Petrogenesis of the Piqiang mafic-ultramafic layered intrusion and associated Fe-Ti-V oxide deposit in Tarim Large Igneous Province, NW China

The Piqiang intrusion is one of the two important mafic-ultramafic layered intrusions that host giant Fe-Ti-V oxide deposits in the Permian Tarim Large Igneous Province, NW China. The intrusion mainly consists of gabbro, anorthosite and minor plagioclase-bearing clinopyroxenite in the marginal zone. Disseminated tomassive Fe-Ti oxide ores occur as layers and lenses within the gabbro. SHRIMP zircon U-Pb results from both a gabbro from the Piqiang intrusion and a granite from the surrounding granitic dyke yield ages of ~270Ma. Geochemically, the Piqiang silicate rocks are enriched in light rare earth elements (LREE) and large ion lithophile elements (LILE), moderately depleted in high field strength elements (HFSE), and have a limited range of Sr-Nd-Hf isotopic compositions. The similar mineralogy, mineral compositions, and trace element characteristics of the layered units suggest that all the rocks are co-magmatic. The parental magma is Fe-Ti-rich and is akin to the most primitive diabasic dyke which is associated with the Piqiang intrusion. Partial melting of the Tarim mantle plume with involvement of a subduction-metasomatized lithospheric mantle source best explains the geochemistry and petrogenesis of the parental magmas of the Piqiang intrusion. We propose that the lithospheric mantle source may have been metasomatized by subduction-related materials and the metasomatic enrichment of this source region which may be correlated with oceanic sediment recycling during southward subduction of the South Tianshan oceanic slab during the Early-Middle Paleozoic. Crystal settling and mechanical sorting is the predominant process responsible for the formation of the massive Fe-Ti oxide ores in the Piqiang intrusion. Central to ore formation is a combination of the protracted differentiation history of a Fe-Ti-enriched parental magma and the later addition of external H2O from the country rocks to the slowly cooling magma chamber. ARTICLE HISTORY Received 2 November 2018 Accepted 17 February 2019


Introduction
Layered intrusions are the most voluminous plutonic expressions of many, if not all large igneous provinces (LIPs) and have been intensively studied for their unique metal endowment [i.e., PGE (platinum group elements), Cr, Ni, V, Ti] (Ernst and Jowitt 2013;Nebel et al. 2013). Notable examples include the Merensky reef and Main Magnetitite Layer of the Bushveld Complex in the Bushveld LIP (Ernst and Jowitt 2013), the Platinova reef of the Skaergaard intrusion in the North Atlantic LIP (Andersen et al. 2002;Holwell and Keays 2014) and the Panzhihua and Hongge Fe-Ti-V oxide deposits in the Emeishan LIP (Zhou et al. 2005Pang et al. 2008;Zhang et al. 2009;Bai et al. 2012). However, there are obvious differences between layered intrusions with respect to metal enrichment, volatile contents, and characteristics of their mantle sources, all of which potentially affect the crystallization sequence and therefore zoning and focusing of ore metals into oxide or sulfide ore horizons (Nebel et al. 2013). Origin of mineral accumulation in specific horizons, differences in chemical composition within minerals, and the roles of source variability and crustal assimilation remain elusive, yet they are crucial for economic prospecting and for the understanding of the formation of the orthomagmatic deposits hosted by layered intrusions.
Early Permian Tarim LIP is a newly recognized LIP in NW China after the recognition of Late Permian Emeishan LIP in SW China (Figure 1(a); Yang et al. 2013;Yu et al. 2017). LIPs are highly prospective for magmatic Fe-Ti-V oxide deposits that occur within mafic-ultramafic layered intrusions, as exemplified by the Bushveld and Emeishan LIPs (Ernst and Jowitt 2013). In the Tarim LIP, Fe-Ti oxide mineralization has been recently identified in the Wajilitag and Piqiang layered intrusions (Zhang et al. 2010Cao et al. 2014Cao et al. , 2017a; these represent a major breakthrough in the exploration for orthomagmatic deposits in this LIP. In contrast to the Wajilitag intrusion, which contains only disseminated Fe-Ti oxide ores in the clinopyroxenite, both massive and disseminated Fe-Ti oxide ores are present in the Piqiang intrusion and closely associated with the gabbroic rocks (Zhang et al. 2018). In this respect, the Piqiang intrusion provides an excellent opportunity to identify the critical factors that led to the formation of the massive ores. Although some work has been done on the Piqiang intrusion (e.g., Zhang et al. 2010Zhang et al. , 2016Zhang et al. , 2018, the origin of the intrusion and its Fe-Ti-V deposits is still a matter of debate.
In this contribution, we carried out detailed petrological, geochemical, and geochronologic investigations on the Piqiang intrusion and associated Fe-Ti oxide ores. These new data are used to constrain the nature and origin of the parental magma and to shed new light on the genesis of massive Fe-Ti oxide ores in layered intrusions.

Geological background
The Tarim Craton in northwestern China is surrounded by the Tianshan orogenic belt to the north and west, and the Kunlun and Altyn orogenic belt to the south (Figure 1(b)). It was amalgamated with the southern part of the Central Asian Orogenic Belt during the Late Paleozoic and is composed of a Precambrian crystalline basement, including volcano-sedimentary and high-grade metamorphic rocks, and a thick Phanerozoic sedimentary cover which comprises Ordovician, Devonian, Carboniferous, Permian and Cretaceous strata (BGMRXUAR 1993;Long et al. 2010). Along the northern margin of the Tarim Craton, Ordovician to Devonian arc rocks (ca. 460−400 Ma) are sporadically exposed, and are generally correlated to the southward subduction of the South Tianshan oceanic lithosphere plate beneath the Craton (Huang et al. 2013;Ge et al. 2014).
The Tarim LIP consists of diverse lithologies including kimberlites, flood basalts, mafic-ultramafic layered intrusions, bimodal dyke swarms, lamprophyres, nephelinites, carbonatites and the full spectrum of volcanic and plutonic silicic rocks (i.e. rhyolites, syenites, granites) (Tian et al. 2010;Liu et al. 2014a;Xu et al. 2014;Cheng et al. 2015Cheng et al. , 2017Zhang et al. 2016;Yu et al. 2017). The estimated areal extent of these magmatic products exceeds 3 × 10 5 km 2 with a thickness ranging from several hundred meters to 3 km (Tian et al. 2010). However, most areas of the Tarim basin is covered by the Taklamakan desert, and thus the present extent of the Tarim LIP is largely deduced from geophysical studies and boreholes. The best outcrops of flood basalts are from Keping, Xiahenan, Qipan and Damusi around the western Tarim basin, and mafic-ultramafic layered intrusions and dykes together with minor kimberlites, Ku luk eta ge K u n l u n n y t l A T ia n sh a n Tu -H    syenites, granites, lamprophyres, nephelinites and carbonatites are mainly from Piqiang and Bachu Cheng et al. 2017). Some of mafic-ultramafic layered intrusions host economically important Fe-Ti-V oxide deposits, including the Wajilitag and Piqiang deposits (Figure 1(b)). Previous field studies and geochronological investigations have identified the time sequence of Permian magmatism of the Tarim LIP is as follows: kimberlites (~300 Ma) → flood basalts and low Nb- Ta (Liu et al. 2014a;Xu et al. 2014;Zou et al. 2015;Zhang et al. 2016;Cao et al. 2017b;Song et al. 2017). This Permian massive magmatism in the Tarim Craton is believed to have resulted from a deep mantle plume activity (Yang et al. 2013;Xu et al. 2014;Cheng et al. 2017;Yu et al. 2017).

