Eoarchean subduction-like magmatism recorded in 3750 Ma mafic–ultramafic rocks of the Ukaliq supracrustal belt (Québec)

Our understanding of the nature of crustal formation in the Eoarchean is limited by the scarcity and poor preservation of the oldest rocks and variable and imperfect preservation of protolith magmatic signatures. These limitations hamper our ability to place quantitative constraints on thermomechanical models for early crustal genesis and hence on the operative geodynamic regimes at that time. The recently discovered ca. 3.75 Ga Ukaliq supracrustal enclave (northern Québec) is mainly composed of variably deformed and compositionally diverse serpentinized ultramafic rocks and amphibolitized mafic schists whose metamorphic peak, inferred from phase equilibria modeling, was below 720 °C. Inferred protoliths to the Ukaliq ultramafic rocks include cumulative dunites, pyroxenites, and gabbros, whereas the mafic rocks were probably picrites, basalts, and basaltic andesites. The bulk-rock and mineral chemistry documents the partial preservation of cumulative pyroxenes and probably amphiboles and demonstrates the occurrence of a clinopyroxene-dominated, tholeiitic suite and an orthopyroxene-dominated, boninite-like suite. Together with the presence of negative μ142Nd anomalies in the boninitic basalts, two liquid lines of descent are inferred: (i) a damp tholeiitic sequence resulting from the fractionation of a basaltic liquid produced by mantle decompression; and (ii) a boninitic suite documenting the evolution of an initially primitive basaltic andesite liquid produced by flux melting. Petrographic observations, thermodynamic modeling, bulk-rock and mineral chemistry, and 142Nd isotopic compositions identify the Ukaliq supracrustal belt as the remnant of an Eoarchean arc crust produced by the recycling of Hadean crust in a similar way as modern-style subduction.


Introduction
Plate tectonics has long governed the mode of crust formation and cooling on our planet, including the long-term operation of the geochemical cycles and, hence, the evolution of the atmosphere, hydrosphere, and biosphere in what has been termed "biogeodynamics" (e.g., Stern 2002;Von Huene and Scholl 1991;Zerkle 2018). Within this biogeodynamic framework, subduction zones generate continental crust through partial melting of the mantle wedge, magmatic accretion beneath island arcs, and refinement via additional remelting and density sorting (e.g., Cawood et al. 2013;Hacker et al. 2011;Ringwood 1974;Schmidt and Jagoutz 2017;Taylor and McLennan 1985). The record of Hf isotopes in zircon is interpreted to show that > 70% of crustal growth occurred in the Archean, or before about 2.5 Ga (e.g., Belousova et al. 2010;McCulloch and Bennett 1994). Based mostly on geochemical and mechanical-structural constraints, the prevailing process associated with Archean Communicated by Timothy L. Grove.
* Thomas Grocolas thomas.grocolas@unil.ch crustal growth asserts that continental drift without subduction as we know it molded the Hadean-Archean Earth's crust (e.g., Bédard et al. 2003;Shirey and Richardson 2011). Considering a mantle potential temperature ~ 300 °C greater than that of today, mantle melting ought to have occurred at a greater depth to produce a relatively thick and buoyant crust (Johnson et al. 2014;Korenaga 2006;McKenzie and Bickle 1988;Sleep 2005). Arguably, the thermal and mechanical properties of such thick crust inhibited subduction processes to instead favor emplacement of a long-lived lithosphere susceptible to reworking via what may have been catastrophic vertical transfer events (e.g., Bédard 2006Bédard , 2018Fischer and Gerya 2016a;O'Neill and Debaille 2014;Van Thienen et al. 2004). At odds with these interpretations are recent numerical models showing that subduction can proceed even under the thermal boundary conditions of very thick and buoyant crust (Maunder et al. 2016;Weller et al. 2019), although other models demonstrate that subduction is inhibited even with a mantle 30 °C hotter in the Proterozoic (e.g., Davies 1992). Polat et al. (2002) present petrological and geochemical constraints from Eoarchean rocks that are compatible with subduction initiation at approximately 3.8 Ga. Similarly, the heavy Si isotope signature of tonalites-trondhjemites-granodiorites (TTG), interpreted to reflect a recycled sedimentary component, lends support to the idea that the onset of plate tectonics occurred at the Hadean-Eoarchean transition around 4 Ga ago rather than sometime later (Deng et al. 2019;Trail et al. 2018). Melt inclusions captured in Hadean zircons analyzed by Hopkins et al. (2008Hopkins et al. ( , 2010 are also used to argue for plateboundary processing of crust before 4 Ga. A competing study of silicate and sulfide inclusions captured in ancient diamonds (Shirey and Richardson 2011), as well as recent studies based on O and B isotopic composition of Paleoarchean to Mesoarchean TTG (Smit et al. 2019;Smithies et al. 2021), instead argues in favor of the initiation of plate tectonics after about 3 Ga, whereas studies of ophiolites and high-pressure metamorphic terranes (Stern 2005) suggest that this process only began as recently as Neoproterozoic time. Therefore, the petrogenetic process associated with the crustal growth trajectory of the Archean Earth remains widely debated because of the poor preservation of crustal material from these early times that could provide direct information on the processes involved in crustal genesis. Hence, resolving these conflicting conclusions about Earth's history of plate tectonics requires analysis of the oldest rocks in addition to numerical modeling. However, this approach is only effective if the oldest preserved rocks can be shown to contain petrological and geochemical characteristics that allow for distinguishing between the competing Archean petrogenetic models cited above. One such early Archean crustal fragment is the ~ 12,000 km 2 Archean Inukjuak domain in the northeast Superior Province of Québec, Canada (Greer et al. 2020). Briefly, the Eoarchean supracrustal enclaves of the Ukaliq (and nearby Nuvvuagittuq) locality are part of the Innuksuac complex (Simard et al. 2003), an association of scattered and variably deformed supracrustal rafts which range in size from < 1 m to > 1 km and are in turn intruded by granitoid gneisses of the Inukjuak domain. As described by Caro et al. (2017), the Ukaliq rocks comprise a series of mafic schists interpreted to have volcanic protoliths chemically similar to those found in a modern forearc environment such as tholeiitic and boninitic lavas. These are also associated with calc-alkaline andesites, the identification of which challenges the exclusivity of a vertical tectonic model for the entirety of the Archean (e.g., Turner et al. 2014), although a subductionless hypothesis has also been advocated by Barnes and Van Kranendonk (2014). We wish to emphasize that rocks from the Innuksuac complex display similar geochemical signatures to rocks documented in the 3.81-3.70 Ga Isua supracrustal belt (ISB; southern West Greenland; Szilas et al. 2015 and references therein) as well as in younger Archean complexes (e.g., Cawood et al. 2006). Yet, as opposed to the ISB rocks which have well documented higher 142 Nd/ 144 Nd values (Caro et al. 2003) relative to bulk silicate Earth (BSE) and reported in the conventional μ 142 Nd notation as positive anomalies, the numerous lithologies of the Innuksuac complex preserve variably negative μ 142 Nd anomalies O'Neil et al. 2008;Roth et al. 2013). Because of the short half-life of 146 Sm (t 1/2 = 103 Ma), such negative 142 Nd anomalies can only be produced in low Sm/Nd (crustal) protoliths differentiated prior to ~ 4 Ga (e.g., Caro 2011).
There are two ways to explain these divergent μ 142 Nd values for what otherwise appears to be synchronous Eoarchean terranes: (i) the negative μ 142 Nd were produced by in situ decay of 146 Sm after emplacement of the rocks, in which case the Nuvvuagittuq belt is of Hadean age (O'Neil et al. 2008(O'Neil et al. , 2019; or (ii) the negative μ 142 Nd signature is inherited from a now-vanished Hadean lithosphere and the μ 142 Nd-Sm/Nd correlation interpreted by O'Neil et al. (2008) as an isochron instead represents a mixing line without geochronological significance Guitreau et al. 2013). Such an inherited signal can be duplicated by assimilation of a Hadean 142 Nd anomaly-bearing crust or foundering of Hadean lithosphere leading to the transfer of a crustal isotopic signature to the overlying mantle wedge.
Two key observations can, however, weaken the assimilation argument. The first of these is that, despite the ubiquitous Hadean crustal signatures, there are no inherited or detrital zircons of Hadean age in rocks of the Innuksuac complex. This is despite thousands of U-Pb zircon analyses performed on samples of igneous and detrital sedimentary protoliths collected from throughout the terrane (Cates et al. 2013;Chowdhury et al. 2020;Greer et al. 2020 and references therein). A second argument lies in the absence of crustal (felsic) contaminants with sufficiently unradiogenic 142 Nd signature to account for the μ 142 Nd values found in mafic rocks of the Nuvvuagittuq and Ukaliq supracrustal belts . Collectively, these observations suggest that the Innuksuac complex inherited the geochemical signature of Earth's primordial crust without significant reworking of Hadean felsic protoliths. To account for these observations, which appear to conflict with the assimilation scenario to explain the variably negative μ 142 Nd anomalies, we propose that crustal recycling through subduction was operative, at least intermittently, in the Eoarchean. We further show how this scenario neatly explains the typical forearc sequence preserved in the Innuksuac supracrustals and the trace element concentrations and the enriched 142 Nd and 143 Nd signatures contained therein.
In this work, we focus our attention on the ultramafic-mafic supracrustal enclave at the Ukaliq locality to describe (1) preserved Eoarchean magmatic features, (2) the cumulate-liquid relationship between the ultramafic and mafic rocks, and (3) the cumulate assemblages that formed during the ascent of differentiating magmas. Based on these observations, we provide an explanation for the chemical evolution of the supracrustal rocks that requires a transition from a damp tholeiitic regime to a wet boninitic sequence. In a modern tectonic setting, such an evolution is most often observed in subduction initiation environments.

