We focus on sediment cores collected from three sites in the eastern North Pacific (2012–2013, R/V Oceanus) and three sites in the western Tasman Sea (2016, R/V Investigator) at water depths of approximately 200, 1500, and 3000 m (Figure 1 and Table 1). For simplicity, we will refer to the 3 North Pacific sites as NP200, NP1500, and NP3000 and the Tasman Sea sites as TS200, TS1500, and TS3000, respectively. Exact water depths, locations, and site names from previous literature are available in Table 1. Detailed descriptions of core and pore water collection are available elsewhere (Abbott et al., 2015b; Abbott, 2019). Briefly, a multi-corer was deployed and cores with a visibly intact sediment-water interface were sectioned on board the ship into HCl cleaned centrifuge tubes in a glove bag filled with an inert (N₂) atmosphere. After centrifugation, pore water was pulled off using an HCl cleaned syringe, filtered with a 0.45 μm syringe filter, and acidified to pH ≤ 2 with distilled HCl. Sediments were frozen and later freeze dried prior to analyses.

Rare earth element analyses for the North Pacific pore waters have been previously reported (Abbott et al., 2015b, 2016a). Tasman Sea pore waters were analyzed using a SeaFAST II inline with a Thermo X-series II Inductively Coupled Plasma-Mass Spectrometer at the Keck Collaboratory for Plasma Mass Spectrometry (as described by Yang and Haley, 2016; Abbott, 2019). To compare the REE patterns of pore waters to potential host phases, we considered all pore water samples collected from the six sites (three North Pacific and three Tasman Sea) and other published marine pore water values (Haley et al., 2004; Abbott et al., 2015b). We did not include pore waters collected using Rhizons© because of the REE fractionation associated with this technique (Abbott et al., 2015b). We also excluded pore water data with less than ten of the REEs reported because these data did not allow for consistent comparison of REE signatures. The included data were grouped into quartiles using the neodymium concentration in each sample. Neodymium was chosen for its applicability to paleoceanographic reconstructions. All concentration data was normalized to PAAS (Taylor and McLennan, 1985) and then to praseodymium (Pr). The normalization to PAAS facilitates inter-REE comparisons despite large differences in natural abundance and the normalization to Pr facilitates direct inter-sample pattern comparison despite differences in concentration.

Grain size was measured on the Tasman Sea samples using a MasterSizer 2000 laser granulometer at Macquarie University. Samples were analyzed pre-leach (freeze-dried) and post a buffered acetic acid decarbonization step (leach procedure after Abbott et al., 2016a, b). Two subsamples were loaded into the MasterSizer for each pre- and post- leach steps, which completed 2 measurement cycles 5 s apart per load for a total of four analyses per sample. All samples were analyzed using deionized water with the sample refractive index set to 1.544 and absorption set to 1. Samples were measured for 20 s (20,000 snaps) and the background was measured between each load for 10 s (10,000 snaps). The ultrasonic probe was set at 50% and remained on while the sample was loaded to a laser obscuration between 5 and 20% and then the probe was turned off at the start of each analysis. The stirrer was set at 500 rpm and the pump was set at 1250 rpm and remained on throughout loading and analyses. Grain size data for the North Pacific sites reported for comparison is from Abbott et al. (2016a).

Bulk mineralogy of representative samples from each site was determined by X-ray diffractometry (XRD) at Macquarie University. Freeze-dried but otherwise untreated samples were homogenized in an agate mortar and pestle, before back-loading into 25 mm internal diameter stainless steel sample holders. X-ray powder diffraction patterns spanning 5 to 90 ∘2θ were collected using a PANalytical Aeris benchtop XRD instrument (0.02∘ step size, Cu-radiation source with 40 kV generator voltage and 15 mA tube current, 1/8 inch divergence slits, and 23 mm beam mask). Diffraction patterns were interpreted using Panalytical HighscorePlus software with the ICSD database for phase identification. The detection limit for bulk XRD analyses depends on the density, Z number, and crystal structure of the compounds in the sample, but corresponds to approximately 1% by volume (Cullity and Stock, 2001).

