Re-assessing the influence of particle-hosted sulphide precipitation on the marine cadmium cycle

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


Introduction
Despite its known toxicity, culturing studies bear abundant evidence that the metal cadmium (Cd) is beneficial to some marine photosynthesisers, partially relieving growth limitation by zinc in both diatoms and coccolithophorids (Price and Morel, 1990;Lane and Morel, 2000;Xu et al., 2007;Morel, 2013). Certainly, regardless of its exact physiological role, dissolved Cd is taken up by phytoplankton in the sunlit surface ocean, reducing surface concentrations and resulting in a large-scale marine distribution that mimics those of the major nutrients. It has long been known that seawater Cd concentrations are closely correlated with phosphate (PO 4 ) at the global scale ( Fig. 1a; Boyle et al., 1976;Bruland, 1980;Boyle, 1988;de Baar et al., 1994), broadly similar to the canonical Redfield correlation between nitrate (NO 3 ) and PO 4 (Redfield, 1934).
However, in contrast to the factor ~2 variation inferred for the nitrogen:phosphorus ratio of particulate export (12-26 mol/mol; Wang et al., 2019), the cadmium:phosphorus (Cd:P) export ratio has recently been shown to vary by a factor of ~7 (0.2-1.5 mmol/mol; Black et al., 2019), while euphotic-zone particulate Cd:P varies by more than 20× (<0.1-2 mmol/mol; Sherrell, 1989;Bourne et al., 2018). This range is likely the result of the extreme plasticity in the stoichiometry of phytoplankton uptake -and, consequently, cellular quota -that is characteristic of Cd and other micronutrients (e.g. Sunda and Huntsman, 2000;Moore et al., 2013;Twining and Baines, 2013). Culture studies with diatoms and coccolithophorids have shown that Cd uptake rates can increase by two orders of magnitude when ambient Cd 2+ concentrations increase to a similar degree Huntsman, 1998, 2000). A similar increase is observed in response to low concentrations of free zinc (Zn 2+ ), likely reflecting upregulation of Cd transport to combat Zn limitation (Sunda and Huntsman, 2000;Xu et al., 2007). Growth limitation by iron (Fe) has also been shown to increase the Cd:P of natural phytoplankton assemblages by a factor of 2-10 (Cullen et al., 2003).
This variable uptake stoichiometry plays an important role in creating a much-discussed feature of the marine Cd-PO 4 relationship: the change in its slope in the Atlantic Ocean (Boyle, 1988;de Baar et al., 1994). The increased data coverage of the GEOTRACES era has allowed confirmation and combination of two early hypotheses that (a) variation in the Cd:P of biological uptake (Saager and de Baar, 1993;de Baar et al., 1994), and (b) the low Cd:PO 4 of upper-ocean waters exported from the Southern Ocean (Frew and Hunter, 1992;Frew, 1995) are dominantly responsible for this slight non-linearity (Baars et al., 2014;. Detailed Southern Ocean studies have shown that elevated phytoplankton uptake of Cd relative to PO 4 south of the Antarctic Polar Front (APF) produces a signal of relative Cd depletion in surface waters of the Subantarctic Zone, a signal that is transmitted to mode and intermediate waters that form in this region (Ellwood, 2008;Abouchami et al., 2014;Baars et al., 2014;Sieber et al., 2019a). Basin-scale meridional transects have shown that these upper-ocean waters transport the Southern Ocean signal of relative Cd depletion into the low-latitude upper ocean, eventually transmitting it to North Atlantic Deep Water (Xie et al., 2015;Middag et al., 2018;Sieber et al., 2019b). As such, the first-order controls on the marine Cd distribution (Roshan and DeVries, 2021) are similar to those governing the distributions of other macro-and micronutrients and their isotopes (Sarmiento et al., 2004;Weber and Deutsch, 2010;de Souza et al., 2012;Martiny et al., 2013;Vance et al., 2017;Fripiat et al., 2021).
While resolving such long-standing questions, GEOTRACES-era data have also revealed new details of the marine Cd-PO 4 relationship, now often studied using the tracer Cd*, which quantifies the deviation of Cd concentration from a linear correlation with PO 4 (Eqn. 1 in Section 3.1). These data have shown that oxygen minimum zones (OMZs) are major exceptions to the open-ocean Cd-PO 4 correlation: in a Cd-PO 4 cross-plot, data from the uppermost OMZ fall below the trend defined by well-oxygenated waters (Fig. 1a), indicating an apparent deficit in Cd relative to PO 4 . It has been proposed that this offset reflects a loss of dissolved Cd at the upper OMZ boundary, through the precipitation of cadmium sulphide (CdS) in sulphidic microenvironments within sinking and decomposing biogenic particles (Janssen et al., 2014;Conway and John, 2015a). Such apparent Cd deficits, identified by a decrease in Cd* values, have been observed in the OMZs of the eastern tropical North and South Atlantic (Waeles et al., 2013;Conway and John, 2015a;Wu and Roshan, 2015;Guinoiseau et al., 2019;Xie et al., 2019a), in the oxygen-poor shallow subsurface of the subarctic Pacific (Janssen et al., 2014;Janssen et al., 2017;Yang et al., 2018), and in the OMZ of the eastern tropical South Pacific (John et al., 2018). The interpretation of these signals as resulting from CdS formation is consistent with evidence for dynamic sulphur cycling within O 2 -depleted oceanic OMZs (e.g. Canfield et al., 2010;Ulloa et al., 2012) and with particulate data showing a complementary enrichment in Cd relative to P in the North Atlantic OMZ (Janssen et al., 2014;Conway and John, 2015a;Waeles et al., 2016;Bianchi et al., 2018).
However, the factor 10-20 variability in the Cd:P stoichiometry of marine particulates (Twining et al., 2015;Bourne et al., 2018;Lee et al., 2018) and particulate export fluxes (based on particles >51 m; Black et al., 2019) introduces a key ambiguity to the interpretation of Cd* in OMZs, which host large inventories of regenerated nutrients that have accumulated through decomposition of organic matter. Within OMZs, lower concentrations of Cd than would be expected from the deep-ocean Cd-PO 4 relationship -i.e., negative Cd* values -may simply result from remineralisation of organic matter with a low Cd:P ratio, and may be further complicated by non-stoichiometric release of Cd and P from decomposing particles (Bourne et al., 2018;Xie et al., 2019a;Cloete et al., 2021;Roshan and DeVries, 2021). Indeed, Wu and Roshan (2015) and Roshan and Wu (2015) have ascribed the apparent Cd deficit in the North Atlantic OMZ to precisely such remineralisation of relatively Cd-poor material, rather than to absolute Cd loss.
A key piece of evidence supporting the inference of particle-hosted CdS precipitation in OMZs has come from the stable isotope composition of dissolved Cd (expressed as  114 Cd relative to the standard NIST-3018). In low-latitude OMZs of the Atlantic and the Pacific, dissolved  114 Cd exhibits an increase that correlates with decreasing Cd* ( Fig. 2a; Janssen et al., 2014;Conway and John, 2015a;John et al., 2018;Guinoiseau et al., 2019;Xie et al., 2019a). Such a  114 Cd elevation would be expected to result from CdS formation, given the preferential incorporation of lighter Cd isotopes into CdS observed in laboratory experiments (Guinoiseau et al., 2018). Indeed, Guinoiseau et al. (2018; have shown that the  114 Cd-Cd* correlations in the low-latitude Atlantic and Pacific OMZs are consistent with fractionation into CdS ( Fig. 2a), and infer that CdS precipitation plays an important role in Cd cycling in all OMZs. Scaling the 6 results of the optimised particle model of Bianchi et al. (2018), Guinoiseau et al. (2019) suggest that the associated marine sink is at least as large as other known sinks of Cd (sub-or anoxic sediments, e.g. Morford and Emerson, 1999;Little et al., 2015), and that it could be the dominant sink in the marine mass balance of Cd.
However, phytoplankton uptake of Cd also favours its lighter isotopes, such that under Cd-replete conditions, euphotic-zone uptake produces an elevated  114 Cd value in residual surface-ocean dissolved Cd (Lacan et al., 2006;Ripperger et al., 2007;Abouchami et al., 2011;Xue et al., 2013). Thus, the same strong Southern Ocean drawdown that slightly decouples the Cd and PO 4 distributions, producing a low-Cd* signal in upper-ocean water masses formed in the Southern Ocean (Baars et al., 2014), also introduces a preformed signal of elevated  114 Cd into these water masses (Sieber et al., 2019a). Their northward transport influences the low-latitude  114 Cd and Cd* distributions (Abouchami et al., 2014;Baars et al., 2014;Conway and John, 2015a;Xie et al., 2015;2017;Middag et al., 2018;George et al., 2019;Sieber et al., 2019b), and John et al. (2018) have pointed out that the  114 Cd-Cd* correlation of the Peruvian OMZ is near-identical to the preformed covariation observed in Southern Ocean waters (Fig. 2b). Conversely, no  114 Cd-Cd* correlation is observed in the subarctic Pacific (Janssen et al., 2017;Yang et al., 2018), although it also hosts extremely low shallow-subsurface O 2 (<20 mol/kg). Janssen et al. (2017) argue that this absence may be a consequence of the high Cd concentrations here, but it is worth noting that this region is not influenced by upper-ocean waters of Southern Ocean origin (Sarmiento et al., 2004).
Together, these observations highlight fundamental ambiguities in water-column evidence that has been interpreted as reflecting particle-hosted CdS precipitation within OMZs: first, the close correlation between Cd* and  114 Cd (Fig. 2) may simply be a distally-produced signal of fractionating biological uptake at high Cd:P, rather than reflecting local processes; and second, the trend towards low or negative Cd* values within OMZs could result from the accumulation of a regenerated signal with low Cd:PO 4 within their large remineralised nutrient pools. This review brings together a range of published datasets in order to deconvolve the competing controls on Cd in low-latitude OMZs. We do so using a data analysis approach that takes into account the large-scale structure of the dissolved Cd* and  114 Cd distributions, as well as the integrated signal of regeneration. A novel element of our synthesis is the consideration of the water-column data in the context of published particulate datasets and their systematics. Our re-assessment reveals little water-column evidence for particle-hosted CdS formation, which likely plays only a small role in the marine mass balance of Cd.

