Immiscible hydrous Fe–Ca–P melt and the origin of iron oxide-apatite ore deposits

The origin of iron oxide-apatite deposits is controversial. Silicate liquid immiscibility and separation of an iron-rich melt has been invoked, but Fe–Ca–P-rich and Si-poor melts similar in composition to the ore have never been observed in natural or synthetic magmatic systems. Here we report experiments on intermediate magmas that develop liquid immiscibility at 100 MPa, 1000–1040 °C, and oxygen fugacity conditions (fO2) of ∆FMQ = 0.5–3.3 (FMQ = fayalite-magnetite-quartz equilibrium). Some of the immiscible melts are highly enriched in iron and phosphorous ± calcium, and strongly depleted in silicon (<5 wt.% SiO2). These Si-poor melts are in equilibrium with a rhyolitic conjugate and are produced under oxidized conditions (~FMQ + 3.3), high water activity (aH2O ≥ 0.7), and in fluorine-bearing systems (1 wt.%). Our results show that increasing aH2O and fO2 enlarges the two-liquid field thus allowing the Fe–Ca–P melt to separate easily from host silicic magma and produce iron oxide-apatite ores.


Review of NCOMMS-17-28343-T
This is a very nice piece of work, which answers a long standing question in Earth Sciences and basically provides a recipe for the genesis of iron oxide -apatite ore deposits. The data is convincing overall, and the interpretations seem to be fully supported by the data. The presentation is of high quality throughout, and the manuscript reads very well. I strongly support publication in Nature Communications pending on some minor revisions based on my comments below: Line 141-143: Sulfur is siderophile only in 2-oxidation state, which is not the case here. For oxidized sulfur, I would rather tie the difference to the degree of melt polymerization (e.g. see Zajacz 2015, GCA).
Line 174-178: This section seems very simplified and somewhat speculative. For example, 2.6 wt% F cannot serve as a ligand for about 20 wt% Ca.
Line 184-185: Does this statement also hold for FeO concentration? Figure 4: On this diagram, the immiscibility field closes at DSi=0.87 instead of DSi=1. This must be a mistake, DSi should equal 1 at the critical point by definition.

Methods:
Line 31-32: It needs to be explained how the intrinsic fO2 of IHPV was defined at FMQ+3.3 at water saturation. If fH2=0 in pure argon pressure medium, this should yield much high fO 2 at aH2O=1 (i.e. as constrained by the equilibrium constant of the water dissociation reaction at the run P, T). No satisfactory explanation is given in the Supplementary Material either, as opposed to the statement in line 36-37. I think it is very important to substantiate this, because fO2 is a critically important variable in this study.
Line 70-73: The quantification of the water concentration from the Raman spectra needs more detailed explanation. Most commonly, the water band area is normalized by the area of one of the major silicate bands for this purpose, but this method generally requires matrix-matched standardization. Therefore, much more specifics are needed on the standards used as well. For example, I don't see how one could use the same standard for the immiscible melts in this study with contrasting compositions. What is the uncertainty on the water concentration determination? Oxygen fugacity calculations section: As pointed out above, the determination of intrinsic fO2 needs more explanation. Also, further down, would 0.3-1.0 wt% dissolved water really correspond to aH2O=0.01?
The language of the last paragraph could use some improvement.

17/12/2017
Thank you very much for your comments and kind help. All comments from the two reviewers are very constructive and helpful. In the following paragraphs, we answer the questions and address point by point the issues pointed out by reviewers.
Before presenting our replies to reviewers, we first want to emphasis that our study is an experimental approach that aims at understanding phase equilibria in magmatic systems related to Iron-Oxide-Apatite (IOA) ores with broad implications for these deposits globally. In other words, our study does not focus on any specific case, and for example on El Laco to which Reviewer #1 continuously refers to. We evaluate the formation of IOA ores on a large-scale perspective rather than on the scale of individual rocks or deposits. The ore-forming processes we propose should not be evaluated from a regional perspective, but from phase equilibria and crystallization conditions as evidenced by our new experiments.