Geology of the Piqiang intrusion and associated Fe-Ti oxide ores
The Piqiang (also known as Puchang) intrusion is located in the northwestern part of the Tarim LIP,~120 km northeast of the Atushi city (Figure 2(a)). It is an~6.1 × 3.5 km oval-shaped body in plan and crops out over an area of 16.7 km 2 . The intrusion was emplaced into limestone and intercalated calcareous sandstone of the Upper Carboniferous Kangkelin Formation and the northwestern part of the intrusion is overlain by Neogene sedimentary rocks (Figure 2(a)). Xenoliths of country rock are found in the Piqiang intrusion close to the contact . Numerous diabasic, dioritic and granitic dykes occur within the intrusion (Figure 2(a-b)); some of these may have been displaced by later faulting. To the southeast lies the neighbouring 282 ± 4 Ma (Cao et al. 2017b) Wajilitag layered intrusion that is the sixth largest Fe-Ti-V oxide deposit identified to date in China (Bai et al. 2018;Pang and Shellnutt 2018).
The Piqiang intrusion mainly consists of gabbroic rocks with variable amount of Fe-Ti oxides, and volumetrically subordinate anorthosite and plagioclasebearing clinopyroxenite (Figure 2). Anorthosite is found only as small separated cupolas in some parts of the intrusion, and plagioclase-bearing clinopyroxenite dominates the margin of the intrusion (Figures 2 and 3(a)). The contact relationship between the anorthosite and gabbro is gradational, whereas the plagioclase-bearing clinopyroxenite and gabbro are in sharp contacts against each other. Anorthosite and plagioclase-bearing clinopyroxenite are coarse-grained massive rocks without prominent layering. Gabbroic rocks are fineto coarse-grained and locally exhibit well-defined layering (Figure 3(b-c)). Igneous layering displays a regular alternation of dark and light layers (Figure 3(c)). Dark layers are enriched in olivine, clinopyroxene and Fe-Ti oxides, whereas light layers consist mostly of clinopyroxene and plagioclase. The layering observed in the Piqiang intrusion is typical of layered intrusions and similar features have been described from the Skaergaard intrusion (McBirney 1996), the Sept Iles intrusion (Namur et al. 2010) and the Panzhihua intrusion (Zhou et al. 2005;Zhang et al. 2009).
In the Piqiang intrusion, Fe-Ti oxide mineralization is mainly concentrated in the gabbroic rocks with fine- grained texture. Oxide bodies occur as occur as layers or lenses, and are in sharp contact with adjacent gabbros ( Figure 3(d-f)). The massive and disseminated ores also have sharp contacts with each other. Some of the massive ore layers occur as centimeter-scale ore bands in disseminated ore layers (Figure 3(g)). The silicate-rich portions of the oxide bodies commonly show planar lamination of elongated plagioclase (Figure 3(h)). This deposit has a total ore resource of 120 million tons with an average ore grade of 20 wt.% total FeO and 11 wt.% TiO 2 .
The plagioclase-bearing clinopyroxenite exhibits a hetero-granular cumulate texture and are composed essentially of euhedral clinopyroxene (up to 2.4 mm in length; 70-85%) and variable amounts of interstitial plagioclase (5-15%) with minor Fe-Ti oxides and amphibole (Figure 4(a). Contiguous clinopyroxene grains typically have 120°triple junctions (Figure 4(a)). Clinopyroxene grains exhibit sporadic and patchy replacement by brown amphibole. Plagioclase is tabular to platy with irregular wavy edges (Figure 4(a)). Oxide minerals including titanomagnetite and ilmenite in roughly sub-equal amounts occur as inclusions in the rim of clinopyroxene or as interstitial phases between silicate minerals ( Figure 5(a)).
Gabbros are fine-to coarse-grained cumulate rocks composed mainly of plagioclase (30-70%), clinopyroxene (20-50%) and small amounts of Fe-Ti oxides, amphibole and biotite ). However, the modal proportions in this rock type vary considerably, and either Fe-Ti oxides or olivine locally reaches as much as 25% in some gabbroic samples ). Plagioclase is an ubiquitous phase with a grain size ranging from 0.1 to 7 mm, forming sub-equant to strongly tabular subhedral to euhedral grains. In fine-grained gabbroic samples (Figure 4(d)), planar foliation and lineation are shown by a marked preferred orientation of plagioclase laths and small elongated clinopyroxene crystals. In places, plagioclase laths may be slightly deformed, displaying kinked twinning (Figure 4(c-d)), which is probably caused by compaction (Godel et al. 2011). Clinopyroxene grains are subhedral and sometimes contain small euhedralsubhedral plagioclase grains (Figure 4(b-c)) and exsolution lamellae of ilmenite along its prismatic cleavages (Figure 4(d)). Olivine appears as very large (up to 2.5 mm) subhedral to locally poikilitic grains (Figure 4(e)) in olivine gabbro. The entrapment of olivine in plagioclase and the abundance of cumulus olivine in these gabbros (Figure 4(e)) indicate that the olivine has crystallized before plagioclase. Similarly, entrapment of plagioclase within clinopyroxene in most gabbroic samples (Figure 4 (b)) indicate that the plagioclase has crystallized before clinopyroxene. Fe-Ti oxide minerals occur either as small (˂1 mm) patches of anhedral titanomagnetite with minor ilmenite or two-phase intergrowths ( Figure 5(b-c)). Irregular Fe-Ti oxide grains have locally been observed as inclusions in silicate minerals (Figure 4(b-e)). Reaction rims of brown amphibole and biotite are in places welldeveloped along contacts between silicate and Fe-Ti oxide minerals (Figure 4(b-c)).
Anorthosite is medium-to coarse-grained, and consists predominantly of euhedral-subhedral plagioclase (˃90%) and minor anhedral olivine, Fe-Ti oxides, amphibole and biotite (Figure 4(f)). Plagioclase occurs either as large (1-7 mm), elongated crystals or aggregates of small (0.1−0.3 mm) grains because of recrystallization. In some cases, it exhibits undulose extinction and deformation twins. Olivine and Fe-Ti oxides typically account for less than 5%, and amphibole and biotite accounts for less than 2%. Orthopyroxene was observed as coronitic rims around olivine in a few anorthositic samples (Figure 4(f)).
Disseminated Fe-Ti oxide ore dominates the ore horizons (Rui et al. 2002) and are composed of Fe-Ti oxides (25-50%), plagioclase (25-40%) and clinopyroxene (10-25%), with minor olivine and amphibole (Figure 4 (g)). The massive Fe-Ti oxide ore is also an important component of the ore horizons; it is fine-grained ( Figure  4(h)), and contains less silicate minerals but more Fe-Ti oxides (˃80%) than the disseminated Fe-Ti oxide ore. The silicate minerals occur as isolated grains, or aggregate of grains, surrounded completely by Fe-Ti oxides. It is worthy to note that both plagioclase and clinopyroxene in the disseminated Fe-Ti oxide ore exhibit the same preferred orientation (Figure 4(g)). In the oxide ores, the major ore minerals consist of titanomagnetite and ilmenite ) and are interstitial to the main rock-forming minerals ), suggesting that they crystallized at a late stage. Titanomagnetite occurs as polygonal grains and accounts for~75−80% of the oxide minerals ( Figure 5(e)). It contains exsolution lamellae including ilmenite and hercynite ( Figure 5 (d-e)). Ilmenite occurs as fine-grained irregular to polygonal crystals and makes up~15% of the oxide minerals. Most boundaries between the oxide grains are straight to slightly curved and meet at distinct triple junctions with~120°interfacial angles ( Figure 5(d-e)). Subordinate hercynite (<2%) is also present as micropatches along the rims of titanomagnetite or as single irregular crystals at the boundaries between titanomagnetite and ilmenite ( Figure 5(d-f)). Occasionally, some inclusions composed of ilmenite and sulphide have been observed within silicate minerals ( Figure 5(f)). Reaction rims of amphibole, and to a lesser extent olivine, are well-developed along contacts between silicates and oxide minerals ( Figure 5(g-h)), indicating disequilibrium textures between oxide minerals and coexisting silicates (Holness et al. 2011;. Also, these silicate grains are typically embayed and resorbed, which appear to have crystallized prior to oxides.