Datasets and sources
This study is based on published open-ocean GEOTRACES-era datasets, available via the GEOTRACES Intermediate Data Product 2017 (IDP2017v2; Schlitzer et al., 2018) or individual publications, with particular focus on dissolved Cd datasets that include  114 Cd observations. Dissolved-phase Southern Ocean data are from the circumpolar datasets of Sieber et al. (2019a) and Janssen et al. (2020). In the South Pacific, we focus on two GEOTRACES transects: GP19, which extends northwards from the Southern Ocean into the western tropical Pacific (Sieber et al., 2019b), and the zonal transect GP16 from the western tropical Pacific to the South American coast (John et al., 2018). Atlantic data are from the GA02 transect that extends northwards from the Southern Ocean along the western Atlantic (Xie et al., 2015;2017;Middag et al., 2018), as well as from tropical waters of the equatorial region (GA11; Xie et al., 2019a), the South Atlantic OMZ (GA08; Guinoiseau et al., 2019), and the North Atlantic OMZ (GA03; Conway and John, 2015a). We do not consider the tropical Atlantic [Cd] data of Waeles et al. (2013;2016), since they report total dissolvable Cd from unfiltered samples. The Indian Ocean Cd concentration, [Cd], data of Vu and Sohrin (2013) are only briefly discussed, since they are not accompanied by  114 Cd or particulate Cd data.

Discussion
Correlations between  114 Cd and Cd* in low-latitude OMZs have been interpreted to unambiguously reflect CdS formation. In the following, we consider the strengths and ambiguities of Cd* (Section 3.1) and review the Southern Ocean control on the Cd distribution in order to assess its influence on low-latitude  114 Cd-Cd* correlations (Section 3.2). Combining dissolved and particulate Cd data, we examine the low-latitude alteration of the [Cd], Cd* and  114 Cd distributions in the tropical South Pacific (Section 3.3) and tropical Atlantic (Section 3.4). We synthesise our findings in Section 3.5, and discuss their implications for the role of particle-hosted CdS formation in the marine Cd budget in Section 3.6.

Definition of Cd*
First applied by Baars et al. (2014), Cd* is an extension to cadmium of a concept previously applied to the major nutrients (Gruber and Sarmiento, 1997;Sarmiento et al., 2004;. The aim of such "star"-tracers, produced by linear combination of two biogeochemically-cycled elements, is to eliminate a major source of large-scale variability, thus bringing into focus variations specific to the element (Gruber and Sarmiento, 1997 Baars et al., 2014). However, a normalisation to local deep water Cd:PO 4 implies a direct connection between local deep-and upper-ocean waters, which is not true over most of the ocean (e.g. Sarmiento et al., 2004).
In our opinion, a single, globally relevant value of (Cd/PO 4 ) ref should be applied regardless of study area: we suggest 0.33 mmol/mol, a value that characterises the Cd-rich deep ocean (mean of IDP2017v2 data >2000m, excluding the Atlantic Ocean: 0.33 ± 0.02, n=474; Schlitzer et al., 2018) and thus assigns a Cd* of 0 nmol/kg to most of the deep ocean, with the (meaningful) exception of much of the Cd-poor deep Atlantic (Fig. 1a). Such a definition of Cd* encodes an expected mean behaviour of Cd, i.e. that its concentration scales with that of PO 4 according to the ratio of their mean deep-ocean concentrations, allowing Cd* to be interpreted as a measure of the decoupling of the distribution of Cd from that of PO 4 . A globally uniform definition of Cd* also permits comparison between basins, revealing highly systematic similarities (Sections 3.2 and 3.4).

Cd* in OMZs: special considerations
It is important to note that the absolute value of Cd* is arbitrary and cannot be directly associated with Cd loss or addition. Preformed variations in Cd* inherited from the surface ocean due to behaviour that deviates from the expected mean (e.g. uptake at Cd:PO 4 > 0.33 mmol/mol) will lead to ocean-interior Cd* variation that is unrelated to local loss/addition processes. Thus, vertical Cd* variations at any one location cannot be attributed to addition or loss without consideration of the preformed structure of the Cd* distribution, especially in the upper ocean. Wu and Roshan (2015) brought to light a second important characteristic of Cd*. These authors noted that, although data from the North Atlantic OMZ fall off the steep Cd-PO 4 trend observed at high [PO 4 ] (Fig.   3b), these waters in fact bear the same dissolved Cd:PO 4 as other shallow waters. Their anomalous behaviour in a Cd-PO 4 cross-plot results simply from the fact that, being OMZ waters with large pools of regenerated nutrients, they plot at relatively high concentrations of both Cd and PO 4 . Thus, Wu and Roshan (2015) argue that the OMZ waters simply extend the shallow-water Cd-PO 4 trend to higher concentrations, due to the regeneration of particles with low Cd:P. Since Cd* is the linear combination of two concentrations, it will become more negative as [Cd] and [PO 4 ] increase along a trend with slope lower than (Cd/PO 4 ) ref (Fig. 3a), documenting remineralisation at a Cd:P below the expected mean of 0.33 mmol/mol. Given the large variability in particulate Cd:P stoichiometry (Twining and Baines, 2013;Bourne et al., 2018), the potential for such a driver of Cd* trends needs to be recognised, especially in OMZs.
Furthermore, the concentration-dependence of Cd* predisposes it to display strong vertical variations in tropical OMZs, where nutrient-rich OMZ waters are typically overlain by nutrient-depleted tropical waters bearing near-zero Cd*.