The mineralogy of the clay fraction in representative samples from each site was determined on oriented preparations of <2 μm separates. The <2 μm fraction was obtained by settling after carbonate removal using sodium acetate buffered acetic acid (for Tasman Sea sites only because North Pacific sites are low carbonate) and ultrasonic dispersal. An aliquot was pipetted onto low background Si samples holders, air-dried and measured using the same instrument (Tasman Sea samples: 2–30 ∘2θ, 0.02∘ step size, 1/8 inch slits, 23 mm beam mask, beam knife in low position) or on a PANalytical Xpert-Pro MPD System (Oregon Margin samples: 2–65 ∘2θ, 0.0167∘ step size, Cu source with 45 kV generator voltage and 40 mA tube current, automatic slits). Clay diffraction patterns were collected both on air-dried samples and after treatment with ethylene glycol, with interpretation following Moore and Reynolds (1997). We did not attempt to directly geochemically characterize the detrital or authigenic clays in the sediments studied here because authigenic clays are typically small, interspersed grains and thus cannot be separated from detrital clays using physical separation techniques. Furthermore, our SEM-based petrographic work shows that even relatively large pellets of authigenic clays contain abundant detrital contaminants including igneous silicates and barite, preventing picked grains from representing a true authigenic signature. Since detrital and authigenic clay phases cannot be confidently separated from <2 μm, non-clay sedimentary fractions, the geochemical characterization of the clay size fraction would represent a mixed detrital clay, non clay, and authigenic clay signature. We therefore use available published characterizations of riverine clay to represent a detrital end member and authigenic phases (fish teeth, Fe-Mn crusts, glauconite) in our phase modeling.

Polished resin mounts of the freeze-dried but otherwise untreated sediments were prepared using standard methods. The samples were not homogenized prior to resin embedding in order to preserve the spatial associations present within the sediment. Resin mounts were subsequently ion polished using a Hitachi IM4000 Argon Ion Mill (30 min at 10° beam incident angle, 5 kV accelerating voltage and continuous sample rotation) to remove surface damage from mechanical polishing which can otherwise obscure phase relations and the presence of nm-scale coatings. Ion milled samples were carbon-coated prior to scanning electron microscope (SEM) analysis on an FEI Teneo LoVac field emission SEM equipped with dual Bruker XFlash Series 6 energy dispersive X-ray spectroscopy (EDS) detectors. High-resolution back-scatter electron (BSE) images and mineral maps of 1 to 3 regions of interest were collected for each sample (13 mm working distance, 15 kV accelerating voltage). BSE image tilesets (100 nm pixel resolution) and EDS spectra (1 μm step size, 8 ms acquisition time) for mineral mapping were collected sequentially using the FEI Maps Mineralogy software, followed by classification of the individual EDS spectra using the FEI Nanomin software (Haberlah et al., 2015). Mineral identification is achieved by comparing EDS spectra collected in the mapped area against reference spectra collected on known mineral standards, and is further constrained by independent XRD-based mineral identification. Unlike earlier SEM-based mineral mapping techniques (e.g., QEMSCAN), the Nanomin mineral classification system can de-convolve mixed X-ray spectra and assign up to three minerals per analyzed spot (Haberlah et al., 2015). This is a critical requirement for the correct interpretation of the mixed phase X-ray spectra characteristic of heterogeneous fine-grained sediments, where the X-ray generating electron interaction volume is commonly larger than the grain size. BSE photomicrographs, EDS elemental maps, and classified mineral maps provide direct visual evidence of grain relationships and common associations. While it is possible to convert these mineral maps into quantified percent compositions, the high heterogeneity of the samples coupled with the relatively small mapped areas results in a large uncertainty relative to bulk sample analyses such as XRD. Mineral mapping is employed here solely to test for the presence of, and establish the identity of trace phases that are too low for detection with the XRD.

We expect the REE pattern of the pore water to be influenced by both the source of REEs to the pore water and any sink of the REEs from the pore water. The fine-grained nature and intermingled distribution of most authigenic clays complicates an assessment of their REE content, and therefore of their potential influence on the pore water REE signature. However, glauconite is a widely occurring authigenic clay mineral which forms easily separated, macroscopic pellets (Baldermann et al., 2013, 2015) for which the REE signature is relatively well constrained (e.g., Huggett et al., 2017). Therefore, to test the feasibility of a clay driven pore water REE signature, we model the impact of glauconite formation on the pore water REE signature. For this calculation, we assume world river average clay (WRAC, Bayon et al., 2015) is the dominant source of pore water REEs (see section “Pore Water REE Patterns Support a Clay Dissolution Source” for a justification). Thus, we set the initial pore water value equivalent to WRAC (PW₀ = WRAC) and we average the Shadwell, Abbey Mills, and top and bottom Victoria glauconites of Huggett et al. (2017) for our glauconite REE pattern. We estimate a partition coefficient, k, by taking the ratio of WRAC to average glauconite concentration of each REE. We model the alteration of the pore water REE pattern with continued uptake into glauconite as:

P ⁢ W = P c × G + ( 1 - P c ) × P ⁢ W 0

where P_c is the percentage of PW₀ lost to glauconite (G) formation.

