Phlogopite-Olivine Nephelinites Erupted During Early Stage Rifting, North Tanzanian Divergence

The North Tanzanian Divergence (NTD, eastern branch of the East African Rift) corresponds to an early stage of continental breakup. In the southern NTD, two quaternary volcanoes of the Manyara-Balangida rift (Labait, Kwaraha) have erupted primary nephelinite lavas (Mg# = 79–57) that allow characterization of their deep mantle source and the alkaline magmas that percolated through the lithosphere during rift initiation. Nephelinites are olivine- and clinopyroxene-rich, and contain up to 4 vol% magmatic phlogopite that crystallized as a liquidus phase with olivine and clinopyroxene. The presence of hydrous mineral (phlogopite) phenocrysts in Kwaraha and Labait lavas strongly suggests that the alkaline melts were H2O-bearing at the time of phlogopite crystallization (1.57–2.12 wt% H2O in phlogopite), demonstrating that phlogopite may have influenced the partitioning of water between the silicate melt and anhydrous silicate minerals (<1 ppm wt H2O clinopyroxene, 1–6 ppm wt H2O in olivine). Geochemical modeling indicates that the nephelinite magmas resulted from a low degree of partial melting (0.2–1%) of a carbonate-rich (0.3%) garnet peridotite containing ∼2 vol% phlogopite. We estimate the depth of partial melting based on primary melt compositions and empirical relations, and suggest that melting occurred at depths of 110–130 km (4 GPa) for craton-edge lavas (Kwaraha volcano) and 150 km (5 GPa) for on-craton lavas (Labait volcano), close to or below the lithosphere-asthenosphere boundary in agreement with the presence of deep refractory mantle xenoliths in Labait lavas. The depth of melting becomes gradually deeper toward the southern NTD: highly alkaline magmas in the north (Engaruka-Natron Basin) are sourced from amphibole- and CO2-rich peridotite at 75–90 km depth, whereas magmatism in the south (south Manyara Basin) is sourced from deep phlogopite- and CO2-rich garnet-peridotite beneath the Tanzania craton (e.g., at the on-craton Labait volcano). Percolation of deep asthenospheric CO2-rich alkaline magmas during their ascent may have produced strong heterogeneities in the thick sub-continental lithospheric mantle by inducing metasomatism and phlogopite crystallization in glimmerite lithologies.

The North Tanzanian Divergence (NTD, eastern branch of the East African Rift) corresponds to an early stage of continental breakup. In the southern NTD, two quaternary volcanoes of the Manyara-Balangida rift (Labait, Kwaraha) have erupted primary nephelinite lavas (Mg# = 79-57) that allow characterization of their deep mantle source and the alkaline magmas that percolated through the lithosphere during rift initiation. Nephelinites are olivine-and clinopyroxene-rich, and contain up to 4 vol% magmatic phlogopite that crystallized as a liquidus phase with olivine and clinopyroxene. The presence of hydrous mineral (phlogopite) phenocrysts in Kwaraha and Labait lavas strongly suggests that the alkaline melts were H 2 O-bearing at the time of phlogopite crystallization (1.57-2.12 wt% H 2 O in phlogopite), demonstrating that phlogopite may have influenced the partitioning of water between the silicate melt and anhydrous silicate minerals (<1 ppm wt H 2 O clinopyroxene, 1-6 ppm wt H 2 O in olivine). Geochemical modeling indicates that the nephelinite magmas resulted from a low degree of partial melting (0.2-1%) of a carbonate-rich (0.3%) garnet peridotite containing ∼2 vol% phlogopite. We estimate the depth of partial melting based on primary melt compositions and empirical relations, and suggest that melting occurred at depths of 110-130 km (4 GPa) for craton-edge lavas (Kwaraha volcano) and 150 km (5 GPa) for on-craton lavas (Labait volcano), close to or below the lithosphereasthenosphere boundary in agreement with the presence of deep refractory mantle xenoliths in Labait lavas. The depth of melting becomes gradually deeper toward the southern NTD: highly alkaline magmas in the north (Engaruka-Natron Basin) are sourced from amphibole-and CO 2 -rich peridotite at 75-90 km depth, whereas magmatism in the south (south Manyara Basin) is sourced from deep phlogopite-and CO 2 -rich garnet-peridotite beneath the Tanzania craton (e.g., at the on-craton Labait volcano). Percolation of deep asthenospheric CO 2 -rich alkaline magmas during their ascent may have produced strong heterogeneities in the thick sub-continental lithospheric mantle by inducing metasomatism and phlogopite crystallization in glimmerite lithologies.