High-latitude influence on the low-latitude  114 Cd-Cd* correlation
Due to the fact that the conversion pathway of upwelling deep waters to upper-ocean waters passes through the surface Southern Ocean, the stoichiometry of biological uptake in this region has an outsize influence on the marine distributions of biologically-relevant elements (Sarmiento et al., 2004;. Recent research has clearly shown that Cd is no exception to this general rule (Ellwood, 2008;Abouchami et al., 2014;Baars et al., 2014;Xie et al., 2015;Middag et al., 2018;George et al., 2019;Sieber et al., 2019b). South of the APF, the shallow-water Cd:PO 4 relationship suggests a highly elevated Cd:PO 4 uptake/regeneration ratio of ~1.05 mmol/mol (Sieber et al., 2019a), consistent with surface-ocean particulate Cd:P north and south of the APF in summer (0.86-2.32 mmol/mol in non-Fe-amended waters during SOFeX; Bourne et al., 2018) as well as winter (0.86-1.07 mmol/mol at 25m; Cloete et al., 2021). This elevated uptake leads to a decoupling of [Cd] and [PO 4 ] in Subantarctic surface waters (Ellwood, 2008;Baars et al., 2014): here, [Cd] drops to close to zero while [PO 4 ] is ~0.5 mol/kg, resulting in a Cd* value of around -0.2 nmol/kg. A recent circumpolar dataset (Sieber et al., 2019a;Janssen et al., 2020) has shown that this Subantarctic Cd* signal is visible in all sampled sectors of the Southern Ocean (Fig. 4a). Due to biological isotope fractionation, the negative Cd* signal is coupled to elevated  114 Cd (Abouchami et al., 2011;Xue et al., 2013;George et al., 2019;Sieber et al., 2019a), with values up to +1‰ in the summer mixed layer (Fig. 4b).
Two GEOTRACES sections extending northward from the Southern Ocean, through the well-ventilated western sides of the South Atlantic (GA02: Xie et al., 2017;Middag et al., 2018) and South Pacific (GP19: Sieber et al., 2019b), have demonstrated the larger-scale influence of Southern Ocean uptake on the [Cd], Cd*, and  114 Cd distributions. Figure 4 plots these data against neutral density ( n ), allowing an isopycnal view that highlights the extremely close similarity between the Cd* distributions in the subtropical South Atlantic and South Pacific (Fig. 4c). Both distributions exhibit a Cd* minimum of about -0.2 nmol/kg centred around  n = 26.8-27 kg/m 3 , the density of Southern Ocean mode waters that are formed in the Subantarctic Zone (McCartney, 1982) and carry this signal of Southern Ocean uptake well into subtropical latitudes (Figs. S1 and S2; e.g. Baars et al., 2014;Xie et al., 2015). The elevated  114 Cd in the subtropical Thus, a  114 Cd-Cd* correlation is not unambiguous proof of subsurface Cd loss. On the contrary, such a correlation seems to be a typical feature of ocean regions influenced by upper-ocean waters of Southern Ocean origin. The presence of such structure in the Cd* and  114 Cd distributions precludes analysis of their systematics in a framework that ignores the spatial dimension, such as the  114 Cd-Cd* cross-plots of Fig.   2. Rather, identifying the influence of low-latitude processes requires an approach that takes into account their preformed distributions. This may be done using models (e.g. Roshan et al., 2017), but due to the fine vertical scale of the signals of interest, we take a data-based approach in this contribution. concentrations are 50-75 mol/kg, while to the east they are below the detection limit of 2 mol/kg (Cutter et al., 2018). 5b,c; Section 3.2). This is not surprising, since Cd* is designed to remain invariant in the face of expected mean regeneration (Eqn. 1), and regeneration has little leverage to alter interior  114 Cd, especially in the Cd-rich Pacific (e.g. Janssen et al., 2017;Yang et al., 2018;Sieber et al., 2019b). As a consequence, the Cd* and  114 Cd distributions more strongly reflect the large-scale Southern Ocean control reviewed in Section 3.2.

Subtropical-tropical comparison
Nonetheless, in the upper tropical ocean, Cd* shows significant differences to the subtropical distribution, deviating from it at densities less than 26.8 kg/m 3 (~350m) in the open tropical Pacific, and at densities less than ~27.25 kg/m 3 (~625m) within the OMZ (Fig. 5b). Especially in the open tropical Pacific and towards the outer OMZ, the Cd* minimum along GP16 is seen at slightly shallower depths than in the subtropics, and Cd* values are more negative as well (Fig. 5b). In addition, there is a steady isopycnal increase in upper-ocean Cd* eastward, towards more poorly ventilated waters. Finally, waters east of 100°W also bear shallow Cd* maxima corresponding to the abovementioned [Cd] maxima (Fig. 5a,b). In these shallow maxima (40-120m water depth), Cd* reaches up to +0.1 nmol/kg, coinciding with the base of the oxycline (Fig. S4). Values of  114 Cd, which are generally very similar to the southern subtropics (George et al., 2019;Sieber et al., 2019b), show little variation except for a slight decrease in shelf/slope waters (Fig. 5c).

Influence of remineralisation in the South Pacific OMZ
For the following, it is important to bear in mind the fact that an isopycnal approach has some limitations in a comparison of the subtropical and tropical Pacific. Tropical dynamics in the zonally expansive Pacific lead to significant cross-isopycnal fluxes in the interior, resulting for instance in the weakening of the AAIW salinity minimum between the subtropics and tropical Pacific ( Fig. 5f; Tsuchiya and Talley, 1996;Fiedler and Talley, 2006). Southern-sourced Cd* minima will similarly be weakened by diapycnal mixing in the tropics, while tropical Cd and nutrient distributions are also influenced by the contribution of North Pacific waters to the equatorial Pacific thermocline (Johnson and McPhaden, 1999;Dugdale et al., 2002;Schott et al., 2004).
Nonetheless, since the only sources of deep water (> 2000m) to the Pacific are the Southern Ocean water masses flowing northward in the abyssal western Pacific (e.g. Reid, 1997), the marked increase in [Cd] over most of the tropical Pacific water column (Fig. 5a) highlights the strong influence of regeneration on the tropical Cd distribution (Roshan et al., 2017). This is especially true for the densities hosting the OMZ, where [Cd] increases by ~0.8 nmol/kg relative to the southern subtropics (Fig. 5a). Some proportion of this increase is likely due to the abovementioned North Pacific contribution. Limited data from the wellventilated subtropical North Pacific (Yang et al., 2018) suggest that these waters bear much higher [Cd] near OMZ densities (~0.4 nmol/kg at 26.5 kg/m 3 ) than corresponding South Pacific waters (~0.05 nmol/kg), and their influence may thus partially explain the higher tropical [Cd].
However, a process that certainly also plays a role, especially in the OMZ and over the shelf/slope, is the regional regeneration of Cd from the large export fluxes of the productive equatorial and coastal upwelling regimes (e.g. Honjo et al., 2008). This is unequivocally shown by the stable isotope composition of dissolved inorganic carbon (DIC),  13 C DIC , which exhibits values ~1‰ lower within OMZ and shelf/slope waters than in the open tropical Pacific ( Fig. 5g; to our knowledge, no  13 C DIC data are available for GP19).
This shift documents the accumulation of isotopically light, regionally regenerated DIC within the OMZ (e.g. Schmittner et al., 2013), and coincides with lower  114 Cd values in shelf/slope waters (Figs 5c,g).
Tellingly, the regeneration-driven subsurface  13 C DIC minimum coincides near-exactly with the shallow subsurface maxima in [Cd] and Cd* east of 100°W (Figs. 5a,b,g), fingerprinting the remineralisation of organic matter with a Cd:P significantly higher than the expected mean of 0.33 mmol/mol (Section 3.1).
Indeed, particulate data compiled by Bourne et al. (2018) show that the Cd:P of euphotic-zone particles along GP16 increases from 0.03 mmol/mol in the oligotrophic open tropical Pacific to as high as 1.42 mmol/mol within the upwelling region above the OMZ. More generally, seasonal and interannual variability documented by Bourne et al. (2018) shows that Pacific euphotic-zone particles have higher Cd:P when productivity is higher. Remineralisation of particles with the high Cd:P observed in the productive Peruvian upwelling will increase subsurface Cd*, as is indeed observed for all waters that bear the tropical remineralisation signal in  13 C DIC (Fig. 5b,g). The slightly lower  114 Cd values in shelf/slope waters over this density range suggest that incomplete Cd utilisation in the upwelling region above the innermost OMZ results in the export of isotopically light particulate Cd, analogous to the low particulate  114 Cd observed in the productive subarctic Pacific (Yang et al., 2018;Janssen et al., 2019).
Thus, the main low-latitude influence on the Cd distribution in the southern tropical Pacific is the remineralisation of organic matter. Within the Peruvian OMZ itself, this organic matter is especially Cdrich relative to the expected mean, resulting in a strong elevation of [Cd] and Cd*, while  114 Cd shows a muted opposite response. Thus, even in the functionally anoxic waters of the Peruvian OMZ, and its associated oxycline, in which a diverse microbial community undertakes a variety of spatially-compressed redox processes including sulphur cycling (Canfield et al., 2010;Ulloa et al., 2012), remineralisation of Cdrich particles has a much larger influence on the Cd distribution than loss of Cd to sulphides (Janssen et al., 2014), deep-living prokaryotes (Ohnemus et al., 2017) or adsorption to particles (Lee et al., 2018). Since remineralisation has limited leverage to alter  114 Cd, the OMZ  114 Cd distribution is largely driven by the preformed upper-ocean  114 Cd gradient (Fig. 5c).
The recognition of this control challenges a recent interpretation of the Peruvian OMZ  114 Cd-Cd* relationship as evidence for widespread CdS formation (Guinoiseau et al., 2018). In fact, as Fig. S5 shows, the main negative  114 Cd-Cd* correlation in the Peruvian OMZ is identical to the preformed correlation observed in the western subtropical Pacific, and thus provides no evidence for CdS formation. The tropical and subtropical  114 Cd-Cd* relationships do differ in the most O 2 -depleted waters, but as a consequence of the high OMZ Cd* values produced by biological uptake and export in the upwelling regime. As we discuss in Section 3.6, this Cd-rich biological export also has implications for recent estimates of the contribution of water-column CdS formation to excess Cd in Peruvian shelf and slope sediments.