Replies to Reviewer #1,
The manuscript by Hou et al is an exciting contribution to the origin of magnetiteapatite rocks, experimentally confirming that they can form from the crystallization of Fe-Ca-P-rich melts -the major goal of the study is that the authors have been able to expand the known immiscibility field of Fe-Ca-P melts by increasing the aH 2 O and the redox state. Thus, the study adds significant information on how the magnetiteapatite deposits are formed, supporting that they can be of magmatic origin. The experiments also confirm many features that have been recently observed by different authors such as the existence of immiscible Fe-P melts at El Laco (Naslund et al 2002, 2003, Mungall et al., 2009, 2014, the existence of melt inclusions with similar compositions to those described here (Clark and Kontak, 2004;Naslund et al., 2009;Kamenetsky et al 2013, the local role of fluorine, the high redox state and water content of those melts and the formation of these systems by crustal contamination of primitive melts (Tornos et al., , 2017 First of all, we have added all the references mentioned by Reviewer 1 into the text in order to give credits to our colleagues from the IOA community. Please see the revised version of the manuscript.
Secondly, although sulfide is always late in this type of mineralization and locally replacing magnetite, it is likely that sulfur got involved initially in the primordial mineralizing-process. Thus, in order to simulate the role of sulfur during magmatic processes, we deliberately added minor sulfur concentrations in three runs.
Since all of our experiments are free of Ni and Cu, there is no possible confusion between IOA-IOCG and Ni-Cu deposits, these last ones being not mentioned at all in our manuscript. It should however be noted that IOCG deposits have considerable amounts of sulfide and IOA deposits are commonly regarded as an end-member of IOCG series (Hitzmann et al., 1992).
2. In fact, in the abstract they include issues that are not discussed in their manuscript such as the validity of the models proposing an hydrothermal genesis.
In this study we are aiming to elucidate the primordial process for the formation of IOA deposits. In order to do so, we conducted experiments simulating magmatic processes. We however find reasonable to include the debate on magmatic vs. hydrothermal genesis in the introduction to put the study into a broader context and to make the reader aware that other processes than magma crystallization have been proposed to explain the genesis of IOA deposits. Our experiments, conducted to constrain the phase equilibria at magmatic temperature, constrain the primordial role of magmas for the enrichment in iron and phosphorous. However, we totally agree that hydrothermal process played an important role in the lower-temperature redistribution of elements but we agree that these late-stage processes did not lead to the formation of deposits (e.g., Knipping et al., 2015;Tornos et al., 2016Tornos et al., , 2017.

Also, it is not true that "Fe-Ca-P-rich and Si-poor melts have never been
observed in natural or synthetic magmatic systems". There is abundant literature on the existence of such Fe-rich melts that are recorded in melt inclusions or in the magnetite-apatite deposits (references above) and Kamenetsky et al (2013) have observed an immiscibility gap identical only slightly smaller than that shown in the figure 4 (DSi values of 02 instead 0.0 as in this study).
Yes, we know the reference and this study is thoroughly discussed in our paper (actually, B. Charlier is co-author in both studies). It is true that Kamenetsky et al. (2013) have observed an immiscibility an important gap but it is not as large as the one reported in our paper. Moreover, the melt inclusions reported in Kamenetsky et al. (2013) are hosted in native iron, which implies that they equilibrated at very reducing conditions close to the iron-wustite (IW) solid buffer. Such reducing environment is not consistent with the formation of IOA deposits which usually record relatively oxidizing conditions above QFM buffer (e.g., Lester et al. 2013;Tornos et al., 2016Tornos et al., , 2017. This is extremely important for iron speciation and the nature of liquidus phases crystallizing from the magmas. This difference regarding oxygen fugacity is an important part of the motivation to conduct new experiments, since the contrasted melt compositions observed by Kamenetsky et al. (2013) are not suitable to interpret the genesis of IOA deposits.

Also, the authors do not present any data that show that assimilation of sediments (by silicate melts?) can produce such Fe-Ca-P-rich magmas. They just highlight an already published statement without adding new data.
First of all, we have added the reference for that statement into the text of our original submission (Tornos et al. 2017). One should keep in mind that this paper is an experimental study aiming to test the model of 'oxide melt' for IOA ores, which had been previously proposed in the literature (e.g., Philpotts, 1967;Frietsch 1978;Nyström and Henriquez 1994;Naslund et al. 2002). We evaluate the formation of IOA ore on a large-scale perspective rather than on the scale of individual rocks or deposits. Hence it makes perfect sense to give global implications for IOA deposits and to test existing hypotheses, without presenting new geochemical analysis of natural samples for individual localities.
Specifically, based on experiments under relevant conditions, one of our major new findings is that elevation of oxygen fugacity could produce Fe-Ca-P enriched melts by immiscibility. Thus any processes occurring in natural systems that could oxidize the ore-forming magmas has the potential to form such melts, including contamination of crustal rocks, e.g. carbonates or evaporates. The O-Sr-Nd isotopes of both igneous rocks and IOA ores are supportive for this inference as discussed in our manuscript . Lester et al (2012Lester et al ( , 2013) that seem to reach similar conclusions.