Fe-Mn oxyhydroxides in marine sediments are a commonly targeted archive for reconstructing paleocirculation because they are assumed to precipitate directly from ambient seawater, preserving a record of seawater REE composition (Rutberg et al., 2000; Bayon et al., 2002, 2004; Frank, 2002; Roberts and Piotrowski, 2015). Although macroscale Fe-Mn crusts and nodules are known to be important REE host phases (e.g., Bau et al., 2014; Conrad et al., 2016), they have a limited spatial distribution in the modern ocean. However, it is widely thought that Fe-Mn oxyhydroxides are ubiquitously present as REE enriched, nanoscale authigenic grain coatings similar to the Fe oxyhydroxides coatings known from terrestrially derived detrital silt and sand size grains (Manceau et al., 2007) or the nano-scale particles and aggregates attached to clays in the suspended fraction of tropical rivers (Poulton and Raiswell, 2005). Crucially, the distribution, preservation, and formation of these coatings remains poorly understood (e.g., Bayon et al., 2004) with no direct, imaging-based identification of their systematic presence to date. This is an important gap because the partitioning of REEs into these phases as well as the susceptibility of these coatings to reductive dissolution means that, where present, they are likely important contributors to pore water REE (e.g., Haley et al., 2004).

Attempts to quantify the abundance of these coatings have mostly relied on indirect, operationally defined chemical extractions rather than direct crystallographic, microscopic, or microbeam identification. This is also true more broadly of our knowledge of sedimentary trace element and REE host phases, which is largely informed by operationally defined chemical extractions; procedures known to be hampered by incomplete dissolution of the target phase, redistribution of elemental species through dissolution and subsequent secondary mineral precipitation, and dissolution of non-target phases (e.g., Taylor and McKenzie, 1966; Schwertmann and Pfab, 1994; Koschinsky and Hein, 2003; Wilson et al., 2013; Homoky et al., 2016; Skinner et al., 2019). Although sediment leachate trace metal ratios are reasonably similar to those measured on macroscopic Fe-Mn crusts (e.g., Gutjahr et al., 2007; Du et al., 2016), the well-established presence of phyllosilicate and phosphate phases in many Fe-Mn crusts (e.g., Rao, 1987; Bau et al., 1996; Hein et al., 1997) and the large range in sediment reductive leachate trace element ratios (e.g., Du et al., 2016 and references therein) leaves significant uncertainty as to the identity of the REE host phases dissolved by commonly used operationally defined leaching procedures. Nd/Fe ratios reported for reductive sediment leachates, for example, range from 0.001 to 0.01 compared to a much tighter range of only 0.001 to 0.002 for Fe-Mn crusts and nodules, whereas Al/Nd ratios in reductive sediment leachates are reported to fall between 50 – 400 compared to only 40 – 250 for Fe-Mn nodules and crusts (Du et al., 2016 and references therein) and Nd recovery associated with high Fe high Al phases has been observed (e.g., Wilson et al., 2013).

In addition, the large proportion of REEs recoverable through the same operationally defined sediment leaching procedures discussed above suggest that the phase or phases driving REE cycling in early sediment diagenesis should be sufficiently abundant to be readily identified by XRD or microscopically. Specifically, mass balance calculations show that 55 to 70% of the bulk sediment REE content is recoverable through combined acetic acid and hydroxylamine hydrochloride sequential leaching, with up to 55% of those REEs recovered during the reductive hydroxylamine hydrochloride leach that is believed to target the oxyhydroxide phase (Abbott et al., 2016b). Based on an average concentration of 58 mg Nd kg^–1 for authigenic Fe-Mn nodules (Bau et al., 2014), 5300 ng Nd cm^–3 recovered through reductive leaching from NP3000 (Abbott et al., 2016a), and assuming a sediment density of 2.6 g cm^–3, Fe-Mn phases would need to make up 3.5% of the bulk marine sediment to account for the Nd recovered in the reductive leaching procedure using:

S p = R N × 100

where S_p is the calculated percent of sediment, R is the recovered Nd through reductive leaching (in mg Nd kg^–1) and N is the measured concentration of Nd in the nodules (in mg Nd kg^–1). This value ranges from 1.5 to 13% if we consider the entire spread of Fe-Mn nodule Nd concentrations reported by Bau et al. (2014), and the lower proportion of bulk sediment Nd concentrations recovered through reductive leaching at NP200 (Abbott et al., 2016a). Although this assessment of Sp is likely subject to a substantial margin of error due to the limited data available with which to constrain the proportion of leachable Nd relative to total Nd, in addition to differences in mineralogy between Fe-Mn nodules, crusts and authigenic Fe-Mn oxyhydroxide coatings, it nonetheless allows us to estimate the likely range of Fe-Mn oxyhydroxide abundance required to account for the Nd recovered by chemical leaching. While the lower limit of the Sp range we calculate is just sufficiently small that Fe-Mn oxyhydroxide phases may not be detected by powder X-ray diffraction (detection limit for Fe-Mn oxyhydroxide phases in siliciclastic matrix is approx. 2 wt%), careful microbeam mineralogical characterization of these samples also fails to identify a ubiquitous presence of authigenic Fe-Mn oxyhydroxides (Figures 4, 5) despite 5 to 9 wt% iron in the bulk sediment from each North Pacific site (Abbott et al., 2016a). This microbeam characterization revealed only trace quantities of 2–4 μm, detrital Fe oxyhydroxide grains (Figure 5). Instead, the bulk of the iron present at all sites appears to be hosted in the clay mineral fraction.