A Cd sink in the open tropical Pacific?
A final -but key -feature of the Cd distribution along GP16 is the strongly negative Cd* minimum observed within the muted oxygen minimum of the open tropical Pacific (Fig. 5a,b). The distributions of N* and the nitrogen isotope composition of nitrate ( 15 NNO 3 ) clearly indicate that this oxygen-minimum layer communicates with the zone of active fixed nitrogen loss within the OMZ (Peters et al., 2018). However, while the N* and  15 NNO 3 signals become weaker away from the OMZ, indicating a mixingrelated attenuation, the Cd* minimum is more extreme in the open tropical Pacific, and extends to shallower density levels. The Cd* minimum is consistently associated with the oxycline above the O 2 -minimum at all open tropical stations, and is centred at a slightly shallower isopycnal,  n = 26.5 kg/m 3 , than the O 2 minimum (Figs. 5b and S6). Numerous mechanisms might cause this stronger Cd* minimum: (a) remineralisation at a Cd:P below the expected mean; (b) North Pacific influence on the equatorial thermocline; or (c) local/regional Cd loss.
The Cd:P of euphotic-zone particles in the productive equatorial Pacific is highly elevated; for example, the median Cd:P of particles <51 m in the equatorial Pacific upwelling (12°S-12°N) is 0.77 mmol/mol (range 0.29-2.0 mmol/mol, n=21; Bourne et al., 2018), and particulate Cd:P is highest (1.18-2.0 mmol/mol, n=4) under more productive upwelling conditions (Bourne et al., 2018). In Supplementary Discussion 1, we argue that this high Cd:P makes it extremely unlikely for remineralisation to reduce Cd*, despite the preferential remineralisation of particulate P in the shallow subsurface (Bourne et al., 2018), the preferential remineralisation of dissolved organic phosphorus over the dissolved organic nitrogen that presumably hosts protein-associated Cd (e.g. Letscher and Moore, 2015;Sipler and Bronk, 2015), and uncertainties associated with temporal variability in particle fields. With regard to North Pacific influence, limited dissolved Cd data from the mid-latitude North Pacific shows that the Cd* minimum there (-0.15 nmol/kg by our definition; Conway and John, 2015b;Yang et al., 2018) is less extreme than in the open tropical Pacific.
The fact that neither remineralisation nor North Pacific influence seem likely to produce the open tropical Pacific Cd* minimum indicates that it must be regionally produced, and could result from Cd loss. This possibility receives strong support from the observation that the open tropical Pacific oxycline is consistently associated with particulate Cd peaks (Ohnemus et al., 2017;. Indeed, a comparison of dissolved and particulate data (Fig. 6) shows that the Cd* minimum is associated with a particulate Cd maximum at every sampled location in the open tropical Pacific. Size-fractionated particulate data (Lee et al., 2018) reveal that this increase in particulate Cd occurs in both the large and small size fractions. Our recognition of these complementary signals in the particulate and dissolved data strongly suggests that the stronger Cd* minimum in the open tropical Pacific oxycline -outside the O 2 -depleted OMZ -is driven by a Cd loss to the particulate phase. Based on their statistical analysis, Ohnemus et al. (2019) attribute the accumulation of particulate Cd to "secondary biomass", which they assume to be heterotrophic and prokaryotic, but further data are required to conclusively identify the processes at work. Certainly, given the low particulate export flux in this region (e.g. Black et al., 2019), it would appear a priori more likely that the Cd loss is associated with some suspended particulate phase (which may be deep-living biota), rather than large sinking particles.
Whatever the process associated with this apparent Cd loss, the open tropical Cd* minimum is not associated with any significant change in dissolved  114 Cd (Fig. 5b,c), an observation that, prima facie, argues against a (fractionating) loss process. As has been previously argued for the North Pacific (Conway and John, 2015b;Janssen et al., 2017), the absence of a  114 Cd signal may be due to the high [Cd] of the tropical Pacific thermocline (~0.4 nmol/kg at the Cd* minimum). Given the strong upper-ocean  114 Cd gradient and the small isotope effects of sulphide formation (-0.32‰; Guinoiseau et al., 2018) and biological uptake (-0.2‰ to -0.45‰; Abouchami et al., 2011;Sieber et al., 2019a), even a fractionating loss of Cd from these Cd-rich waters is unlikely to result in a resolvable  114 Cd signal: as detailed in Supplementary Discussion 2, Rayleigh distillation predicts a maximum expected  114 Cd increase of only 0.05-0.10‰, while deep-ocean data from GP16 suggest a limit of resolvability of ±0.10‰ (2 SD for  n ≥ 27.9 kg/m 3 , excluding 3 hydrothermally-influenced stations). This inference is supported by contrasting  114 Cd behaviour in the Cd-poorer tropical Atlantic (Section 3.4.4).

Summary
This section has demonstrated the importance of remineralisation of Cd-rich organic matter for the Cd distribution in the tropical Pacific, and especially within the Peruvian OMZ. Remineralisation raises Cd* over the entire water column, and especially in the shallow OMZ, where positive Cd* values are observed at the base of the oxycline. A second feature is the strong negative Cd* minimum that is observed above the milder oxygen minimum (50-75 mol/kg) of the open tropical Pacific, which appears to be related to Cd loss to particulates. The tropical oxycline is thus associated with both an excess of Cd (in the OMZ) and a deficit (in the central tropics). We consider it possible that the subtle loss process whose consequences are visible in open tropical Cd* is in fact ubiquitous in the tropical oxycline, but that its influence on Cd* within the OMZ is overprinted by the strong signal of remineralisation.