Also, the authors do not discuss the results of
The results of Lester et al. (2013) have now been taken into consideration and discussed at the end of Introduction section: "For example, in the simple system of SiO 2 -FeO-Al 2 O 3 -K 2 O with presence of volatile, Lester et al. (2013) produced Ferich melts with SiO 2 content of >20 wt% by immiscibility, which is still not a silicafree oxide melt." It is worth noting that this clearly means their conclusions are significantly different to ours with respect to the formation of IOA deposits.

Line 33. Discordant? Some of the major magnetite-apatite deposits are stratabound such as Kiruna or El Laco
We have deleted "discordant" from that sentence.

Line 41. Unconsistent with what said above
Here we present the current debate for the formation of IOA deposits, i.e. hydrothermal vs. magmatic models. Accordingly, we first described the hydrothermal model, and then followed by those for the magmatic model. Thus it is logically correct and the two parts do not mutually agree with each other as expected.
There are papers that discuss this problem in much more detail (e.g., Velasco et al., 2016) We have replaced these two references by Velasco et al. (2016).
10. Line 48. The inclusions described by these authors are in ortho-and clinopyroxene but also (and dominantly) in plagioclase.
We changed this sentence into "The development of immiscibility is supported by the coexistence of two types of melts in glassy matrices and inclusions hosted by phenocrysts in the ore and in andesitic wall rocks." Yes, we combined some of the representative compositions of these immiscible melts coexist as illustrated in Figs. 2 and 4. However, as we stated clearly in the text, the Fe-rich melt inclusions still contain considerable amount of silica and they are not compositionally similar to the high grade, i.e., almost SiO 2 -free, IOA ores. This was our motivation to conduct this study. In contrast to the melts observed in the inclusions, the Fe-rich immiscible melts that we produced experimentally in this study are almost SiO 2 -free which and are capable of crystallizing the high-grade IOA ores.
To the best of our knowledge, similar melts had not been observed either in natural melt inclusions or in experiments. Even in Lester et al. (2013) such a melts were not reported although the authors of this study used a simple system which is more favorable to extreme, but not geologically relevant compositions.
In summary, we present the first results that prove experimentally that almost SiO 2 -free melts can be produced by immiscibility under geological environment relevant for IOA deposits in a multi-component, geologically-realistic system. As pointed out by Reviewer #1, we are "the first ones obtaining such extreme end-members", which endorses the novelty of our study.

Line 61 and onwards. The authors do not even quote the presence of actinolite/diopside, a common component of these rocks.
We now clearly quote the presence of actinolite/diopside in the revised version, as stated below, "Extreme enrichment of apatite and iron oxide over silicate minerals, as observed in IOA deposits, cannot simply result from differential crystal settling in an iron-rich silicate melt. This is because, with the exception of plagioclase, common silicate minerals (actinolite and diopside) are denser than the melt and would sink along with the oxides" 13. Line 78. Pyrite is not so common in IOA, which are characterized by the almost lack of sulphides. Pyrite is always late in this mineralization, postdating and replacing magnetite.
Yes, this is true. However, as mentioned by Reviewer 1, some IOA deposits contain sulfides. In order to thoroughly investigate and understand the general mechanisms by which IOA deposits formed, it is necessary to take into account that they contain (even only a little bit) sulfur. This is why we added sulfur in the experiments simulating magmatic stage. Although sulfides appear as being late phases in the crystallization sequence, there is still a possibility that these sulfur played a role during the primordial process and were re-worked during post-magmatic hydrothermal processes. We however strongly emphasize that the presence of sulfur in our experiments has no implications for the compositional range of immiscible melts that we produced. Adding sulfur was therefore important to understand how sulfides form in IOCG deposits but does not have implication for the development of immiscibility and the formation of apatite and oxide ore. Mountains, Missouri (Kisvarsanyi and Kisvarsanyi, 1990). Moreover, fayalite is a common mineral phase in IOCG deposits (Porter et al., 2010), for example, Carajas district in Brazil (Xavier et al., 2008).
15. Line 84 and 104. Ilmeno-magnetite is another of your products while, as the authors quote above, magnetite-apatite deposits are characterized by the lack of Ti-rich magnetite.
As discussed by Velasco et al. (2016), low-Ti magnetite in IOA deposits does not necessarily mean hydrothermal in origin, because considerable reequilibration between the magnetite and the parental melt may occur during the onset or during the oxy-exsolution at shallow depths prior to magma ascent. This seems to be the case for the magnetite in IOA deposits which usually shows significant variations in the highly compatible transition elements such as Ti, V, Cr and Ni (e.g., Knipping et al., 2015). It is therefore reasonable to suggest that all or part of the low-Ti magnetite crystals observed in IOA/IOCG deposits initially had a much higher Ti content when the formed at the magmatic stage. Our experiments of course only record the magmatic stage and not the late-stage processes during which Ti may be lost from the magnetite lattice. In order to simulate the potentially high-Ti feature during the magmatic stage, we may have added excessive Ti in the starting mixtures.
However, there's nothing necessarily wrong with such a methodology. For example, in the Se-Chahun IOA deposit, Bafq district, Iran (Bonyadi et al., 2011), magnetite crystals are always featured with Ti-rich core (~2.5 wt.%) and ilmenite exsolution is quite common in magnetite, which is notable, as being of an indicator for high-Ti bulk composition of primary oxide minerals.

