The global ocean holds one of Earth's major carbon reservoirs as dissolved organic matter (662 ± 32 PgC). Most of this material (>95%) is termed refractory dissolved organic carbon (RDOC) as Williams and Druffel (1987) found it to be old relative to the circulation time of the ocean. While RDOC within the modern ocean is thus perceived as vast and only slowly renewed, its mobilization has been implicated by Sexton et al. (2011) to explain Earth's transient warming events (i.e., hyperthermals) of the Paleocene and Eocene epochs (65–34 million years ago). Assessing this proposed function of RDOC as a rapidly (~5–10 kyr) exchangeable carbon reservoir is presently limited by insufficient knowledge of the responsible processes. Here we investigate the dynamics of RDOC in the deep Pacific Ocean, previously characterized by concentration gradients thought to be established by slow but systematic RDOC removal with circulation and aging of the water masses. We demonstrate that RDOC is instead conserved during much of its circulation, but that there exist localized sinks in the deep, far North Pacific and at mid depth in the subtropical South Pacific. Water mass mixing into these sink regions creates the observed RDOC gradients. Together, the Pacific sinks remove 7–29% of the 43 Tg RDOC added to the deep global ocean each year with overturning circulation, and point to an important but still unidentified control on the RDOC inventory of deep marine systems.
 Marine DOC exhibits a spectrum of reactivity, from very fast turnover of the most bioavailable forms in the surface ocean to long-lived materials entrained in abyssal circulation [Hansell, 2013]. These disparate reactivities differentiate DOC fractions with distinctive functions in the carbon cycle, ranging from rapid (daily) turnover in support of vast marine heterotrophic microbial populations, to decadal/centennial sequestration of carbon, to a hypothesized major source/sink of atmospheric CO2 driving paleoclimate variability [Sexton et al., 2011]. The DOC fractions that exhibit daily to decadal turnover sum to a relatively small mass (20 ± 3 PgC) [Hansell, 2013; Hansell et al., 2012], so these pools are important in terms of carbon flux and as substrates to microbial processes [Carlson, 2002] but not as Earth's major carbon sequestration reservoirs. The 642 ± 32 Pg pool of refractory DOC (RDOC), in contrast, plays a relatively modest role in the carbon cycle of the modern ocean over decadal time scales, but when paleoclimatic variations require a large, readily exchangeable reservoir of carbon, this pool is a plausible candidate. Ultimately, understanding the past and future roles of RDOC in climate depends on illumination of its dynamics in the modern ocean.
 RDOC is present at concentrations ranging from ~34 to <45 μmol kg-1 [Hansell et al., 2012]. The pool is highly resistant to microbial turnover [Barber, 1968], yet a concentration gradient of ~14 μmol kg-1 exists between the North Atlantic and the North Pacific [Hansell and Carlson, 1998], indicating that sinks (i.e., removal processes) exist. When DOC concentrations were correlated with radiocarbon ages of deep Pacific waters, a removal rate of ~0.003 µmol kg-1 yr-1 was determined, consistent with the modeled rate required to reproduce observed deep ocean DOC on the global scale [Hansell et al., 2009]. However, inconsistencies between observed and modeled DOC in some locales were noted, particularly in deep equatorial waters where observed DOC concentrations typically exceeded modeled concentrations. Resolving these discrepancies led to the analysis reported here, with our finding that the RDOC removal rate previously described depicts the system incorrectly.
2 Data Employed
 All data used in this analysis are publicly available: Key et al.  provided salinity at the 4000 m isopleth in the Pacific Ocean, as shown in Figure 1. Feely et al.  provided the data shown from the meridional CLIVAR Repeat Hydrography section P16S/N, the location of which is shown in Figure 1.