Geological setting
The variably deformed Eoarchean supracrustal enclaves of the dominantly Neoarchean Inukjuak domain (Minto block, northeast Superior Province, Canada) principally comprise plutonic and volcano-sedimentary schists; these range in age from 3.8 to 3.5 Ga and were metamorphosed at amphibolite facies conditions (0.4 GPa, 640 °C; Cates and Mojzsis 2009;Greer et al. 2020) with local retrogressions. Although less well known than cognate Eoarchean rocks of the ISB, the ca. 3.75 Ga Nuvvuagittuq supracrustal belt (NSB) was the first to show evidence of anomalous depletions in 142 Nd/ 144 Nd relative to BSE (negative μ 142 Nd) that seem to correlate to Sm/Nd (O'Neil et al. 2008). The subject of our study is another neighboring body of metamorphosed volcanosedimentary rocks also displaying this characteristic 142 Nd signature: the Ukaliq supracrustal belt (USB; Caro et al. 2017). The largest USB enclave is a poly-metamorphosed and intensely deformed NNW-trending flat, ellipsoidal body of about 100 m × 7 km ( Fig. 1) a few kilometers north from the NSB. The USB is composed of three main lithologies: (i) a massive amphibolite composed of hornblende-or cummingtonite-rich rocks inferred to have volcanic protoliths; (ii) ultramafic boudins and enclaves, mainly serpentinized; and (iii) intercalated siliceous units comprising layered and strongly tectonized quartz + magnetite ± amphibole ± pyroxene ± olivine rocks interpreted as banded-iron formations (BIF), and quartz + biotite schists and massive to banded  143 Nd are from Caro et al. (2017) and U-Pb ages for the Voizel and Boizard suites are from Greer et al. (2020). Modified after Caro et al. (2017) quartzite (± fuchsite) of detrital origin Greer et al. 2020). Two forms of ultramafic rocks can be distinguished: (i) a thick layer (~ 30 m) parallel to the massive NNW-dipping foliation; and (ii) lenses (1-5 m) included within the amphibolites (Fig. 2a). The contact between ultramafic and mafic rocks is sharp (Fig. 2b) and concordant with the foliation. Several thin (~ 10 cm) BIF layers occur within the mafic rocks and are parallel to the regional structural trend of the USB enclave. The quartzites are mainly located near the eastern contact between USB rocks and the Voizel suite granitoids (3550 Ma; Greer et al. 2020; Fig. 1). Ultramafic rocks range from pure serpentinite on the eastern side to a more pyroxene-rich composition in the west. The NSB differs from the USB in the occurrence of the Ca-poor amphibole cummingtonite; this lithology is rare at Ukaliq, where the amphibole is hornblende as opposed to the NSB where cummingtonite can be the dominant amphibole in amphibolite. Earlier U-Pb thermal ionization mass spectrometry (TIMS) geochronology on detrital zircons recovered from micaschists in the NSB yields an age of 3825 ± 18 Ma (Darling et al. 2013;David et al. 2002), whereas zircons extracted from intrusive trondhjemitic orthogneisses lead to a minimum emplacement age of 3751 ± 10 Ma (Cates and Mojzsis 2007;Greer et al. 2020). Elsewhere in the NSB, detrital zircons from detrital fuchsitic quartzites and micaschists interpreted to be quartz-pebble conglomerates provide a maximum age of emplacement for the various volcanic protoliths of ca. 3.78 Ga (Cates et al. 2013;Darling et al. 2013). For a review of the geology of the wider region and a debate over the interpretation of the geochronology, we refer the reader to the synthesis provided in Greer et al. (2020).

Methods
This study uses 35 samples from Caro et al. (2017) and 22 collected in the USB in 2016. The latter were cut into small blocks and only fresh material was further used for bulkrock analysis. The blocks were crushed in a hydraulic press and ground to a fine powder in an agate mill. Minéraux (SARM, CRPG) facility using the inductively coupled plasma optical emission spectrometer (ICP-OES) iCap6500 and the inductively coupled plasma mass spectrometer (ICP-MS) iCapQ, respectively. The detection limits and standard deviations are reported in the Electronic Supplementary Material ESM 1.
In situ major element compositions of minerals from a subset of 16 samples (8 ultramafic rocks and 8 amphibolites) were determined using the Cameca SX100 electron microprobe at GeoRessources laboratory (Université de Lorraine). Measurements were corrected with the PRZF method (Armstrong 1995). The acceleration voltage was 15 keV and beam current 12 nA. Counting times were 30 s on the peak and 15 s on the background. Trace element analyses were also performed for 6 samples using an Agilent 7500 sector field ICP-MS coupled with a Geolas platform hosting a 193 nm excimer laser housed at GeoRessources laboratory. Prior to the session, the instrument was tuned in linear scan mode on SRM-612 silicate glass. Oxide generation was mitigated to be < 0.07% for 248 ThO + / 232 Th + . SRM-612 was also measured during the analytical session twice every 16 analyses to correct for instrumental drift. A 10 Hz repetition rate was used for sample analysis, and laser output energy density was 6 J cm −2 . The ablation spot size was 60-80 μm for clinopyroxenes (cpx) and amphiboles (amph), and 120-150 μm for orthopyroxenes (opx). The acquisition time for background was 80 s and laser ablation 35 s. The raw data were processed and transformed into elemental contents using the LAMTRACE software (Jackson 2008). The detection limits and standard deviations are reported in the Electronic Supplementary Material ESM 1.