Atlantic low-latitude cycling
The OMZ systematics of the tropical Pacific are contrasted almost diametrically by those of the tropical Atlantic OMZs. As reported by Janssen et al. (2014) and Conway and John (2015a), the North Atlantic OMZ bears strongly negative Cd* values in the shallow subsurface, coinciding with the tropical oxycline; Guinoiseau et al. (2019) have recently shown the same for the South Atlantic OMZ. In the following, we aim to understand the drivers of the differences in the Cd* and  114 Cd distributions between Atlantic and Pacific OMZs. We do so by placing tropical Atlantic observations (Conway and John, 2015a;Guinoiseau et al., 2019;Xie et al., 2019a) in the context of the Cd* and  114 Cd distributions of the well-ventilated western South Atlantic ( Fig. 4; Xie et al., 2017;Middag et al., 2018). Our inclusion of the North Atlantic OMZ (Conway and John, 2015a) in this framework is based on the recognition that the Cd* anomaly here is hosted in waters of primarily South Atlantic origin (Supplementary Discussion 3), allowing us to consider it together with the southerly-ventilated equatorial Atlantic (Xie et al., 2019a) and South Atlantic OMZ (Guinoiseau et al., 2019). The upper-ocean [Cd] increase is much weaker than that of [PO 4 ] (Fig. 7a,d), reflecting the low Cd:P of remineralisation Roshan and Wu, 2015;Middag et al., 2018). In the North Atlantic tropics, concentrations of Cd (but not PO 4 ) are significantly lower at intermediate depths ( n = 27.25 -27.75 kg/m 3 ) than elsewhere in the tropics (Fig. 7a,d), likely due to the influence of Cd-poor and saline North Atlantic water masses (Conway and John, 2015a;Jenkins et al., 2015) visible in the salinity distribution ( Fig. 7f).

Subtropical-tropical comparison
The distributions of Cd* and  114 Cd (Fig. 7b,c) are much less affected by tropical remineralisation than [Cd], analogous to the South Pacific (Fig. 5b,c; Section 3.3.2). Their distributions are mainly determined by the preformed structure set by the Southern Ocean (Section 3.2) and visible in the southern subtropical data (Fig. 7b,c), i.e. the steady upward increase in  114 Cd, and the Cd* minimum of around -0.2 nmol/kg at the density of Southern Ocean mode waters ( n = 26.8-27 kg/m 3 ). Nonetheless, remineralisation below the expected mean of 0.33 mmol/mol (0.17-0.26 mmol/mol; Roshan and Wu, 2015;Middag et al., 2018) results in a slight general Cd* decrease in the tropics relative to the subtropics (Fig. 7b), with no apparent effect on the  114 Cd distribution (Fig. 7c). As with the [Cd] distribution, the intermediate North Atlantic ( n = 27.25-27.75 kg/m 3 ) is an exception here, with Cd* ~0.1 nmol/kg lower than other tropical waters, and a hint of slightly elevated  114 Cd values, although offsets at densities > 27 kg/m 3 are only seen at a single station, and are on the order of the inter-laboratory reproducibility (±0.07‰). The most coherent Cd* difference between the tropics and the southern subtropics is the stronger tropical Cd* minimum in the upper thermocline, at  n ≈ 26.5 kg/m 3 (grey band in Fig. 7). This stronger Cd* minimum is visible in the western equatorial Atlantic, and becomes more negative in the North and South Atlantic OMZs (Fig. 7b), where [PO 4 ] is higher (Fig. 7d) but [Cd] barely increases relative to the equatorial Atlantic (Fig. 7a). These Cd* minima are associated with the tropical oxycline ( However, the shallow OMZ Cd* minima of the Atlantic contrast strongly with the Cd* maximum observed at the oxycline above the OMZ of the tropical South Pacific (Fig. 5b).

Influence of low-latitude remineralisation on Cd*
Two sets of major-element isotope data from the North Atlantic OMZ indicate that remineralisation drives the difference in OMZ Cd* systematics between the Atlantic and the Pacific (Fig. S7). Firstly, the shallow subsurface Cd* minima of the North Atlantic OMZ (at 51-89m water depth) coincide with shallow minima in  13 C DIC (Quay and Wu, 2015), in sharp contrast to the coincidence of Cd* maxima with  13 C DIC minima in the Peruvian OMZ (Fig. 5). Secondly, the North Atlantic Cd* minima also coincide with minima in  15 NNO 3 , which unambiguously document the regional regeneration of organic matter bearing low  15 NNO 3 due to nitrogen fixation in the North Atlantic (Marconi et al., 2015). These isotopic and Cd* minima are all hosted within a shallow subsurface O 2 minimum at around 100m depth ( n = 26.5 kg/m 3 ) that is physically separate from the deeper, O 2 -poorer minimum at 400m ( n = 27 kg/m 3 ; Fig. S7). Together, these observations indicate an upwelling-associated remineralisation signal at the level of the shallow Cd* minimum in the North Atlantic OMZ, and likely elsewhere in the tropical Atlantic as well, although colocated  13 C DIC and  15 N-NO 3 are not available.
What is the likely effect of this remineralisation on Atlantic Cd*? Limited particulate data from above the North Atlantic OMZ indicate a mean euphotic-zone particulate Cd:P of 0.36 mmol/mol (n=3; Bourne et al., 2018), much lower than Cd:P >1 mmol/mol above the Peruvian OMZ (Bourne et al., 2018), and similar to the expected mean of 0.33 mmol/mol. The scant Atlantic data and significant seasonal Cd:P variability observed elsewhere in the ocean (Bourne et al., 2018) result in some uncertainty here, but bulk particle remineralisation at the observed Cd:P ratios would not be expected to lower Cd* in the North Atlantic OMZ. Following our discussion in Supplementary Discussion 1, it is likely that the preferential shallow remineralisation of P over Cd (Bourne et al., 2018;Cloete et al., 2021) will result in a Cd:P signal from remineralisation that is lower than bulk particle Cd:P in the very shallow subsurface. This reduction is difficult to quantify, but regardless of its extent, the existing particulate Cd:P data suggest that remineralisation is unlikely to raise Cd* in the shallow North Atlantic OMZ -in strong contrast to the Peruvian OMZ (Section 3.3.2; Fig. 5). Together with the well-documented low Cd:P of Atlantic remineralisation inferred from independent dissolved data Roshan and Wu, 2015;Middag et al., 2018;Roshan and DeVries, 2021), this leads us to postulate that the tendency of remineralisation to either lower or not alter Cd* applies to the entire tropical Atlantic.
Thus, one reason for the difference in the Cd* systematics between the South Pacific and Atlantic OMZs may be the fact that, while remineralisation will certainly raise Cd* in Pacific OMZs, it does not seem to in the Atlantic. Given this near-neutral influence of remineralisation on Atlantic Cd*, the upper-OMZ Cd* distribution may be influenced, in near-coastal waters, by excess PO 4 released from anoxic shelf sediments at depths close to the Cd* minimum (Schroller-Lomnitz et al., 2019). Benthic PO 4 fluxes are mechanistically coupled to those of Fe (e.g. Ingall and Jahnke, 1994;Noffke et al., 2012), and dissolved Fe stable isotopes Klar et al., 2018) fingerprint sedimentary release of reduced Fe to the shallow North Atlantic OMZ waters that host the Cd* minimum. Together, the recognition of remineralisation and potential benthic flux signals in these waters suggests that the evidence in the North Atlantic OMZ Cd* distribution for widespread particle-hosted CdS formation is more ambiguous than has previously been inferred (see also Section 3.6; Janssen et al., 2014;Conway and John, 2015a).

