peridotites.Hence, the mineral separates and corresponding bulk rocks provide further evidence for metasomatism that may have brought a heavy crustal signature of variable magnitude to the analysed peridotites.

Ultrapotassic rocks
Due to a lower solidus, metasomatised mantle domains preferentially melt during subsequent tectonic processes, giving rise to ultrapotassic magmatism (Foley, 1992).The analysed ultrapotassic volcanic rocks are enriched in heavy Ti isotopes, displaying the highest d 49 Ti values recorded in mantle-derived primary magmas (Fig. 7).Metasomatised domains are preferentially mobilised during low-degree partial melting (Foley, 1992) and, hence, Ti largely partitioned into the melt phase.Titanium isotope fractionation related to partial melting of the metasomatically enriched mantle is, therefore, not expected.Since the rocks are primitive, fractional crystallisation should not have affected their Ti isotope composition.Furthermore, the high Mg# and high Ni and Cr contents of the samples argue against crustal assimilation during magma ascent (Table S2).Ultrapotassic rocks from the Alpine and Variscan orogenic belts are assumed to originate from mantle domains strongly metasomatised by subducted terrigenous sediments or continental crust, respectively (e.g.Conticelli et al., 2009;Soder and Romer, 2018).Hence, their Ti budget is largely dominated by recycled crustal material.Whilst the heavy Ti isotope compositions are likely inherited from the recycled crustal components, i.e. sediments and differentiated rocks that are characterised by high d 49 Ti values (Fig. 5), isotopic fractionation during subduction might have induced further enrichments in isotopically heavy Ti (section 5.1.1).The heavy Ti isotope compositions of the South African ultrapotassic rocks support the presence of an ancient recycled component within the subcontinental lithospheric mantle beneath the Kaapvaal craton (e.g.Becker and Roex, 2006;Coe et al., 2008).

Titanium isotope fractionation during partial melting of the mantle
Peridotites from the Horoman massif represent residues of partial melting and melt extraction from a MORB-mantle in the garnet stability field at P < 3 GPa (e.g.Takazawa et al., 2000).Whole-rock major and trace element compositions indicate variable degrees of partial melting of up to 25 % (Takazawa et al., 2000;Yoshikawa and Nakamura, 2000).The Horoman harzburgites and lherzolites have highly variable d 49 Ti values ranging from extremely light to heavy Ti isotope compositions (Fig. 4a).As discussed above, isotopically heavy, continental crust-like Ti isotope compositions are likely related to subduction-zone metasomatism.The extremely low d 49 Ti values, on the other hand, are restricted to the most meltdepleted peridotites (0.46-1.42 wt% Al 2 O 3 , 0.61-1.24wt% CaO) with notably low Ti contents (0.001-0.013 wt% TiO 2 ) (Fig. 6a-b).Although so far no evidence has been provided for significant Ti isotope fractionation during partial melting of the mantle (Deng et al., 2018a;Millet et al., 2016), we briefly evaluate this scenario in the following.
For the assessment, we combined the ab-initio-derived force constants for various mantle minerals and melt-equivalent, and calculated melt-mineral Ti isotope fractionation factors.We then modelled the elemental and isotopic behaviour of Ti during melting of a garnet peridotite source using an incremental batchmelting model.
Melt-mineral fractionation factors: In the peridotitic mantle, Ti is hosted by clino-, orthopyroxene, spinel and garnet, with 6-fold coordinated Ti at upper mantle P-T conditions (Ackerson et al., 2017;O'Neill and Navrotsky, 1983;Waychunas, 1987).Titanium is a trace element in olivine and may be incorporated in the 4and 6-fold coordination site depending on olivine water content (Berry et al., 2005(Berry et al., , 2007)).In mafic silicate melts, however, Ti is in 5-and 6-fold coordination (Farges and Brown, 1997;Leitzke et al., 2018) with an average CN of 5.5 for natural terrestrial basaltic glass (Farges and Brown, 1997).While a difference in CN can provide a first order estimate for the magnitude of the equilibrium isotope fractionation, a quantitative evaluation requires the knowledge of the mean force constants for chemical bonds as well as experimental data.From the mean force constants, we determined equilibrium isotope fractionation factors between coexisting phases as a function of the temperature (Appendix).We used ab initio-derived force constants for fresnoite ( [5] Ti, 468 N/m), an analogue for 5-fold coordinated Ti in silicate melts, and diopside ( [6] Ti, 383 N/m) from Aarons et al. (2021) and for pyrope ( [6] Ti, 447 N/m) from Wang et al. (2020).In addition, Wang et al. (2020) calculated the mean force constant for olivine with Ti in 4-fold coordination.However, since natural hydrous olivine may contain both 6-and 4-fold coordinated Ti (Berry et al., 2005), we assumed a mixture of 50 % [6] Ti and 50 % [4] Ti and used the average force constants for [6] Ti (358 N/m) and [4] Ti (590 N/m) (Aarons et al. 2021; derived from data provided by Aarons et al. 2021;Leitzke et al. 2018;Wang et al, 2020), corresponding to a mean value of 474 N/m.For a basaltic melt with 50 % [6] Ti and 50 % [5] Ti (Farges and Brown, 1997), we estimated a force constant of 413 N/m.Assuming a temperature of 1300 °C, the following melt-mineral fractionation factors were derived: a melt-cpx = 1.000046, a melt- grt = 0.999948 and a melt-ol = 0.999906 (i.e 10 3 lna melt-mineral of 0.046, À0.052 and À0.094‰, respectively).As discussed in 5.1.2,Ti isotope fractionation between ortho-and clinopyroxene is not expected.Therefore, a melt-opx can be equated with a melt-cpx , which is in agreement with experimentally derived fractionation factors between orthopyroxene and mafic melt at terrestrial fO 2 and 1260 °C (Rzehak et al., 2021).
Melting model: We utilised a non-modal incremental batchmelting model, following the approach of Williams and Bizimis (2014) (Appendix).Using the above specified mineral-melt fractionation factors, the modelling predicts negligible Ti isotope fractionation during partial melting of the mantle with a melt- peridotite $ 1 (1.000033-0.999998or 10 3 lna melt-peridotite of 0.033 to À0.002‰, Fig. 8).These results are in line with (i) the limited Ti isotope variations between peridotites from the Ivrea Zone and the Beni-Bousera massif and MORBs (Fig. 4a; Mandl, 2019;Millet et al. 2016) and (ii) the similar Ti isotope data of komatiites and MORBs (Deng et al., 2018a;Greber et al., 2017b), each formed by different degrees of partial melting.An unreasonably high meltperidotite fractionation factor a of 1.00034 (or 10 3 lna melt- peridotite = 0.340‰) would be required to approximate the Ti isotope compositions of the refractory peridotites (Fig. 8).Titanium isotope fractionation during partial melting of the mantle can, therefore, not account for the anomalously light Ti isotope compositions of the highly depleted Horoman peridotites.