Structure and petrography
Three main lithologies were identified within the main USB body (see Caro et al. 2017): (i) ultramafic rocks (pyroxenerich to pure serpentinite) present as enclaves or as decameter-sized layers; (ii) hornblende-or cummingtonite-bearing amphibolites; and (iii) quartzitic and micaceous rocks of sedimentary protolith with foliation parallel to that expressed in the amphibolites. Therefore, contacts between these lithologies are transposed as the foliation never transects layer contacts, even in highly folded areas, such as the southeastern contact between quartzites and amphibolites ( Fig. 1).

Ultramafic rocks
The dark green serpentine-bearing ultramafic rocks (Fig. 2c) are located in the center of the Ukaliq supracrustal belt and display an average foliation striking N70° ± 10° (Fig. 1). These serpentinites are mainly composed of antigorite and chlorite, as well as magnetite in the groundmass. At the microscopic scale, a cpx-bearing series and a cpx-absent, opx-bearing series are identified. Only one ultramafic sample is composed of cpx (IN16098b); it is located in a single, separate ultramafic lens and contains > 90 area% of antigorite and chlorite. This sample contains limited amounts (< 1 area%) of submillimeter-sized, fractured cpx in inclusion within magnesio-hornblende (Mg-hbl) or in the antigorite + chlorite groundmass (Fig. 3d). These anhedral Mg-hbl form millimeter-sized porphyroblasts preferentially altered along their cleavage planes, separating them into several subgrains. They are systematically surrounded by a tremolite rim typical of disequilibrium between serpentine and amphibole. Magnetite is ubiquitous in the cpx-bearing ultramafic rock and is in equilibrium with antigorite and chlorite.
The cpx-absent ultramafic rocks are also dominated by the antigorite + chlorite + magnetite + tremolite paragenesis, but centimeter-sized plates of subhedral to euhedral opx may represent > 60 area% of a thin section. These opx can be in contact with or surrounded by Mg-hbl ( Fig. 3b), surrounded by antigorite and chlorite (Fig. 3c), or completely transformed into serpentine (Fig. 3a). Antigorite pseudomorph lamellae following opx cleavage can occur and separate opx into several, optically continuous subgrains originally forming large, centimeter-sized opx (Fig. 3c). We interpret these large, partially altered opx as the remnants of an altered cumulative texture (Campbell 1968), while opxassociated antigorite probably forms through fluid circulation along cleavage planes. The Mg-hbl from these samples are similar to those described in the cpx-bearing ultramafic rock. In these cpx-absent ultramafic samples, dark, subhedral to euhedral spinels (spl) occur as inclusion within opx and hbl (Fig. 3b), reminiscent of a typical magmatic relationship (e.g., Bouilhol et al. 2015). Brownish to greenish hercynitic spl are only present in cpx-absent samples and are locally surrounded by magnetite (Fig. 3c), which is typical of disequilibrium between hercynite and serpentine. Finally, talc is locally observed overgrowing antigorite.

Amphibolites
Volumetrically, the USB is dominated by a dark, massive unit and many other smaller deformed enclaves of amphibolites and paragneisses scattered throughout the complex (Chowdhury et al. 2020;Greer et al. 2020). At the centimeter scale, the mafic rocks display a fine-grained (< 0.5 mm) texture ( Fig. 2d) with a typically metamorphic amph + plagioclase (plag) + quartz ± cpx paragenesis. Light gray to beige amphibolite corresponds to the cummingtonite-rich amphibolite much more widespread throughout the neighboring NSB (David et al. 2002). The typical feature of these cummingtonite-bearing amphibolites is the absence of cpx; they will be hereafter referred to as cpx-absent amphibolites. At the microscopic scale, amphibolites exhibit a typical isogranular texture with a foliation marked by millimetersized cpx and amph (Fig. 3e, f), which may be green (Mghbl, tremolite, cummingtonite) to bluish green (pargasite). Except for their colors, Mg-hbl and pargasite share the same textural features. Tremolite often surrounds and grows on top of Mg-hbl, pargasite, and cpx. These overgrowths and the presence of cummingtonite may be attributed to the metamorphic history of the massif. Most of the Mg-hbl and pargasite are in equilibrium with (1) millimeter-sized, partially sericitized plag, (2) quartz with undulose extinction Fig. 3 Photomicrographs in cross-polarized light (a, b, d, and f) and plane-polarized light (c and e) illustrating mineralogy and textural features of USB rocks. a Ultramafic sample exhibiting an opx pseudomorph metamorphosed into antigorite containing an inclusion of hbl. b Subhedral to euhedral Cr-Al spl inclusion within millimetersized opx displaying a sharp contact with hbl in an ultramafic rock. Euhedral, millimeter-sized hbl also occur as inclusion within opx. c Centimeter-sized opx crystal altered to serpentine along cleavage planes within an ultramafic rock. The brownish to greenish hercynitic spl are located in the antigorite groundmass and systematically have a magnetite rim. d Submillimetric cpx inclusions in hbl being surrounded by antigorite and chlorite (> 90% of ultramafic rocks modal abundance) within the cpx-bearing ultramafic sample (IN16098b). e Amphibolite exhibiting a granoblastic texture associated with plag + amph + cpx + quartz + oxides paragenesis. f Fractured millimetric cpx in contact with amph + quartz within an amphibolite. The foliation is marked by elongated cpx and hbl. The subgrain boundaries of quartz are typical of metamorphic recrystallization. atg antigorite; chl chlorite; hrc hercynitic spl; mgt magnetite; qtz quartz. Other abbreviations as in the text and subgrain boundaries that are features characteristic of plastic deformation, and (3) cpx having higher relief than amph (Fig. 3e, f). Subhedral titanite and ilmenite are present at grain boundaries (< 0.1 mm). The texture and the mineral relationships suggest that amph cores and cpx cores surrounded by tremolite may be relics of a relatively high-grade metamorphic event alongside plag and cummingtonite, whereas tremolite probably formed as a result of retrograde metamorphic re-equilibration.

Rocks of sedimentary protolith
Banded-iron formations are mainly composed of alternating bands of Fe-oxides and silicates, which renders them a characteristic reddish color at weathering. Silicate layers record an NNW-striking mineral foliation parallel to the USB main structural grain. At the microscale, the Fe-oxides are associated with Fe-olivine, Fe-pyroxene, and Fe-amphibole, interpreted to form through the isochemical transformation of Feoxides and quartz during amphibolite facies metamorphism (e.g., Klein 2005).
Micaschists of probable detrital origin share the common NNW-striking foliation. They exhibit a grano-porphyroblastic texture with numerous aluminous phases such as garnet and biotite that can be used as a geothermometer. Porphyroblastic garnet (~ 2 mm) containing biotite and quartz inclusions is surrounded by often chloritized, prismatic, millimeter-sized biotite. Millimeter-sized plag and quartz with undulose extinction, as well as small amounts of ilmenite and magnetite, also occur.

