Deformation, hydration, and anisotropy of the lithospheric mantle in an active rift: Constraints from mantle xenoliths from the North Tanzanian Divergence of the East African Rift

Article history: Received 7 July 2014 Received in revised form 31 October 2014 Accepted 3 November 2014 Available online 15 November 2014


Introduction
Continental rifting is a complex process that results in localized thinning and, in some cases, in disruption of a continental plate. While the surface expression of this deformation is clear and usually well understood, little is known about how the lithospheric mantle deforms to accommodate rifting. The widely differing surface expression of continental rifting has led to contrasting lithospheric extension models. McKenzie (1978) proposed a symmetrical rift model, where the lithosphere deforms by homogeneous thinning and stretching in response to far field extensional forces. To account for the observations in the Basin and Range, Wernicke (1981Wernicke ( , 1985 proposed an asymmetrical extension model, in which the deformation is localized on a lithospheric-scale detachment fault. However, such models cannot account for the narrow rift valley and the strong mantle lithosphere thinning observed in East Africa (Dugda et al., 2007(Dugda et al., , 2009. The latter observations are in better agreement with the Nicolas et al. (1994) model, where rifting occurs via lithospheric rupture and rise of an asthenospheric wedge within the lithospheric mantle. The later model has been further developed by Vauchez et al. (1997) and Tommasi and Vauchez (2001), who, based on the analysis of the influence of inherited structures on the localization of continental breakup, suggested that most major rifts start forming through a transtensional deformation regime produced by the reactivation of the olivine crystallographic fabric frozen in the lithospheric mantle. Numerical models in which the upper mantle has an anisotropic viscosity controlled by the evolution of olivine crystallographic orientations corroborate this Tectonophysics 639 (2015)   assumption . Rheological heterogeneities, both in the crust and in the mantle, also have a major effect on the localization of rifting (e.g., Dunbar and Sawyer, 1989;Nyblade and Brazier, 2002;Vauchez et al., 1997).
In addition to being deformed by an extensional regime, the lithosphere within an active rift is often subjected to extensive magma percolation. Dyke intrusions have been proposed to help initiate extension in a thick continental lithosphere (Buck, 2006). The major role of magmas in rifting has been corroborated by seismic anisotropy data in the Afars, which unambiguously point to aligned melt pockets throughout the crust and lithospheric mantle (Bastow et al., 2010;Kendall et al., 2005). At smaller scales, the presence of melt may result in weakening of mantle rocks (Hirth and Kohlstedt, 2003;Zimmerman and Kohlstedt, 2004). It may trigger strain localization if melt is heterogeneously distributed (e.g., Le Roux et al., 2008) or promote homogeneous deformation if it is homogeneously distributed through a large volume . Melt-or fluid-rock reaction may also result in softening through crystallization of weaker phases and/or associated grain size reduction and phase mixing (e.g., Dijkstra et al., 2002;Soustelle et al., 2010).
Mantle xenoliths provide a valuable means by which to study the deformation of mantle lithosphere during rifting. They allow for quantification of the hydration state and characterization of melt-rock reactions and of their timing relative to the deformation. Moreover, the analysis of the microstructures and crystallographic preferred orientations (CPO) bring constraints on the deformation mechanisms and conditions in the lithospheric mantle below active rifts. In the present study, we explore the relations between deformation, melt or fluids percolation, and hydration in a series of mantle xenoliths from four localities in the North Tanzanian Divergence region (East African Rift). This region, still in the early stages of rifting, offers favorable conditions to study the expression of rifting on the lithospheric mantle. In addition, we estimate the seismic anisotropy of these rocks based on their CPO and mineralogical composition and compare these results to seismic anisotropy measurements performed within and around the East African Rift.

Geological setting
The East African Rift is one of the few active continental rifts on Earth. It extends over~4000 km, from the Afar triple junction in the Red Sea to the Gulf of Mozambique (Fig. 1), mostly following the trend of older orogenic belts (Nyblade and Brazier, 2002;Vauchez et al., 1997 and references therein). Rifting and volcanism started 35 Ma ago in Ethiopia and northern Kenya (MacDonald et al., 2001;Morley et al., 1992). It migrated southwards, reaching southern Kenya 8-5 Ma ago (Cerling and Powers, 1977;Crossley and Knight, 1981). The youngest section of the East African Rift splits into two branches around the Tanzania craton. In the Eastern branch, extension is accompanied by intense magmatic activity concentrated within the rift valley. The region experiences relatively intense present-day seismic activity, with earthquakes mainly located within the rift valley or in its immediate surroundings (e.g., Albaric et al., 2010;Nyblade et al., 1996). A major faulting episode at 1.0 ± 0.2 Ma (MacIntyre et al., 1974) gave rise to the present-day rift valley morphology. In addition, off-axis volcanic activity is observed in the North Tanzanian Divergence, south of which rifting occurs in a more diffuse manner, with deformation accommodated in many branches (Fig. 1). Seismic studies show that crustal and lithospheric thicknesses in this young rifting domain vary between 36-44 km and 100-150 km, respectively (Dugda et al., 2009;Julià et al., 2005).
Mantle xenoliths occur in both in-and off-axis volcanoes, providing an exceptional opportunity to study the tectono-thermal evolution of the mantle lithosphere in response to the progression of the East African Rift along the boundary between the Tanzanian craton and the Neoproterozoic Mozambique Belt. The present study focuses on 53 mantle xenoliths from four localities from the North Tanzanian Divergence (  Ryan et al. (2009) showing the North Tanzania Divergence and the location of the xenolith localities studied here (P: Pello Hill, E: Eledoi, L: Lashaine, and O: Olmani), as well as the Tanzanian craton and main volcanoes. Labait is another locality containing abundant mantle xenoliths, which microstructures and crystal preferred orientations were studied by Vauchez et al. (2005). two within the transverse volcanic belt (Lashaine and Olmani). The xenoliths from the cratered tuff cone of Pello Hill and from the Eledoi explosion crater were collected in fine-grained scoria with a bulk-rock chemistry similar to olivine-melilitites and olivine-nephelinites (Dawson and Smith, 1988). The formation of these craters postdates the major rift faulting episode at 1.2 Ma (Dawson and Smith, 1988). The Lashaine tuff cone and the Olmani cinder cone are ankaramite rift-related volcanoes erupted during the Neogene. They are located in a part of the Mozambique Belt characterized by Archean crustal material reworked in the Neoproterozoic by the Mozambique orogeny (Mansur et al., 2014 and references therein;Möller et al., 1998).

Pressure and temperature estimates
Major elements composition of olivine, orthopyroxene, clinopyroxene, garnet, and spinel were analyzed in 15 samples using a Cameca SX100 electron microprobe at Microsonde Sud facility, in Montpellier (France). Analytical conditions were a 20 kV accelerating voltage and a 10 nA probe current. Core and rim composition of 3 to 4 grains were measured for each mineral (mineral compositions of the analyzed samples presented in the online Supplementary Material Table 1).
When orthopyroxene and clinopyroxene were both present, the sample equilibrium temperature was calculated using the two pyroxenes Fe-Mg exchange geothermometer from Brey and Köhler (1990), which has an uncertainty of ±50°C. However, many of the samples are clinopyroxene-free harzburgites to dunites. To verify if other thermometers (Brey and Köhler, 1990;Fabriès, 1979;Li et al., 1995;Sachtleben and Seck, 1981;Wells, 1977;Witt-Eickschen and Seck, 1991) could be used to calculate the equilibrium temperatures for the clinopyroxene-free peridotites, the consistency between the temperatures calculated with the two pyroxenes geothermometer of Brey and Köhler (1990) and these thermometers was tested for the clinopyroxene-bearing peridotites. Because the discrepancies between the equilibrium temperatures obtained using different thermometers were large, we choose not to present temperatures calculated for the clinopyroxene-free samples. The equilibrium pressure was estimated in one garnet-bearing harzburgite from Lashaine (LS11) using the orthopyroxene-garnet barometer of Nickel and Green (1985), which has uncertainties of 0.2 GPa.