INTRODUCTION
Silica-undersaturated magmas are abundant in continental rifts, including highly alkaline lavas such as melilitite, nephelinite, basanite, and phonolite. Low-silica lavas are commonly CO 2 -rich and have been described as the product of CO 2 -bearing mantle domains (e.g., Brey, 1978;Dasgupta et al., 2007) associated with carbonatite, suggesting a cogenetic origin (Woolley and Kjarsgaard, 2008). The oldest volcanism of the East African Rift (EAR, ∼45 Ma) occurred in Ethiopia associated with the propagation of the EAR toward the south (Ebinger et al., 2000) and produced abundant highly alkaline rocks and carbonatites (e.g., Pouclet et al., 1981;Foley et al., 2012). Melilitites and nephelinites were erupted at the tip of the propagating rift and around the Tanzanian craton (i.e., Tanzania, Uganda; Foley et al., 2012). In northern Tanzania, primitive lavas, erupted generally in monogenetic volcanic fields (e.g., the Engaruka-Natron monogenetic field), are characterized by small eruptive volumes, little fractional crystallization during ascent, relatively high ascent rates, and xenolith-bearing lavas (Mattsson et al., 2013).
Primary alkaline lavas erupted on and at the edge of the Tanzanian craton sampled mantle xenoliths that indicate the presence of heterogeneous sub-cratonic lithospheric and asthenospheric mantle domains related to metasomatism by carbonated (Jones et al., 1983;Rudnick et al., 1993;Lee and Rudnick, 1999;Foley et al., 2012) and/or H 2 O-rich fluids (Dawson and Smith, 1988;Koornneef et al., 2009). The xenolith data suggest that primary lavas originate from deep asthenospheric mantle processes. Furthermore, alkaline lavas are volatile-rich (e.g., Ivanikov et al., 1998;Keller et al., 2006;Métrich and Wallace, 2008;Hudgins et al., 2015) and could only derive from deep, low-degree partial melting of H 2 O-CO 2bearing peridotites. The depth of origin, degree of partial melting, and mantle source control the alkalinity of the primary melt (e.g., Maaløe et al., 1992;Rogers et al., 1992). The diversity of alkaline lavas (melilitite, nephelinite, and basanite) erupted in the North Tanzanian Divergence (NTD) clearly indicates partial melting at various levels in the sub-cratonic lithosphere and asthenosphere (e.g., Rosenthal et al., 2009;Foley et al., 2012;Mana et al., 2015). Further differentiation by fractional crystallization, liquid immiscibility, and assimilation produced Mg-rich nephelinite, Mg-poor nephelinite, and carbonatite at different levels in the mantle and continental crust during ascent (Klaudius and Keller, 2006;Zaitsev et al., 2012;Baudouin et al., 2016).
In this paper, we performed a petrological and geochemical investigation of the most primitive lavas of the NTD, i.e., nephelinites from the Manyara basin (Kwaraha and Labait volcanoes) to determine the mantle conditions of silicaundersaturated magma genesis at the first stage of continental break-up. We discuss the presence of magmatic phlogopite in terms of alkaline magma composition and crystallization environment of these magmas, and model the partial melting of a deep metasomatized mantle source of the highly alkaline magmas beneath the eastern part of the Tanzanian Craton.

Geological Background
The EAR is divided into two branches which correspond to different stages of plate boundary extension, from rift initiation (eastern branch initiation: Manyara-Balangida basin, Tanzania, and western branch initiation: Toro-Ankole basin, Uganda) to oceanic stage in the Afar Triple Junction. In the eastern branch, volcanism began 30 Ma in northern Kenya, 15 Ma in central Kenya, and 6 Ma in northern Tanzania (Mana et al., 2012(Mana et al., , 2015. The north Tanzanian rift splays to form the NTD, which includes large volcanic complexes (e.g., Ngorongoro, Meru, and Kilimanjaro) and small volcanic cones (e.g., Lashaine, Olmani) (Dawson et al., 1970(Dawson et al., , 1997Jones et al., 1983). The southern NTD is divided into three rift areas: Eyasi, Manyara-Balangida, and Pangani, from west to east (Le Gall et al., 2008; Figure 1). The Eyasi and Pangani fault systems are amagmatic and represent rift propagation from the Natron basin to the western and eastern parts of the NTD, respectively. The Manyara-Balangida basin, representing southward rift propagation, lies along the rift escarpment and includes two volcanic centers (Hanang and Kwaraha volcanoes, Supplementary Figures A1, A2) surrounded by small volcanic cones (Labait and Sora Hill, respectively, Dawson, 2008). East of the Manyara basin, several volcanic centers have erupted relatively evolved nephelinite lavas, whereas small volcanic cones in the Monduli-Meru area have erupted primitive lavas. The most primitive lavas were erupted at Kwaraha and the small cones around Hanang and Meru. We sampled Labait and Kwaraha for this study (Figure 1).

Major and Volatile Element Analyses
Whole-rock major element compositions were measured by inductively coupled plasma optical emission spectroscopy (iCap 6500 Thermo Fisher) at the Service d'Analyse des Roches et des Minéraux (SARM) at the Centre de Recherches Pétrographiques et Géochimiques (Nancy, France) following the protocol established by Carignan et al. (2001). One gram of whole-rock powder was dissolved with HNO 3 and the mixture (with LiBO 2 ) was fused. The reference standard was SLRS-5 and errors are estimated to be <2% (1σ).
Complementary whole-rock analyses has been performed by wide-angle X-ray fluorescence using sequential spectrometer Bruker S4 Pioneer at the analytical services of the Instituto Andaluz de Ciencias de la Tierra (IACT, University of Granada, Spain) using Rh X-ray tube (160 kV, 159 mA). Rock powders (1 g) are weighed with di-lithium tetraborate flux, and then the mixture is fused at 1000 • C for 15 min. The concentrations of major elements in the samples are measured by comparing the X-ray intensity for each element with nine reference geological standard samples. Whole-rock sulfur and carbon contents were determined for each sample via elemental analyser, F and Cl contents were determined by wet precipitation-ferrithiocyanate spectrophotometry using a Varian Cary 50 spectrophotometer. High LOI (up to 8 wt%), and CO 2 content (up to 4.6 wt%) in nephelinitic lavas are due to the presence of small size secondary minerals within the groundmass from hydrothermal alteration (i.e., calcite, zeolite). Such minerals are common in nephelinites as reported at Mount Etinde (Cameroon; Etame et al., 2012), Engaruka volcanic field (Neukirchen et al., 2010), Cape Verde (Mourão et al., 2012), and as well worldwide nephelinites and melilitites (Georoc database 1 ; Supplementary Table A1).
Mineral major and volatile element concentrations were determined via electron microprobe (Cameca XS100 at Geoscience Montpellier, France). Analyses were performed with an accelerating voltage of 20 keV, a 10 nA beam current, and a focused (1 µm) beam. The counting time was fixed at 20 and 40 s for major and volatile element (S, Cl) analyses, respectively. The standards used for major and volatile element analyses were wollastonite for Si and Ca, Al 2 O 3 for Al, TiO 2 for Ti, forsterite for Mg, hematite for Fe, orthoclase for K, albite for Na, apatite for P, native metal for Ni, Mn, and Cu, barite for S and Ba, fluorite for F, and chlorapatite for Cl.