Bulk-rock chemistry
Major and trace element concentrations allow us to distinguish between five different protoliths of magmatic origin in the USB (Table 1). The ultramafic rocks can be divided into two different groups according to their mineral assemblage and bulk-rock chemistry, whereas the amphibolites can be separated into three groups based on their bulk-rock chemistry and isotopic signature.
Apart from these three groups, one sample (IN12032) has an atypical chemistry that differs markedly from the other amphibolites. Although its SiO 2 content (52.2 wt.%) and Mg# (52.5 mol.%) are similar compared to previously described rocks (Fig. 4c), the TiO 2 content (0.61 wt.%) is as low as in boninitic amphibolites. However, this sample is different from the boninitic amphibolites by having higher REE concentrations (∑REE = 52.1 ppm) typified by a strong LREE enrichment (La N /Sm N = 4.0) and a HREE depletion compared to N-MORB. When normalized to primitive mantle, this sample is shown to be enriched in large ion lithophile elements (LILE; Th N + U N > 30) and exhibits depletion in high field strength elements (HFSE) with a negative Nb-Ta anomaly (Nb* = 0.19; Fig. 5h). The major and trace element concentrations of this sample are characteristic of calc-alkaline basalts (Fig. 4b) and will, therefore, be referred to as calc-alkaline amphibolite. In addition, this calc-alkaline sample carries the most negative 142 Nd signature (μ 142 Nd = -9.4; Caro et al. 2017).

Mineral chemistry
The representative major and trace element analyses of cpx, opx, and amph from the ultramafic and mafic rocks are presented in Table 2. The complete dataset can be found in Electronic Supplementary Material 1 (Table S2).
Amphiboles from the tholeiitic ultramafic rock are dominantly Mg-hbl and tremolite, the latter forming rims around Mg-hbl. The transition from hbl core to tremolite rim is sharp, with Si increasing from 7.15 to 7.95 a.p.f.u., while (Na + K) A and Al IV decrease from 0.25 to 0 a.p.f.u. and from 0.75 to 0 a.p.f.u., respectively (Fig. 7). Mg-hbl delineate two distinctive trends of increasing TiO 2 (0-0.89 wt.%) contents with decreasing Mg# (Fig. 7c), probably representing two generations of amph. In both Mg# groups, high TiO 2 (> 0.1 wt.%) amphiboles also have high Cr 2 O 3 (> 0.2 wt.%) contents, contrasting with the low Cr 2 O 3 contents of tremolite. When chondrite normalized, Mg-hbl exhibit a smooth, nearly parallel pattern compared to cpx except for HREE being less fractionated (Dy N /Yb N ≈ 1.20) and a small positive Eu anomaly (Fig. 7d). When primitive mantle normalized, Mg-hbl display slight U-Th enrichment and little to no HFSE depletion. Amphiboles from the boninitic ultramafic rocks are Mg-hbl and display increasing TiO 2 concentration with decreasing Mg# (Fig. 7c), which is negatively correlated with Cr 2 O 3 , trending from 1.0 to 0.2 wt.% with decreasing Mg#. Chondrite normalized, these hbl have the same U-shaped REE pattern as their host rock typified by their low (Dy/Yb) N values in the range of 0.68-0.69 and a small negative Eu anomaly (Fig. 7d). Normalized to primitive mantle, they display strong Nb, Ta, and Sr negative anomalies and U and Pb enrichments.
Amphiboles from tholeiitic and transitional amphibolites range from Mg-hbl to pargasite and are often surrounded by retrogression tremolite (Fig. 7a). The Mg-hbl and pargasite display a broad range of Mg# from 36.1 to 69.2 mol.% uncorrelated to TiO 2 (Fig. 7c). Chondrite normalized, Mghbl exhibit a nearly parallel pattern compared to cpx from mafic rocks, but with higher concentrations. Pargasites were not analyzed for their trace elements. Amphiboles from boninitic amphibolites are cummingtonite, corresponding to CaO-poor (0.40-1.17 wt.%) amph. Overall, cpx and amph from amphibolites are chemically zoned with Al 2 O 3 -, TiO 2and Cr 2 O 3 -rich cores that differ from tremolite rims which usually display depletion in the aforementioned elements.

Pressure-temperature constraints
In this section, we use phase equilibria modeling and geothermometers to assess the equilibrium conditions of the USB rock phases. This allows distinguishing between  Sun and McDonough (1989). CI CI-type chondrite; S.D. standard deviation medium-to-high temperature metamorphism and higher temperatures, magmatism-related phase crystallization. The mineral assemblage of the micaschists allows the determination of pressure and temperature conditions of the last metamorphic peak experienced by the Innuksuac complex. The paragenesis of sample IN16164 is biotite + garnet + plag + quartz + chlorite + ilmenite + magnetite and contains 7-9 vol.% garnet with an almandine-rich composition (Alm 71 , Prp 15 , Grss 10 , Spss 4 ). Pseudosections were calculated by Gibbs energy minimization using Perple_X (version 6.9.1; Connolly 2005) and the self-consistent thermodynamic database of Holland and Powell (2011) in the system NCKFMASHTO. The selected solution models and the chosen H 2 O and O 2 contents are discussed in the Electronic Supplementary Material ESM 2. The pressure-temperature equilibrium phase diagram calculated for sample IN16164 reveals that the observed assemblage is stable from 3 to 7 kbar and 580-630 °C (Fig. 9). Garnet isopleths and isochores further narrow the domain of equilibrium of our sample to 4-5.5 kbar and 615 ± 10 °C. To strengthen the thermodynamically inferred temperatures, paired analyses on garnet and biotite have been performed on the same sample. Temperatures are calculated based on the garnet-biotite thermometer of Ferry and Spear (1978) and are consistent within error (593 ± 35 °C) with the modeled pseudosection. Similar garnet-biotite temperatures (580 ± 20 °C) are reported in the NSB (Cates and Mojzsis 2009). Amphibole compositions have been shown to be temperature and/or pressure dependent (e.g., Holland and Blundy 1994). However, Al VI and Ti contents from USB amph display no correlation with Al IV , indicating that the pressuredependent Al-Tschermak and temperature-dependent Ti-Tschermak exchanges only played a minor role. As shown in Fig. 7b, the dominant vector controlling amph chemistry is the temperature-dependent edenite exchange, suggesting that the evolution in (Na + K) A is controlled by cooling at constant pressure. As such, Ti-dependent amph thermometers (e.g., Putirka 2016) cannot be used to constrain the crystallization temperature of amph from the USB ultramafic rocks. We thus applied the edenite exchange-based amph-plag thermometer of Holland and Blundy (1994) which gives temperatures of 704 ± 47 °C for the amphibolites, slightly higher than the amph-plag temperatures (≤ 680 °C) reported for the NSB (Cates and Mojzsis 2009).
To determine the equilibrium conditions of opx from the boninitic ultramafic rocks, we used the Ca-in-opx thermometer of Brey and Köhler (1990). Assuming that the opx from USB equilibrated at 0.5 GPa, we obtain temperatures ranging from 820 to 980 °C. Despite its pressure dependency, the Ca-in-opx thermometer is almost insensitive to pressure changes, with an increase of 10 °C for a pressure of 0.8 GPa. The two-pyroxene thermometers are not employed as cpx and opx are never found in equilibrium in USB rocks.

Discussion
In this section, we use phase relationships, bulk-rock and mineral chemistry, and thermometry to unravel the metamorphic signal from the original magmatic signal and reconstruct the cumulate sequence followed by tholeiitic and boninitic melts by characterizing the most primitive magmas and their associated liquid lines of descent.