Electron-backscattered diffraction (EBSD)
The crystallographic preferred orientation (CPO) of olivine, pyroxenes, pargasite, phlogopite, spinel, and garnet was measured by indexation of electron back-scattered diffraction (EBSD) patterns at the SEM-EBSD facility, Geosciences Montpellier. These patterns are produced by interaction of a vertical incident electron beam and a carefully polished thin section tilted at 70°to the horizontal. Measurements were performed in a JEOL JSM 5600 scanning electron microscope using an acceleration voltage of 17 kV and a working distance of 23 mm. For each sample, maps covering nearly the entire thin section were obtained using steps between 100 and 30 μm, depending on grain size. Indexing rates varied between 60 to 90% depending of the mineral species and degree of fracturing present in the thin section. Phlogopite, when present, was usually poorly indexed. Orthopyroxene was sometimes misindexed as clinopyroxene. Careful post-acquisition data treatment controlled by comparison between EBSD maps and microscopic observations was performed to reduce inaccurate mineral determination as well as misindexation due to olivine pseudosymmetry. Modal composition, grain sizes, and shape-preferred orientations were also obtained from the EBSD maps.
Crystal-preferred orientation data are displayed in pole figures, presented as lower hemisphere stereographic projections. Data were plotted as one point per grain to prevent over-representation of large grains. When the foliation and lineation could be identified, the orientation of the main crystallographic directions: [100], [010] and [001] for olivine and pyroxenes, was plotted relatively to the principal axes of the deformation ellipsoids X, Y, and Z. Because the foliation and lineation could not be identified in coarse-grained samples, thin sections were cut in random orientations. To make comparison between different samples easier, we rotated the CPO of randomly oriented samples into a common orientation in which the maximum concentration of olivine [100] and [010] axes are parallel to the E-W and the N-S directions of the pole figure, respectively.
To characterize the olivine CPO symmetry we computed the dimensionless BA index (where B and A stand for [010] and [100] axes, respectively, Mainprice et al., 2014) defined as: where P and G are the Point and Girdle fabric indices (Vollmer, 1990) of the olivine principal axes [100] and [010]. These indexes were calculated from the eigenvalues of the normalized orientation matrix using the MTEX texture analysis Matlab toolbox (Hielscher and Schaeben, 2008;Mainprice et al., 2011). The BA index allows a classification of the olivine CPO symmetry into 3 types: axial- The strength of the fabric was quantified using the dimensionless J-index, which is the volume-averaged integral of squared orientation densities: where f(g) is the orientation distribution function (ODF) and dg = dφ 1 dφdφ 2 sinφπ 2 (Bunge, 1982). φ 1 , ψ, and φ 2 are the Euler angles that define the rotations allowing for coincidence between the crystallographic and external reference frames. Olivine CPO in natural peridotites is characterized by J-indexes mostly between 2 and 20, with a peak at 8-10 (Ben Ismaïl and Mainprice, 1998;Tommasi et al., 2000). The J-index for each sample was calculated based on the mean orientation of each grain using the MTEX texture analysis Matlab toolbox (Hielscher and Schaeben, 2008).

Seismic properties
Seismic properties of Tanzanian xenoliths were calculated using the CPO of all major phases and their respective modal content estimated from EBSD maps (Mainprice, 1990). For olivine, orthopyroxene, clinopyroxene, garnet, amphibole, and phlogopite single-crystal elastic constant tensors at ambient conditions were used (Abramson et al., 1997;Bezacier et al., 2010;Chai et al., 1997aChai et al., , 1997bCollins and Brown, 1998). A Voigt-Reuss-Hill averaging was applied in all calculations. The seismic anisotropy parameters and the elastic constants of all samples are presented in Table 1 and online Supplementary Material 2, respectively.
Average seismic properties were also calculated for Lashaine, Olmani, and rift axis localities by averaging the individual samples elastic constant tensors for each locality. This calculation results in a maximum estimation of the seismic anisotropy for the locality, as it relies on the assumption of a common orientation of the foliation and lineation for all samples.

Fourier transform infrared spectroscopy (FTIR)
Fifteen double-polished thin sections were prepared for unpolarized FTIR analysis. Before analysis, the sections were immersed in pure acetone for at least 12 h to dissolve any intergranular CrystalBond glue. FTIR spectroscopy analyses were performed at the Laboratoire des Colloïdes, Verres, Nanomatériaux at Montpellier University using a Bruker IFS66v coupled with a Bruker HYPERION microscope and 36 V. Baptiste et al. / Tectonophysics 639 (2015) 34-55 mercury-cadmium-telluride (MCT) detector cooled by liquid nitrogen. A Globar light source and a Ge-KBr beam splitter were used to generate unpolarized mid-infrared radiation. A background measurement was performed at the beginning of the analysis of each sample and repeated when necessary. Measurements were done on olivine, orthopyroxene, and garnet crystals with beam spots of 30 to 100 μm. No measurement could be made on clinopyroxene crystals, which were too small and too altered. Two hundred scans were accumulated with a resolution of 4 cm −1 for each measurement. A baseline correction was applied on each spectrum using the OPUS software. Spectra were then normalized to a sample thickness of 1 cm. The sample thickness was measured using a micrometer with a tolerance of ±1 μm, and was always near 500 μm (Table 2). Fractures and inclusions were strictly avoided. The calibration of Paterson (1982) was used to quantify the OH concentration: where C OH is the hydroxyl concentration (in mol H/l), ζ is an orientation factor (1/3 for unpolarized measurements), and k(ν) is the absorption coefficient in cm − 1 for a given wavenumber v. X i is a density factor, which depends on the mineral phase (X i(ol, Fo90) = 2695 wt. ppm H 2 O, X i(orthopyroxene) = 2812 wt. ppm H 2 O). The uncertainty in the resulting OH concentrations is~30% . We also converted the olivine OH concentrations (See Table 2) obtained using the calibration of Paterson (1982) to the new calibration of Withers et al. (2012). The conversion factor between the two calibrations is a function of the sample thickness and of the OH concentration obtained with the calibration of Paterson (1982). Here, the sample thickness ranges between 476 and 567 μm and the olivine OH concentrations range between 1 and 18 wt. ppm H 2 O. Therefore, the olivine OH concentrations must be multiplied by a factor of 1.8 (see Withers et al., 2012, Supplementary material Fig. 1).