Trace Element Analyses
Whole-rock trace element analyses of lavas from Kwaraha and related small volcanic cones (samples Kw1-Kw8), and Labait were performed by ICP mass spectrometry (ICP-MS) after HNO 3 and HF digestion of 0.1 g of sample powder in a Teflonlined vessel at 180 • C and 200 psi for 30 min, evaporation to dryness, and subsequent dissolution in 100 mL of 4 vol% HNO 3 . Triplicate measurements were performed with a NexION 300d (Perkin Elmer) ICP-MS at the Instituto Andaluz de Ciencias de la Tierra (University of Grenada) using Rh as an internal standard, and using multi-element calibration solutions for external calibration. Analytical precision was better than ± 5% for concentrations > 10 ppm. Whole-rock analyses of samples Kw10, Kw11, and Kw12 were performed using a quadrupole 7700x ICP-MS at the Analyse des Eléments en Trace dans l'Environnement (AETE platform, OSU-OREME, University of Montpellier) where 0.1 g of whole-rock powder was dissolved with acid (HF-HNO 3 ). Blanks spiked with In and Bi were prepared to monitor instrumental drift. Solutions were analyzed at a final dilution factor of 4000. The sensitivity of the ICP-MS in this configuration was 200 × 10 6 /ppm 115 In. Analytical accuracy was estimated from measurements of international rock standards UBN and BEN for both analytical procedures (Supplementary Table A1).
Mineral trace element concentrations were determined by laser ablation ICP-MS at AETE using a GeoLas Q+ Excimer CompEx 102. A 26 and 56 µm diameter laser beam was used for apatite and silicate minerals, respectively, with a laser repetition rate of 6-10 Hz and laser power of 0.5 mJ (5 J·cm −2 ). The spot size was chosen as a compromise between signal intensity and the size of the minerals of interest in the samples. Concentrations were calibrated with glass standard NIST612 and SiO 2 and CaO concentrations previously determined by electron microprobe. The BIR-1 standard was used as an external 11 http://georoc.mpch-mainz.gwdg.de/georoc/ standard. Glitter Software (Griffin et al., 2008) was used to process the raw data files (signal intensity vs. time) into elemental concentrations. This allows precise selection of blanks and signals, and rapid visualization of the intensity data. Instrumental drift was compensated by internal standard calculations using Glitter; no other drift corrections were performed.

Water Content Analyses
The water contents of clinopyroxene and olivine (ol) were determined using Fourier transform infrared (FTIR) spectroscopy. Handpicked minerals were prepared as doubly polished sections with thicknesses between 140 and 325 µm. The hydrogen concentration was measured in transmission mode at the Laboratory Charles Coulomb (University of Montpellier, France). For olivine crystals, FTIR analyses were performed in the center of crystals far from iddingsite rims. Unpolarized infrared spectra were acquired using a Bruker IFS 66v coupled with a Hyperion optical microscope. A Globar light source and a Ge-KBr beam splitter were used to generate unpolarized mid-infrared radiation (4000-400 cm −1 ). Water contents were calculated by integration of spectra between 3770 and 3000 cm −1 for clinopyroxene and between 3610 and 3000 cm −1 for olivine (for the detailed analytical method, see Denis et al., 2015). Water concentrations were calculated using the calibration of Paterson (1982), considering the water concentration as a function of the density of clinopyroxene (χ clinopyroxene ( (Paterson, 1982) and uncertainties on the sample thickness and background correction are ± 15% (Denis et al., 2015). The integrated unpolarized absorbances normalized to 1 cm sample thickness are also reported to use alternative mineral-specific FTIR calibrations.
The water content of phlogopite from sample Kw3 was measured on mineral separates by Karl-Fischer titration (KFT) using air as the transporting gas and muscovite as a standard (σ = 0.1 wt% H 2 O) at the University of Hannover (Behrens et al., 1996). Whole rock H 2 O content were measured by Karl Fischer titration at SARM. Standard solutions were used to check the accuracy of spectrophotometry analyses, and standard deviations are less than 5%.