This is not to say that nanoscale Fe-Mn oxyhydroxides are not present. Recent work by Blaser et al. (2016), for example, repeatedly applied a modified “weak” reductive leach to sediments from over ten sites in the North Atlantic and consistently identified a rapid decline in Nd and Mn concentrations from the first leach to subsequent leaches. Leachate elemental ratios and Nd isotopic composition indicated only minimal contamination from non-target siliciclastic phases during the first leach step. Overall, this suggests that labile Nd-bearing Mn phases may be widespread but relatively low-abundance phase in marine sediments. Mössbauer spectroscopic analysis of a range of marine sediments further confirms the presence of widespread, low abundance reactive Fe oxyhydroxides in marine sediments, and identifies these to be dominantly nanoscale (<12 nm diameter) goethite (van der Zee et al., 2003). Interestingly, goethite is considered to be relatively resistant to reductive leaching (Poulton and Canfield, 2005). The imaging approach used here does not have the resolution to identify nanoscale Fe-Mn oxyhydroxide, identifying only aggregates >200 nm in size. It is therefore likely that the Fe-clay association we identify in our samples is partly due to nano-scale intermingling and coating of clays by Fe-Mn oxyhydroxides. However, the presence of iron-bearing clays such as chlorite and Fe-smectite (nontronite) and the failure to identify Fe-Mn oxyhydroxides by XRD suggests that the major portion of clay-associated Fe is hosted within the clays, rather than representing discrete nanoscale oxyhydroxide coatings. This interpretation is consistent with the findings of Homoky et al. (2011) that Fe-Mn bearing clays may be a thermodynamically favorable explanation to the presence of nano-scale iron and manganese phases in pore water. We therefore argue that while Fe-Mn oxyhydroxides maybe an ubiquitous trace component of marine sediments, they appear to be insufficiently abundant to represent the primary source of Nd in reductive leachates or pore fluids, consistent with recent observations that the REEs recovered by leaching procedures are not necessarily sourced from oxyhydroxide phases (e.g., Vance et al., 2004; Tachikawa et al., 2013; Osborne et al., 2017; Skinner et al., 2019).

The complex, but commonly observed relationship between Fe and Nd in marine pore waters and the range of observed Nd/Fe ratios (e.g., Haley et al., 2004; Gutjahr et al., 2007; Du et al., 2016; Abbott et al., 2016a) leads us to suggest that the REEs are remobilizing from iron containing phases other than Fe oxyhydroxides. This suggestion is consistent with recent research revealing the same sedimentary origin for REEs and iron in the Southern Ocean (Blain et al., 2008; Zhang et al., 2008; Grenier et al., 2018). Our EDS elemental mapping suggests the primary iron host in the sediments studied here to be the clay mineral fraction (Figure 5). Indicators more commonly interpreted as evidence for oxyhydroxide coating recovery during reductive leaches, including trace metal ratios within the leachate (e.g., Gutjahr et al., 2007; Du et al., 2016), REE patterns (e.g., Gutjahr et al., 2007), and the agreement between the recovered neodymium with the overlying seawater (e.g., Elderfield and Greaves, 1982; Palmer and Elderfield, 1985; Rutberg et al., 2000; Roberts and Piotrowski, 2015) are also consistent with a benthic control model of REEs (Haley et al., 2017) that could be driven by clay minerals as we propose here. Further supporting this hypothesis is the correlation between finer average grain sizes, higher pore water REE concentrations (e.g., Abbott et al., 2016a), and higher sediment REE concentrations (e.g., Sa et al., 2018); consistent with higher reactivities of fine-grained, high surface clays than of coarser sediment component. We propose that clay dissolution and authigenesis may provide the universal controls on the sedimentary release (e.g., reverse scavenging; Elderfield and Greaves, 1982; Jeandel et al., 1995) and uptake (e.g., authigenic clay formation; Michalopoulos and Aller, 1995; Rahman et al., 2017) of REEs, and may therefore be the primary driver of REE distribution in seawater, pore water, and marine authigenic phases.