Magmatic signal preservation
It is evident from the mineralogy and pseudosection modeling that the USB underwent at least one metamorphic episode that reached amphibolite facies conditions. In detail, the Innuksuac complex experienced at least two metamorphic episodes (3622 ± 46 Ma and 2738 ± 25 Ma) at the amphibolite facies Mojzsis 2007, 2009). These episodes probably correspond to magmatic intrusions of the Voizel and Boizard suites, respectively (e.g., Greer et al. 2020). These metamorphic events also likely modified the most fluid-mobile element concentrations of the USB rock suites. To assess the degree of preservation ) and isochores (6, 8, and 10 vol.% garnet) are used to constrain the peak pressure-temperature conditions underwent by the USB. bi biotite; cd cordierite; chl chlorite; ep epidote; g garnet; ilm ilmenite; L silicate melt; mt magnetite; mu muscovite; pl plagioclase; q quartz. Other abbreviations as in the text of the original magmatic signal, we used the correlationbased method of Cann (1970). This method relies on the degree of correlation of any given element with Zr, considered fluid-immobile (e.g., Fraser et al. 1997). A theoretical crystallization vector can be drawn for each plot and each lithological unit, allowing the identification of uncorrelated, altered samples (Pearce 2014b). It is important to note that ultramafic rocks cannot be considered in the crystallization vector because they do not have a liquid composition (see below). Nevertheless, an identification of the altered samples will be attempted. As shown in Fig. 10a, a good correlation is found between Zr and Th, except for 2 tholeiitic amphibolites characterized by slightly lower Th contents. Overall, the REE also present a good correlation with Zr except for 3 tholeiitic ultramafic rocks that are La-rich and Yb-poor (Fig. 10b, c). The Nb vs. Zr diagram also shows a good correlation for every unit (Fig. 10d). Similar conclusions can be drawn for SiO 2 and MgO. In contrast, fluid-mobile elements (not plotted) display poor correlations with Zr, which can be interpreted as a magmatic signal, reflecting variable fluid-mediated LILE enrichments in the source or as a late metasomatic overprint. These processes are not mutually exclusive but cannot be distinguished based on trace and major element chemistry alone. As such, only REE, HFSE, Y, and Th concentrations will now be considered representative of the magmatic signal and used further. These conclusions are consistent with the expected immobility of Th, Y, REE, and HFSE, and mobility of LILE during fluid circulation (Pearce 2014a, b).

Interpretation of mineral textures and chemistry
Mineral textures and chemistry and thermometric constraints described in the previous sections allow us to infer the metamorphic peak and discuss the origin (metamorphic vs. magmatic) of the different minerals constituting the USB rocks.
For the ultramafic rocks, the equilibrium paragenesis consisting of antigorite + chlorite + magnetite + tremolite is typical of greenschist to lower amphibolite facies conditions (< 600 °C), witnessing the retrogression of the ultramafic rocks. At the pressure-temperature conditions calculated using thermodynamic modeling (Fig. 9) and garnet-biotite and amph-plag thermometers, secondary olivine and opx could be present (Spear 1993). As such, the origin of the d Zr (ppm) vs. Yb (ppm). Theoretical vectors show the covariation expected during crystallization observed opx must be addressed. The large size (> 5 mm) and optical continuity of opx, the inclusion of Cr-spl, and their calculated equilibrium temperatures (~ 980 °C) much above the metamorphic overprint point to a magmatic origin. In addition, the opx from the fractionation experiments of Krawczynski et al. (2012) at 0.5 GPa have similar compositions and crystallize in the same range of temperatures (950-1000 °C) as the opx from USB. As such, we interpret the opx to be of magmatic origin and their trace element concentrations representative of the liquid from which it crystallized. The trace element concentrations observed here are also similar to those found in modern arc to forearc environments (e.g., Jagoutz et al. 2007), strengthening the magmatic origin for the opx. Furthermore, the original magmatic signal is probably preserved for the slow-diffusing elements present in tetrahedrally coordinated sites (e.g., Cr, Al, Ti, and REE; Cherniak and Liang 2012;Ganguly et al. 2007). Indeed, the Cr diffusion closure temperature in opx has been defined at 870 °C for an initial temperature of 900 °C (Ganguly et al. 2007). The origin of cpx and Mg-hbl in the ultramafic rocks is more challenging to assess because of the lack of inferred equilibrium temperatures. The cpx major element compositions show a wide range of Al 2 O 3 , Cr 2 O 3 , and TiO 2 contents for a near-constant Mg#. Low TiO 2 and Cr 2 O 3 cpx are greenschist facies re-equilibrated cpx, but the high-TiO 2 grains point towards a magmatic origin. Indeed, if they were metamorphic, temperatures above 800 °C would be needed to stabilize cpx (Spear 1993), which is much higher than the peak temperature recorded here. The cpx major element compositions are relatively common in cumulative rocks, and elemental variations at a constant Mg# can reflect a reactional origin of the grains (melt-rock reaction) whereby the Mg# is buffered by the percolated rock, and, for example, the high TiO 2 contents reflect the reacting melt. If correct, then the trace element concentrations can be used to evaluate the liquid in equilibrium with the cpx. Interestingly, the observed cpx compositions are similar to those found in cumulative reactional pathways of magmatic arc systems that have drained damp tholeiitic melts (e.g., Chilas complex, Kohistan; Jagoutz et al. 2007).
Two types of amph can be observed in the ultramafic rocks, whereby Mg-hbl almost systematically show a retrogression to tremolite. These Mg-hbl are often found as inclusion or intimately related to the pyroxenes (Fig. 3) and have magmatic characteristics, such as high Cr 2 O 3 and TiO 2 contents showing trends correlated with Mg#. Their major element compositions are often found in arc cumulates (Bouilhol et al. 2015;Cooper et al. 2016;Jagoutz et al. 2007) and experimental studies (Feig et al. 2006;Krawczynski et al. 2012;Nandedkar et al. 2014). Interestingly, Mg-hbl have near-parallel REE patterns and similar concentrations to the opx in the boninitic ultramafic rocks and the cpx in the tholeiitic ultramafic rocks. Such feature is symptomatic of a co-magmatic origin for Mg-hbl and the pyroxenes (Adam and Green 2006;Bouilhol et al. 2015;Tribuzio et al. 2000). On the other hand, the slight positive Eu anomaly of Mghbl from the tholeiitic ultramafic samples could reflect a metamorphic origin whereby Mg-hbl replace plag resulting in high Eu contents and a negative REE slope (Helmy et al. 2008). In all cases, the lack of calculated equilibrium temperatures for these Mg-hbl hampers our ability to constrain their origin unambiguously.
The amphibolites display a granoblastic texture associated with amph-plag temperatures of ~ 700 °C, which is typical for mafic rocks that underwent amphibolite facies metamorphism. Further, both cpx and Mg-hbl show negative Eu and Sr anomalies, relatively flat HREE segments, and have nearly parallel REE patterns, with Mg-hbl being more enriched than cpx by an order of magnitude. Together with a well-equilibrated microtexture, these features call for a global chemical equilibrium between the amphibolitebearing phases.
Overall, opx, cpx, and probably Mg-hbl from ultramafic rocks most likely represent the only primary igneous crystals in the USB rocks that can further be used to infer a crystallization sequence. Finally, the upper bound of the metamorphic peak experienced by the USB might be placed at 720 °C, which corresponds to the onset of partial melting of the metasediments, a feature not observed throughout the Innuksuac complex.