RESULTS
All samples from Kwaraha (lava flows) and Labait (scoria) have microlitic porphyric textures with large phenocrysts of olivine and clinopyroxene (>1 mm) (Figure 2). Following the nomenclature of Le Bas (1989), the bulk rock compositions of Kwaraha lavas and Labait scoria are foidites with compositions between melilitite and nephelinite, similar to nephelinites reported in north Tanzania (Figure 3; Dawson et al., 1970Dawson et al., , 1997Mana et al., 2012). Because of the similar mineral assemblages and chemical compositions of all samples, as well the absence of melilite minerals, we refer to them herein as nephelinites.
In the following sections, we discuss the mantle partial melting environment leading to the genesis of silica-undersaturated alkaline magmas, the role of volatiles (H 2 O, CO 2 ) in magma differentiation and metasomatism, and the interaction between magmas and the Tanzanian lithospheric mantle. We conclude with a comparison of the characteristics of magmatism during the early stages of rift initiation in the western and    Lee and Rudnick, 1999;Koornneef et al., 2009) and north Tanzania Smith, 1988, 1992) and from Bunyaruguru volcanic field (Uganda; Muravyeva and Senin, 2018). Gray and black rectangles for phlogopite composition in MARID and PIC, respectively, from Dawson and Smith (1977); Jones et al. (1982), Kramers et al. (1983); Waters (1987), and Grégoire et al. (2002). Lab: Labait; pyr: pyroxenite; pdt: peridotite.
eastern branches of the EAR (e.g., Uganda-Rwanda and north Tanzania, respectively).

Phlogopite-Bearing Primary Alkaline Magma
Primary magmas as product of partial melting of mantle peridotite are very rare at the Earth surface as the composition of magmas can be modified in their source deeper in the mantle by crystallization and/or during ascent by interaction with or assimilation of lithospheric mantle (Huppert and Sparks, 1985;Streck et al., 2007). The compositional range observed in Labait and Kwaraha nephelinites may suggest olivine accumulation (<10%ol; Mg# = 76-79) ( Table 1 and Figure 4) and fractional crystallization during ascent (Kwaraha lavas with Mg# < 65), respectively (e.g., Frey et al., 1978). However, the lavas erupted at Labait and Kwaraha volcanoes are the more primitive magmas of the southern part of the NTD (i.e., Mg# > 65) and their major and trace element composition is consistent with experimental primary alkaline melt from partial melting of mantle peridotite (Table 1 and Figures 3, 4, 8; e.g., Mattsson et al., 2013;Green and Falloon, 2015). In nephelinites, olivine and clinopyroxene are the liquidus phases and coexist with phlogopite phenocrysts. The presence (B) Representative chondrite-normalized whole-rock REE pattern of Kwaraha lavas, and the silicate liquids in equilibrium with phlogopite (phl) calculated using partition coefficients for kimberlite (Fujimaki et al., 1984) and basanite (Adam and Green, 2006). The chondrite-normalized REE pattern of kimberlite (Eccles et al., 2004) is shown for comparison with the calculated liquids. Chondrite and primitive mantle values from Sun and McDonough (1989).
of phlogopite in nephelinite lavas carrying phlogopite-bearing mantle xenoliths has led to the consideration that phlogopites are xenocrysts assimilated from the lithospheric mantle (e.g., Johnson et al., 1997;Lloyd et al., 2002). However, our analyses of phlogopite crystals in lavas and mantle xenoliths show that they have different major and trace element compositions, suggesting different crystallization environments. Phlogopite from Labait lavas have lower Mg# and Cr 2 O 3 and Al 2 O 3 contents, and higher trace element concentrations than phlogopite from glimmerite (igneous rocks mainly composed of mica) and phlogopitebearing peridotite mantle xenoliths from the NTD (Mg# = 79-83 and 90-93, respectively, Table 2 and Figure 6), suggesting that they crystallized early from primary trace-element-rich magmas. In Kwaraha nephelinites, phlogopites occur as inclusions in clinopyroxene and as phenocrysts in the groundmass (Figure 2) and all phlogopites have the same range of composition with high Mg# (up to Mg# = 83) suggesting an early crystallization ( Figure 2D). The variable phlogopite compositions in Kwaraha lavas (Mg# = 83-76, Figure 6) correspond to their crystallization from slightly more evolved magmas with higher Al 2 O 3 and trace element concentrations than phlogopite from Labait. Using available experimentally determined partition coefficients (D mineral/melt = X mineral /X melt , where X is the concentration of a given element in each phase) between phlogopite and alkaline and potassic magmas (i.e., basanite, kimberlite, lamproite; Fujimaki et al., 1984;Schmidt et al., 1999;Adam and Green, 2006, respectively), the calculated silicate liquid compositions in equilibrium with phlogopite from Kwaraha and Labait have high REE ( Figure 7B) and trace element concentrations similar to the whole-rock nephelinite compositions. This suggests that early phlogopite crystallization occurred at the liquidus with clinopyroxene and olivine, possibly from primitive K-rich silicate liquid. These observations agree with partial melting and phase equilibria experiments in which phlogopite crystallized with clinopyroxene at high pressure (1-5 GPa) in potassic magma (Parat et al., 2010;Förster et al., 2018).
We note that the early crystallization of phlogopite in alkaline magmas may have slightly lowered the potassium content of these magmas during crystallization (5 vol% phlogopite corresponds to ∼0.5 wt% K 2 O). Therefore, their primary magma may have had a composition close to that of potassic melt produced by incongruent melting of a mica + amphibole + rutile + ilmenite + diopside assemblage and harzburgite at 5 GPa, which provides all the constituents for rapid growth of phlogopite and clinopyroxene (Förster et al., 2018). The high CaO contents of nephelinite NTD magmas may be related to the presence of clinopyroxene and lherzolite, rather than harzburgite, in their mantle source.