3 Observational Context
 That RDOC sinks exist in the deep Pacific Ocean is evidenced by concentration gradients [Hansell et al., 2009]. To evaluate those, salinity is employed here, as its distribution conservatively traces water mass circulation and mixing; deviations between salinity and DOC distributions illuminate the sinks. In the Pacific, net circulation of deep and bottom waters constitute a meridional cell, with high-salinity Circumpolar Deep Water (CDW) of the Southern Ocean (with a strong North Atlantic Deep Water component) flowing northward along the bottom, thus ventilating the system as a deep western boundary current east of New Zealand (Figure 1). The water then moves north across the equator near the dateline. During its flow into the far North Pacific, the water is warmed and freshened (note lowered salinity; Figure 1), increasing buoyancy and rising (Figure 2a). Freshening occurs by vertical mixing with the low-salinity, subarctic intermediate waters (IW, located at <1500 m) that mix downward by tidally induced turbulence over rough topography including the continental margins [Ledwell et al., 2000]. Mixing of IW and CDW forms the intermediate-salinity, high-silicic-acid water mass North Pacific Deep Water (NPDW; Figure 2a), which moves southward at mid depths [Schmitz, 1996]. As the deep basin of the Pacific Ocean does not experience exchanges with basins other than the Southern Ocean, water mass mixing is largely between those waters described above (i.e., CDW, IW, and NPDW).
4 Water Mass Mixing and RDOC Sinks
 Using the conservation of salinity, NPDW (salinity ~34.65) is estimated to consist of ~97% high-salinity CDW (~34.71) and ~3% low-salinity IW (~34). Conservative mixing of DOC held in these water masses (end-member values: CDW DOC ~40 µmol kg-1; IW DOC ~45 µmol kg-1 DOC [Hansell et al., 2002]) would result in NPDW with DOC ~40 µmol kg-1 (this product of conservative mixing, as estimated from salinity, is here termed DOCsal). The meridional distribution of DOCsal at >4000 m on P16 (Figure 3a) is invariably ~40 µmol kg-1 because CDW dominates (>97%) those near-bottom waters. But observed DOC concentrations (DOCobs) in the far North Pacific are ~37 µmol kg-1 (Figures 2b and 3a), indicating a DOC deficit (i.e., deviation from conservation) of ~3-4 µmol kg-1 in that system (Figure 2c).
 The responsible DOC sink exists either within and during the centennial-scale, basin-wide circulation and mixing [Stuiver et al., 1983; Van Aken, 2007] of the near-bottom waters (such that the DOC is slowly removed with time [Hansell et al., 2009]) or, alternatively, in more localized regions. The sink dynamic can be isolated by comparing DOCobs to DOC concentrations expected by conservative mixing between CDW and, in this case, NPDW (the conserved product is again termed DOCsal). A sink that exists within the basin-wide circulation will appear as a deviation between DOCobs and DOCsal, as previously observed in Figure 3a. If on the other hand the distributions of DOCobs and DOCsal are identical (within analytical uncertainty), then the deep DOC gradient must result from conservative mixing of DOC within the water mass end-members (i.e., CDW and NPDW).
 In Figure 3b, DOCsal is indistinguishable from DOCobs (see statistical analyses in section 7 below and in Figure 4), indicating that conservative mixing between high-DOC CDW and low-DOC NPDW creates the deep DOC gradient. This result requires that the DOC sink be regionally localized in the far North Pacific, perhaps associated with the formation of NPDW. It is this sink (and subsequent conservative mixing of NPDW with CDW) that creates the DOC concentration gradient observed in the deep waters (Figures 2b and 3b).
 A DOC sink is evident in the midwater column (~1500–3500 m) of the subtropical South Pacific as well, where concentrations reach <35 µmol kg-1 (Figure 2b). The dynamics of DOC there are similarly tested using salinity as the conservative tracer. Following formation, NPDW is transported to the South Pacific (Figure 2a), isopycnally mixing with waters originating in the southern hemisphere. DOCsal is estimated from two end-member mixing between those two water masses (end-members: northern component NPDW salinity ~34.65 and DOC ~37 µmol kg-1; southern component salinity 34.73 and DOC ~ 40 µmol kg-1); mixing is evaluated on the isopycnal surface 41.43 < σ3 < 41.445 (~3000 m; Figure 2a) as it lies near the core of NPDW. There is consistency between DOCobs and DOCsal north of the equator (Figures 3c and 4b), but DOCsal at 10–46°S is ~38 µmol kg-1 while DOCobs reaches <35 µmol kg-1, with their trends against latitude unequal (Figure 4c). The DOC deficit in this zone reaches as high as ~4 µmol kg-1 (Figure 2c), or an ~10% reduction in DOC relative to its CDW source. Between 40°S and 60°S, DOCobs and DOCsal are similar (Figure 3c), indicating conservative downward mixing of the DOC-enriched upper-layer waters.