Protoliths of Ukaliq mafic and ultramafic rocks
Ukaliq ultramafic rocks display major element concentrations typical of dunite (SiO 2 = 40-42 wt.%), pyroxenite (SiO 2 = 48-52 wt.%) and (hbl-)gabbro (SiO 2 = 46-48 wt.%, Mg# ≈ 70 mol.%). More specifically, the tholeiitic ultramafic rocks have SiO 2 /MgO (1.0-3.8) and Al 2 O 3 /CaO (0-0.7) ratios, as well as Al 2 O 3 contents (0-6.8 wt.%) indicative of the former presence of olivine, cpx, plag, and amph in the protolith (Fig. 11). In detail, one sample has high SiO 2 /MgO and Al 2 O 3 contents, and most likely crystallized plag, and two samples located between the cpx and amph fields probably formed limited amounts of amph. On the other hand, the boninitic ultramafic rocks span a wide range of Al 2 O 3 / CaO values (1.0-11.1) and Al 2 O 3 contents (4.1-15.9 wt.%) and have lower SiO 2 /MgO values (1.2-2.4) rather pointing to large amounts of olivine, opx, and spl in their protoliths (Fig. 11). The former presence of olivine in both tholeiitic and boninitic ultramafic samples is further evidenced by the high bulk-rock and antigorite NiO contents. Such observations agree well with the observed opx-cpx dichotomy in the ultramafic rocks and characterize these ultramafic rocks as former dunite, wehrlite, orthopyroxenite, and (hbl-)gabbro. The presence of opx phenocrysts in ultramafic rocks and the occurrence of dunite-, pyroxenite-, and gabbro-type rocks are usually related to cumulates during melt evolution. This cumulative origin is well demonstrated by the bulk-rock Mg# of the ultramafic rocks that range from 72 to 90 mol.%, whereas the mantle has higher Mg# (e.g., Bodinier and Godard 2014). The compatible element contents, such as Ni, Cr, and V, also indicate a cumulate origin for the ultramafic rocks. Indeed, refractory mantle rocks have Ni contents up to 3000 ppm, whereas pyroxenitic cumulates have concentrations ranging from 200 to 1400 ppm (e.g., Bodinier and Godard 2014;Bouilhol et al. 2009Bouilhol et al. , 2015. As with the bulk-rock chemistry, the mineral compositions also point to a cumulative origin rather than a mantle origin. Clinopyroxenes from the tholeiitic ultramafic rock have REE concentrations up to ten times the chondrite values, while cpx from the sub-arc mantle and from abyssal peridotites have LREE content values ten to a hundred times lower than chondrite (Bodinier and Godard 2014;Bouilhol et al. 2009). Consequently, the two identified ultramafic groups are now considered as two cumulate suites.
Amphibolite samples have basaltic to basaltic andesite major element compositions with Mg# ranging from 38.2 to 68.1 mol.%, which cannot be used to distinguish between an intermediate cumulate and a lava. Gabbroic rocks usually display a cumulate signal, either cpx-dominated or plag-dominated, leading to a REE pattern with LREE depletion and variable Eu anomaly (e.g., Pallister and Knight 1981). Such a signature is not observed in the USB amphibolites. Although the presence of melt-like gabbros cannot be completely ruled out, these observations, together with the alternation of amphibolites and metasediments, point to a volcanic rather than a cumulate protolith for USB amphibolites. Specifically, we interpret the protoliths of the tholeiitic amphibolites as tholeiitic basalts, the boninitic amphibolites as boninitic basalts to basaltic andesites, and the transitional amphibolites as transitional basalts.

Relationship between ultramafic cumulates and lavas
Mineralogical and geochemical features suggest that the ultramafic cumulates and the basaltic melts are cogenetic. Indeed, tholeiitic suites are characterized by extensive cpx crystallization, whereas boninitic melts are usually opxnormative, which is directly observed within the ultramafic rocks. Geochemically, the tholeiitic and boninitic samples constitute two distinct trends (Fig. 4d) that can be interpreted as differentiation trends and exhibit two distinct REE pattern types (Fig. 5). To further test the cogenetic character of the tholeiitic cumulates and tholeiitic lavas on one hand, and the boninitic cumulates and boninitic lavas on the other, we calculated the REE concentrations of the melts in equilibrium with cpx and opx in the two different cumulate series. The parental melt calculated for cpx from tholeiitic cumulates using partition coefficients from Wood and Blundy (1997) exhibits a REE pattern parallel to that observed in transitional basalts, albeit with slightly lower HREE concentrations (Fig. 12a). Such a discrepancy is probably caused by the non-primary character of the most primitive USB tholeiitic lava and could be alleviated by fractionation of olivine and/or cpx. The same approach has been applied to the transition metals using partition coefficients from Bédard (2014) and yields the same results (Electronic Supplementary Material ESM 2). Given that, the cumulative cpx, and thus the tholeiitic cumulates, represent a cumulative phase of the tholeiitic-to-transitional suite (Fig. 12a).
The calculated liquid in equilibrium with magmatic opx from boninitic ultramafic cumulates using partition coefficients of Bédard (2007) reproduces the typical U-shaped pattern observed in boninitic melts (Fig. 12b). Compared to USB boninitic amphibolites, the calculated melt has a pronounced REE depletion. Once again, the most primitive USB boninitic melt has probably crystallized a certain amount of olivine and/or opx prior to extraction, which would explain its REE enrichment compared to the calculated REE composition. The same modeling for the transition metals results in similar compositions between the liquid in equilibrium with opx and the boninitic lavas (Electronic Supplementary The mineral compositions are compiled from modern arc cumulates (Bouilhol et al. 2015;Greene et al. 2006;Jagoutz et al. 2007;Otamendi et al. 2010;Walker et al. 2015) and crystallization experiments (Feig et al. 2006;Krawczynski et al. 2012;Villiger et al. 2007). Legend as in previous figures. Abbreviations as in the text Material ESM 2). We, therefore, interpret the boninitic ultramafic rocks as cumulates of the boninitic basalts.

Primitive melts and liquid lines of descent
As previously discussed, most amphibolites from the USB have characteristics of liquids, and some of them have chemical attributes of primitive liquids that can be primary melts in equilibrium with mantle olivine with an Mg# of 87-91 mol.% and a Ni content of 1800-4500 ppm (Electronic Supplementary Material ESM 2). However, the REE modeling showed that these melts cannot be considered primary magmas but rather as parental melts. This is confirmed by the calculated melt Mg# from cpx compositions, following the method described by Wood and Blundy (1997), being higher (> 75 mol.%) than Mg# from the amphibolites. The most primitive tholeiitic melt (IN14009) has SiO 2 = 46.5 wt.%, MgO = 9.4 wt.%, Mg# = 60 mol.%, Ni = 394 ppm, and Cr = 356 ppm (Fig. 13). The most primitive boninitic melt (IN12034) has SiO 2 = 50.4 wt.%, MgO = 12.4 wt.%, Mg# = 68 mol.%, Ni = 162 ppm, and Cr = 807 ppm (Fig. 13). These parental melts can be used to model the evolution of these two series.
Tholeiitic(-to-transitional) and boninitic liquids SiO 2 , Mg#, Al 2 O 3 /TiO 2 , and (Dy/Yb) N variations are modeled following the method of Jagoutz (2010) using a rearrangement of the mass balance equation. The model assumes fixed cumulate compositions subtraction, although compositions may vary with fractionation, rather than mineral-liquid partition coefficients to allow major element modeling. The proportions of removed phases are adjusted to match the major and trace-element data (Fig. 13).