Sample preparation and analytical methods
A total of 55 samples were collected from the exposed section of the open-pit mine in Piqiang, and one sample (PQ1126) was collected from a granitic dyke crosscutting the Piqiang intrusion. Their locations are indicated in Figure 2. Weathered or altered surfaces were removed from the samples before jaw-crushing. Fresh chips were then selected for analysis using a binocular microscope and pulverized into powders using agate mortars. Zircons were separated from a gabbro sample (PQ1113) and granite sample (PQ1126) using conventional heavy liquid and magnetic techniques and purified by handpicking under a binocularmicroscope. They were mounted in an epoxy resin, and then polished. Cathodoluminescence and back-cattered images were used to select undeformed zircon crystals that show regular growth patterns plus lack of inherited cores and secular compositional zoning for U-Pb dating. U-Pb isotopes of the selected zircon crystals were determined using a SHRIMP-II in the Chinese Academy of Geological Sciences, Beijing, following procedures of Compston et al. (1992) and Williams (1998).
Whole-rock major elements were analysed on fused glass discs using a Rigaku RIX 2000 X-ray fluorescence spectrometer in the Guangzhou Institute of Geochemistry, Chinese Academy of Sciences (GIG-CAS). The analytical uncertainties are mostly between 1% and 5%. Trace elements were analysed by solution inductively coupled plasma mass spectrometry (ICP-MS) in the GIG-CAS. The solutions were prepared by complete digestion of rock powders in HF + HNO 3 in Teflon bombs at 100°C for 7 days. The precision for most elements was typically better than 5% RSD (relative standard deviation), and the measured values for Zr, Hf, Nb and Ta were within 10% of the certified values of the two employed standards (diabase W-2 and basalt BHVO-2).
Electron microprobe analyses of major rock-forming minerals were obtained using the JEOL Superprobe JXA-8100 electron microprobe in the GIG-CAS. The operating conditions are 15 kV accelerating voltage, 20 nA beam current, 1-2 μm beam diameter. Elements were analysed with wavelength-dispersive spectrometers and were calibrated by reference to oxide and mineral standards using the ZAF correction routine. The precision for oxide concentrations is better than 1%. The counting times were 20 s on the peak and 10 s on the background.
Trace element concentrations in clinopyroxene were determined on thin sections by laser ablation ICP-MS in the GIG-CAS, using an Agilent 7500a ICP-MS system coupled with a Resolution M50-HR 193 nm ArFexcimer laser sampler. The analytical procedures, operation conditions, calibration, and data reduction are the same as those given in Tu et al. (2011).

Zircon U-Pb ages
Zircon U-Pb isotope data for the Piqiang gabbro and the surrounding granite dyke are given in Supplementary Table 1. Zircon crystals of a gabbro sample PQ1113 are mostly subhedral with variable lengths of 30 to 200 μm. Most of them are dark in CL images. Some of them show oscillatory zoning and some others are structureless. Thirteen anaylses on different zircon grains have a wide range of Th (158-8775 ppm) and U (1262-8699 ppm) contents with Th/U ratios between 0.05 and 1.04. Except for Spot 9, 11 and 14 with an exceptionally high common Pb, the analytical results (Supplementary Table 1) yield a concordia U-Pb isotope age of 270.8 ± 2.3 Ma (mean square weighted deviation (MSWD) = 0.76; Figure 6(a)). Zircon crystals of a granite sample PQ1126 are mostly euhedral and prismatic, with lengths between 100 and 300 μm. These zircon grains are characterized by either sector or oscillatory zoning in CL images. The Th/U ratios of 12 analyses on the selected zircon grains range from 0.22 to 0.85. Except for Spot 11 with a slightly young apparent 206 Pb/ 238 U age of 245.2 ± 4.4 Ma (1σ; Supplementary Table 1), the U-Pb results form a coherent cluster and yield a concordia age of 270.3 ± 2.6 Ma (MSWD = 1.3; Figure 6(b)). The concordia ages, which consider the 207 Pb/ 206 Pb age and both Pb/U ages (Ludwig 1998), are interpreted as the best age estimates for these rocks. The new age data from this study (Figure 6(ab)) reveal that the Piqiang intrusion is nearly   Table 2). The anorthosite samples are characterized by high SiO 2 , Al 2 O 3 , (Na 2 O+K 2 O) and relatively low contents of total Fe 2 O 3 and TiO 2 compared with the plagioclase-bearing clinopyroxenites and gabbros. Generally, the compositional variations can be attributed to the varying proportions of olivine, clinopyroxene, plagioclase, titanomagnetite and ilmenite ( Figure 7). In particular, the disseminated and massive Fe-Ti oxide ores plot out of the trajectories of the olivine, clinopyroxene and plagioclase (Figure 7), suggesting that the titanomagnetite and ilmenite within these samples are cumulus phases that crystallized directly from the magma rather than being intercumulus phases formed as a result of trapped liquid saturation. Total Fe 2 O 3 (Figure 7(c)) and TiO 2 (Figure 7(d)) contents negatively correlate with SiO 2 contents, whereas Al 2 O 3 (Figure 7(e)) and CaO (Figure 7(b)) contents positively correlate with SiO 2 . This is consistent with a crystal sorting process that caused the segregation of dense Fe-Ti oxides and less dense silicate minerals within the Piqiang magmatic system (Bai et al. 2018).
In the TiO 2 (Figure 8(a)) and V (Figure 8(b)) versus total Fe 2 O 3 diagrams, two trends indicative of titanomagnetite control and titanomagnetite-ilmenite control are observed in the Piqiang samples. It is notable that the disseminated and massive Fe-Ti oxide ores follow the titanomagnetite-only trend, suggesting that titanomagnetite is the only dominant oxide mineral. Representative trace element compositions of the Piqiang intrusive rocks are given in Supplementary  Table 2 and illustrated in Figure 9. The intrusive rocks are all enriched in the light rare earth elements (REE) relative to the heavy REE and have positive Eu anomalies (Eu/Eu* = 0.88−4.30) (Figure 9(a-b)). In the primitive mantle normalized immobile trace element diagram (Figure 9(c)), the plagioclase-bearing clinopyroxenite, gabbro and anorthosite samples display moderate negative Nb, Ta, Zr, Hf anomalies and variable Ti anomaly. In comparison, the disseminated and massive Fe-Ti oxide ores have strongly positive Nb-Ta and weakly positive Zr-Hf anomalies (Figure 9(d)). These differences in Nb-Ta and Zr-Hf anomalies are consistent with the fact that Nb and Ta are compatible but Zr and Hf are slightly incompatible in ilmenite within basaltic systems (Klemme et al. 2006;Dygert et al. 2013).