Compositions and equilibrium conditions
All samples are garnet-free peridotites, except for four Lashaine peridotites (LS11, LS4, L1, and 89-680). Most samples have harzburgitic to dunitic compositions ( Fig. 2 and Table 1). Orthopyroxene contents vary mainly between 0 and 20%, but several samples from Lashaine, Olmani and Pello Hill display higher modal orthopyroxene contents (22 to 32%) and two samples from Olmani have N 50% orthopyroxene (OL3 and 03TZ14K). Secondary minerals present in diffuse veins or pockets include clinopyroxene, amphibole, phlogopite, and spinel. Veins may comprise up to 15% of the total rock (modal contents including the veins are given in parentheses in Table 1).
We have been able to estimate equilibrium temperatures for only 9 of the 15 samples analyzed, because the others lacked clinopyroxene. Independently of their provenance or microstructure, all these samples have core equilibrium temperatures (calculated from core compositions in orthopyroxene and clinopyroxene) between 1000 and 1100°C ± 50°C (Table 1). Rim equilibrium temperatures (calculated from rim compositions in orthopyroxene and clinopyroxene) tend to be slightly, but systematically lower (by 5-21°C), except for sample PEL22 from Pello Hill, where rim equilibrium temperature is lower than core equilibrium temperature by N100°C (960°C versus 1090°C). In a garnet harzburgite from Lashaine (LS11), core and rim equilibrium pressures of 3.3 GPa and 3.1 GPa were obtained, respectively.

Microstructures
Because of their very different characteristics, the xenoliths microstructures will be described as a function of their provenance: the rift axis (Pello Hill and Eledoi) or the transverse volcanic belt (Lashaine and Olmani).

Rift axis samples
The xenoliths from volcanic localities within the rift axis display variable microstructures. Among the 27 studied xenoliths, 2 are mylonitic, 22 are porphyroclastic, and 3 are coarse-granular to tabular.
Mylonitic microstructures are only observed in 2 samples from Pello Hill (PEL15 and PEL17; Fig. 3a). These rocks are characterized by centimeter-sized orthopyroxenes, which are highly elongated (aspect ratios of up to 15:1), defining the lineation. These orthopyroxenes always display well-defined kink bands and undulose extinctions. They have indented grain boundaries, with olivine grains filling the embayments or even forming vein-like inclusions parallel to the orthopyroxene elongation (Fig. 4a). Exsolution in orthopyroxene is ubiquitous. Olivine is present as plurimillimetric porphyroclasts and as recrystallized grains with sizes ranging from 0.4 to 2 mm. It displays curvilinear boundaries, evolving sometimes to polygonal shapes. Subgrain boundaries in olivine are highly oblique to perpendicular to the grains elongation. The frequency and extent of deformation features in the porphyroclasts vary, however, between the 2 samples. In mylonite PEL17, subgrain boundaries are well developed and closely spaced. In mylonite PEL15, there are fewer subgrain boundaries. In both mylonites, the olivine neoblasts exhibit few subgrain boundaries and sometimes form 120°triple junctions.
The porphyroclastic peridotites (Fig. 3b, c) are the most common textural type at Pello Hill and Eledoi. They display plurimillimetric elongated olivine porphyroclasts (Fig. 3b), with common subgrain boundaries, undulose extinction, and sutured grain boundaries. Subgrain boundaries in olivine are dominantly at high angle to the foliation, but in many grains 2 orthogonal subgrain boundary families are observed (Fig. 4b). Small olivine neoblasts have fewer intracrystalline deformation features and tend to have more polygonal shapes, sometimes forming triple junctions. The frequency of the triple junctions in the recrystallized matrix and the subgrain boundaries spacing in porphyroclasts vary from sample to sample. For example, in sample PEL7, olivine porphyroclasts exhibit closely spaced subgrain boundaries and serrated grain boundaries (Fig. 4b), whereas olivine porphyroclasts in sample PEL1 display fewer subgrain boundaries and have polygonal grain boundaries (Fig. 3c). In the porphyroclastic peridotites from Eledoi, olivine porphyroclasts are rare, giving to the rock a more homogeneous microstructure. Interpenetrating olivine-olivine grain boundaries are common (Fig. 3b, c), suggesting grain boundary migration. In both sites, when present, orthopyroxene is anhedral, with intracrystalline deformation features, such as kink bands and undulose extinction. Exsolutions and corrosion embayments filled with olivine are common (Figs. 3b and 4c). Mylonitic 3.29 0.84 3.3434 9.6 6.5 2.1 6.1 PEL17B Mylonitic 4.23 0.79 3.3428 9.7 6.9 2.7 5.9 PEL20 PEL20     Paterson (1982) have ±30% uncertainty . b OH concentrations corrected by a factor of 1.8 according to Withers et al. (2012).
All coarse-granular to tabular peridotites are dunites (Fig. 3d, e). They exhibit coarse olivine grains (1-8 mm) with equant to tabular shapes. Olivine has curvilinear to polygonal boundaries and is often devoid of any substructure. 120°triple junctions are common. Three coarsegranular to tabular dunites (EL14, PEL11 and PEL41) are crosscut by diffuse veins composed of millimeter-sized clinopyroxene, phlogopite, amphibole and spinel (Figs. 3d and 4d). Intracrystalline deformation features are very rarely observed in phlogopite and amphibole grains. The orthopyroxene crystals in contact with the vein are partially replaced by diopside.
In the porphyroclastic and mylonitic samples, the distribution of metasomatic minerals (phlogopite, amphibole, clinopyroxene, and spinel) is usually more diffuse. They form irregular pockets (Fig. 4e) or occur as small interstitial grains dispersed in the rock. When these minerals are dispersed in the rock, they may display subgrain boundaries. They are usually associated with fine-grained olivines, forming "bands" with smaller grain sizes. In the pockets, spinel is associated with clinopyroxene, pargasite or phlogopite (Fig. 4e).