Tholeiitic(-to-transitional) suite
Sample IN16098b represents the only cpx-bearing ultramafic rock in our sample collection. Nevertheless, and as discussed in previous sections, the primitive parental melt probably crystallized olivine as the liquidus phase, followed by cpx leading to the formation of dunite, wehrlite and (olivine-)clinopyroxenite. This observation contrasts with the ISB cumulates sequence where no cpx has been observed nor inferred from the bulk-rock compositions (Szilas et al. 2015). Olivine and cpx fractionation induced enrichment of SiO 2 , Al 2 O 3 , TiO 2 , and REE coupled with a rapid Mg# decrease and slow Al 2 O 3 /TiO 2 and (Dy/Yb) N decrease in the residual melt (Fig. 13). A similar evolution is observed within the crystallizing cpx. The inferred crystallization of limited amounts of plag and amph then leads to the formation of plag-and hbl-bearing pyroxenite and hbl-gabbro (Mg# ≈ 65 mol.%, SiO 2 ≈ 46 wt.%) and drives the melt towards higher SiO 2 values, and to lower Mg#, Al 2 O 3 /TiO 2 and (Dy/Yb) N values (Fig. 13). The relatively high SiO 2 /MgO and Al 2 O 3 contents coupled with the positive Eu anomaly of several tholeiitic cumulates corroborate the appearance of plag in the cumulate assemblage of the tholeiitic suite. The rather delayed onset of plag crystallization allows the melt to evolve towards higher SiO 2 values early during its differentiation. Finally, the most evolved andesitic compositions of the tholeiitic suite can be reproduced from a primitive melt undergoing > 60 wt.% crystallization of olivine + cpx + plag ± hbl. Such a cumulate line of descent is symptomatic of damp tholeiitic systems and is consistent with the mineral assemblage  (Wood and Blundy 1997) are calculated for pressures of 0.4-1.0 GPa and temperatures of 1150-1200 °C and are also represented. b Orthopyroxene compositions from boninitic ultramafic rocks, and their calculated parental melts using the partition coefficients of Bédard (2007). The top right panel of b shows the opx compositions used for calculation and the calculated partition coefficients. The blue, green, and brown arrays represent the USB tholeiitic, transitional, and boninitic basalts compositions. Normalization values are from Sun and McDonough (1989). Abbreviations as in the text derived from Feig et al. (2006) experiments performed at 0.5 GPa using a water-undersaturated tholeiitic magma.

Boninite-like suite
Boninitic cumulates are well represented in the collected samples, but no olivine has been observed in these samples. Nonetheless, as previously discussed, olivine probably appears as the liquidus phase. The primitive parental liquid thus probably formed dunite, websterite and orthopyroxenite whose fractionation led to a rapid Mg# (< 60 mol.%) and Al 2 O 3 /TiO 2 decrease at constant SiO 2 content (SiO 2 ≈ 50 wt.%) in the residual liquid (Fig. 13). In addition, olivine and opx fractionation enhanced the REE, and especially the LREE, concentration in the melt resulting in a (Dy/Yb) N increase (Fig. 13c). With increasing differentiation, opx become Al-and Ti-rich, reflecting the progressive enrichment of these elements in the melt. Subsequent plag and amph fractionation is documented by the gabbroic composition of the most evolved boninitic cumulates and leads to (hbl-)norite crystallization after ~ 30% fractionation, which is characteristic of silica-rich primitive melts (e.g., Grove et al. 2002). Although the exact onset of plag and hbl crystallization is uncertain, their role in the final differentiation of the boninitic melts remains limited, with a slight decrease in Al 2 O 3 /TiO 2 and (Dy/Yb) N . This observation agrees with the low bulk-rock SiO 2 /MgO ratios of the boninitic cumulates (Fig. 11). Interestingly, the range of boninitic melt compositions is probably controlled by the composition of the primitive liquids. Indeed, some samples display SiO 2 contents ranging from 48 to 52 wt.% at a constant Mg# of ~ 60 mol.%, which, in a differentiation sequence dominated by olivine + opx, can only be explained by primary melt compositions with a large spectrum of SiO 2 contents. Finally, boninitic melt compositions are best explained by 40-50 wt.% fractionation of subsequent olivine + opx ± plag ± amph. The experiments of Van der Laan (1989) reproduce the olivine + opx liquidus assemblage in a boninitic system at 0.3-0.4 GPa, 1200-1260 °C, and initial H 2 O contents of 1.0-3.1 wt.%.

Age and origin of the inherited Hadean component of the Innuksuac complex
We demonstrated that magmatic rocks of the USB formed following two liquid lines of descent, one from a damp basaltic primitive melt and the other from a wet boninitic primitive melt that are ubiquitous and best produced during low-pressure mantle melting in subduction systems. As shown in Fig. 14, boninitic lavas are chemically more depleted than tholeiitic lavas and carry an isotopically  (Fig. 14d). Previous studies also showed that 142 Nd/ 144 Nd and Sm/Nd ratios measured in the diverse amphibolite groups of the NSB and USB are positively correlated, reflecting either a mixing or an isochronous relationship (O'Neil et al. 2008;Roth et al. 2013;Caro et al. 2017). However, our results show that the μ 142 Nd-Sm/Nd array is defined by at least two non-cogenetic series which, considered separately, have homogeneous μ 142 Nd regardless of their Sm/Nd ratio . Chemical sediments (BIFs) interleaved within the Ukaliq and Nuvvuagittuq mafic-ultramafic sequences also have homogeneous μ 142 Nd despite a wide range in Sm/Nd, which conflicts with a Hadean deposition age. These observations suggest that the μ 142 Nd-Sm/Nd array reflects contamination of Eoarchean mantle domains or mantle-derived melts by a Hadean crustal component. The differentiation age of this Hadean reservoir cannot be derived from the slope of the μ 142 Nd-Sm/Nd array but can be estimated at ~ 4.4 Ga using coupled 146,143 Nd chronometry . This result demonstrates that the presence of negative 142 Nd anomalies coupled with LREE enrichment in amphibolites of the Innuksuac complex reflects geochemical inheritance from a crustal reservoir differentiated during or shortly after crystallization of the terrestrial magma ocean (Morino et al. , 2018. The presence of unradiogenic 142,143 Nd signatures in boninitic lavas may thus result from direct assimilation of 4.4 Ga crust by tholeiitic melts or from melting of a metasomatized mantle source triggered by the recycling of Hadean lithosphere into the mantle. As shown in Fig. 14a, boninitic melts are more depleted in incompatible elements than tholeiitic melts, which rules out a simple crustal assimilation scenario. Their hydrous character, typified by high Th, LREE, and Th/ Yb (0.2-2.0) contents at constant Nb/Yb (1-2) provide further evidence in favor of fluid-induced melting of chemically depleted mantle domains (Fig. 14c). These combined petrological, geochemical, and isotopic observations suggest that tholeiitic melts were produced by decompression melting of near-primitive mantle domains. In contrast, melts displaying boninitic affinities were extracted from a more refractory mantle overprinted by a LREE-enriched fluid component carrying a crustal 142,143 Nd signature. As mentioned in the bulk-rock chemistry section, our collection includes a unique sample (IN12032) characterized by a marked LREE enrichment and a strongly negative μ 142 Nd. Given the low solubility of Nd in aqueous fluids (Brenan et al. 1995), these features are unlikely to result solely from fluid fluxing and point towards the addition of an enriched silicate melt component to the USB mantle source. In modern arc settings, the unradiogenic 143 Nd-and LREE-enriched compositions of calc-alkaline lavas result primarily from partial melting of subducted sediments in the mantle wedge (e.g., Class et al. 2000;Yogodzinski et al. 2010). This stems from the fact that subducted sediments have unradiogenic 143 Nd/ 144 Nd values that contrast with the radiogenic composition of the depleted mantle (e.g., Plank and Langmuir 1998). In contrast, the oceanic crust is isotopically similar to the depleted mantle and has a low propensity to melt during subduction. As shown by Class et al. (2000) in the context of the Aleutian arc, partial melting of recycled sediments produces correlated Th/Nd-ε 143 Nd variations in arc lavas, which are similar to the Th/La-μ 142 Nd array reported by Caro et al. (2017) in USB amphibolites. The recycling of Hadean sediments may, therefore, provide a plausible mechanism for transferring unradiogenic 142,143 Nd-and LREE-enriched signatures to the mantle source of Ukaliq and Nuvvuagittuq calc-alkaline lavas. This scenario would imply accumulation of erosion products of ~ 4.4 Ga protocontinents on the Hadean seafloor and the development of differentiated felsic crust shortly after magma ocean crystallization. Alternatively, higher lithospheric buoyancy may have favored long-term preservation of Earth's primordial lithosphere (e.g., Korenaga 2006), allowing the development of unradiogenic 142,143 Nd compositions even in a predominantly mafic crustal reservoir. The elevated temperature of the Eoarchean mantle would then provide the necessary conditions for partial melting of this recycled Hadean crust and the production of high Th/La-and LREE-enriched melts capable of generating the isotopic and trace element signature of calc-alkaline amphibolites of the Innuksuac complex.