Mantle metasomatism and kinetic isotope fractionation
Kinetic effects during melt/fluid infiltration may induce large isotope fractionations.Kinetic, diffusion-driven isotope fractionation was proposed to explain anomalous isotope compositions in peridotites for fast-diffusing elements such as Li (Lai et al., 2015), and slower-diffusing elements like Fe (Weyer and Ionov, 2007) and Ca (Zhao et al., 2017).Since mantle melting cannot produce significant Ti isotope fractionation, the low d 49 Ti values of the Horoman harzburgites and lherzolite likely result from diffusiondriven kinetic isotope fractionation during fluid/melt percolation.As discussed in 5.1.1 and 5.1.2,there is evidence for multiple metasomatic agents that passed through the Horoman mantle section.Depleted, olivine-rich peridotites are highly permeable for silicate melts, facilitating liquid migration (Toramaru and Fujii, 1986;Zhu and Hirth, 2003).The two harzburgites and lherzolite with light Ti isotope compositions have very low Ti contents (0.001-0.013 wt% TiO 2 ).Thus, the metasomatic media would likely be relatively enriched in Ti, causing a concentration gradient.Due to a higher diffusivity, isotopically light Ti will preferentially diffuse from the liquids into the wall-rock, resulting in low d 49 Ti values in the peridotites.In addition, reaction with a fluid or melt characterised by a distinct, light Ti isotope signature could provide a further explanation for the low d 49 Ti.Depleted peridotites are particularly susceptible to metasomatic overprint.The extremely low d 49 Ti values could, therefore, alternatively reflect the Ti isotope composition of a distinct, but so far unidentified, metasomatic agent.