Xenoliths from the transverse volcanic belt
The xenoliths from the two localities (Olmani and Lashaine) within the divergence exhibit different microstructures. We will thus describe them separately.
4.2.2.1. Lashaine. The xenoliths from Lashaine show mainly very coarse granular microstructures (Fig. 3f). Olivine grain size ranges from few millimeters to more than a centimeter. They have curvilinear boundaries and widely spaced subgrain boundaries. Interpenetrating grain boundaries are common. Orthopyroxene is usually smaller (1-8 mm) than olivine, with rare kink bands or undulose extinction. Exsolution in orthopyroxene is only observed in sample LS4. Lashaine is the only location where garnet peridotites (samples LS4, LS11, 89-680, and L1) were sampled. Garnet is present as irregularly-shaped crystals, several mm long, most often rimming orthopyroxene (marked by arrows in Fig. 3f). Kelyphitic rims are always present. In sample L1, garnet has been entirely replaced by kelyphite. Clinopyroxene is present as exsolutions in orthopyroxene or as very small interstitial crystals associated with pargasite and spinel. Olivine with interpenetrating boundaries is commonly observed (Figs. 3f and 4g), as well as orthopyroxene embayments filled with olivine. Peridotite LS9 displays slightly different characteristics: it contains recrystallized zones with finer-grained olivine showing curvilinear to polygonal boundaries (Fig. 4f). Olivine tablets free of any substructure are present on other olivine crystals (Fig. 4f). Orthopyroxene often contains olivine inclusions.
Two samples have granular microstructures but smaller grain sizes: the coarse-granular harzburgite 89-680 and the granular dunite LS15. In harzburgite 89-680, olivine grain size ranges mainly between 1 and 4 mm. Olivine displays curvilinear to polygonal boundaries, and few subgrain boundaries. Orthopyroxene grain size varies between 0.5 and 2 mm. It has irregular shapes with curvilinear boundaries and may contain kink bands. Small interstitial clinopyroxene, phlogopite and amphibole grains are present. Dunite LS15 displays even smaller olivine grain sizes, ranging between 0.5 and 2 mm. Olivine has curvilinear, sometimes interpenetrating grain boundaries and a higher proportion of subgrain boundaries than the other samples from Lashaine.
The porphyroclastic harzburgites display elongated olivine porphyroclasts with millimetric grain sizes (Fig. 3g), which mark the foliation, and small olivine neoblasts (0.2-1 mm) with curvilinear to polygonal boundaries. The frequency of subgrain boundaries and the degree of polygonization of the grain boundaries vary from sample to sample. For instance, sample 03TZ14H shows interpenetrating olivineolivine grain boundaries, whereas polygonal olivine grains characterize sample OL-NN4. Orthopyroxene occurs as small (b1 mm), interstitial crystals displaying rare kink bands.
In coarse-porphyroclastic harzburgites, olivine grains are millimetric to centimetric and elongated. They commonly display closely spaced subgrain boundaries and interpenetrating grain boundaries. Small olivine grains rim the porphyroclasts. These neoblasts display curvilinear to polygonal grain boundaries. Orthopyroxene usually displays irregular shapes, rare kink bands and undulose extinction, as well as common exsolutions and corrosion embayments filled with olivine. Olivine crystals included in orthopyroxene grains are sometimes observed. In sample 03TZ14M, spinel-pyroxene symplectites are observed.
In coarse-granular harzburgites, olivine and orthopyroxene grain sizes vary between 0.5 and 3 mm. Two of these samples (OL3 and 03TZ14K) display very high orthopyroxene content (N 50%). Olivine is characterized by curvilinear to polygonal boundaries and common subgrain boundaries. In 2 samples (03TZ14K and 03TZ14L), olivine is occasionally included in orthopyroxene. Orthopyroxene grains are anhedral. They display well-defined kink bands and undulose extinction, as well as exsolutions. In several orthopyroxene grains, corrosion embayments filled with olivine are present.
The very coarse-granular peridotites from Olmani are similar to the samples from Lashaine (Fig. 3h). They display millimetric to centimetric olivine grains. The density of subgrain boundaries and their spacing vary considerably from one sample to another. Interpenetrating olivine grain boundaries are always observed. Orthopyroxene is present in only 1 of these samples . In this sample, orthopyroxene has millimeter grain sizes, curvilinear boundaries, and displays well-defined kink bands.

Crystallographic preferred orientations (CPO)
As for the microstructures, we describe separately the CPOs of peridotites from the rift axis, Lashaine, and Olmani.

Rift axis
Most rift-axis samples have low to moderate olivine CPO strength, with J-indexes varying mainly between 2-6 ( Fig. 5). There is no correlation between the olivine CPO symmetry, its strength, and the microstructure. Peridotites from Pello Hill and Eledoi exhibit olivine CPO symmetries ranging from axial-[100] to orthorhombic (Fig. 5) Olivine CPO strength of Lashaine samples is more variable than for the rift axis peridotites. J-index ranges between 2 and 8 for those samples where more than 100 grains could be measured (Fig. 5). In some very-coarse grained peridotites, the number of measured grains was b 100, leading to a possible overestimation of the Jindex (Ben Ismail and Mainprice, 1998). These values are marked by dashed arrows in Fig. 5 (Fig. 5, sample 03TZ14E1).

Seismic properties
Average anisotropic seismic properties for rift-axis and off-axis samples are illustrated in Fig. 6. To a first order, all localities share common seismological characteristics. The P-wave propagation is fastest close to the olivine [100] maximum, which corresponds to the lineation in the samples where it has been identified, and slowest close to [010] maxima, then normal to the foliation. The slowest propagation directions of the slow S-wave (S 2 ) are in the YZ structural plane. The fast split shear wave (S 1 ) is polarized in a plane containing the main concentration of olivine [100] and the propagation direction. The highest Vp/Vs 1 ratio is also parallel to the preferred orientation of olivine [100] and the lowest Vp/Vs 1 is observed for waves propagating normal to this direction.
Changes in olivine CPO symmetry result in second order variations of the seismic anisotropy pattern. Within the rift axis and at Olmani, where axial-[100] and orthorhombic olivine CPOs predominate, P-wave velocities are slow and S-wave azimuthal anisotropy is high within the YZ plane. The fastest velocities of S 2 -waves are observed for directions close to the lineation. The slowest propagation of the S 1 -wave occurs for directions either close or normal to the lineation, while the fastest S 1 propagation are expected for directions at 45°to the lineation.
At Lashaine, where axial-[010] to orthorhombic olivine CPOs dominate, slow P-wave velocities are found for propagation directions normal to the foliation plane. In contrast to the average properties computed for the rift axis, the S-wave polarization anisotropy is very low for propagation directions at high angle to the plane containing the concentrations of [100] and [001] (corresponding to the foliation when observed), with a minimum within the XZ plane at~45°to the lineation. The maximum S-wave polarization anisotropy lies within the foliation at 90°to the lineation. The fastest propagation directions of S 1 -wave are contained in the foliation plane, while the slowest directions are parallel to the Z structural direction. Slow and fast S 2 -wave propagation directions are observed in a plane normal to the lineation and in the XZ plane, at 45°to the lineation, respectively.
Seismic anisotropy intensities for individual samples are highly variable (Fig. 7, Table 1). Maximum P-wave anisotropies are comprised between 3.3 and 18.4%, while maximum S-wave polarization anisotropies range between 2.3 and 13.3%. Porphyroclastic peridotites, which have the strongest olivine CPO, tend to have the highest anisotropies, whereas granular peridotites yield the lowest anisotropies (Fig. 7). Within a microstructural group, however, there is no systematic variation in anisotropy as a function of the provenance.
We observe no simple relationship between P-and S-wave anisotropy intensities and the olivine CPO symmetry (characterized by the BA index; Fig. 7c, d). Most porphyroclastic peridotites, which have axial-[100] olivine CPO, are strongly anisotropic (Fig. 5).
Crystallization of large volumes of metasomatic phases (N25% vol.) may considerably reduce P-, S 1 -and S 2 -wave velocities, and the maximum P-and S-wave anisotropy, as illustrated by Fig. 8, in which we compare the seismic properties of sample EL15 calculated by taking into account or not a phlogopite-bearing clinopyroxenite vein that crosscuts this sample. It reduces the Vp/ VS 1 range, by increasing the minimum values, but it has no effect on Vp/VS 2 ratios.

Orthopyroxene and garnet
Orthopyroxene spectra are more homogeneous (Fig. 9b) than those of olivine. They display up to seven absorption bands. The four major bands are found at 3600, 3544, 3517 and 3410 cm −1 . Their intensities are variable. Minor peaks are also present around 3473, 3324 and 3060 cm −1 . Several millimeters long clear garnet grains were present in only one of the analyzed samples (harzburgite xenolith from Lashaine LS11). Their spectra are flat, indicating anhydrous garnet.
Although both samples are coarse-granular, in the whole set of analyzed samples, there is no systematic correlation between olivine OH concentration and microstructure. There is also no correlation between the olivine OH concentration and Mg# (Fig. 11). The homogeneity of OH concentrations in olivine was tested by analyses with 50 to 150 μm step sizes along transects (850 to 3450 μm long) on randomly oriented olivine grains in three samples: one mylonite from Pello Hill (PEL17), one garnet-bearing harzburgite from Lashaine (LS11), and one veinbearing sample from Pello Hill (PEL41). These profiles indicate homogeneous concentrations plateaus within grains, incompatible with significant dehydration (Thoraval and Demouchy, 2014).
Average orthopyroxene OH concentrations range between 97 and 212 wt. ppm H 2 O (Table 2), being in the range of concentrations reported by previous studies (Falus et al., 2008;Peslier, 2010;Peslier et al., 2002). In rift axis samples, orthopyroxene OH concentrations are more heterogeneous, but there is no systematic correlation between orthopyroxene OH concentration and xenolith locality or microstructure. Indeed, the lowest orthopyroxene OH concentrations are measured in mylonite  PEL17 and the highest, in mylonite PEL15. There is also no correlation between the olivine and orthopyroxene OH concentrations.