A Cd sink in the tropical Atlantic?
However, there is also the evidence of the  114 Cd distribution to consider. As in the South Pacific (Fig. 5), the isopycnal structure of tropical Atlantic  114 Cd is largely set by the preformed distribution visible in the subtropical South Atlantic (Fig. 4; Fig. 7c; Xie et al., 2017). However, at upper thermocline depths hosting the tropical Cd* minimum,  114 Cd values are consistently higher than in the subtropics (Fig. 7c). The isopycnal  114 Cd offset from southern subtropical waters averages 0.1‰ (n=7) and ranges from 0.03-0.18‰, with the largest offsets observed in the North Atlantic (Fig. 7c). We acknowledge that southern subtropical data around this density range are limited, and the  114 Cd differences are small relative to the inter-laboratory reproducibility of ±0.07‰. However, most of the Atlantic data (with the exception of the North Atlantic) were analysed in a single laboratory (Xie et al., 2017;Guinoiseau et al., 2019;Xie et al., 2019a) with a better external reproducibility of ±0.04-0.05‰ (Guinoiseau et al., 2019). Thus, considering that marine  114 Cd profiles exhibit consistency within oceanographic regions (e.g. Yang et al., 2018;George et al., 2019;Sieber et al., 2019b), the consistent association of higher  114 Cd values with the tropical Cd* minimum (Fig. 7b,c) suggests a true elevation of  114 Cd in the upper tropical Atlantic thermocline relative to the subtropics. While this observation requires confirmation by future analyses, it receives support from the fact that many of the  114 Cd values observed in the Cd* minimum are as high as, or higher than, the values in the overlying surface waters (an obvious exception are the North Atlantic data, but the fidelity of the high 114Cd values in overlying waters here remains debated; e.g. Xie et al., 2017;Guinoiseau et al., 2019;Sieber et al., 2019b).
Together, the stronger Cd* minimum and elevated  114 Cd in the tropical Atlantic oxycline seem to indicate a Cd loss process here. Crucially, however, this potential loss signal is visible not only in the OMZs, but also within better-ventilated tropical waters of the western equatorial Atlantic (Fig. 7b, c,e;Xie et al., 2019a). This suggests that the loss process reflected in Cd* and  114 Cd is not specific to OMZ waters, but may in fact be a general feature of the tropical Atlantic oxycline -analogous to the stronger Cd* minimum observed in the oxycline of the open tropical Pacific (Section 3.3.3). This inference receives support from data across the western subtropical and tropical Atlantic (Middag et al., 2018) showing that a stronger Cd* minimum, associated with the tropical oxycline, appears at the transition from the southern subtropics to the tropics (Fig. S8). This tropical Atlantic Cd* minimum is slightly shallower than the O 2 minimum, as in the South Pacific (Fig. S6), and shoals isopycnally towards the equator (Fig. S8). Figure 8 shows that at comparable latitudes, the shallow tropical Atlantic Cd* distribution is near-identical to that of the open tropical Pacific (Fig. 8a,b), despite higher O 2 saturation in the better-ventilated Atlantic (Fig. 8c,d). As discussed in Section 3.3.3, the open tropical Pacific Cd* minimum is consistently associated with a subsurface particulate Cd maximum ( Fig. 6; Ohnemus et al., 2019), indicating a Cd loss there. No cosampled particulate Cd data are available for the Atlantic, but here we instead have the evidence of the  114 Cd distribution (Fig. 7c), which suggests a fractionating Cd loss process associated with the tropical Cd* minimum.
Thus, two independent lines of evidence, from two different oceans, appear to provide support for Cd loss in the shallow tropical oxycline. This inference is consistent with the limited data available for the Indian Ocean (Vu and Sohrin, 2013) which, though not complemented by isotopic or particulate data, show the subtropical preformed Cd* minimum giving way to a shallow tropical minimum at ~100m depth (Fig. S9).
Furthermore, the presence of a resolvable  114 Cd signal in the Atlantic (Fig. 7c), but not in the Pacific (Fig.   5c), is entirely consistent with the same loss process operating in both oceans. Our calculations in Supplementary Discussion 2 indicate that while Cd loss from the Cd-rich tropical Pacific would not produce a resolvable  114 Cd elevation at the Cd* minimum, the same Cd* decrease in the nutrient-poorer Atlantic would increase  114 Cd by 0.11-0.26‰, entirely in keeping with the observations (Fig. 7c). Thus, although the tropical Atlantic Cd*- 114 Cd correlation primarily results from the preformed distribution rather than Cd loss (as in the Pacific; Fig. S5), tropical Cd loss does extend this correlation to slightly more extreme values than in the southern subtropics ( Fig. S10; Conway and John, 2015a;Xie et al., 2019a;Guinoiseau et al., 2019). However, this signal is not restricted to O 2 -poor OMZ waters, but is visible throughout the Atlantic tropics (Figs. 8, S8). Furthermore, the  114 Cd shift is too small for it to be diagnostic of any one process such as sulphide formation or biological uptake.

Synthesis: Cd systematics in the low-latitude Pacific and Atlantic
We have shown that the covariation of Cd* and  114 Cd observed in low-latitude OMZs of the Atlantic and Pacific comes about primarily due to the coupling of Cd* and  114 Cd by phytoplankton uptake in the Southern Ocean, and not because of CdS precipitation. Nonetheless, particulate elemental and limited dissolved isotopic data appear to support the existence of a Cd loss process that creates a stronger Cd* minimum in the tropical oxycline of the Pacific and Atlantic (Sections 3.3 and 3.4). However, this process is not limited to eastern-boundary OMZs, as has been previously proposed (Janssen et al., 2014;Conway and John, 2015a;Janssen et al., 2017;Guinoiseau et al., 2019), but instead seems to extend throughout the tropics (Figs. S8 and S9). We suggest that the influence of this loss process is masked in the Peruvian OMZ by the remineralisation of Cd-rich biogenic particles (Section 3.3.2; Black et al., 2019), while in the Atlantic OMZs, the lower Cd:P ratio of sinking particles (Bourne et al., 2018) either exacerbates the reduction in Cd* or plays a neutral role (Section 3.4.2). The difference in the Cd* systematics between the Atlantic and South Pacific OMZs can thus be largely explained by differences in the Cd:P stoichiometry of exported particles. Below, we consider the drivers and consequences of this difference (Section 3.5.1) before discussing possible mechanisms of Cd loss in the tropical oxycline and identifying questions to be addressed by future research (Section 3.5.2).