Crystallization Environment and Volatile Concentrations
Temperature, pressure, and volatile contents are key parameters controlling phase equilibria, mineral, melt, and gas compositions, and further magmatic evolution including fractional crystallization, immiscibility, and degassing processes (e.g., Carmichael and Ghiorso, 1990;Toplis and Carroll, 1995;Berndt et al., 2005). The composition of clinopyroxene in mafic magmas is strongly dependent on melt composition, pressure, and temperature during crystallization, and can be used as a geothermobarometer (e.g., Putirka, 2008;Masotta et al., 2013). Clinopyroxene and their associated bulk nephelinites have Kd (Fe 2+ /Mg) values ranging from 0.2 to 0.41 for Kwaraha, suggesting that clinopyroxene are near equilibrium with the silicate liquid and thus suitable for calculation of the pressure   Figure A9). Simonetti et al. (1996) proposed a correlation between the Al VI /Al IV ratio of clinopyroxene and their crystallization pressure, and the high clinopyroxene Al VI /Al IV ratios (0.89-0.56) from Kwaraha are consistent with crystallization at 1000 MPa. Volatiles such as CO 2 and H 2 O affect the conditions of partial melting (e.g., Dasgupta et al., 2007;Baasner et al., 2016), play a fundamental role during magmatic evolution and ascent, and control eruptive style (e.g., Oppenheimer et al., 2003;Berndt et al., 2005). The presence of phlogopite in Kwaraha and Labait lavas suggests that the silicate melts were H2Obearing at the time of phlogopite crystallization (1.57-2.12 wt% H2O in phlogopite, KFT method). However, the abundant water in crystal lattice of phlogopite (OH site) prevents us from quantifying precisely the amount of water present in the melt (i.e., non-Henrian behavior). In experimental study, hydroxyl- Only nominally anhydrous minerals (NAMs) such as olivine and clinopyroxene incorporate trace concentrations of hydrogen that can be used to determine the H 2 O content of the silicate melt during their crystallization (D clinopyroxene/melt = 0.01-0.013, D olivine/melt = 0.0013; Aubaud et al., 2004;Hauri et al., 2006). The measured water contents are very low: < 1 ppm wt H 2 O in clinopyroxene and olivine from Kwaraha lavas ( Figure 5) and 3-6 ppm wt H 2 O in olivine from Labait. Using the partition  (Condamine et al., 2016). Although primary melts from Labait and Kwaraha are hydrated, the water content was probably not high enough to favor phlogopite crystallization over olivine (Foley, 1993). CO 2 contents in silicate melts are generally investigated through melt inclusion studies (e.g., Métrich and Wallace, 2008;Hudgins et al., 2015;Baudouin et al., 2018). The lack of melt inclusions in minerals from Labait and Kwaraha lavas prevents  us from precisely constraining the CO 2 content of the nephelinite magmas. However, northern Tanzania contains some of the most concentrated carbonatite magmatism on Earth (e.g., Dawson, 2012) and the occurrence of calciocarbonatite lava (Kw1) at Kwaraha volcano strongly suggests the presence of abundant CO 2 in the primary alkaline magmas. CO 2 solubility in alkaline magmas (e.g., melilitite, phonolite) was experimentally calibrated at 1.5-2 GPa, and strongly depends on the ratio of non-bridging oxygens to tetrahedrally coordinated anions (NBO/T) in the melt (Brooker et al., 2001;Moussallam et al., 2015). Based on the observed compositions of nephelinite lavas (NBO/T = 1.4-1.67 for Labait and 0.93-1.26 for Kwaraha), their CO 2 solubilities (i.e., the maximum amount of CO 2 that can be dissolved in the magma) range from 7.5 to 15.8 wt% CO 2 (Supplementary Figure A10). The primary magma from Labait evolved to a Mgpoor nephelinitic liquid containing up to 6 wt% CO 2 at 800 MPa (Hanang volcano; Baudouin et al., 2018); CO 2 contents of Labait nephelinites may indicate that fractional crystallization and immiscibility buffered the CO 2 content during differentiation (<6 wt% CO 2 ).