Rift Axis (Pello Hill and Eledoi)
The peridotites from the rift axis localities show a variety of microstructures, from mylonitic to porphyroclastic and coarse-granular. All exhibit evidence for plastic deformation and both dynamic and for static recrystallization, but the deformation conditions and the extent of static recrystallization vary from one sample to another. The obliquity between the lineation and subgrain boundaries in olivine suggests simple shear deformation in both porphyroclastic peridotites and mylonites, but the higher elongation of orthopyroxenes imply deformation under higher stresses in the mylonites. The heterogeneity in microstructures may indicate a spatial or a temporal variation of deformation conditions and intensity within the mantle below Eledoi and Pello Hill.
In both mylonites and porphyroclastic peridotites, the frequency and spacing between subgrain boundaries in olivine porphyroclasts and the extent of grain boundary polygonization vary from sample to sample ( Fig. 3a and b). These microstructural features support a deformation followed by variable degree of annealing. The well-preserved intracrystalline deformation features in most porphyroclastic peridotites suggest, however, that the time span between the deformation and the extraction by the magma was too short to allow for complete annealing of the deformation microstructures. The recent character of the deformation is further supported by the rather high equilibration temperatures (1050-1100°C) recorded by these peridotites. The variable degree of annealing observed in these rocks may be explained by variable interaction with fluids or melts, by sampling of different depths, or by transient and spatially heterogeneous heating events.
Mylonites also record a temporal variation of deformation conditions. Highly stretched orthopyroxenes are a characteristic feature of the mylonites (Fig. 3a). Similar highly stretched orthopyroxenes have been described, in mylonitic peridotites from the Lanzo (Nicolas et al., 1972), Ronda (Soustelle et al., 2009;Tubìa et al., 2004), and Beni Bousera massifs (Frets et al., 2014), and were interpreted as the result of deformation under high stress at moderate temperature (850-950°C estimated from thermobarometry in the peridotites and associated pyroxenites; Frets et al., 2014;Garrido et al., 2011). Closely spaced kinks in orthopyroxene also indicate deformation under high stress conditions. However, the stretched orthopyroxenes from the mylonites have indented shapes, with embayments filled by vein-like olivine crystals, which are often subparallel to the orthopyroxene elongation (Fig. 4a). In porphyroclastic peridotites, orthopyroxene embayments filled with olivine and olivine inclusions in orthopyroxene (Fig. 4c) are also present. These microstructures suggest reactions leading to consumption of orthopyroxene and crystallization of olivine. In both mylonites and porphyroclastic peridotites, the olivine crystals that replace orthopyroxene display ubiquituous undulose extinction and subgrain boundaries dominantly oriented normal to the crystals' elongation. This suggests that this reaction is synkinematic. Three processes may produce replacement of orthopyroxene by olivine: (1) incongruent melting of orthopyroxene (Kubo, 2002); (2) reaction between the peridotite and a Si-undersaturated melt (Kelemen, 1990;, (3) percolation of aqueous fluids (Padrón-Navarta et al., 2010). The last process may occur at low temperatures (650-700°C), but implies very large volumes of aqueous fluids (Padrón-Navarta et al., 2010), which are plausible in a subduction zone, but not in a continental rift.
In summary, the recent deformation event recorded by the rift axis xenoliths started under high stress and probably low temperature. The deformation was later coupled with partial melting or reactive percolation of Si-undersaturated melts. Both processes imply high, near-solidus temperatures, which are at odds with the high stresses inferred in the  mylonites from the stretching of orthopyroxenes. Moreover, the ubiquituous exsolutions seen in orthopyroxenes suggest significant cooling between this melt-assisted deformation and the extraction of the xenoliths. Together these observations imply transient and probably spatially heterogeneous temperature fields, marked by fast heating and cooling episodes, which probably accompanied deformation and reactive melt percolation (Kourim et al., in press;Kruckenberg et al., 2013). Interestingly, similar observations have been recently reported for peridotites samples from the boundary of the Middle Atlas (Marocco) in a context of progressive exhumation of the mantle lithosphere (El Messbahi et al., in press). Coarse-grained dunites (EL14, PEL11, and PEL41) contain rare subgrain boundaries and common triple junctions (Fig. 3e). They display, however, a clear crystallographic fabric. Altogether this suggests that efficient annealing followed deformation by dislocation creep. Dunite may form when a harzburgite interacts with Si-undersaturated   melts (Berger, 1985;Berger and Vannier, 1984;Kelemen, 1990;Morgan and Liang, 2003). The coarse-grained dunites from Pello Hill and Eledoi contain olivines with significantly lower Mg# (84-88, while the average for the other microstructures in these localities is~92, Table 1). Such a strong enrichment in Fe suggests interaction with very large volumes of melts. The coarse-grained dunites might thus represent melt channels or melt accumulation levels within the mantle lithosphere (Berger, 1985;Berger and Vannier, 1984;Kelemen and Dick, 1995;Tommasi et al., 2004). In mylonitic, porphyroclastic and coarse-granular peridotites, the well-defined olivine CPO indicates dislocation creep as the dominant deformation process. The olivine CPO patterns range from axial-[100] to orthorhombic. These olivine CPOs are consistent with deformation by simple or pure shear with dominant activation of (010)[100] or (0kl)[100] slip systems under high temperature, low pressure and anhydrous conditions (Tommasi et al., 1999. Axial-[100] CPO symmetry likely denotes a strain regime producing a well-defined stretching direction, such as transtension; Kruckenberg et al. (2014) have recently shown the correlation between axial-[100] CPO symmetry and prolate strain ellipsoids due to constructional deformation. Orthopyroxene CPOs indicate deformation by dislocation creep with activation of the [001] (100) slip system. They are always correlated to the olivine CPO, indicating that both minerals underwent the same deformation event.
In addition to the olivine-forming reactive melt percolation, textural evidence for metasomatism by hydrous melts has also been identified in many peridotites from Pello Hill and Eledoi. For instance, Dawson and Smith (1988) first analyzed the composition of the veins from Pello Hill mantle xenoliths and concluded that they bear asthenospheric Nd and Sr isotopic signatures and imparted REE, K, Fe, Ti metasomatism on the surrounding peridotite. It encompasses: (1) veins with diffuse margins that comprise clinopyroxene, amphibole, phlogopite, and spinel that crosscut deformation microstructures (Fig. 3d), (2) irregularlyshaped pockets of clinopyroxene, amphibole, phlogopite, and spinel (Fig. 4e). In these minerals, deformation substructures are uncommon. However, clinopyroxene crystals in veins do exhibit a clear CPO that is coherent with the CPO of olivine and orthopyroxene from the surrounding peridotite. Two hypotheses may explain this observation. First, vein intrusion may be synkinematic, but this hypothesis is at odds with the orientation of the veins that crosscut the foliation and with the lack of plastic deformation of the metasomatic phases. Thus, we propose that the vein intrusion was post-kinematic and the minerals crystallized in a preferential orientation controlled by the orientation of the preexisting orthopyroxene.