Sr-Nd-Hf isotopes
The analysed Piqiang rocks show limited variations in Sr, Nd and Hf isotopic compositions (Supplementary Table 3). The age-corrected 87 Sr/ 86 Sr ratios (t = 270.8 Ma) range from 0.70492 to 0.70662, and ε Nd (t) values vary from −3.2 to +0.9; ε Hf (t) values range from −6.1 to −0.5. Our data (Supplementary Table 3) are consistent with those of previous isotopic studies on the rocks from the layered series (Zhang et al. 2010(Zhang et al. , 2018. The Sr-Nd data for our samples overlap the field of Group 2 basalts from the Tarim LIP as well as the Bachu diabasic dykes, and plot above the fields of the Group 1 basalts in the Tarim LIP (Figure 10(a)). The Mazaertag and Wajilitag intrusions in the Tarim LIP have much higher ε Nd (t) and lower ( 87 Sr/ 86 Sr)i ratios. Notably, the Piqiang samples have a more enriched Sr-Nd isotopic signature compared with the uncontaminated Tarim plume-derived melts represented by Bachu diabasic dyke-hosted clinopyroxene macrocrysts (Wei et al. 2015). The Nd-Hf data for the Piqiang intrusion plot within the OIB field and but slightly below the mantle array line (Figure 10(b)) , showing weak decoupling (Δε Hf = −4.1 to −0.6).

Major elements
Representative chemical compositions and structural formulae of olivine, clinopyroxene, plagioclase, titanomagnetite and ilmenite from Piqiang intrusion are listed in the Supplementary Tables 4−8 and are shown in Figure 11. Olivine within the Piqiang intrusion has forsterite (Fo) contents that range from 61 to 79 mol%. The olivine crystals from the disseminated and massive Fe-Ti oxide ores have higher Fo values (70−79) than those of the silicate rocks (Fo 61−69 ) (Supplementary Table 4). Individual crystals in the massive Fe-Ti oxide ores are reversely zoned with high Mg rims, probably reflecting late-stage subsolidus reequilibrium between olivine and the surrounding oxide   minerals (Shellnutt and Pang 2012). This is supported by a positive correlation between the olivine Fo content and whole-rock total FeO content (Figure 11(a)).  (Deer et al. 1978). Piqiang clinopyroxenes have low octahedral Al consistent with their formation in situ, at low pressure, and the excess tetrahedral Al is probably due to the presence of substitutions such as CaCrAlSiO 6 and CaFe 3+ AlSiO 6 (Chambers and Brown 1996). TiO 2 concentrations (0.28-2.66 wt.%) generally correlate with the change of Mg# (Figure 11(b)). Plagioclase within the intrusion belongs to the bytownite to andesine series with anorthite (An) contents from 48.4 to 88.8 mol% (Supplementary Table 6). Plagioclase in the plagioclase-bearing ( Figure 11(c-d)). Titanomagnetite shows a wide compositional range with TiO 2 contents ranging from 1.2 to17.5 wt. %, 0.0-2.9 wt.% MgO, 0.6-7.4 wt.% Al 2 O 3 and 0.0-0.6wt.% MnO (Supplementary Table 7). Titanomagnetite in the disseminated and massive Fe-Ti oxide ores tend to have higher TiO 2 , MgO, Al 2 O 3 and MnO than in silicate rocks, a phenomenon also known in the Panzhihua intrusion (Pang et al. 2008). Besides, a positive correlation exists between MgO contents of titanomagnetite and Fo contents of olivine (Figure 11(e)). It worth noting that the titanomagnetite in the oxide inclusions has a low Cr 2 O 3 content (0.0-0.9 wt.%) similar to the interstitial titanomagnetite (Figure 11(f); Supplementary Table 7). Ilmenite in the disseminated and massive Fe-Ti oxide ores also have higher MgO (4.3−6.3 wt.%) than the silicate rocks (Supplementary Table 8). These compositional trends of Fe-Ti oxide minerals are presumably caused by the subsolidus Fe-Mg equilibration between mafic silicates and Fe-Ti oxides as suggested for the Panzhihua intrusion (Pang et al. 2008(Pang et al. , 2009).

Trace elements
Clinopyroxenes in the three units of the Piqiang intrusion have total REE contents ranging from 32.6 to 96.1 ppm (Supplementary Table 9) and shows similar, 'hump-shaped' light REE-enriched chrondritenormalized REE patterns (Figure 12(a)), coupled with negative Nb-Zr anomalies in primitive mantlenormalized trace element patterns (Figure 12(a)). 6. Discussion

Parental magma composition
The similarities in mineralogy, mineral compositions, chondrite-normalized REE patterns (Figure 9(a-b)) and Sr-Nd-Hf isotopic compositions ( Figure 10) observed in the three units of the Piqiang intrusion suggest that all the rocks were derived from a common parental magma by similar processes of magmatic differentiation. Unfortunately, no chilled margins have been found that could help to constrain the compositions of the parental magma. A suite of alkali diabasic dykes are exposed in the vicinity of the Piqiang intrusion and they were thought to have compositions similar to the Piqiang parental magma (Zhang et al. 2010(Zhang et al. , 2018. Though the ages of these dykes are unknown, they are spatially associated with the Piqiang layered intrusion and their textural resemblance to the mineral phases within the layered sequences strongly suggest that they belong to the same magmatic event. Applying a K D of 0.3 ± 0.03 to the distribution of Fe and Mg between olivine and liquid (Roeder and Emslie 1970) indicates that the most Mg-rich olivine (Fo = 76) from the Piqiang intrusive rocks were in equilibrium with a liquid with an Mg# [= molar 100× MgO/(MgO+FeO)] of between 38 and 42. The Mg# value quoted for the least fractionated alkali diabasic dyke (Sample 08KT01-14) discussed by Zhang et al. (2010), (2018)) is 38, again supporting the contention that such a dyke may share similar compositions with the parental magma of the Piqiang intrusion. The parental magma was thus a ferrobasaltic melt that was relatively low in SiO 2 (51.56 wt.%), highly enriched in total FeO (12.48 wt.%) and TiO 2 (2.28 wt.%) (Zhang et al. 2010). This composition is close to the estimate of the Panzhihua layered intrusion (c.51.72 wt.% SiO 2 ; 13.03 wt.% total FeO; 4.37 wt.% TiO 2 ) by . On the basis of the partition coefficients of trace elements between clinopyroxene and silicate melt (Supplementary Table 10), we calculated the trace element composition of the melt in equilibrium with the most primitive clinopyroxenes from both the massive Fe-Ti oxide ore and barren silicate units. The calculated trace element patterns (Figure 12(c)) are similar to those of the surrounding diabasic dykes (Zhang et al. 2010). Therefore, we propose that the parental magma of the Piqiang intrusion is compositionally similar to that of the surrounding diabasic dykes.