Chemical transfer reactions among terrestrially derived particles and seawater include adsorption/desorption, ion exchange, and dissolution/precipitation reactions (e.g., Elderfield and Sholkovitz, 1987; Jeandel et al., 1995; Rousseau et al., 2015; Homoky et al., 2016). Desorption and ion exchange tend to be relatively rapid processes, driven by salinity increases and changes in fluid major ion composition occurring as soon as river-borne particles come into contact with higher salinity estuarine waters (e.g., Rousseau et al., 2015). The conservative nature of major seawater ions and uniform salinity of bottom water versus pore water therefore suggest that exchange reactions are unlikely to play a significant role in the transfer of REEs from clays to pore water at our study sites. Dissolution and precipitation reactions, on the other hand, are longer-term processes that continue as long as the minerals are out of equilibrium with the adjacent fluid. The rate of dissolution and precipitation will change through time as a function of the saturation of the surrounding pore waters, as seen in batch reactor experiments (Oelkers et al., 2011). While the rates and extent of dissolution are hard to quantify due to simultaneously occurring secondary precipitation reactions (Oelkers et al., 2011; Pearce et al., 2013; Rousseau et al., 2015; Homoky et al., 2016), these experiments and investigations of Nd isotope evolution along estuarine salinity gradients do show that dissolution progresses rapidly (days to weeks) and is of sufficient magnitude to impact on oceanic REE budgets. Closed-system dissolution experiments, for instance, show 0.4% of the Nd contained in natural basaltic particles is released after only one month, sufficient to alter the initial ε_Nd composition of the seawater to that of the bulk basalt (Pearce et al., 2013). Modeling studies constrained by oceanic ε_Nd distributions suggest up to 1–3% of the continentally derived sediments may dissolve into seawater (Lacan and Jeandel, 2005; Arsouze et al., 2009). This estimated dissolution is several orders of magnitude greater than the amount of dissolution required (<0.001%) to account for elevated (relative to seawater) pore water Nd concentrations at the North Pacific sites (Abbott et al., 2016a; we do not estimate this for Tasman Sea sites as bulk sediment REE data is currently unavailable).

A disproportional contribution to this 0.001 to 3% total sediment dissolution from clay minerals is likely because mineral dissolution rates are commonly a function of surface area (e.g., Schott et al., 2009, 2012). For comparison, the specific surface areas of common clay minerals [N₂ sorption Brunauer–Emmett–Teller (BET) surface areas of illite and smectite are 76–91 m²g^–1 and 32–97 m²g^–1, respectively; Van Olphen and Fripiat, 1979] are up to two orders of magnitude greater than that of the ground mineral powders used in typical laboratory dissolution experiments (e.g., Chaïrat et al., 2007; Oelkers et al., 2008). In addition, ubiquitously distributed Fe³⁺ containing clays (including chlorite, illite, smectite, and mixed-layer clays) are susceptible to microbially mediated alteration via reductive dissolution (Kim et al., 2004; Vorhies and Gaines, 2009; Liu et al., 2012; Zhang et al., 2012; Ijiri et al., 2018). Given that microbially mediated clay mineral dissolution occurs because of the reduction of structural Fe³⁺ (Dong et al., 2009), it stands to reason that these same minerals would be targeted by the reductive leachates used to isolate Fe-Mn oxyhydroxide phases. Evidence for dissolution of silicate phases, including clays, by reductive hydroxylamine hydrochloride leaches can be found in a recent sequential leaching study of North Atlantic sediments, which documented substantial Al and Fe mobilization from the silicate fraction after repeated exposure to a “weak” reductive leach (Blaser et al., 2016). This could reconcile the operational defined leachate evidence for a significant Fe-Mn oxyhydroxide REE host phase in the absence of visible authigenic Fe-Mn oxyhydroxide coatings. Furthermore, clay mineral dissolution could at least partly explain the relationship between pore water dissolved iron and REEs in some sedimentary settings and a lack of a clear relationship in others (e.g., Haley et al., 2004; Abbott et al., 2015b), depending on the Fe content of the clays in question.

If we assume that a single process, or source, dominates the benthic REE cycle then the REE pattern and concentrations of the pore water should be a function of the source of REEs and any simultaneous sink, such that:

P ⁢ W = S - P

whereas PW is the pore water, S is the host phase of REE in the sediments acting as the source and P is the loss of REEs through adsorption or authigenic mineral precipitation. In our considerations of P, we ignore the diffusive flux out of the sediments as this flux is not considered a mechanism for fractionation and thus will not change the REE pattern or isotopic signature of the pore water. Using this basic framework, we can examine the range of REE patterns observed in pore waters and authigenic phases in order to test the hypothesis that clays are the primary drivers of benthic REE cycles.

We expect the pore water REE signature to be most similar to the source phase REE signature at depths that have the highest overall pore water REE concentrations, i.e., the depth at which P is most likely to approach 0 (“source” depth of Abbott et al., 2015b). We find that the most Nd enriched pore waters show REE signatures similar to the pattern of World River Average Clay (WRAC, Figure 6; Bayon et al., 2015), consistent with clay dissolution as the source of REEs. Pore waters with lower REE concentrations, by contrast, have more variable REE patterns that do not closely match the WRAC pattern, and are generally more HREE enriched than the higher concentration pore waters (Figure 7). This could reflect REE sourced from a host phase other than clays (i.e., a different S-term in framework above), but we argue that this is more likely caused by a modification of the pore water pattern by fractionation from REE sorption or incorporation in authigenic minerals (P term in framework above).