Line 99. This statement is really interesting. Droplets of an immiscible sulphide-rich liquid… and magnetite apatite deposits lack of sulphides coeval with magnetite. In line 131 you quote it as close to stoichiometric FeS (pyrrhotite?). In line 78 you quote pyrite as the stable phase.
Please see reply #13 for the question "magnetite apatite deposits lack of sulphides coeval with magnetite." Additionally, we had already replaced 'pyrite' in line 78 with pyrrhotite as it is close to stoichiometric FeS.

Ca-Ti melts?
First of all, these melts are low in Ti, i.e. as low as < 1wt.% TiO 2 (please see row 5 in Table 1). This is consistent with the relatively low Ti signature of the IOA magnetites. In 'Fe-Ca-P' melts we defined, the MgO content (up to 7.5 wt.%) is the highest among other elements except FeO tot , CaO and P 2 O 5 , which is still two times below the lowest content of FeO tot (18.88 wt.%). Hence, FeO tot , combined with CaO (~27wt.%) and P 2 O 5 (~26wt.%), are the major constitutes in the 'Fe-Ca-P' melts.
Moreover, except MgO, these melts also contain ~5 wt.% Al 2 O 3 , which is not really minor components. Nevertheless, considering the major mineral assemblage in the iron oxide apatite ores, i.e. magnetite + apatite, we believe the name (Fe-Ca-P) better reflect the characteristics of such an immiscible Fe-rich melt. Please also see reply #25 for the more discussion for the names of the immiscible Fe-rich melts.

Line 150. This is not true. Your experiments quote abundant MgO and TiO2
also.
Thank you very much for the suggestion, we changed the sentence into "With cooling, immiscible melt pairs become increasingly contrasted in composition, but dry Fe-rich melts reported so far have never approached the extreme, Si-free, composition of IOA ores (predominantly iron, calcium, phosphorous, with minor magnesium and titanium)". Minor magnesium and titanium may be incorporated into magnetite and exsolution of ilmenite in some cases. Moreover, except magnetite and apatite, our Fe-Ca-P melts may inevitably crystallize some phases containing all the other elements (e.g., MgO, MnO, Al 2 O 3 and SiO 2 etc.), for example, diopside, actinolite, amphibole, epidote, titanite and allanite. These phases are commonly seen in IOA deposits but may be interpreted as secondary phases formed by extensive alteration.
19. Line 173. These are the conclusions of the previous papers of Tornos et al (2017).
We now cite Tornos et al. (2016 and2017) in the revised version.
Accordingly, we changed the sentence into "Thus, increasingly hydrous and oxidizing conditions can explain the formation of Fe-P-enriched silicate liquids (Fig.   4), which is consistent with the recent studies on El Laco deposit  2017)". 1) Si-rich rocks, including rhyolite and dacite are widely recognized in IOA deposits, even in the districts like Andes. For example, dacitic rocks are well exposed in Marcona in Peru (Chen et al., 2010), and a recent study on El Laco had also revealed that a high-K subalkaline rhyodacitic melt is present in the groundmass of the andesite and in most melt inclusions (Tornos et al., 2017).