Crustal contamination
Crustal contamination is a potential process for mantle-derived melts during their ascent through continental crust or their evolution within a magma chamber in a continental environment (He et al. 2016). All the silicate rocks in the Piqiang intrusion are characterized by negative Nb-Ta anomalies and enrichment in light REE and LILE in the primitive mantle-normalized trace element diagram (Figure 9(c)). The depletion of these elements could potentially be explained by crustal contamination, because continental crust is poor in these elements (e.g., Rollison 1993). The relatively high δ 18 O values in zircon from the various layered units of the Piqiang intrusion (5.6 to 7.1‰ with mean value of 6.4 ± 0.06‰; Zhang et al. 2016) also indicate the possibility of crustal contamination. However, extensive crustal contamination would have produced positive Zr-Hf anomalies and high (La/Nb) PM and (Th/Nb) PM ratios (Sun and McDonough 1989). Thus, the weak negative Zr-Hf anomalies of the Piqiang silicate rocks (Figure 9(c)), their low (La/Nb) PM and (Th/Nb) PM ratios indicate that crustal contamination was not an important process in the formation of this body. Morevoer, contamination of crustal materials can result in variable isotopic compositions due to the inhomogeneity of crustal rocks. However, the Piqiang rocks have relatively uniform ( 87 Sr/ 86 Sr)i and ε Nd (t) values, and follow a crystal fractionation (FC) rather than an assimilation crystal fractionation (AFC) trend (Figure 10(c-d)).
The country rocks of the intrusion (limestones of the Kangkelin Formation; Figure 2(a)) and deep crustal materials (e.g., Neoproterozoic gabbros and granites, and Early-Middle Paleozoic granites) are the main potential contaminants. We use the average Sr-Nd isotopic compositions of Bachu diabasic dyke-hosted clinopyroxene macrocrysts [( 87 Sr/ 86 Sr)i = 0.70352; ε Nd (t) = 4.6; Sr = 39.28 ppm; Nd = 3.74 ppm, Wei et al. 2015] to represent the composition of the uncontaminated mantle-derived magma. Using the Sr-Nd isotopic compositions of both the Tarim lower and upper continental crust (e.g., Neoarchean granitic gneiss, and Neoproterozoic-Early Paleozoic granite) as the the composition of the possible contaminant, modelling calculations show that ˂5% of upper crustal contamination is required to explain the observed isotopic composition of the studied intrusion (Figure 10(a)). Nevertheless, a limited degree of crustal contamination (7-13%) at source is sug-

Modelling of magma evolution
The parental magmas of the Piqiang intrusion may have experienced variable degrees of fractional crystallization, en route from source to surface. Compared with the mantle-derived olivine (Fo =~90; Gibson et al. 2000), the low Fo content in most magnesian olivines (Fo 69 ) in silicate rocks indicates that the parental magma has experienced extensively fractional crystallization before it emplaced into the current magma chamber. Moreover, Cr depletion in titanomagnetite in the Piqiang intrusion compared with the Mazaertag and Wajilitag intrusions in the Tarim LIP Cao and Wang 2017) most probably indicate a Cr-depleted magma resulting from a previous Cr-rich titanomagnetite fractionation process of a primitive magma.
The Piqiang magma intruded into the slightly older Upper Carboniferous Kangkelin Formation and Lower Permian Bieliangjin Formation, the maximum thickness of which is estimated as~2.7 km (Zhang et al. 2013b). This is the best constraint on the final emplacement depth of the Piqiang intrusion and gives a maximum value of~l kbar. H 2 O has a significant effect on the differentiation paths of basaltic magmas, mainly by suppressing plagioclase crystallization and therefore decreasing its cotectic proportion in the cumulus mineral assemblage (Sisson and Grove 1993;Botcharnikov et al. 2008). Compared to the absence of plagioclase as a cumulus phase in the Mazaertag intrusion , the presence of abundant cumulus plagioclase throughout much of the Piqiang intrusion suggest that the Piqiang parental magma contained much less initial H 2 O content than the Mazaertag intrusion (~1.5 wt.% H 2 O in the Mazaertag parental magma; Wei et al. 2014). This is also consistent with the fact that the Piqiang gabbros contain much less primary hydrous phases (generally ˂5% principally amphibole and to a lesser extent biotite) than the Mazaertag wehrlites, in which the interstitial amphibole and biotite are up to 8−10% Cao and Wang 2017). Low water content (˂0.5%) in the Skaergaard parental magma enhances the liquidus volume of plagioclase and suppresses those of mafic silicates and oxides, and is thought to be responsible for producing significant volume of cumulate rocks (59.7% of the Skaergaard intrusion; Nielsen 2004) before the appearance of Fe-Ti oxides (Botcharnikov et al. 2008). By contrast, the Piqiang intrusion was characterized by the occurrence of Fe-Ti oxides throughout the intrusion (Zhang et al. 2010(Zhang et al. , 2018 this study), implying that the Piqiang parental magma contained significant amounts (e.g., ≥0.5 wt.%) of H 2 O. Evidence presented here indicates that the initial H 2 O content of the Piqiang parental magma was likely 0.5 wt.%.
The initial redox state of the Piqiang magma is uncertain. Application of the magnetite-ilmenite thermometer and oxygen barometer [using the QUILF program of Andersen et al. (1993)] to the magnetite-ilmenite mineral pairs from the Fe-Ti oxide ores of the Piqiang intrusion yields temperatures ranging between 394 and 592°C and fO 2 ranging between FMQ-5.6 and FMQ-1.8 (Figure 13; Supplementary Table 11). All the data follow trends of decreasing fO 2 with decreasing temperature that lie close to the U-30 isopleth (Figure 13). The result also indicates that the end of subsolidus re-equilibration for magnetite and ilmenite of the Piqiang and Wajilitag Fe-Ti oxide ores occurred at similarly low fO 2 . In addition, the magnetite within the Piqiang intrusion contain low concentrations of V (3331−6594 ppm) that are indicative of formation from magmas under relatively low fO 2 conditions (˂FMQ+0.5; Xing et al. 2013). In addition, experiments by Toplis and Carroll (1995) have indicated that magnetite-dominated oxide phases crystallize from a ferrobasaltic system at high fO 2 conditions (FMQ−FMQ+1.5), and those containing cumulus magnetite and ilmenite crystallize at fO 2 conditions at FMQ. Data from this study and from Cao et al. (2014) show that the Fe-Ti oxide minerals from the Piqiang intrusion are dominated by titanomagnetite, whereas both magnetite and ilmenite are present in the Wajilitag intrusion (Figure 7). This means that it is probable that fO 2 of the Piqiang parent magma is somewhat higher compared with the Wajilitag parent magma, despite the relatively dry nature of the Piqiang magma. The occurrence of coexisting ilmenite and sulphides gives an indication of moderate fO 2 , likely close to FMQ, since sulphides are stable only at fO 2 condition ˂FMQ+1.5 in various magmatic systems (Jugo et al. 2005).
We use the geochemical modelling program MELTS (Ghiorso and Sack 1995) to test if a parent magma composition similar to an alkali diabasic dyke from the vicinity of the Piqiang intrusion (Sample 08KT01-14 of Zhang et al. 2010) could model successfully the evolution of the Piqiang magma. The result of the MELTS modelling is shown in Figure 14. In the model, we use fO 2 = FMQ, a starting temperature of 1200°C, a final temperature of 900°C, H 2 O = 0.5 wt.% and a pressure of 1 kbar. The MELTS modelling shows that the crystallization sequence of the melt is orthopyroxene → plagioclase → clinopyroxene+Fe-Ti spinel → apatite. This is not consistent with the observations from the Piqiang intrusion, where orthopyroxene is only found as a minor intercumulus phase in anorthosites (Figure 4 (f)). Notably, olivine, which is a significant cumulus phase in most of the Piqiang rocks (Figure 4), is not present in the modelled assemblage. It is probable that the high SiO 2 content in the model parent magma results in orthopyroxene rather than olivine appearing on the liquidus at the very early stage of differentiation. To justify this, we conducted another model run using the same starting composition with a low SiO 2 content (49.2 wt.% according to sample ZK4-2-15-3; Zhang et al. 2018) under similar conditions. The modelling of the evolution of the parental magma suggests that the sequence of mineral crystallization is olivine → plagioclase → clinopyroxene → Fe-Ti spinel → apatite, which is generally consistent with petrographic observations. The calculated compositions of olivine (Fo = 67-70), plagioclase (An = 34-63), and clinopyroxene (Mg# = 60-74) are roughly comparable to the measured range of cumulus mineral compositions (Fo = 61-69, An = 48-68, and Mg# = 58-77).