Lashaine
Mantle xenoliths from Lashaine mainly display very coarse-grained microstructures characterized by millimeter-to centimeter-sized grains with curvilinear, interpenetrating boundaries and low densities of subgrain boundaries (Figs. 3f and 4g). These observations point to efficient grain boundary migration. Together with the well-defined olivine and orthopyroxene CPOs, they support deformation by dislocation creep at high temperature and effective annealing. In LS9, the frequent olivine-orthopyroxene interpenetrating boundaries suggest reaction with percolating melts. In addition, the olivine tablets growing on other crystals (Fig. 4f) have been previously interpreted as resulting from fast grain growth in presence of fluids (Drury and Van Roermund, 1989).
Among the studied samples, the xenoliths from Lashaine are the only ones containing garnet, implying that deeper parts of the Tanzanian lithosphere have been sampled. Garnet crystals in peridotites LS4 and LS11 display interstitial shapes and usually rim orthopyroxene. This morphology led Gibson et al. (2013) to interpret these pyrope garnets as the product of exsolution from orthopyroxene due to cooling.
Well-defined olivine and orthopyroxene CPOs in Lashaine peridotites indicate that dislocation creep was the main deformation process. Orthopyroxene CPOs are consistent with deformation by dislocation creep with dominant activation of [001](100), as usually observed in deformed mantle peridotites and pyroxenites (e.g., Frets et al., 2012Frets et al., , 2014Vauchez et al., 2005). Olivine CPO patterns are dominantly orthorhombic to axial-[010] (Fig. 5). Axial-[010] olivine CPO symmetry, though less common than orthorhombic and axial-[100] symmetries , has been described in many xenoliths suites from continental and oceanic environments (e.g., Bascou et al., 2008;Tommasi et al., 2008;Vauchez et al., 2005;Zaffarana et al., 2014) and in peridotite massifs (e.g., Frets et al., 2014;Soustelle et al., 2010). It may result from similar activation of [100](010) and [001](010) slip systems as a result of changes in physical parameters (e.g., olivine water contents, pressure, differential stresses, presence of melt) during deformation or from a deformation regime characterized by an oblate strain ellipsoid ( transpression (Tommasi et al., 1999). A dispersion of olivine [100] axes has also been observed in rocks undergoing dynamic and static recrystallization (Falus et al., 2011;Tommasi et al., 2008). Experiments on olivine single crystals and polycrystals have shown that [001]-glide is favored at low temperature and high differential stresses (Demouchy et al., 2009Durham and Goetze, 1977;Phakey et al., 1972;Raleigh, 1968 Jung et al. (2006). Transition from [100] to [001] glide have also been observed as a result of increasing pressure in deformation experiments at high confining pressure (Couvy et al., 2004;Jung et al., 2006;Raterron et al., 2009). Occurrence of axial-[010] CPO in garnet-bearing harzburgite LS11 may suggest a change in dominant mechanism with increasing pressure, an interpretation similar to the one proposed by Vauchez et al. (2005) for the deepest Labait peridotites. However, this explanation cannot account for the common axial-[010] patterns in the Lashaine spinel peridotites. Deformation in presence of melt, accompanied by refertilization reactions, also results in development of axial-[010] olivine CPO Tommasi, 2012, 2014;Le Roux et al., 2008). Yet, there is no clear evidence for synkinematic refertilisation reactions in the studied peridotites. Thus axial-[010] olivine CPO in Lashaine peridotites probably results from either transpression or recrystallization. Transpression should also result in dispersion of orthopyroxene [001] axes in the foliation plane, but the small number of orthopyroxene crystals present in most samples does not allow corroborating or excluding this hypothesis. Evidence for static recrystallization and grain boundary migration is, on the other hand, widespread in these peridotites.

Olmani
Xenoliths from Olmani display microstructures and CPOs intermediate between those found in peridotites from rift-axis localities and from Lashaine (Figs. 3 and 7). Half of the Olmani samples display coarsegrained to very coarse-grained sizes, similar to the microstructures observed in Lashaine. The other half of Olmani samples exhibits coarseporphyroclastic to porphyroclastic microstructures, with characteristics similar to those observed in rift axis peridotites. All textural types from Olmani show evidence for olivine crystallization at the expenses of orthopyroxene. Chemical evidence for interaction with carbonatites was identified in several of the studied peridotites (89-772, 89-774, 89-776, and 89-773) by Rudnick et al. (1993). These samples display coarsegranular microstructures, suggesting that the carbonatitic metasomatism is not directly related to the recent deformation (probably rift-related), recorded by the porphyroclastic microstructures.
The well-defined olivine CPOs measured in Olmani peridotites suggests that dislocation creep was the main deformation process. All three olivine CPO symmetries are observed. Interestingly, Olmani granular peridotites tend to display axial-[010] olivine CPO, similar to those of the granular peridotites from Lashaine, whereas porphyroclastic to coarse-porphyroclastic samples show dominantly axial-[100] patterns, similar to rift axis peridotites. Moreover, the weak annealing observed in the porphyroclastic peridotites suggests that deformation occurred shortly before extraction. The orthopyroxene CPO is consistent with deformation by dislocation creep with dominant activation of [001](100) and [001](010) slip systems. Olivine and orthopyroxene CPOs are always correlated, implying that they record the same deformation event.
A variation in microstructures and olivine CPO patterns between the localities of Olmani and Lashaine was not expected, as both localities are within the Tanzanian Divergence and distant of less than 30 km. This variation may be explained by: (1) younger eruption ages in Olmani than Lashaine, in which case Olmani peridotites would record the recent deformation associated with the rifting, (2) a bias in xenolith sampling, (3) Olmani and Lashaine sample lithospheric domains with different tectonic ages, or (4) heteregeneous deformation of the lithospheric mantle at the scale of few tens of km within the transverse volcanic belt. The eruption age of Olmani and Lashaine is poorly constrained. However, neighboring volcanic centers with similar morphologies have eruption ages between 2.5 and 1.56 Ma (Evans et al., 1971;Le Gall et al., 2008;MacIntyre et al., 1974;Wilkinson et al., 1986). Hypothesis (1) is therefore improbable. A bias in xenolith sampling (2) cannot be definitely excluded, but is unlikely, since in spite of the large number of xenoliths collected, Lashaine xenoliths have very homogeneous microstructures. The presence of lithospheric domains of different ages (3) within the Mozambique belt is possible. Indeed, Lashaine and Olmani are located in a domain of the Mozambique belt that is interpreted as a remnant of cratonic lithosphere reworked during the Neoproterozoic Orogeny (Mansur et al., 2014;Möller et al., 1998). Similarities between Lashaine and cratonic peridotites are numerous. Rudnick et al. (1994) first highlighted similarities in mineral compositions between Lashaine and the Kaapvaal craton peridotites. Minimum ages of 3.4 and 2.9 Ga were obtained from Re-Os studies of sulfides in mantle xenoliths from both Lashaine and Labait, a Quaternary volcano located on the eastern boundary of the Tanzanian craton ( Fig. 1; Burton et al., 2000;Chesley et al., 1999). Comparison of the microstructures and the olivine CPOs of the Labait and Lashaine peridotites provides additional evidence for a common origin. Vauchez et al. (2005) described three textural types in the Labait peridotites, among which the garnet-free, coarse-grained peridotites display microstructures and olivine CPO very similar to those observed in the Lashaine peridotites. However, the similarities between Lashaine and Olmani coarse-granular peridotites make hypothesis (3) unlikely. Heterogeneous rift-related deformation within the volcanic belt is the most probable explanation to Lashaine and Olmani microstructural and CPO difference. Indeed, the deformation recorded by the Olmani porphyroclastic peridotites might be related to localized deformation in a strain transfer zone connecting the main rift to the Pangani Graben ( Fig. 1) or to reactivation of the neoproterozoic Aswa shear zone (e.g., Corti et al., 2007;Ruotoistenmäki, 2014 and references therein) in a transtensional regime that resulted in the opening of the Pangani Graben.