Line 185. Do you mean that all the dacite-rhyolite found near magnetite
2) During magmatic evolution, once the liquid line of descent (LLD) crossed the two-liquid field, immiscibility that split the homogenous melt into Fe-rich and Si-rich conjugates will develop. The compositions of immiscible pairs define a locus between which immiscibility develops. Any composition that plots on the mixing trend between the equilibrium immiscible pairs would unmix, the proportion of the two liquids being defined by the lever rule. If the liquid line of descent left the two liquid field by simple fractional crystallization and compositional evolution of the bulk liquid, the most Si-rich, i.e., evolved liquids can be produced. Thus, the Si-rich magmas could be only immiscible Si-rich melts or a combination of immiscible Sirich melts and most evolved magmas after LLD out of the solvus. Hence, relative volume of evolved melt and ore material could be varied significantly, i.e., large volumes of dacite/rhyolite in some places and very small volumes in other places with a single process.

Line 186. Here there is a major confusion with tectonic environments and magma geochemistry. IOA deposits are only located in convergent margins in
the Andes and in the lower Yangtze. In other settings they are in extensional intraplate settings (Hitzman et al., 1992;Williams et al., 2005). The genesis of hydrous melts is not only related to slab subduction but also melting of crustal rocks.
Thank you very much for pointing this out, we added "melting of crustal rocks" into the statement, "First, IOA deposits are commonly located in convergent settings where slab dehydration leads to the formation of water-bearing magmas 44 .
There are also commonly observed in extensional intraplate setting where crustal rocks melting (Hitzman et al., 1992;Williams et al., 2005)* 45-46 may also produce hydrous magmas 10 ." *Note: the format of citation will be kept consistent as superscript numbers in the text.

Line 192. Which kind of isotopic studies? The work of Barton et al is on IOCG deposits and not magnetite-apatite and proposes the involvement of basinal
brines, not melts, in the genesis of these deposits. They just provide isotopic evidence that the sulphur derives from the evaporites but that does not mean that the rocks are crustal. This needs a much more rigorous approach.
Recent papers published by Tornos and his group had already provided a rigorous approach in a support for our suggestion which is mainly based on experimental results. For example, in the case study of El Laco, Tornos et al. (2016 and2017) presented isotope data from the host andesite ( 87 Sr/ 86 Sr: 0.7066-0.7074; εNd: -5.5 to -4.1; δ 18 O whole rock : 7.2-9.6‰; δ 18 O magnetite : 5.1-6.2‰) and an underlying andesite porphyry ( 87 Sr/ 86 Sr: 0.7075-0.7082; εNd: -5.9 to -4.6). These value reflect the interaction of a primitive mantle melt with Andean crustal rocks. It worthy noting that a rock with evaporate can hardly be anything else than crustal.
Since we had already cited Tornos et al. (2017) in this sentence, we thus just changed it into "Indeed, systematic O-Sr-Nd isotopic studies of IOA ores suggest a significant crustal component 48 ". Tornos et al (2017a and b).

Line 194 and onward. This is the model by the recently published papers of
We checked the publications from Tornos and his group. We could only find one paper published on Economic Geology in 2017. Reviewer 1 is presumably referring to the paper published in Geology in 2016 . Since we had cited Tornos et al. (2017) here in the last version, and in this revised version, we simply added the another reference to Tornos et al. (2016).

Line 199. Panzhihua is not a magnetite-apatite deposit. It is a Ni-Cu deposit
with magnetite intergrown with sulphides. As in many other Ni-Cu deposits, crustal contamination is the key for forming an immiscible sulphide-rich melt.
We deleted the reference to the Panzhihua deposit and cited Tornos et al. (2017), in which presence of anhydrite in El Laco had been reported.  Mungall et al (2009Mungall et al ( , 2014. See also the description of such melts by Naslund et al 2002Naslund et al , 2003. We cite Mungall et al. (2018) in which his systematic experiments are presented and the two papers of Naslund in the revised version. Naslund et al. (2002)   34. As pointed out above, the determination of intrinsic fO 2 needs more explanation. Also, further down, would 0.3-1.0 wt% dissolved water really correspond to aH 2 O=0.01?
We re-calculated the aH 2 O based on the model of Burnham (1994). aH 2 O of 0.1 corresponds to 0.3-1.0 wt.% dissolved water. Please see the new section on oxygen fugacity in the Supplementary material for additional details.
35. The language of the last paragraph could use some improvement.
The manuscript has been checked and improved by an English native speaker.