Partial Melting of Metasomatized Deep Mantle
Alkaline mafic lavas such as melilitites and nephelinites represent more primitive magmas erupted in intracontinental settings, and are characterized by silica-undersaturation and very high Mg contents (Dasgupta et al., 2007;Foley et al., 2012). Their high incompatible element concentrations and strong REE fractionations (Ivanikov et al., 1998;Platz et al., 2004;Keller et al., 2006) have been proposed to result from very low degrees of partial melting of CO 2 -bearing lherzolite (Green and Falloon, 1998;Foley et al., 2009Foley et al., , 2012Foley and Fischer, 2017) or the partial melting of peridotite with contributions from recycled oceanic crust (e.g., Hofmann, 1988) or pyroxenite (Eggler and Holloway, 1977;Dasgupta et al., 2007;Baasner et al., 2016). Around the Tanzanian craton, nephelinite volcanoes are concentrated at the propagating tips of the rift and mafic and carbonate-rich alkaline melts are associated with thick lithosphere (Foley and Fischer, 2017). The presence of deep garnet-bearing mantle xenoliths in Labait scories indicate that partial melting occurred at least at 150 km (Lee and Rudnick, 1999).
Geochemical modeling results of the partial melting of mantle peridotite to produce Labait and Kwaraha nephelinite magmas corroborate the presence of refractory HREE-rich minerals such as garnet in the mantle source (Figure 9). Trace elements such as Y and Rb are compatible in garnet and phlogopite, respectively, which can account for the high Ce/Y (>5) and Rb/Sr ratios in the nephelinite melt (Rogers et al., 1992;Platz et al., 2004;Mana et al., 2015). Partial melting of a garnet peridotite with 62% olivine, 16% opx, 12% clinopyroxene and 10% garnet (theoretical mineral assemblage in mantle from Sun and McDonough, 1989) produces melts with high Ce/Y (∼3), low Rb/Sr (<0.04), and low Zr/Hf ratios that cannot represent the composition of nephelinite (i.e., Ce/Y > 5, Rb/Sr > 0.04, Zr/Hf = 39.2-46). The high Rb/Sr ratios (>0.04) in the nephelinites strongly suggest a contribution of phlogopite (Rogers et al., 1992;Platz et al., 2004), whereas their high Zr/Hf ratios indicate the contribution of a carbonated source (Rudnick et al., 1993). We modeled the partial melting of carbonatite-free and carbonated (0.3 and 1% carbonatitic component) garnet-rich and phlogopite-bearing peridotite mantle (62% olivine, 16% opx, 10% clinopyroxene, 10% garnet, 2% phlogopite, proportion modified from Sun and McDonough, 1989), as well as a natural garnet-bearing xenolith from Lashaine volcano in the north part of the NTD (Gibson et al., 2013).
The best fit to the observed data suggests that nephelinites result from a very low degree of partial melting (<1%) with nonmodal fractional melting (Figure 9 and Supplementary Figure  A11). The low volumes of melts (<1%) is generally assume to be motionless in the mantle rocks due to their small permeability and wetting angle characteristics (Maumus et al., 2004). However, recent study based on physical properties and melt-rock reaction suggest that small volume of melts (i.e., 0.75%) could be extracted from the mantle (Soltanmohammadi et al., 2018). The extraction of low volume melt will be effective for a melt with high content in alkaline and volatile elements (Keller and Katz, 2016;Soltanmohammadi et al., 2018), the volatile content being a key parameter to enhance the melt extraction process due to their effect on the melt viscosity properties (Condamine and Médard, 2014;Gardès et al., 2020). In Labait and Kwahara lavas, early phlogopite crystallization suggests that the primary magmas have higher K content than the erupted lavas (nephelinites), and high volatile contents, especially in CO 2 , allowing low melt volume 0.5-1 vol% to be extracted from the mantle (Keller and Katz, 2016;Gardès et al., 2020).
Small amounts of carbonate in the source reproduce the high Zr/Hf fractionation observed in Labait lavas (Zr/Hf = 42.6-46) (Figure 8). The trace element variability of lavas from Labait and Kwaraha may suggest that the lavas originate from different degrees of partial melting (0.2-1%) and/or variable carbonate components in their mantle sources (0.3%, Figure 9B). The mantle source may be homogeneous (i.e., phlogopite-bearing lherzolite) or mixed heterogeneous source (i.e., 90% phlogopitefree lherzolite + 10% phlogopite-pyroxenite), as suggested by isotopic signatures (Paslick et al., 1996;Aulbach et al., 2011), both leading to similar trace element characteristics in primary melts. It should be noted that pyroxene and phlogopite (i.e., mica pyroxenite) as main component of the mantle source would lead to the genesis of highly alkaline melt with trace element content higher than those reported in Labait and Kwaraha lavas.
Using the empirical equation of Albarède (1992), calibrated for partial melting producing basaltic melts, the SiO 2 and MgO contents of the primary melts suggest that the melting pressure was 110-130 km (3.7-4.3 GPa) for Kwaraha, and 150 km (5 GPa) for Labait, below or close to the lithosphereasthenosphere boundary (LAB at 146 km close to the Tanzanian craton beneath Labait and 135 km beneath Kwaraha, Craig et al., 2011; Figure 10). The depth and temperature of partial melting are consistent with (i) the presence of garnet in the mantle source (>90 km; McKenzie and O'Nions, 1991;Sato et al., 1997), (ii) the presence of mantle xenoliths equilibrated at 100-147 km depth (3.2-4.9 GPa; Lee and Rudnick, 1999), and Ce/Y  Gibson et al., 2013) has also been modeled. The degree of melting (F) is indicated for each composition. Nephelinites (neph.) and melilitites (mel.) from Engaruka-Natron basin (Mattsson et al., 2013) and nephelinites from Essimingor (Mana et al., 2012) are also reported. Mafic nephelinites from Kwaraha (Mg# > 65) are reported in blue circles, and less mafic lavas from Kwaraha (Mg# < 65, Table 1) are reported in light blue circles.
(iii) magnetolleruric data by Selway et al. (2014) suggesting that beneath Labait, the geotherm was initially cratonic at 44 mW/m 2 , before been impacted by the plume ascent, increasing the mantle temperature of 300 • C at 5 GPa, which may lead to melting of fertile lherzolite (Figure 10). The presence of phlogopite in Kwaraha and Labait magmas indicates that the asthenospheric mantle source is hydrated, which may considerably decrease the solidus temperature (e.g., Green et al., 2014). The water content of the nephelinite magma estimated from the water content of olivine and clinopyroxene was 0.2-0.5 wt% H 2 O ( Table 4). Considering 1% partial melting, the garnet-rich and phlogopite-bearing peridotite source may have contained around 300 ppm H 2 O (Novella et al., 2015;Demouchy and Bolfan-Casanova, 2016).
The presence of carbon in mantle source is an important parameter to consider for partial melting at the lithosphere-asthenosphere boundary because (i) it induces a decrease of the solidus temperature of peridotite (Dasgupta et al., 2007) and (ii) below 5 GPa (below LAB), carbon is stable as graphite and it will be oxidized at 4-5 GPa to produce carbonate melt (redox melting, Figure 10; Green and Falloon, 1998;Stagno et al., 2013;Hammouda and Keshav, 2015). Experiments of partial melting of CO 2 -and phlogopite-bearing garnet peridotite at high pressure (>4-5 GPa) would help to better understand the  (Putirka, 2008) for Kwaraha (Mg# > 65) and P-T conditions of partial melting (Albarède, 1992) for Kwaraha and Labait lavas. P-T conditions for Labait and Lashaine mantle xenoliths are from Lee and Rudnick (1999) LR99 and Gibson et al. (2013) G13, respectively. Crystallization temperatures of olivine are represented by color bars (blue: Kwaraha, and orange; Labait; calculation from Putirka, 2008). We include the dry peridotite and peridotite+CO 2 solidi and melt CO 2 isopleths with CO 2 content in weight percent from Dasgupta et al. (2013), the spinel-garnet boundary from Lee et al. (2000), and the lithosphere-asthenosphere boundary (LAB) from Craig et al. (2011). The curve of phlogopite stability (phl) correspond to experiments (KMASH) from Yoder and Kushiro (1969); Sato et al. (1997), and Trønnes (2002). The reaction (A) F-phl+cpx+grt = ol+opx+melt has been determined from melting experiments of phlogopite-garnet peridotite by Condamine et al. (2016). Cratonic geotherm (44 mW/m 2 ) and geotherm modified by plume thermal anomaly are also reported (Selway et al., 2014), as well as the different melt types produced by low partial melting of CMAS ± H 2 O and CO 2 : Colored balloons represent P and T conditions of partial melting for basanite (blue balloon), nephelinite, melilitite, and leucitite melt compositions (pink balloon) (Frey et al., 1978;Adam, 1990;Green and Falloon, 1998).
melting conditions of primary nephelinite melt as observed at Labait and Kwaraha volcanoes in the southern part of the NTD.