Hydration state of the Tanzanian lithosphere
Diffusion data from experimentation and dehydration profiles from basalt-borne peridotite xenoliths point out to a very fast hydrogen ionic diffusion in olivine at high temperature (Mackwell and Kohlstedt, 1990;Demouchy and Mackwell, 2006;Peslier and Luhr, 2006;Denis et al., 2013). However, the homogeneity of OH concentrations along profiles in olivine in three samples suggests that the measured concentrations were not modified during xenolith extraction Thoraval and Demouchy, 2014).
Average OH concentrations in olivine measured in the Tanzanian samples vary from very low to moderate (up to 12 wt. ppm H 2 O; Fig. 10, Table 2). Average orthopyroxene OH concentrations range between 97 and 212 wt. ppm H 2 O. In xenoliths from Pello Hill, the olivine OH concentrations are more heterogeneous than in those from Eledoi and Olmani. However, this difference might be related to the fact that we analyzed more samples from Pello Hill than from other localities. We also observe no clear relationship between the microstructure or the olivine CPO pattern and OH concentrations (Table 2, Fig. 10). We also do not observe significant differences in olivine OH concentrations between in-and off-axis samples (Fig. 10). The 2 samples that exhibit the highest OH concentrations in olivine are a garnet-bearing harzburgite LS11 from Lashaine and a vein-bearing dunite PEL41 from Pello Hill. We discuss below the potential origin of these relatively higher OH concentrations.
The harzburgitic to dunitic compositions and high olivine Mg# in most samples suggest extensive partial melting, which might be as old as 3.4 Ga, based on Os isotopic dating of sulfides in a garnet lherzolite from Lashaine (Burton et al., 2000). Since H behaves as an incompatible element during partial melting (Bolfan-Casanova, 2005;Dixon et al., 2002;Hirschmann et al., 2005), olivine with high Mg# should have the lowest OH concentrations. However, we observe no correlation between olivine Mg# and OH concentrations (Fig. 11), implying that the Tanzanian samples were potentially hydrated or rehydrated during later metasomatic events.
Geochemical studies have provided evidence for multiple metasomatic events in mantle xenoliths from the sampled localities. Carbonatite metasomatism led to enrichment in REE, increase in Mg#, and the crystallization of clinopyroxene and phosphate in Olmani xenoliths (Rudnick et al., 1993(Rudnick et al., , 1994. However, this carbon-rich metasomatism should not result in extensive hydration of Nominally Anhydrous Minerals (NAM) because a high CO 2 fugacity will lower H 2 O fugacity in the system and thus minimize the hydration of NAMs (Dixon et al., 1995;Sokol et al., 2010). In Lashaine peridotites, an episode of metasomatism, which enriched the peridotites in K, Fe Ca, Ti, Rb and REE, was dated at~2 Ga based on Re-Os isotope data on sulfides (Burton et al., 2000;Dawson, 2002). A more recent metasomatism led to addition of Si, K, Ti, Ca, Fe, Nb, and Ta (Dawson, 2002). Rudnick et al. (1994) suggested that the SiO 2 enrichment in Lashaine garnet peridotites was caused by the interaction with silicic melts derived from partial melting of a subducting slab, during the major subduction on the eastern edge of the Kaapvaal craton around 2 Gy ago (Möller et al., 1995). Such a process may modify the trace element contents of NAMs and allow for OH incorporation in olivine. It might therefore explain the relatively high OH concentration in olivine from garnet-bearing harzburgite LS11 (Fig. 10). Moreover, LS11 harzburgite is a garnet-bearing peridotite. Its deeper origin may also explain the higher OH concentration, since higher pressure implies higher water fugacity, favoring OH incorporation in olivine (Férot and Bolfan-Casanova, 2012;. The presence of undeformed hydrous phases in veins or diffuse pockets (Figs. 3d and 4d,e) in xenoliths from all sites implies extensive post-kinematic metasomatism by hydrous fluids or melts. Indeed, the vein-bearing dunite PEL41 from Pello Hill shows the highest OH concentrations in olivine from all analyzed samples (Fig. 10). However, measurement of the OH concentration in olivine at different distances from the veins in dunite PEL41 shows no systematic increase in olivine OH concentrations in the vicinity of the vein (Fig. 12). Similar measurements were performed for samples EL15 and PEL40. These data suggest that hydration of olivine is not related to the metasomatic event that formed the veins. The actual nature of the metasomatic agent and the timing of the hydration of olivine remain therefore undetermined.