Implications on mantle Ti isotope heterogeneity and conclusions
Komatiites and basalts of different tectonic settings (i.e.MORBs, OIBs, island arc basalts) indicate a rather homogenous mantle source with respect to Ti isotopes (e.g.Greber et al., 2017b;Millet et al., 2016).Minor systematic variations in d 49 Ti (<0.1‰) are reported for komatiites and MORBs (N-MORBs vs E-MORBs) (Deng et al., 2018a;Greber et al., 2017b).Our study on orogenic peridotites, however, suggests the presence of local mantle domains characterised by highly variable Ti isotope compositions.Significant Ti isotope heterogeneity of 2.07‰ occurs on scales of several tens of meters and is likely induced by metasomatism and crustal recycling, as well as kinetic effects during fluid/melt percolation.The observed range towards high, continental crustlike d 49 Ti values in peridotites suggests that Ti is mobilised from the subducted slab and transported to the mantle wedge.Hence, subduction-zone metasomatism may generate substantial Ti isotope variations within the mantle.The homogeneous d 49 Ti values of Western Alps serpentinites formed at different P-T conditions during subduction, however, may imply that Ti is not significantly mobilised or fractionated during prograde dehydration of mafic lithologies (Millet et al., 2016).In the light of our results, the overall homogeneous Ti isotope composition of island arc basalts, which is within uncertainties indistinguishable from the MORB average (Millet et al., 2016), may be surprising.This indicates that the elemental and isotopic Ti budget of arc basalts is largely derived from the peridotitic mantle and the contribution of the subducted material is minor.Small-scale variations in d 49 Ti are likely diluted during flux melting of the mantle wedge.The heavy d 49 Ti signature is, however, preserved in ultrapotassic magmas produced by low-degree melting of highly metasomatised mantle domains.Consequently, heavy Ti isotope compositions may be expected for high-K basalts and primitive shoshonites, which are the most enriched arc magmas with a high contribution of recycled material.
In conclusion, peridotites from the Horoman massif exhibit a wide range in d 49 Ti from -1.523 ± 0.029 to 0.547 ± 0.015‰, recording extreme small-scale Ti isotope heterogeneity within the upper mantle.Heavy, continental crust-like Ti isotope compositions are likely associated with subduction-related metasomatism.While isotopically heavy Ti was inherited from the subducted material, Ti isotope fractionation during mobilisation from the slab may have further contributed to the heavy signature.Heavy Ti isotope compositions of 0.114 ± 0.040 to 0.290 ± 0.030‰ are further evi-Fig.8. Titanium isotope fractionation during partial melting of a garnet peridotite source using an incremental non-modal batch-melting model.The model (black curves) predicts no significant Ti isotope fractionation during partial melting of the mantle if using ab initio-derived melt-mineral fractionation factors.This is in line with the similar Ti isotope compositions of basalts and komatiites, with d 49 Ti values scattering around 0‰ (grey bar).An unreasonably high melt-peridotite fractionation factor a melt-peridotite of 1.00034 (i.e. 10 3 lna melt-peridotite = 0.340‰) is required to fit the d 49 Ti compositions of the depleted peridotites (green solid curve).The grey solid line indicates the Ti isotope composition of the source (0.005‰).(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)dent in the investigated orogenic and South African ultrapotassic rocks, which originate from a mantle source contaminated by high amounts of isotopically heavy recycled continental crust.By contrast, extremely low d 49 Ti values of À1.523 ± 0.029 to À0.677 ± 0. 036‰ are identified in highly refractory peridotites and might have resulted from the preferential diffusion of light Ti isotopes into the host peridotite during fluid/melt percolation.This indicates diffusion-driven kinetic fractionation may further contribute to Ti isotope variability in mantle-wedge peridotites.

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.

Fig. 4 .
Fig. 4. Titanium isotope data of a) peridotites and serpentinites analysed in this study (green symbols) and fromMandl (2019) andMillet et al. (2016) (grey symbols).The Horoman peridotites have highly variable Ti isotope compositions, covering a d 49 Ti range of 2.07‰.Horoman data are ordered according to the stratigraphic position in the Bozu section.The grey shaded area indicates the range of terrestrial basalts and komatiites.Uncertainties are 2SD (this study;Mandl, 2019) or 95 % CI(Millet et al., 2016) and smaller than symbol size.b) Titanium isotope compositions of mineral separates (empty symbols) and respective whole rocks (filled symbols).Subsamples processed for mineral separation are indicated by the suffix S. Uncertainties (2SD) are smaller than symbol size, unless indicated otherwise.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 1
Titanium isotope compositions of samples measured in this study.
The Ti isotope data were measured relative to the Ti Alfa Aesar wire standard (d 49 Ti Ti-AA Wire ) and are reported relative to the OL-Ti standard (d 49 Ti) for comparison.R: replicate (repeated digestion and analysis) n: number of analyses

Table 2
Titanium isotope compositions of mineral separates and respective whole rocks.