Melt-Rock Interaction and Lithospheric Metasomatism
Percolation of CO 2 -and water-bearing alkaline nephelinite magmas from the LAB through the lithospheric mantle may have induced metasomatism and phlogopite crystallization in garnet and spinel lherzolite and glimmerite lithologies. The large variety of mantle xenoliths sampled by nephelinite lavas at Labait and other localities of NTD strongly indicates that the lithospheric mantle beneath the Tanzanian Craton edge is highly heterogeneous, including fertile and refractory mantle peridotite (i.e., Cr spinel with high Cr#; Cr# up to 94), and metasomatized amphibole-and/or phlogopite-bearing peridotite (Dawson and Smith, 1988;Lee and Rudnick, 1999;Koornneef et al., 2009;Gibson et al., 2013;Baptiste et al., 2015). The study of magmatic phlogopite in nephelinite and mantle phlogopite in xenoliths from Labait and Kwaraha corroborates previous studies that established the nature and the timing of the metasomatism events. Xenoliths from Labait, Lashaine, Pello hill, and Eledoi (Natron basin) suggest the percolation of a H 2 O-rich metasomatic agent within the lithospheric mantle and the presence of heterogeneous lithosphere related to different metasomatic events (Jones et al., 1983;Dawson and Smith, 1988;Rudnick et al., 1993;Koornneef et al., 2009;Baptiste et al., 2015). NAMS minerals in lithospheric peridotites have a low water content (<50 ppm wt H 2 O whole rock peridotite; Baptiste et al., 2015;Hui et al., 2015) although the presence of hydrous minerals as amphibole and phlogopite indicates the presence of K 2 O-MgO-H 2 O-rich silicate melt in the lithosphere. The composition of phlogopite in mantle xenoliths from Labait (i.e., glimmerite, core, and rim of phlogopite in lherzolite, Figure 6) indicate at least 3 metasomatic events within the continental lithosphere. Multiple stage of metasomatism in the lithosphere beneath Labait have already been suggested from textural, chemical and isotopic data (Koornneef et al., 2009): Phlogopite, Cr-rich diopside and spinel pockets were produced by an old metasomatism event (i.e., Pan-African Orogeny, 610-650 Ma) associated with a liquid rich in Ni, Cr, Na, K, H 2 O, Ba, Fe, Ti. Glimmerite lithologies could be formed by recent metasomatism (rift event) and melt differentiation in veins in the lithosphere, induced by the percolation of nephelinite liquid rich in K, Fe, Ti, and poor in Cl, Mg, Al, Ba (Figure 10 and Table 3). Rift related metasomatism event was previously suggested as anhydrous from textural observation for a spinel lherzolite (Koornneef et al., 2009). However, it can be noted that glimmerite xenoliths (or glimmerite) have been poorly sampled and represent only 2% of Labait xenoliths (Koornneef et al., 2009). This may lead to a sample bias for phlogopite interpretation and hydrous metasomatism.
In the NTD, volcanism of different alkalinities occurs along two perpendicular axes (Figure 1): the N-S axis erupted highly alkaline magmas such as nephelinite (e.g., Oldoinyo Lengai-Labait; this study; Mana et al., 2012;Mattsson et al., 2013), whereas the W-E axis erupted dominantly alkaline magmas (Essimingor-Kilimanjaro; Nonnotte et al., 2011;Mana et al., 2012Mana et al., , 2015 and sub-alkaline magmas in the western part close to the Tanzanian Craton (i.e., Ngorongoro-Olmoti; Nonnotte, 2007;Mollel et al., 2008Mollel et al., , 2009. Although there is a clear correlation between the alkalinity of magmas and their geographical alignment, we found no obvious correlations between eruptive age, alkalinity, depth of partial melting, or mantle source. Along the N-S axis of the NTD, highly alkaline magmas are the product of very low degrees of partial melting (<1%; this study; Mattsson et al., 2013) of different mantle sources at various depths. From north to south, the depth of partial melting increases and the mantle sources change from (1) CO 2rich amphibole-lherzolite at shallow depth (∼75-90 km) in the Engaruka-Natron basin (Oldoinyo Lengai; Mattsson et al., 2013), to (2) amphibole-bearing garnet lherzolite at intermediate depth (∼110-140 km at Burko; Mana et al., 2012), and finally to (3) CO 2 -rich phlogopite-bearing garnet lherzolite near or below the LAB at 150-160 km depth (Labait; this study; Dawson et al., 1997). The estimated depth of partial melting in the southern part of the NTD is well supported by the presence of deep refractory mantle xenoliths in Labait lavas (>150 km, Figure 10; Lee and Rudnick, 1999) and the thick cratonic lithosphere (on-craton eruption; e.g., Dawson et al., 1997;Craig et al., 2011). Alkaline and sub-alkaline magmas erupted along the W-E axis are related to high degrees of partial melting (2-6%) of amphibole-bearing garnet lherzolite at 85-140 km (Nonnotte, 2007;Mana et al., 2015). They erupted at the craton-edge where asthenospheric upwelling may have proceeded to shallow levels, leading to common alkali basalts similar to those observed north of the NTD in central Kenya, which represents a more mature stage of rifting (e.g., Ebinger et al., 2000;Roex et al., 2001).
The relationships between magmatism and rift dynamics observed in the eastern EAR are similar to those observed in the western EAR despite their opposite rift propagations. In the western EAR, magmatism evolved from basalt-basanite at Kivu (south) to K-nephelinite at Toro-Ankole (north) during early stage rifting (Pouclet et al., 1981(Pouclet et al., , 2016Chakrabarti et al., 2009;Rosenthal et al., 2009) are also associated with peridotite, pyroxenite and glimmerite xenoliths (Muravyeva and Senin, 2018). Alkaline rocks are potassic, attesting to the presence of a potassium-rich mantle source beneath the western part of the Tanzanian Craton (up to 7 wt% K 2 O in melt; Pouclet et al., 1981;Rogers et al., 1998). Primary magmas from the north of the western rift originate from the greatest melting depths (>140 km in the presence of phlogopite at Toro-Ankole), and the depth of melting becomes gradually shallower toward the south of the western branch (<90 km in the presence of both amphibole and phlogopite), similar to the observed depth-of-melting in the east branch of EAR Foley et al., 2012). The mantle source of potassic magmatism has been characterized as two metasomatic assemblages of phlogopite-clinopyroxenite and a carbonate-rich garnet-free assemblage resulting from kimberlitic-like melt impregnation and crystallization (e.g., Rosenthal et al., 2009;Foley et al., 2012). The asthenospheric source beneath the West branch of EAR is CO 2 -rich and highly metasomatized leading to the genesis of ultramafic and potassic primary melt (Foley et al., 2012;Foley and Fischer, 2017).
The low volume of highly alkaline volcanism in both the western and eastern rift branches may be related to the low degrees of partial melting and the deep thermal anomaly beneath the thick cratonic lithosphere (e.g., Ebinger and Sleep, 1998). The occurrence of thin, rifted lithosphere in close proximity to thick cratonic heterogeneous lithosphere may generate melting at shallow depth, likely related to rift propagation and/or the plume-related thermal anomaly. The most likely mechanism for producing sub-cratonic lithospheric heterogeneities is the accumulation of low-degree partial melts from the convecting depleted asthenospheric mantle (e.g., this study, McKenzie, 1989;Lee and Rudnick, 1999;Vauchez et al., 2005;Baptiste et al., 2015;Chin, 2018).
The variability of the mantle sources of potassic and alkaline volcanism between the western and eastern branches of the EAR may also reflect different sub-cratonic lithospheric lithologies beneath the Paleoproterozoic and Neoproterozoic belts underlying the western and eastern branches, respectively (e.g., Corti et al., 2007;Katumwehe et al., 2015).