Upper mantle deformation and seismic anisotropy in the North Tanzanian Divergence
Microstructures and CPO analysis of the Tanzanian xenoliths show that the lithospheric mantle in the Northern Tanzanian Divergence is pervasively deformed. However, the strong heterogeneity in microstructures and olivine CPO suggests that this deformation was acquired during multiple tectonic events, probably separated by quiescence episodes, which allowed for annealing. The observed microstructural heterogeneity could also be explained by spatial variations in strain rate. The mylonitic to granular microstructures of the rift axis peridotites, despite rather high equilibrium temperatures, underwent limited annealing, which suggests that they were deformed shortly before extraction. Therefore, these samples likely record the deformation associated with the rift propagation in this region.  (Tommasi et al., 1999). High temperature (1200°C), high strain rate (b 10 −6 s −1 ) torsion experiments in dry olivine aggregates (Bystricky et al., 2000;Hansen et al., 2014) have shown, however, that [100]-axial CPO also forms during the initial stages of simple shear deformation, with a progressive transition from axial-[100] symmetry to orthorhombic symmetry for γ N 5. The obliquity between subgrain boundaries and the foliation observed in the mylonites and porphyroclastic peridotites are consistent with a simple shear component of deformation. However, the extreme stretching of the pyroxenes in the mylonites (Figs. 3a and 4a) implies high finite strains. Altogether, these observations support that the deformation, which developed the mylonites and porphyroclastic peridotites in the mantle lithosphere beneath the southern part of the East-African rift, results from a transtensional strain regime with a strong simple shear component. Comparison between the seismic properties of the studied xenoliths and seismological data may help to further constrain the deformation regime in the region. Anisotropy measurements in the Tanzania Divergence show fast SKS polarization directions parallel or slightly oblique to the rift axis (Albaric et al., 2014;Bagley and Nyblade, 2013;Gao et al., 1997;Walker et al., 2004). Oblique fast polarization directions for teleseismic shear waves are expected for a rift formed by transtensional deformation of the lithosphere, which should result in axial-[100] olivine CPO patterns with a concentration of olivine [100] axes close to horizontal and slightly oblique to the rift axis (Vauchez et al., 2000). These data, together with the predominance of axial-[100] olivine CPO in the studied mantle xenoliths from Pello Hill and Eledoi, are in good agreement with the suggestion that the East African Rift was initiated by a transtensional reactivation of inherited crystallographic fabrics, as suggested by Tommasi and Vauchez (2001) and Tommasi et al. (2009) who showed, using numerical models, that a CPO-induced mechanical anisotropy in the mantle imposes deformation with a strong strike-slip component parallel to the reactivated mantle structure.
Away from the rift axis, the coarse-granular, highly annealed microstructures of the Lashaine peridotites probably record an older deformation event. The olivine CPO in these peridotites also differs markedly from those from the rift axis: axial-[010] to orthorhombic patterns predominate. The deformation microstructures and CPOs frozen in these peridotites might be related to the last major compressive event in the region: the formation of the Mozambique Belt during the Neoproterozoic East African Orogeny (e.g., Pinna et al., 1993). Considering the similarity in microstructure and composition between the xenoliths from Lashaine, Labait (Vauchez et al., 2005), and the Kaapvaal craton (Baptiste et al., 2012), it may also be hypothesized that the lithosphere beneath Lashaine is a remnant of a cratonic domain embedded and preserved in the Mozambique Belt, as suggested by Gibson et al. (2013) based on the geochemical study of ultra-depleted garnets. In this case, the structures and CPO of Lashaine xenoliths might result from even older deformations, such as those recorded in the neighboring Tanzanian craton (e.g., Bagley and Nyblade, 2013;Walker et al., 2004). Although localized within the transverse volcanic belt, the Olmani mantle xenoliths show both types of microstructures and CPOs: rift-axis or Lashaine ones. This might result from localized deformation reworking an older lithosphere (similar to the Lashaine mantle, or affected by the Aswa transcurrent shear zone) leading to the development of a subordinate branch of the East African Rift southeast of Mount Kilimandjaro (see Fig. 1).
The present study shows that lithospheric mantle beneath the southern part of the East African Rift is anisotropic. Below the rift axis, calculated seismic anisotropy intensities (Fig. 7, Table 1) range between 3.3 and 18.4% for P-waves, and between 2.3 and 13.3% for S-waves. The contribution of the mantle lithosphere cannot therefore be neglected in the interpretation of shear wave splitting data, even if the coherence of the large-scale seismic anisotropy pattern has lead investigators to favor a sublithospheric flow interpretation (e.g., Bagley and Nyblade, 2013). Estimates of the crustal and lithospheric thicknesses in this region vary between 36-44 km and 100-150 km (Dugda et al., 2009;Julià et al., 2005), respectively. Upper mantle S-wave velocities of 4.6 km/s were measured in the region (Julià et al., 2005). Assuming that the rift formed by transtension, we expect the foliation to be near vertical and the lineation close to horizontal and slightly oblique to the rift axis (Vauchez et al., 2000). SKS-waves will then sample the maximum S-wave polarization anisotropy (Y direction in Fig. 6), leading to delay times around 0.65 s for a 60 km-thick mantle, and of 1.2 s for a 110 km-thick mantle. In the North Tanzanian Divergence, Walker et al. (2004) measured SKS delay times of 0.3-0.8 s, while Bagley and Nyblade (2013) and Albaric et al. (2014) obtained mean delay times of 0.7 and 1.2 s, respectively. Therefore, the anisotropy induced by olivine CPOs in the Tanzanian peridotites can explain the SKS delay times measured in the region. Within the main Ethiopian rift, SKS studies also report polarization directions subparallel to the rift trend and variable delay times (0.5-1.7 s: Gashawbeza et al., 2004;1-3 s: Kendall et al., 2005). However, SKS splitting delay times as high as those reported by Kendall et al. (2005) in the Afars cannot be explained by the anisotropies recorded in Tanzanian xenoliths, in agreement with the interpretation that they reflect the presence of oriented melt pockets (Ayele et al., 2004;Bastow et al., 2010;Hammond et al., 2014;Kendall et al., 2005;Sicilia et al., 2008).

Conclusions
Xenoliths sampling the lithospheric mantle beneath the North Tanzanian Divergence in the East African Rift display a high variability in microstructures and olivine CPO depending on their location. Beneath the rift-axis and Olmani, within the volcanic transverse belt, the occurrence of mylonitic to porphyroclastic microstructures suggests recent deformation. Variable microstructures and grain sizes in these rocks suggest lateral and/or vertical variation of the deformation conditions within the mantle, as well as variable degrees of annealing, that may be related to variable interaction with fluids or melts, or to different time spans between deformation and xenolith extraction. The presence of mylonites point to strain localization, but there is no evidence of dominant grain boundary sliding in any of the studied rocks: the ubiquitous dislocation-related intracrystalline deformation features in olivine and orthopyroxene and the strong axial-[100] olivine CPO point to dislocation creep with dominant activation of the [100](010) slip system. The mylonites also record evidence for marked changes in temperature, probably due to transient heating events. Highly stretched orthopyroxene crystals in mylonites from rift axis localities suggest that the deformation initiated under high stress and probably low/moderate temperature conditions, but the evidence of synkinematic olivine crystallization at the expenses of orthopyroxene observed in both mylonitic and porphyroclastic peridotites from rift axis localities and Olmani indicate that deformation continued in the presence of melt, under near-solidus conditions. Finally, ubiquitous exsolutions in orthopyroxene in the mylonites suggest significant cooling between this melt-assisted deformation and the xenoliths extraction. Late, postkinematic metasomatism by hydrous melts is evidenced by the occurrence of veins crosscutting the microstructure, as well as the presence of interstitial clinopyroxene and phlogopite in rift axis peridotites.
In the volcanic transverse belt, the coarse-granular microstructures and well-defined CPOs that characterize the Lashaine and part of the Olmani peridotites likely result from an older deformation event. The well-defined axial-[010] to orthorhombic CPO patterns in these peridotites may be associated with the formation of the Mozambique Belt or with an even older event. Finally, the variability of microstructures and CPOs in Olmani peridotites, which are either Lashaine-like or rift-axis-like, may indicate a heterogeneous deformation within the volcanic transverse belt, probably related to localized deformation associated to the development of the Pangani Graben through transtensional reactivation of the Aswa shear zone (Fig. 1).
Tanzanian xenoliths are variably hydrated (OH contents in olivine vary between 2 and 18 ppm H 2 O wt., Paterson calibration), but olivine OH concentrations do not vary systematically neither between in-and off-axis samples, nor as a function of the microstructure or CPO patterns. Occurrence of OH in olivines with high Mg# implies hydration by metasomatism after partial melting. However, the lack of spatial correlation between OH contents in olivine and the veins containing hydrous phases, implies that the olivine hydration is not directly due to the vein-forming metasomatism.
Maximum P wave azimuthal anisotropy (AVp) ranges between 3.3 and 18.4% and the maximum S wave polarization anisotropy (AVs), between 2.3 and 13.2%. The change in olivine CPO symmetry from a locality to another results in a variation in the seismic anisotropy patterns. Comparison between olivine CPOs and polarization direction of the fast SKS wave is consistent with a rift formed by a transtensional deformation involving reactivation of inherited tectonic fabric.