Particulate Cd:P systematics and their influence on the Cd distribution
Euphotic-zone particulate Cd:P ratios reach much higher values (up to 1.42 mmol/mol) in the upwelling region above the South Pacific OMZ than above the North Atlantic OMZ (up to 0.36 mmol/mol; Bourne et al., 2018). These data are from GEOTRACES sections that extend from upwelling regions into the open oligotrophic tropics (GP16) or subtropics (GA03), where euphotic-zone particulate Cd:P ratios are low, with median values of 0.21 mmol/mol (n=11) and 0.09 mmol/mol (n=8) respectively (Bourne et al., 2018). Figure 9 compiles these euphotic-zone particulate compositions with euphotic-zone average dissolved [Cd] from seawaters sampled on the same cruises, calculated from IDP2017v2 data (Schlitzer et al., 2018). Given the simplicity of this comparison and the complexity of the phytoplankton ecosystem changes between upwelling regions and the oligotrophic ocean, Fig. 9 reveals an astonishingly coherent relationship between [Cd] and particulate Cd:P. In general, particulate Cd:P increases with [Cd], with a non-linear dependence that can be statistically described as either logarithmic or following saturating Michaelis-Menten dynamics (curves in Fig. 9a,b). However, the relationship is most systematic at low [Cd], below ~10 pmol/kg in both oceans (Fig. 9c); at higher [Cd] there is significant scatter. While culturing studies have shown an increase in eukaryotic Cd quota with dissolved inorganic Cd Huntsman, 1998, 2000), limited observations indicate that ~70% of Cd is organically chelated in the subtropical surface (Bruland, 1992). The systematic relationship between particulate Cd:P and [Cd] in the oligotrophic waters may thus reflect relatively constant concentrations of Cd-chelating ligands, especially in the North Atlantic subtropics where euphotic-zone Cd:P and [Cd] are tightly coupled (Fig. 9b,c).
More generally, the oligotrophic systematics suggest that phytoplankton take up more Cd as its dissolved concentration increases, perhaps due to the non-specificity of divalent metal transporters, or targeted Cd uptake at the low dissolved [Zn] of the low-latitude ocean (Sunda and Huntsman, 2000;Lane et al., 2009).
At higher [Cd], both the shift towards higher particulate Cd:P as well as the larger scatter most likely result from ecosystem shifts between oligotrophic and upwelling-influenced regimes, since (a) there are clear group-level differences in phytoplankton Cd quota (Lane et al., 2009), and (b) within upwelling-influenced regimes, phytoplankton community structure shifts towards larger eukaryotes (e.g. Franz et al., 2012) with elevated Cd quotas (Twining and Baines, 2013), as reflected by higher particulate Cd:P in the equatorial Pacific during upwelling conditions (Bourne et al., 2018). However, most broadly, the factor ~4 higher particulate Cd:P observed in the South Pacific is associated with euphotic-zone [Cd] an order of magnitude higher than in the North Atlantic (Fig. 9a,b,c). Similar covariation between co-sampled euphotic-zone [Cd] and particulate Cd:P has recently been observed in the Southern Ocean (Cloete et al., 2021). Thus, while inter-basin differences in species and ecosystem structure as well as (micro)nutrient status Huntsman, 1998, 2000;Cullen et al., 2003;Lane et al., 2009;Bourne et al., 2018) likely play some role in determining the difference in particulate Cd:P between the Atlantic and Pacific, it appears that the difference in [Cd] in these two upwelling regions is a first-order driver of differences in particulate Cd:P stoichiometry.
The inference of a [Cd] control on particulate Cd:P in the low latitudes of the Atlantic and Pacific is directly analogous to Middag et al.'s (2018) invocation of a [Cd] control to explain differences in the Cd:P of export between the high latitudes of the Southern Ocean and subarctic North Atlantic. Indeed, biological Cd:P plasticity may be by far the most important process shaping the global-scale marine Cd (and Cd*) distribution. As has been previously argued (Saager and de Baar, 1993;Sunda and Huntsman, 2000;Cullen et al., 2003;Baars et al., 2014;Xie et al., 2015;Roshan et al., 2017;Roshan and DeVries, 2021) the high Cd:PO 4 uptake of Southern Ocean phytoplankton decouples the global Cd and PO 4 distributions, through its influence on preformed Cd and PO 4 . This preformed structure is primarily modified by low-latitude productivity in the tropics, where export fluxes are high. Here, biological Cd:P plasticity plays a role once again: in the Cd-poor tropical Atlantic, the remineralisation of particles with low Cd:P (Twining et al., 2015;Bourne et al., 2018;Middag et al., 2018) tends to lower Cd* in the interior (Roshan and DeVries, 2021); while in the Cd-richer Pacific, exported particulates generally have higher Cd:P (Bourne et al., 2018;Black et al., 2019) and tend to raise Cd*, especially below upwelling regions where ample Cd is supplied to surface ecosystems. Thus, variability in the Cd:P of biological uptake and export at both high and low latitudes, potentially driven by [Cd], will tend to decouple the Cd and PO 4 distributions, producing signals in the interior Cd* distribution that -though solely reflecting biological cycling -may mimic Cd loss.

Mechanisms driving tropical Cd loss
Identifying the process that drives the putative Cd loss we identify in the tropical oxycline is exceedingly difficult based on available data. What is clear is that the Cd* minimum is associated with particulate Cd maxima in the open tropical Pacific ( Fig. 6; Ohnemus et al., 2019) as well as the North Atlantic OMZ (Conway and John, 2015a). Ohnemus et al. (2019) attributed the open tropical Pacific particulate maxima to the presence of heterotrophic prokaryotes, while Janssen et al. (2014) and Conway and John (2015a) attributed the North Atlantic maxima to CdS precipitation within decomposing particles. In the anoxic Peruvian OMZ, where active sulphur cycling is known to occur (Canfield et al., 2010), Ohnemus et al. (2017) argue that the stoichiometric ratio of ≤1 mol:mol observed between particulate Cd and acid-volatile sulphides is consistent with, but not convincing evidence for, a sulphide carrier of particulate Cd there.
However, continued cryptic sulphur cycling in the O 2 -richer oxygen minima (50-150 mol/kg) of the wider tropical Atlantic and Pacific is highly unlikely. Furthermore, the Cd* minimum with its associated particulate Cd maximum is observed even in oligotrophic regions, where sinking particulate fluxes are low (Honjo et al., 2008). Given the apparent dependence of particle-hosted sulphate reduction on productivity (Raven et al., 2021), it appears that CdS precipitation within the sulphidic cores of large sinking particles, as proposed by Janssen et al. (2014) and simulated by Bianchi et al. (2018), cannot be invoked as an explanation for the shallow Cd* minimum associated with the oxycline throughout the tropics.
On the other hand, the attribution of the open tropical Pacific particulate Cd maxima to heterotrophic prokaryotes is based on statistical analysis of particulate chemistry data (Ohnemus et al., 2019), and remains to be confirmed through more detailed particle characterisation. Although speculative, it is worth noting that the tropical Cd* minima are sometimes shallow enough to overlap with elevated fluorescence in the tail of the deep chlorophyll maximum (e.g. Fig. S11). Genetic data indicate the presence of low-lightadapted ecotypes of the ubiquitous low-latitude cyanobacterial phytoplankton Prochlorococcus at depths of ~100m throughout the tropical Atlantic (Johnson et al., 2006) and in the Peruvian OMZ (Franz et al., 2012), i.e. within the depth range of the Cd* minimum in regions of upwelling-related isopycnal uplift.
This raises the possibility that autotrophic prokaryotic shade flora may have a role to play in tropical Cd loss (Ohnemus et al., 2017), at least at shallower levels within the euphotic zone. While prokaryotic phytoplankton generally take up only small amounts of chalcophilic metals like Cd (e.g. Saito et al., 2003), laboratory cultures have shown that under low light, the Cd:P of a cyanobacterial species (Cyanothece sp.) increases to ratios similar to those of eukaryotes (Finkel et al., 2007). It is unknown whether this result translates to Prochlorococcus, or whether it would apply to low-light-adapted ecotypes. However, in the subtropical North Atlantic, Twining et al. (2015) also report higher labile particulate Cd:P ratios in the deep chlorophyll maximum than in the mixed layer. Clearly, we are at an early stage of understanding tropical Cd loss mechanisms, and future work that enables the simultaneous characterisation of microbial ecosystem functioning (e.g. through 'omics approaches or analysis of environmental DNA; e.g. Ruppert et al., 2019;Debeljak et al., 2021), of dissolved phase behaviour, and of particulate hosting phases, will be required to conclusively identify the driving process(es).