CONCLUSION
Volcanoes of the Manyara-Balangida rift, associated with early rifting in the East African Rift (northern Tanzania), have erupted primary nephelinites composed of olivine, clinopyroxene, and up to 4 vol% phlogopite. Trace models of Kwaraha and Labait lavas from a low degree of partial melting (0.2-1%) of a garnet-phlogopite-bearing peridotite with a carbonate-rich component (0.3%), consistent with others northern Tanzania lavas. The presence of garnet-bearing mantle xenoliths and the primary melt compositions indicate that melting occurred at depths ≥150 km, below the lithosphere-asthenosphere boundary. The partial melting at the earliest stage rifting is the deepest source reported for East branch of East African Rift. The percolation of deep, CO 2 -rich and H 2 O-bearing highly alkaline magmas may have produced strong heterogeneities through metasomatism process in the thick sub-continental Tanzanian lithospheric mantle. H 2 O-bearing magmas percolation has induced phlogopite crystallization in spinel lherzolite and glimmerite lithology weakening the rheology of the Tanzanian lithosphere. These compositional changes are thus major parameters that have influenced the rift propagation near the craton edge, and should be taken into account to explain tectonic deformation at the surface.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found online at https: //tel.archives-ouvertes.fr/tel-01563231/.

AUTHOR CONTRIBUTIONS
CB carried out the microprobe, icpms, and FTIR analyses. CB and FP contributed to the interpretations and writing. All authors contributed to the article and approved the submitted version.

FUNDING
This research was financially supported by the ANR project CoLiBrEA CTN • LS 104568 and the HATARI project (Tellus-RIFT INSU).