5 RDOC Removal Rates
 The rate of RDOC removal in these two sinks can be estimated from water mass transport rates and DOC concentration changes. Estimates for the net northward transport of bottom water into the North Pacific, supporting the formation of NPDW, range from 1.5 to 5 Sv (1 Sv = 106 m3 s-1) [Schmitz, 1996; Ganachaud and Wunsch, 2000; Macdonald, 1998]. DOC removal associated with this formation (at ~3 µmol kg-1) lies between 1.7 and 5.7 TgC yr-1. In the South Pacific, NPDW is transported at 4–9 Sv [Schmitz, 1996] and experiences an additional ~1–2 µmol kg-1 loss of DOC (1.5–6.8 TgC yr-1). Summing these sinks (3.2–12.5 Tg yr-1) indicates that 7–29% of the 43 Tg yr-1 of DOC introduced to the deep ocean each year to maintain its global inventory [Hansell et al., 2012] is removed in these two systems.
 Other RDOC sinks must exist elsewhere in the deep global ocean, with anomalously low DOC concentrations indicating those locations. The deep Mediterranean Sea, for example, holds DOC concentrations of <38 µmol kg-1 [Santinelli et al., 2012]. As those deep waters are formed locally with surface waters of North Atlantic origin (with elevated DOC concentrations), and as the concentrations at <38 µmol kg-1 are lower by 2–3 µmol kg-1 than found in the adjacent deep North Atlantic [Carlson et al., 2010], an RDOC sink is apparent in that system.
6 Sinks of Unknown Mechanisms
 UV oxidation of the RDOC pool has been described as an important sink in the modern ocean [Mopper and Kieber, 2002], but it is a process limited to the surface layer and so not involved with the deep sinks described here. The mechanisms for deep RDOC removal observed here are unknown. Both Pacific sink regions are bordered by ocean ridges and/or margins (Figure 1), and both are within gyre circulations [Reid, 1997]. In the deep North Pacific, the lowest DOC concentrations coexist with the lowest salinity (Figures 2a and 2b), with this low-salinity feature emanating from the continental margin of the NE North Pacific (Figure 1). Downward mixing of low salinity IW is likely enhanced at the margins by tide-induced turbulence [Ledwell et al., 2000]. In the subtropical South Pacific, the low-DOC waters circulate within an anticyclonic gyre [see Reid, 1997], guided southward along the ocean ridge located near the dateline and then eastward by a ridge (the Chatham Rise) at 45°S (Figure 1). Turbulence associated with these circulations likely suspends seafloor particles, which may serve as surfaces for DOC adsorption [Druffel et al., 1998].
 A strong sinking particle flux from highly productive surface oceans, applicable to the subarctic North but not the subtropical South Pacific, offers three mechanisms for removing RDOC. The release of exoenzymes by particle-attached microbial assemblages [Karner and Herndl, 1992; Arnosti, 2011] could lead to the breakdown and subsequent uptake of recalcitrant compounds [Williams, 2000]. Alternatively, solubilization of sinking particles would add bioavailable DOC to the deep-water column [Kiørboe and Jackson, 2001; Azam and Long, 2001; Nagata et al., 2010] that could support cometabolism (priming). It has been hypothesized that labile solubilized substrates provide energy to the free-living or particle-attached heterotrophic prokaryotes to produce hydrolytic exoenzymes for the cometabolism of RDOC [Madigan et al., 1997]. This priming effect is observed in soils and aquatic systems, where increased remineralization of recalcitrant organic matter occurs after input of biologically labile materials [Carlson et al., 2002; Bianchi, 2011]. However, incorporation of RDOC into microbial biomass is not discernable in the deep North Pacific [Ingalls et al., 2006; Hansman et al., 2009], indicating that if cometabolism is occurring, then little is assimilated into microbial biomass. A third mechanism is that sinking particles scavenge RDOC as they sink, but there is a much larger and deeper flux of particles in the subarctic than the subtropical Pacific [Buesseler et al., 2007], so exported particles should not play equivalent roles in the two systems.