The question of Eoarchean subduction
The petrological, geochemical, and isotopic characteristics summarized in Fig. 14 provide a unique view of a magmatic system associated with the recycling of the Hadean lithosphere in the mantle. The petrogenetic history inferred from these observations can, therefore, shed some light on Eoarchean geodynamics, which otherwise strongly rely on thermomechanical and analog modeling studies (e.g., Sizova et al. 2010;Fischer and Gerya 2016a, b;Van Hunen and Van den Berg 2008;Piccolo et al. 2019). Based on these models, two contrasting views are provided, nevertheless supported by the incomplete geological record of the Archean Earth (e.g., Bédard 2018; Van Kranendonk 2010). They are: (i) the active-lid model, in which crustal recycling occurs by subduction (e.g., Fischer and Gerya 2016b; Van Hunen and Moyen 2012); and (ii) the stagnant-lid model, where crustal recycling proceeds via dripping of eclogitized lower crust (e.g., Fischer and Gerya 2016a;Moore and Webb 2013;Piccolo et al. 2019). These processes are not mutually exclusive, and transitional regimes involving episodic or plumeinduced subduction in a predominantly stagnant-lid regime have also been suggested for the Archean Earth (e.g., O'Neill et al. 2007;Gerya et al. 2015). In the subduction model, slab-induced corner flow leads to H 2 O-assisted decompression melting, generating tholeiitic melts. Harzburgitic residue can further melt in a forearc position at low pressure in fluid-saturated conditions to yield boninitic melts (Grove et al. 2002;Schmidt and Jagoutz 2017). These two melting regimes can be juxtaposed in a forearc sequence and are thought to represent subduction initiation (see Stern et al. 2012), as was proposed for the nearby Nuvvuagittuq supracrustal belt (Turner et al. 2014). In the stagnant-lid model, the dripping of hydrothermally altered mafic crust induces the production of TTG-like melts and recycling of refractory eclogitic residues in the mantle (Piccolo et al. 2019). Asthenospheric uplift triggered by lower crustal dripping can, in turn, produce tholeiitic melts by decompression melting. Although recycled eclogitic residues are expected to be relatively dry, providing limited opportunity for further dehydration and fluid-induced melting of the mantle, hydrated material may be dragged along lower crustal drips (Piccolo et al. 2019) and potentially create petrogenetic environments similar to those observed in subduction settings. In contrast to subduction, however, dripping would only occur due to crustal overthickening and, therefore, does not provide a direct pathway to producing low-pressure boninitic melts. Lower crustal dripping would also favor hybridization of newly formed TTG and tholeiitic melts with preexisting (Hadean) crustal components (e.g., Fischer and Gerya 2016a;Piccolo et al. 2019), which conflicts with the lack of geochemical relationship reflecting direct assimilation of Hadean crust, and the absence of inherited Hadean zircons in the Innuksuac complex. Therefore, the most parsimonious way to explain the ubiquitous presence of Hadean geochemical crustal signatures in the absence of relict Hadean crustal components is to recycle this ancient lithosphere through subduction.
In contrast to modern arcs, the USB is dominated by tholeiitic lavas and includes only minor boninitic and calcalkaline components. This difference may reflect the inherently short-lived nature of subduction systems in a hotter mantle, as modeled by Van Hunen and Van den Berg (2008). Alternatively, Fischer and Gerya (2016b) showed that with increasing mantle potential temperatures, magmatic activity in the mantle wedge would become dominated by decompression melting of dry asthenospheric mantle rather than fluid-fluxed melting characterizing modern arcs. Therefore, the dominant tholeiitic and subordinate boninitic and calcalkaline compositions of the USB may reflect the prevailing magmatic expression of subduction in a hotter mantle rather than the embryonic stages of arc development. Going beyond these different scenarios, our results suggest that subduction systems were operative, at least intermittently, as early as 3.8 Ga and thus contributed to the genesis of Earth's oldest crust. Conflicting evidence for stagnant-lid or subduction-associated tectonic processes in the Archean can be reconciled by hybrid models involving either concomitant (e.g., Gerya et al. 2015) or alternating (O'Neill et al. 2007) plume and subduction-induced magmatism.

Conclusions
The Eoarchean (ca. 3.78-3.75 Ga) Ukaliq supracrustal belt is part of the Innuksuac complex within the ca. 12,000 km 2 Inukjuak domain (Québec, Canada). The Ukaliq supracrustals host mafic and ultramafic rocks which can be subdivided into five categories according to their phase relationships and bulk-rock chemistry: (i) tholeiitic basalts; (ii) boninitic basalts to andesites exhibiting TiO 2 depletion and U-shaped REE pattern; (iii) transitional basalts representing a continuum between the two categories above; (iv) cpx-bearing, tholeiitic(-to-transitional) cumulates; and (v) cpx-absent, boninitic cumulates. We show through bulk-rock and mineral analyses coupled with melt composition calculations that the ultramafic rocks represent cumulate products of the mafic lavas and used these data to model a damp tholeiitic liquid line of descent consisting of olivine + cpx + plag ± hbl fractionation, and a wet boninitic differentiation sequence that crystallized olivine + opx ± plag ± hbl. The liquid lines of descent inferred from both bulk-rock and mineral chemistry suggest that the Eoarchean Ukaliq supracrustal belt originated in an environment capable of reproducing today's subduction zone petrological processes. Such environment produces covariation between 142 Nd/ 144 Nd and Sm/Nd ratios in amphibolites of the Innuksuac complex through intake of a 142 Nd anomaly-bearing Hadean slab that imprinted the mantle wedge and its melt derivatives via fluids and/or melts carrying the 142,143 Nd-and HFSE-depleted and LREEenriched signatures. This suite of interpretations would not be possible without the identification of primary magmatic signatures preserved in the Eoarchean rocks of the Ukaliq locality.