Implications for marine Cd mass balance
Our investigation of the ocean-internal cycling of Cd has two sets of implications for the role of watercolumn CdS formation in the marine Cd budget. But first, we must distinguish between two marine CdS formation regimes. One regime exists within functionally anoxic OMZs, in which H 2 S may be present in the water column in trace to micromolar amounts. For example, Plass et al. (2020) have documented Cd loss from bottom waters above sulphidic sediments on the Peruvian shelf (0.4-9.5 M H 2 S in near-surface porewater). This observation, presumed to result from CdS precipitation in the presence of trace H 2 S, is consistent with previous observations of Cd loss associated with episodic H 2 S release into the water column here (Schunck et al., 2013;Xie et al., 2019b) and, more generally, with enhanced accumulation of Cd in upwelling-margin and euxinic-basin sediments (Little et al., 2015;Chen et al., 2021). The second regime involves CdS precipitation in sulphidic microenvironments within sinking particles (Janssen et al., 2014), and it is this regime to which our discussion is directly pertinent. Plass et al. (2020) estimate that CdS precipitation (either in near-bottom waters or via shallower particleassociated processes) supplies a minimum of 28-67% of excess Cd in the sediments they studiedincluding in non-sulphidic sediments situated in oxygenated waters below the depth of the Peruvian OMZ (750m water depth, >5 M bottom-water O 2 ). A major supply of CdS to non-sulphidic sediment suggests a rain of particle-hosted CdS formed at the upper oxycline (Janssen et al., 2014), 700m above the seafloor.
However, Plass et al.'s (2020) CdS flux estimate hinges critically on the estimated supply of Cd from biogenic particles, calculated by scaling organic carbon (C) fluxes with an "average phytoplankton" Cd:C of 1.69 mol/mol (Moore et al., 2013). This value, derived from a Cd:P of 0.21 mmol/mol (Ho et al., 2003), stands in stark contrast to the Cd:P of ~0.95 mmol/mol in euphotic-zone particles (Bourne et al., 2018) and the export flux (Black et al., 2019) over the Peruvian shelf. Recalculation using this site-specific stoichiometry suggests that biogenic particles can in fact easily account for all excess Cd found in nonsulphidic sediment deeper than the OMZ (Table S1). Thus, there is little need to invoke a sulphide carrier of Cd to non-sulphidic sediments of the Peruvian margin.
Two complexities must however be noted with regard to our recalculation. The first is that particulate fields vary temporally and spatially; e.g., the Cd:P of export over the Peruvian shelf and slope varies from 0.52 mmol/mol to 1.11 mmol/mol (Black et al., 2019; Table S2), variability that propagates into our calculations (Table S1). Secondly, particulate Cd:P is not conserved through the water column: differential remineralisation decreases Cd:P with depth (Bourne et al., 2018;Cloete et al., 2021), while within the OMZ, particulate Cd:P may increase due to prokaryotic trace metal accumulation (Ohnemus et al., 2017) or Cd adsorption (Lee et al., 2018). How strongly such secondary processes affect the Cd:P of the sinking flux is difficult to quantify, but they will certainly have a larger effect over the deeper water column above the non-sulphidic slope sediments than on the shelf. Large particle (>51 m) data from the Peruvian OMZ suggest that Cd:P decreases by ~50% between 50m and 500m (Lee et al., 2018), perhaps explaining why the Cd:P of export (Black et al., 2019) can explain up to ~4 times the excess Cd in slope sediments (Table   S1).
Most broadly, however, our inference that no CdS carrier is needed to explain excess Cd in non-sulphidic sediments of the Peruvian margin is consistent with observations in similar sediments in the South Atlantic (Bryan et al., 2021). A major biological contribution to excess Cd in Peru margin sediments is also in line with the finding by Chen et al. (2021) that organic matter can account for 50-100% of excess Cd in analogous sediments of the eastern tropical North Pacific. Furthermore, the fact that the low "average phytoplankton" Cd:P of 0.21 mmol/mol can explain excess sedimentary Cd in the Cd-poor Atlantic (Bryan et al., 2021) but not the Pacific (Plass et al., 2020) is consistent with the systematics of Fig. 9 as well as the inter-basin differences in particulate Cd:P observed directly (e.g. Bourne et al., 2018) or inferred from water-column data Roshan and DeVries, 2021). These observations highlight the importance of considering biological metal-quota plasticity in sedimentary studies, as has recently been recognised for Cd and other metals (Chen et al., 2021;Plass et al., 2021).
The fact that there is no need to invoke a sulphide carrier of Cd to sediments deeper than OMZ extent, when combined with the potential biological mechanism of particulate Cd accumulation within tropical OMZs (Ohnemus et al., 2017), has implications for a recent estimate of particle-hosted CdS formation (Bianchi et al., 2018) and the associated marine Cd sink (Guinoiseau et al., 2019). The parameters of Bianchi et al.'s (2018) particle model, such as the rate constants for CdS formation and dissolution, were optimised against a particulate Cd profile from the North Atlantic OMZ (Janssen et al., 2014), and converged to a dissolution rate constant 1.5-33× higher than experimental or theoretical values, as noted by Guinoiseau et al. (2019).
This high value suggests either that marine CdS is more labile than CdS of other origins in other media, or that the optimisation required a high dissolution rate to rapidly attenuate an overestimated CdS poolperhaps because some portion of the observed particulate Cd is actually due to prokaryotic Cd uptake (Ohnemus et al., 2017). Each of these possibilities suggests that particle-hosted CdS formation will lead to only a small net loss of marine Cd to sediment. Indeed, calculating an internally-consistent estimate of the net CdS sink resulting from the CdS formation flux estimated by Bianchi et al. (2018) yields a value of 25mol/yr -1.8×10 7 mol/yr (Table S3). These estimates range from entirely negligible for the marine Cd budget to the magnitude of other estimates of Cd loss to suboxic or anoxic sediments (1.5-12.0 x 10 7 mol/yr; Bryan et al., 2021;Chen et al., 2021). Thus while it may be non-negligible, the net loss of Cd associated with particle-hosted CdS formation is unlikely to be the dominant Cd sink in the modern ocean, and is at most at the lower end of the range estimated by Guinoiseau et al. (2019). While there is no simple connection between the two, this inference is in keeping with our analysis in Sections 3.2-3.4, which shows that dissolved data previously interpreted as documenting widespread particle-hosted CdS formation in fact mainly reflect large-scale controls on the Cd distribution, with only a minor role for low-latitude loss processes.

Conclusions
Correlations between Cd* and  114 Cd previously interpreted as evidence for widespread particle-hosted CdS precipitation in OMZs in fact primarily reflect the biologically-controlled preformed distributions of these tracers. Indeed, given such oceanographic structure in their distributions, non-dimensional cross-plots of Cd* against  114 Cd have little value for process identification. We have also shown that remineralisation of biogenic particles, and specifically the large range in their Cd:P stoichiometry, is the most important modifier of the preformed distributions of Cd and Cd* (its influence on  114 Cd is very limited). In particulate elemental and dissolved isotopic data, we find subtle evidence for Cd loss from the oxyclines of the tropical Pacific and Atlantic. An assessment of the global extent of this signal, and its origin, would be aided by higher-resolution dissolved Cd data from the tropical Indian Ocean or the eastern tropical North Pacific OMZ, especially if complemented by particulate and/or isotopic Cd data. Although the process(es) driving this loss remains unclear given the data currently available, it is conceivable that it is biological in origin (Ohnemus et al., 2017;. Regardless of its driving mechanism, the data indicate that Cd loss is ubiquitous in the tropical oxycline, and not confined to the oxygen-poor or -depleted waters of the eastern basins. This realisation implies that the evidence of OMZ particulate Cd data for particle-hosted CdS precipitation is more ambiguous than previously recognised (Janssen et al., 2014;Conway and John, 2015a).
Global extrapolations of these data in order to estimate the magnitude of the water-column CdS sink (Bianchi et al. 2018) and its importance for the marine Cd budget (Guinoiseau et al., 2019) may have overestimated the influence of particle-hosted CdS formation. Furthermore, by considering the extreme Cd:P plasticity in phytoplankton, we show that even in the anoxic Peruvian OMZ underlain by sulphidic sediments that scavenge bottom-water Cd as CdS, Cd supply to the sediment by sinking biogenic particles may contribute significantly (15-90%) to sedimentary Cd, consistent with recent findings in Atlantic and Pacific sediments (Bryan et al., 2021;Chen et al., 2021). Together, this evidence indicates a small or negligible role for particle-hosted sulphide formation in the marine mass balance of Cd.
2017v2. Paul Quay kindly allowed us to use his Pacific  13 C DIC data in the GEOTRACES IDP2017v2, and Mathieu Waeles generously shared his Angola Basin data. We thank Reiner Schlitzer for development of Ocean Data View 5 (Schlitzer, 2019) which aided exploratory data analysis and preparation of figures.
Careful and constructive reviews by Ruifang Xie, Damien Guinoiseau and two anonymous reviewers, and the editorial oversight of Claudine Stirling, helped to improve the manuscript. This research was supported by ETH Zurich and did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. SHL is currently supported by a NERC independent research fellowship (NE/P018181/2).

Author contributions
GFdS conceived the study together with MS, TMC and DV, undertook the data analysis, and wrote the first draft. SHL contributed especially to Section 3.6; DV contributed especially to development of the manuscript. All co-authors contributed to the development of the manuscript in preparation for submission.

Data statement
The data considered in this study have been previously published by other workers, and almost all are freely available. Most data are to be found in the GEOTRACES Intermediate Data Product 2017v2 at https://www.bodc.ac.uk/geotraces/data/idp2017/ or the GEOTRACES Intermediate Data Product 2021 at https://www.bodc.ac.uk/geotraces/data/dp/. Data from Guinoiseau et al. (2019) are freely available at https://dx.doi.org/10.1029/2019GB006323, the North Pacific data of Yang et al. (2018)