As noted in section “Mechanisms of REE Release From Clay Minerals,” experimental quantification of the rates and extent of dissolution of detrital sedimentary material is complicated by simultaneously occurring precipitation reactions producing a range of secondary minerals (e.g., Oelkers et al., 2011), mostly aluminosilicates. A wide range of secondary minerals including phyllosilicates, Fe-Mn oxyhydroxides and phosphates are known to form during early diagenesis, each of which are known to preferential scavenge LREEs, leading to HREE enrichment in the pore water (Cantrell and Byrne, 1987; Byrne and Kim, 1990; Byrne and Kim, 1993; Jeandel and Oelkers, 2015). Of these, authigenic clays are likely the most widespread and quantitatively significant authigenic phase. We explore the impact of secondary mineral precipitation on pore water REE patterns using a simple mass balance calculation, assuming that the starting REE pattern of pore water resembles WRAC (in agreement with section “Pore Water REE Patterns Support a Clay Dissolution Source”), and that the secondary phase responsible for REE uptake is an authigenic glauconite clay (Figure 8, see also section “Fractionation Feasibility Modeling”). We find that increased removal of REEs by glauconite produces an increasingly seawater-like pattern in the residual dissolved phase (Figure 9). Specifically, we observe an increase in HREE/LREE and a decrease in the MREE anomaly. Because experimental and modeling work suggests dissolution of detrital silicates releases up to 0.4 to 3% of the solid phase REEs, but pore water has several orders of magnitude less REEs per volume then the accompanying solid phase (<0.001%, Abbott et al., 2016a), we consider the high authigenic phase REE uptake (∼90%) required to produce a seawater-like HREE/LREE and MREE signal from a WRAC dissolution source to be reasonable. However, the resulting MREE anomaly is slightly lower than that observed in pore water, fish teeth, foraminifera, and Fe-Mn nodules. This discrepancy is likely because while WRAC is a useful starting point for our assessment of a clay source, as an average value WRAC masks the broad range of REE signature of clays sourced from different bedrocks (e.g., Bayon et al., 2015). The dissolution of clay size fraction during early diagenesis is unlikely to be congruous, with some clay phases dissolving more rapidly or more completely than others. For example, clays in suspended load of rivers draining volcanic (basaltic) terrains are known to have greater MREE anomalies than WRAC and this detrital volcanic material may preferentially influence pore water (e.g., Dessert et al., 2003; Vance et al., 2004; Wilson et al., 2013). Rivers draining basaltic terrains contribute heavily to clays throughout the Pacific Ocean, compatible with the MREE enrichment in pore water relative to WRAC even at source depths in the pore water profiles (Figure 9). The deep-sea clay fraction likely contains detrital clays sourced from a range of source areas, and disproportionate dissolution of a particular clay fraction, such as fine-grained, high surface smectite clay that is commonly of volcanic origin, may be what results in MREEs higher than WRAC. Equilibrium modeling identified Fe-bearing smectite clay as the most likely nano-colloidal phase present in pore water (Homoky et al., 2011) supporting the plausibility of this mechanism.

Several other independent lines of evidence support our inference that the HREE enrichment of the lower concentration pore waters is largely due to fractionation from clay authigenesis rather than the formation of authigenic oxyhydroxide or phosphate phases. Firstly, while not conclusive, microbeam characterization of the samples reveals only trace quantities of Fe oxyhydroxide and calcium phosphates, neither of which are of authigenic origin suggesting that these phases are not abundant enough to drive the REE composition of pore water. Secondly, findings that authigenic clays are the primary sink for Li in the ocean (e.g., Vigier et al., 2008) support the importance of authigenic clay formation in marine elemental cycling. Pore water Li concentrations decrease with depth and with decreasing REE concentration at the same North Pacific sites (Abbott et al., 2015b). Together with the presence of Fe-smectite and glauconite pellets at NP1500, the pore water Li concentrations at sites with and without visible glauconite formation are evidence of authigenic clay formation (e.g., Stoffyn-Egli and Mackenzie, 1984). Furthermore, while we are unable to quantify authigenic clays at our sites, recent studies have proposed a significantly increased magnitude and spatial extent of authigenic clay formation/reverse weathering than previously believed (Michalopoulos and Aller, 1995; Rahman et al., 2017). Baldermann et al. (2015) for example, have shown that authigenic clays (glauconite) are a potentially more significant Fe sink in low-oxygen settings than authigenic Fe sulfides, whereas Rahman et al. (2017) use cosmogenic ³²Si to estimate that authigenic clay formation is responsible for the burial of an estimated 4.5–4.9 × 10¹² mol/yr Si, representing almost 50% of the total dissolved Si inputs to the ocean (Frings et al., 2016). Finally, our simple model demonstrates that clay authigenesis (reverse weathering) has the potential to push the pore water REE pattern away from the REE pattern of the primary host phase toward more seawater-like compositions.