Magma generation and nature of the mantle source
It has been demonstrated that the Tarim LIP has two main magmatic pulses, i.e.~290 Ma and~280 Ma. The~290 Ma basalts were generated from shallow melting of the metasomatized sub-continental lithospheric mantle (SCLM) by conductive heating from the deep-seated plume, whereas the~280 Ma Bachu layered intrusions (such as Wajilitag and Mazaertag) and contemporaneous diabasic dykes are products of deepmelting of the plume head Xu et al. 2014;Yu et al. 2017). In addition, Zhang et al. (2013a) and Xu et al. (2014) suggested the existence of small-volume~300 Ma magmatism represented by Wajilitag kimberlites and it might represent the earliest eruption resulting from the melting of the metasomatized base of the Tarim SCLM during plume-lithosphere interaction. As noted earlier, the Piqiang layered intrusion has been dated at 270.8 ± 2.3 Ma. At Wajilitag, there is also a limited number of carbonatitic, lamprophyric and nephelinitic dykes with an age between 266 and 272 Ma (Wang 2014;Cheng et al. 2015;Song et al. 2017). These ages suggest that a weak but significant mantle-derived magmatism indeed occurred~10-20 Ma later than the main stage of Tarim plume magmatism. Campbell and Griffiths (1990) have proposed that plume magmatism lasts a relatively short time, with the bulk of magmas emplaced over a period of 1-3 Ma and then followed by less voluminous mantle melting products during a further 5-25 Ma. This scenario agrees well with time scale of the Tarim LIP.
Generally, the cumulate rocks of the Piqiang intrusion has REE and incompatible trace element patterns similar to those of Bachu alkali diabasic dykes and OIB (Figure 9 (a-c)). The Sr, Nd and Hf isotope compositions of the Piqiang intrusion fall into the ranges for Bachu alkali diabasic dykes and OIB (Figure 10(a-b)). These lead us to conclude that the Piqiang intrusion was genetically related to the Tarim mantle plume activity. However, the cumulate rocks also have negative Nb-Ta and Zr-Hf anomalies, possibly indicating a fingerprint of lithospheric mantle or crustal materials. As stated earlier, the Piqiang rocks show minor crustal contamination, and these signatures probably reflect the true character of their mantle  Figure 13. logfO 2 (ΔFMQ)−temperature diagram constructed from compositions of coexisting titanomagnetite and ilmenite pairs of the Piqiang rocks. Oxygen fugacity is normalized to that of the fayalite-magnetite-quartz (FMQ) buffer, where log fO 2 (ΔFMQ) = log fO 2 −log fO 2 (FMQ). Solid lines labeled 'I' or 'U' are isopleths of ilmenite and ulvöspinel, respectively, adapted from Frost et al. (1988). Dashed lines are extrapolated lines representing possible temperature and oxygen fugacity conditions. Data of the Mazaertag and Wajilitag intrusions are shown for comparison and from Cao et al. (2014) and Cao (2015), respectively. source. Thus, the negative Nb-Ta and Zr-Hf anomalies of the Piqiang mafic-ultramafic rocks could be attributed to the involvement of lithospheric mantle. It can further be noted that the negative to slight positive whole-rock ε Nd (t) values (-3.2 to +0.9), enrichment of LILE (e.g. Rb, K, Th, and U) and light REE, and depletion of HFSE (e.g. Nb, Ta, Zr, and Hf) of the Piqiang mafic-ultramafic rocks suggest a chemically enriched SCLM. On the other hand, the Hf-Nd isotopic systems are decoupled (Figure 10(b)) with negative Δε Hf values (−4.1 to −0.6), signatures typically observed from zircon-bearing sediments (Bayon et al. 2009;Guo et al. 2014). The rather high (Ta/La) PM and (Hf/Sm) PM ratios indicate that mantle source metasomatized by subduction-related processes (Figure 10(e)). As stated above, the northern margin of the Tarim Craton, where the Piqiang intrusion is located, was strongly influenced by the southward subduction of the South Tianshan oceanic plate during Ordovician to Devonian times (Huang et al. 2013;Ge et al. 2014). This is evidenced by the occurrence of Early-Middle Paleozoic Andean-type continental arc magmatism on the northern margin of the Craton (460−400 Ma; Ge et al. 2014;Kong et al. 2019). Hence, the enriched SCLM source was probably metasomatized by subduction-related melts or fluids derived from subducted zircon-bearing marine sediments during the Early-Middle Paleozoic when the South Tianshan oceanic slab was subducted beneath the northern margin of the Tarim Craton. The subduction and metasomatic events affecting lithospheric mantle during the Early-Middle Paleozoic are also supported by the noble gas isotope studies on olivine and clinopyroxene separated from the~268 Ma Wajilitag nephelinite, which is~150 km southeast of Piqiang (Figure 1(b); Kong et al. 2017). We thus propose that the parental magmas of the Piqiang intrusion were generated by plume-lithosphere interaction or ascending plume-derived melts contaminated by the enriched SCLM, analagous to models proposed for the origin of Group 2 basalts from the Tarim LIP (Yu et al. 2017) and Panzhihua intrusion from the Emeishan LIP (Hou et al. 2013).
The elevated (Tb/Yb) PM ratios (Figure 10(f)) indicate that the magmas parental to the Piqiang intrusion were  probably derived from a garnet-bearing source region rather than a spinel-bearing source (Xu 2001 (Furman and Graham 1999). The mafic-ultramafic rocks of the Piqiang intrusion have low Rb/Sr values (˂0.07) and high Ba/Rb (mostly ˃20) values, consistent with melting of an amphibole-bearing lherzolite. Thus, the Piqiang magmas could have been derived from an amphibole-bearing garnet lherzolite mantle source. Class and Goldstein (1997) have discussed evidence for the presence of amphibole and phlogopite in the mantle sources for some ocean island basalts, and suggest that metasomatism of the oceanic lithosphere by small volume silicate melts plays an important role in ocean island magmatism. As discussed previously, the lithospheric mantle underneath the Tarim Craton has likely been metasomatized by slab-derived melts or fluids during Early-Middle Paleozoic subduction. The subducted oceanic slab likely underwent eclogitefacies metamorphism (Qu et al. 2011) and generated slab melts, a scenario supported by the presence of the ca. 430 Ma Baicheng dioritic pluton on the northern margin of the Tarim Craton . On the Ba versus Nb/Y diagram (not shown), the Piqiang silicate rocks have the relatively large variation of Nb/Yb (0.12-2.13) with restricted Ba contents (29.6−267 ppm), suggesting that the source of the Pi qiang intrusion was plausibly metasomatized by melts derived from the subducted slab and overlying sediments as well (Kepezhinskas et al. 1996). The infiltration of subduction-related melts into the lithospheric mantle would lead to the formation of metasomatic minerals, such as pyroxene, amphibole, garnet and/or phlogopite (e.g. Yaxley 2000;Herzberg 2006). This scenario is supported by the presence of peridotite and clinopyroxenite mantle-derived xenoliths as well as clinopyroxene, garnet, amphibole and phlogopite entrained xenocrysts in the Wajilitag kimberlitic rocks (Cheng et al. 2014).
Experimental studies have revealed that the formation of primitive Fe-rich magmas is genetically related to the presence of garnet pyroxenite in the source under high pressure (~5 GPa) and temperature (~1550°C) (Tuff et al. 2005). The garnet pyroxenite, which may occur as blocks or veins in the shallow lithospheric mantle (Hirschmann and Stolper 1996), may be produced by the reaction of subduction-related melts with lithospheric peridotites (Herzberg 2006). Therefore, the Piqiang ferrobasaltic parental magmas could have been generated by partial melting of a mixture of an ascending mantle plume and the garnet pyroxenite-loaded, enriched SCLM that was metasomatized by subduction-related melts prior to magma generation.

