One of many powerful applications of oxygen isotope geochemistry is the detection of otherwise cryptic water-rock interaction. Taylor and Sheppard (1986, p. 238) put it succinctly: “All relatively 18O-rich or 18O-depleted silicate melts on Earth…must have in part been derived from, or have exchanged with, a precursor material that once upon a time resided on or near the Earth’s surface.” Oxygen isotope signatures of such supracrustal reservoirs are definitive because Earth’s mantle is a generally homogeneous (and large) oxygen reservoir with an overall isotope ratio that is expressed (in delta notation) as δ18O = 5.1 ± 0.2‰ (Vienna standard mean ocean water [VSMOW]), or 0.51% enriched in the rare 18O isotope relative to the ocean water standard, based on a survey of mid-oceanic ridge basalts (Eiler et al., 2000). Likewise, analyses of lunar samples and terrestrial peridotites consistently yield the same, primitive ratios (e.g., Epstein and Taylor, 1971; Mattey et al., 1994). Interaction with water at or near the surface of Earth, whether by hydrothermal alteration, weathering, or biological action, shifts oxygen isotope ratios by (most commonly) enriching 18O (e.g., Eiler, 2001). High-temperature hydrothermal alteration, or weathering in extreme icehouse environments can also cause δ18O to drop below 5‰ (e.g., Rumble et al., 2002).

Oxygen isotopes are partitioned between minerals by the thermodynamics of the system at their time of crystallization, and, depending on the diffusivity of oxygen within a given crystal lattice, can be reset by interaction with fluids. The power of oxygen isotope ratios to trace contrast between supracrustal rocks and mantle materials is improved by choosing a mineral that is unlikely to exchange oxygen with other reservoirs after crystallization. Because of its extremely robust nature, slow diffusion characteristics, and amenability to isotopic dating, the mineral zircon has extended the utility of the oxygen isotope system by preserving its original isotope signatures through subsequent events (Valley, 2003). Elevated δ18O (∼6–7.5‰) in the most ancient zircon samples of Earth require the presence of liquid water at the surface of a “Cool Early Earth” to weather materials that were subsequently melted and crystallized zircons by ca. 4.3 Ga (e.g., Peck et al., 2001; Wilde et al., 2001; Cavosie et al., 2005).

The oxygen isotope system’s sensitivity for non-mantle sources has been applied as a filter in Hf isotope studies of zircon. By ensuring that only zircons with a “mantle” signature are included in studies that estimate mass transfer from mantle to crust and those that document reworking of existing continental material are excluded, workers have placed important new constraints on the growth of continental crust over time (Hawkesworth and Kemp, 2006; Kemp et al., 2007; Dhuime et al., 2012). However, high-precision examination of primitive mantle materials reveals that they can be heterogeneous beyond analytical uncertainty, suggesting the presence of minor supracrustal oxygen in certain tectonic settings. The study of the oxygen isotope record of mantle heterogeneity is one that requires a careful analytical approach. Subtle (<1‰) variability in δ18O values has been used to estimate only minor overall contributions (typically 2%, but as much as 8% in the most enriched samples) of supracrustal materials in mantle reservoirs (Eiler et al., 1995; Eiler, 2001; Workman et al., 2008). Zircons found as xenocrysts in kimberlite are also broadly homogeneous, but limited variability from these deep samples has likewise been interpreted as representing a minor subducted supracrustal component (Valley et al., 1998; Page et al., 2007). Supracrustal oxygen is clearly present in the mantle, but it leaves only hints of its presence.

Spencer et al. (2017, p. 979 in this issue of Geology) report evidence of supracrustal oxygen in the mantle that is anything but subtle. Zircon crystals with δ18O of 14–28‰ formed in a peraluminous granite that intruded into the mantle section of the Oman-UAE (United Arab Emirates) ophiolite prior to its obduction. These are the most 18O-enriched values yet reported for zircon, and are drastically different than the “mantle” values previously described in oceanic igneous zircon (Cavosie et al., 2009; Grimes et al., 2011). Although the possibility of a sediment source for these granites had been suspected based on major and trace-element systematics (Rollinson, 2015), the extreme isotope signature preserved in these zircons requires the wholesale melting of a pelagic sediment, as there is no other reservoir that can both form a peraluminous granite and impart this extreme isotope signature. It is all the more remarkable that these ultra-high δ18O melts crystallized within the mantle, apparently untainted by its oxygen isotope signature (although less robust minerals than zircon were partially reset by later fluids). The simple U-Pb history documenting pre-obduction intrusion of these melts, coupled with their truly jaw-dropping isotopic composition, effectively precludes the alternative hypothesis of post-emplacement contamination by country rock (e.g., Belousova et al., 2015), and points, unambiguously, to an origin of pelitic sediments, melted mud, in the mantle.

Although Spencer et al. document the first evidence of unadulterated sediment melting in the mantle, extreme oxygen isotope ratios have previously been used to demonstrate the deep subduction of near-surface materials. High δ18O coesite (7–16‰) and garnet (6–8‰) inclusions in eclogite-suite diamonds confirm the suspected supracrustal origin of the generally elevated δ18O values found in mantle eclogite (Schulze et al., 2003, 2013). It has become apparent that supracrustal oxygen, shielded by a diamond carapace or the robust structure of zircon, can persist as extreme isotope heterogeneities within an otherwise broadly homogeneous mantle. These observations bring to mind other evidence of tiny volumes of crustal material (including zircon) found within podiform chromitites that have been deeply subducted and preserved for long periods of time in the mantle, largely without interaction or equilibration (Liou et al., 2014), including within the Oman-UAE ophiolite (Robinson et al., 2015). We know from these and other studies that “bread crumbs” of supracrustal material enter the mantle through subduction, and are sometimes preserved, but it remains to be seen what role they play in the long-term evolution of mantle reservoirs and crustal evolution.

It is difficult to reconcile the presence of these extreme heterogeneities and billions of years of sediment subduction with the broadly homogeneous mantle that shares a presumably primordial δ18O value with the moon. Evidence of elevated δ18O (8–11‰) in suprasubduction-zone mantle samples has been documented in melts (Eiler et al., 1998), and, more recently, in peridotite (Liu et al., 2014), but is rare, and is not as extreme as that found in the Oman granites. The Oman zircons (and other high-δ18O “breadcrumbs”) may be the exception, with preservation a result of the robustness of zircon and, perhaps, an unusual tectonic setting. Most extremely high-δ18O material is likely tempered by fluid processes as it enters the mantle.

One such process could be the hybridization of high-δ18O sediments and altered basalt with serpentinite in the fluid-rich subduction mélange environment (e.g., Bebout and Penniston-Dorland, 2016). Kilometer-scale homogenization of δ18O by fluids can take place in the subduction channel (Bebout and Barton, 1989), and could temper extreme δ18O not preserved in zircon to more reasonable values. Recent models based on trace-element data that implicate that these hybrid lithologies in subduction zone magmatism are entirely consistent with a homogenization of the various oxygen inputs to the mantle (Marschall and Schumacher, 2012; Nielsen and Marschall, 2017). As ever, more work is needed to determine the role of unusually high δ18O heterogeneities in the mantle, but it is becoming clear that, just as on the carpet, mud can also leave tracks in mantle.

ACKNOWLEDGMENTS

I thank William Peck, Andrea Goltz, and editor Brendan Murphy for constructive comments on an earlier draft that greatly improved the present article. All remaining errors and oversights are my own. Partial support was provided by National Science Foundation grant EAR-1249778.