Observations of the non-conservative behavior of neodymium isotopes in the ocean have largely been concentrated around the margins (e.g., Lacan and Jeandel, 2005; Arsouze et al., 2009; Grasse et al., 2012; Wilson et al., 2013). In line with these observations, several additional but geographically limited sources of Nd to the ocean have been proposed, including submarine groundwater discharge (e.g., Johannesson et al., 2011), hydrothermal vents (e.g., Douville et al., 1999), and even localized detrital inputs (e.g., Howe et al., 2016). The relatively strict geographic limits of these inputs and the non-conservative behavior of ε_Nd associated with these has allowed proxy records collected from outside of these regions to be interpreted under the assumption of quasi-conservative behavior of ε_Nd. However, the ubiquitous distribution of clays throughout the global ocean means that a benthic source driven by clay minerals would not be limited by proximity to land or as a function of water depth. This supports the proposal of Abbott et al. (2015b) that a widespread benthic source of REEs from the pore water could be the major source of REEs to the ocean, thus balancing the oceanic budget. While studies have noted the potential influence of reactive, REE-enriched sedimentary components with distinctive Nd isotopic composition (relative to the bulk sediment) on the oceanic REE budget (e.g., Tachikawa et al., 2004; Wilson et al., 2013; Abbott et al., 2015b; Du et al., 2016), particularly at continental margin sites (e.g., “Boundary Exchange,” Lacan and Jeandel, 2005; Jeandel et al., 2007; Rousseau et al., 2015; Jeandel, 2016), our results suggest the margins are not necessarily the regions with the largest flux. Indeed, the more distal sites with smaller grained sediments likely have larger fluxes (e.g., Holdren and Berner, 1979; Abbott et al., 2016a; Sa et al., 2018). A higher flux from distal sites could explain the isotopic agreement between modern core top leachates, pore water, and bottom water particularly if such a benthic flux is the dominant source of ocean Nd (e.g., Elderfield and Greaves, 1982; Abbott et al., 2015a; Haley et al., 2017). From this perspective, the absence of an anomalous isotopic signature in deep water regions is not indicative of a modest benthic flux, rather it is a predictable consequence of a long-term benthic flux from a major proportion of the seafloor. In other words, the absence of an anomalous signature in these regions is expected if sediment diagenesis is the dominant source of Nd to bottom waters (Elderfield and Greaves, 1982; Abbott et al., 2015b). On the other hand, regions near the margin are more prone to the additional influence of fresh reactive phases making the ε_Nd of the flux in these areas more likely to be unique, and therefore noticeable.

Significant variation in ε_Nd between suspended and dissolved riverine loads, dissolved riverine load and local bedrock, different sized sediment grains, as well as between organic matter and detrital material (VanLaningham et al., 2008; Viers et al., 2008; Freslon et al., 2014; Bayon et al., 2015; Hindshaw et al., 2018) all demonstrate the importance of host phase identification in balancing the oceanic REE budget and for the application of REEs as tools in biogeochemical studies. For instance, in some large river systems there is more than a 1 epsilon unit offset between silt and clay sized fraction, with the smaller clay sized particles being systematically more radiogenic (Bayon et al., 2015). Attributed to preferential weathering (Bayon et al., 2015), this suggests that some of the most labile components of the sediment flux to the ocean are not necessarily representative of the average composition of the corresponding drainage basin. Minor components of the bulk sediment are therefore able to shift pore water ε_Nd away from the bulk sediment if these components are reactive and enriched in REEs, making for an ε_Nd of the flux that is noticeably different to the bulk sediment and the expected bottom water value. This discrepancy is most likely to occur proximal to the source (e.g., continental margins), especially in regions with more recently eroded materials, consistent with existing observations of boundary exchange concentrated near the margins (e.g., Lacan and Jeandel, 2005; Arsouze et al., 2009; Jeandel, 2016). Generally, these considerations result in the actual source being more radiogenic than the bulk rock (e.g., Viers et al., 2008; Grenier et al., 2018; Hindshaw et al., 2018). This model could explain North Pacific pore water neodymium isotopes that were more radiogenic than the bulk sediment or leachable phases (Abbott et al., 2016a) and why near shore sites were more radiogenic off the Kerguelen Plateau (Grenier et al., 2018). If the glass shards at NP200 (Figure 5) are relatively easily dissolved after deposition, they may release a more radiogenic signature than that of the bulk sediments at that site.