Origin of the Fe-Ti oxide ores
Compared with Fe-Ti oxide mineralization hosted by the Wajilitag intrusion in the Tarim LIP, massive Fe-Ti oxide ores are more common in the Piqiang intrusion (Cao et al. 2017a;Zhang et al. 2018). Cumulates with very high Fe-Ti oxide modes (>85 vol.%) approaching monomineralic facies have been reported from other Fe-Ti-V oxide deposits in mafic-ultramafic layered intrusions such as the Bushveld complex in the Bushveld LIP (Cawthorn and Ashwal 2009) and Panzhihua and Hongge intrusions in the Emeishan LIP . Two competing models have been proposed for the genesis of massive Fe-Ti oxide ores: (1) crystallization of a distinct Fe-Ti-(P) rich immiscible liquid segregated from the evolved basaltic magma (Reynolds 1985;Wang and Zhou 2013;Liu et al. 2014b;Zhou et al. 2013), and (2) gravitational settling and sorting of Fe-Ti oxide minerals from the Fe-Tienriched magmas (Pang et al. 2008;Bai et al. 2012;Cawthorn 2013;Song et al. 2013;Luan et al. 2014).
Silicate liquid immiscibility has been proposed for the genetic mechanism for some Fe-Ti-V-(P) oxide deposits in various magmatic intrusions, including the Upper Zone of the Bushveld Complex (VanTongeren and Mathez 2012), Skaergaard intrusion (Jakobsen et al. 2011), Sept Iles layered intrusion (Charlier et al. 2011;Namur et al. 2012), Panzhihua-type layered intrusions Zhou et al. 2013;Liu et al. 2014b), and Duluth Complex (Ripley et al. 1998). This has been verified by laboratory experiments showing that immiscibility usually develops at very late stage of tholeiitic magma differentiation and Fe, Ti and P are preferentially partitioned into the immiscible Fe-rich liquids during immiscible separation of Si-rich and Ferich silicate liquids (Charlier and Grove 2012). However, in these cases the immiscible Fe-rich liquids commonly contained 33.4 to 52.6 wt.% SiO 2 , 5.60 to 36.0 wt.% total Fe 2 O 3 , 0.08 to 12.2 wt.% TiO 2 and 0.07 to 10.2 wt.% P 2 O 5 . In the Piqiang case, the massive Fe-Ti oxide ores are extremely enriched in total Fe 2 O 3 (62.8−72.5 wt.%), TiO 2 (14.0−17.6 wt.%) but poor in SiO 2 (3.0−10.2 wt.%), P 2 O 5 (˂0.01 wt.%). P strongly partitions into the Fe-rich liquid (D P LFe / LSi ≈8−10, Watson 1976;Schmidt et al. 2006). If massive Fe-Ti oxide ores were formed from an Fe-rich liquid, they would have had cumulus apatite. The coexistence of apatite with Fe-Ti oxides is indeed observed in the Bushveld Complex (Eales and Cawthorn 1996), Skaergaard intrusion (Jakobsen et al. 2005) and Sept Iles layered intrusion (Namur et al. 2012). Therefore, the absence of coexisting apatite and Fe-Ti oxides indicates liquid immiscibility is not the mechanism for producing Piqiang massive Fe-Ti oxide ore layers. Moreover, our field observation that the base of massive Fe-Ti oxide ore layers is typically gabbro, rather than anorthosite as is common in the Bushveld Complex (Cawthorn and Ashwal 2009), which is also a strong argument against silicate liquid immiscibility at Piqiang. Thus, silicate liquid immiscibility is incapable of explaining the formation of Piqiang massive Fe-Ti oxide ores, and alternative genetic model involving crystal settling and mechanical sorting of dense Fe-Ti oxide minerals is preferred.
According to our studies, the Piqiang rocks exhibit a large range of major element compositional variation (Figure 7), yet all have similar Sr-Nd isotopic and trace element characteristics (Figures 9 and 10), suggesting that the parental magmas have experienced varying degrees of fractional crystallization and crystal accumulation after emplacement. The order of crystallization can be deduced from the field relations, petrographic observations and MELTS modelling. The following crystallization sequence in the Piqiang intrusion is proposed: Olivine is the earliest phase on the liquidus, followed by plagioclase, clinopyroxene and then Fe-Ti oxides. Olivine, plagioclase and clinopyroxene began to crystallize during cooling of the magma chamber. Continuous fractionation of silicate minerals (in particular plagioclase) may have driven the elevated concentrations of Fe and Ti in the residual magma (Figure 15(a)). In this way, the residual magma becomes strongly enriched in the total FeO, TiO 2 and depleted in SiO 2 (Figure 14), which may promote subsequently extensive Fe-Ti oxide crystallization. This is in accordance with the occurrence of massive Fe-Ti oxide ores in the middle-upper parts of the Piqiang gabbroic section (Figure 2(c)) and ilmenite exsolution lamellae in the clinopyroxene crystals (Figure 4(d)). The anorthositic rocks were probably formed at this stage by flotation of plagioclase to the top of the magma chamber (Figure 15(a)).
As fO 2 is thought to significantly influence the Fe-Ti oxide saturation of mafic magmas (Toplis and Carroll 1995), the formation of the massive Fe-Ti oxide ores in the Panzhihua and Hongge intrusions is therefore attributed to the early crystallization and extensive accumulation of Fe-Ti oxides as a result of elevated fO 2 in magmas (Ganino et al. 2008;Pang et al. 2008;Bai et al. 2012). Ganino et al. (2008) suggested that the higher fO 2 conditions recorded by the Panzhihua intrusion were associated with the release of abundant CO 2 from carbonate wall-rocks during contact metamorphism. Although assimilation of the footwall limestones might result in significant elevation of fO 2 and mass crystallization of the Fe-Ti oxides immediately after emplacement of the Piqiang intrusion, such a hypothesis cannot explain that why the massive Fe-Ti oxide ore horizons in this intrusion are not in direct contact with the limestone wall-rocks ( Figure 2). Furthermore, we found no evidence for significant bulk assimilation of limestone wall-rocks in the Piqiang intrusion based on our calculations, suggesting that oxidation of basaltic magma by CO 2 -rich fluids released from limestone wall-rocks did not play a major role in the formation of the massive Fe-Ti oxide ores in the Piqiang case.
Recently,  and  have suggested that Fe-Ti oxides crystallizes late in dry (<0.5 wt.% H 2 O) magmas and early in wet (>1.5 wt.% H 2 O) magmas and that hydration is a key factor affecting the stability of Fe-Ti oxides over silicate phases (in particular plagioclase) in the evolution of the ferrobasaltic magmas. This in turn suggests addition of H 2 O would result in extensive crystallization of Fe-Ti oxides prior to silicates and resultant consumption of previously crystallized silicate phases . The hypothesis could be reconciled with the silicate disequilibrium textures observed in the Piqiang Fe-Ti oxide ores. The widespread planar foliation and lineation shown by the orientation of plagioclase and clinopyroxene in the Piqiang Fe-Ti oxide ores (Figure 4(g)) may indicate that magmatic currents and a laminar flow regime may have resulted from the replenishment of magma (e.g. Wager and Brown 1967;Irvine 1987). Furthermore, the high abundances of the interstitial amphibole (~2−5%) in the Piqiang Fe-Ti oxide ores indicated that the pre-existing residual magma was H 2 O-enriched (2-3 wt.%), probably when Fe-Ti oxides crystallized (Luan et al. 2014). Likewise, symplectites between clinopyroxene and Fe-Ti oxides or plagioclase ( Figure 5(c)) clearly indicated the presence of 'new' hydrated melts introduced during intrusion solidification, which is likely associated with multiple replenishment of Fe-Ti-magma from a deeper source (Gao et al. 2017). Thus the Piqiang Fe-Ti oxide ores could be produced by large scale Fe-Ti oxide crystallization at the expense of silicates because of an introduction of additional H 2 O to the initially dry (~0.5 wt.% H 2 O) magma . Considering that the wall-rocks of the Piqiang intrusion are limestone and sandstone and the intrusion is directly intruded into them, it is likely that external H 2 O was introduced into the Piqiang intrusion during the assimilation of the wall-rocks, although the amounts of crustal contamination are small (Ganino et al. 2008;Luan et al. 2014). Addition of H 2 O is known to reduce melt viscosities (Giordano et al. 2008), which could effectively facilitate rapid transport and accumulation of the Fe-Ti oxides in the H 2 O-enriched magma during the late stages of magma differentiation. This process, coupled with successive Fe and Ti enrichment in the Piqiang fractionating magma, would have created favorable conditions for the precipitation of large quantities of Fe-Ti oxides at late-stage and resulted in the generation of quantities of dense, Fe-Ti oxide-rich slurries . These dense Fe-Ti oxide crystal slurries with low viscosity infiltrated downwards through the unconsolidated crystal pile that occurs below them. During this stage, they sorted during settling to form massive Fe-Ti oxide ore layers at various stratigraphic levels in the gabbroic section. It is implied, therefore, in this model that massive Fe-Ti oxide ores should be be intrusive into lower lithologies within the intrusion. In fact, the main Fe-Ti oxide ore bodies in Piqiang generally show clear intrusive boundaries with the host gabbros ( Figure  3(d-e)). The compaction of the crystal mush by gravitational accumulation further leads to expulsion of residual interstitial liquid (Figure 15(b)) and formation of essentially monomineralic titanomagnetite layers that exhibit polygonal grain boundaries which meet in triple junctions with interfacial angles of 120°( Figure 5(d-e); Reynolds 1985). The residual liquid from the main magma body was likely silicic and might be responsible for the intrusion of a series of granitic dykes that intersected the Piqiang layered intrusion (Zhang et al. 2010;Cao et al. 2013).
Therefore, it is reasonable to infer that a lengthy period of fractional crystallization of a Fe-Ti-enriched parental magma, involving later addition of water to the fractionating magma after its final emplacement, finally results in large amounts of Fe-Ti oxide precipitation and gave rise to the growth of distinct massive Fe-Ti oxide ore layers in this case.

Conclusions
The Piqiang intrusion was formed at~270 Ma and may represent the last pulse of the Tarim LIP magmatism. The parental magmas are deduced to be ferrobasaltic in composition. Partial melting of the Permian Tarim plume that interacted with the garnet pyroxenite component in an enriched lithospheric mantle source could explain the petrogenesis of the ferrobasaltic parental magma. Lithospheric mantle enrichment probably resulted from metasomatism associated with oceanic sediment recycling during southward subduction of the South Tianshan oceanic slab during Early-Middle Paleozoic. The Piqiang rocks are clearly cumulates and we consider that crystal settling and mechanical sorting is the predominant process responsible for their formation. The presence of silicate disequilibrium textures and symplectites between clinopyroxene and plagioclase indicates later addition of external H 2 O from the wallrock to the more fractionated magma during the late-stage of magma differentiation. We propose a model in which abundant dense Fe-Ti oxide crystals formed from the fractionated parental magma during the late stages of magma differentiation and then rapidly settling to the underlying unconsolidated silicate crystal pile. The dense Fe-Ti oxide crystal slurries further tended to effective accumulate Fe-Ti oxides to form massive Fe-Ti oxide ores under a H 2 O-rich environment. The formation of massive Fe-Ti oxide ores could plausibly result from a combination of the protracted differentiation history of a Fe-Ti-enriched parental magma and the later addition of water to the fractionating magma after its final emplacement.