The array of REE patterns associated with authigenic phases such as those measured in sediment leaches, fish teeth, and pore waters (Figure 9; the “authigenic array” of Du et al., 2016) can largely be explained as a balance between a WRAC REE source and the preferential uptake of LREEs by authigenic clays as demonstrated by the REE pattern of glauconite. While this combination of clay dissolution and clay authigenesis can explain the HREE/LREE variation observed in pore water and seawater, the low MREE anomaly of these HREE enriched waters may indicate the role of additional processes but the formation of a MREE enriched authigenic phase (Figures 8, 9; e.g., Kang et al., 2014; Bayon et al., 2016; Trotter et al., 2016) would further decrease, rather than increase the MREE anomaly in the fluid phases. However, the MREE anomaly may be a result of local variability in the clay source, preferential dissolution of more labile clay fractions (i.e., volcanic), or fractionation associated with clay authigenesis not captured by the glauconites considered above. WRAC only considered large rivers covering an estimated 30% of the area that drains into the ocean (Bayon et al., 2015) and does not allow for the locally dominant, chemically distinct contribution of smaller river systems as mentioned above. The potential importance of the local to regional variability in the clay supply to the ocean is illustrated by the range of REE patterns associated with clays coming from volcanic landscapes versus large rivers (Figure 8; Bayon et al., 2015). This sensitivity to clay supply also means that the magnitude and signature of the benthic source may vary as a function of continental weathering, ocean circulation, and sediment provenance influences on the overall sediment supply.

The association of iron and REEs with clay phases in the sediments emphasizes the need for caution in interpretations of the marine authigenic record. We suggest the leaching protocols typically employed to recover this authigenic record are not only targeting oxyhydroxide phases, but also attacking clay host phases. Conventional leaching techniques mobilize significant quantities of silicate hosted REEs particularly in carbonate poor sediments and can release up to 16% of sedimentary iron (e.g., Wilson et al., 2013; Blaser et al., 2016). Recent developments in leaching techniques using a gentler approach can minimize the “contamination” from a silicate fraction, but even these refined methods mobilize some of the silicate fraction in the absence of a carbonate buffer (Blaser et al., 2016, 2019). For instance, the recent sequential leaching experiments on a wide range of North Atlantic sediments that used a refined reductive leaching approach demonstrated progressively increased Al/Nd, accompanied by shifts in leachate ε_Nd away from a foraminiferal/authigenic value toward a volcaniclastic/detrital silicate signature (Blaser et al., 2016). Several clay minerals contain structural Fe³⁺, which is prone to reductive dissolution especially in smectite (Vorhies and Gaines, 2009) and thus could be a source of the iron recovered during reductive leaching protocols. Even our simple model of the effect of glauconite formation on a WRAC-like REE pattern in the pore water demonstrated that uptake by glauconites can plausibly lead to the HREE enrichment commonly observed in pore waters, seawater, and other authigenic phases (Figure 9, e.g., fish teeth, Fe-Mn nodules). We argue that the similarity of the REE pattern of authigenic phases such as Fe-Mn nodules (e.g., Bau et al., 2014) and fish teeth (e.g., Huck et al., 2016) to lower concentration pore waters (Figures 8, 9) is likely due to continuing exchange (and homogenization) between these authigenic phases and the pore waters, and not an original Fe-Mn or phosphatic source of REEs to the pore water. This argument is consistent with diagenetic transfer of REE from clay minerals to biogenic phosphates in Ce anomaly mass balances (Grandjean et al., 1987) and would mean that the exchange between the sediments and pore water is continuing through early diagenesis. Thus, the authigenic phases in the sediments are going to record the balance of fluid-particle exchange in the pore water rather than a true bottom water signature, consistent with observed pore water-like REE patterns in Fe-Mn oxyhydroxide and biogenic phosphates (Figures 8, 9) with implications for our application of authigenic records to paleocirculation reconstructions.

The magnitude and character of the REE flux is likely to change in response to changing climate and the resulting changes in sediment supply. In this way, the apparent correlation between authigenic ε_Nd records and major climatic events (e.g., Böhm et al., 2015; Howe et al., 2016; Abbott et al., 2016b; Deaney et al., 2017) makes sense. Ocean circulation is going to, at least in part, be responsible for sediment provenance at any given site with the climate and tectonic activity both exerting control over the sources of those sediments (e.g., continental weathering; VanLaningham et al., 2008; Viers et al., 2008); thus, we would still expect a correlation between the timing of ε_Nd value excursions and intervals of climatic or oceanic change. This means that in a clay-driven system, the changes in REE and ε_Nd in authigenic phases may reflect both changes in sediment provenance and changes in continental weathering regime. If widely applicable, clays as a source of pore water REEs may make clay type and distribution potentially useful factors for estimating the diagenetic overprinting of authigenic neodymium isotope records. The resulting authigenic signature in sediments would be a function of the REE content and reactivity of the solid phases present as well as the location, mineral association, and nature of the host phase as observed with strontium (McKinley et al., 2007). Furthermore, if reductive dissolution is an important component of overall sediment dissolution, then we may see a relationship between organic matter production and REE remobilization. If this is the case, a relationship between organic matter supply and the spatial and temporal variability of a benthic flux may exist which could link the benthic source of neodymium to the ocean and changes in continental weathering regimes through time.