 Hydrothermal systems expelling water that has been altered within Earth's crust should influence RDOC distributions in the deep ocean, although the extent of influence has not been established. Hydrothermal fluids circulate through the upper oceanic crust, especially on ridge flanks [Johnson and Pruis, 2003] where fluids flow through porous basalts. Outflows on the Juan de Fuca Ridge have low DOC concentrations (~10–15 μM) [Lang et al., 2006], indicating RDOC removal during the fluid's 10 kyr crustal residence time [Mottl, 2003]. Alternatively, as hydrothermal plumes release chemically reduced fluids into an oxygenated water column, redox disequilibria may provide energy for free-living chemoautotrophs [Lam et al., 2004; McCarthy et al., 2011] whose production could prime the removal of RDOC.
 Finally, self-assembling organic gels that contribute to particle formation may constitute an RDOC sink. Biopolymers such as DOC, gels, and transparent exopolymers [Wells, 1998; Carlson, 2002; Passow and Alldredge, 2004] can move organic molecules up the particle size spectrum to sizes capable of sinking through the water column [Verdugo et al., 2004; Engel et al., 2004] and/or of being mineralized by the resident microbes.
7 Statistical Analyses of Correlations
 The slopes and intercepts of the linear regression of DOC observed (DOCobs) and DOC modeled from salinity (DOCsal) against latitude were compared. For data from >4000 m, Figure 4a shows that both linear regression models were significant at p < 0.0001. The estimated slope of the linear regression of DOCsal versus latitude (DOC = 38.4 − 0.03*Latitude, n = 381) lies within the 95% confidence interval (−0.026 to −0.034) of the slope of the linear regression of DOCobs versus latitude (DOC = 38.6 − 0.03*Latitude, n = 382). The intercepts of the two regression lines are within 0.2 µmol kg-1 of each other. The agreement between these two linear regression models indicates that the observed DOC distributions are consistent with basin-wide conservative mixing of RDOC held within CDW and NPDW. The observed values have more scatter than those predicted, reflecting less analytical precision for DOC (coefficient of variation (CV) of 2–3%) than for salinity (CV 0.003%).
 The same analysis was conducted within the isopycnals that define the core of the NPDW. We examined how the models compared for the stations north of the equator (0–56°N) and from 10°S to 46°S. In the North Pacific, the linear regression model of the observed DOC (DOCobs = 37.7 − 0.015*Latitude; n = 71, p = 0.0035) and salinity-modeled DOC (DOCsal = DOC = 37.9 − 0.017*Latitude; n = 70, p < 0.0001) were both significant (Figure 4b). The slopes of the modeled DOC concentrations were within the 95% CI (−0.023 to −0.005) of the observed DOC versus latitude, with the intercepts being within 0.2 µmol kg-1 of each other. Agreement between these two linear regression models indicates that the observed DOC distributions are consistent with conservation of RDOC in NPDW north of the equator.
 In the South Pacific between 10°S and 46°S, the linear models of observed and predicted DOC concentration versus latitude were not consistent with each other (Figure 4c). The linear model between DOCsal versus latitude (DOC = 37.51281 − 0.0149233*Latitude [°N], n = 23, p = 0.0007) was significant while the DOCobs versus latitude (DOC = 36.390234 − 0.0130115*Latitude [°N], n = 24, p = 0.485) was not; there was no overlap of the 95% CI. The lack of agreement between these two linear models indicates that the observed DOC did not behave conservatively in this latitudinal range in the South Pacific.
 Elevated DOC concentrations were found associated with the high-salinity waters of CDW in the Southern Ocean. The elevated salts in CDW ultimately derive from saline North Atlantic Deep Water (NADW) [Reid, 2005], so it is likely that the elevated DOC in CDW has the same source. Much of the RDOC concentration gradient in the global deep ocean may exist by conservative mixing between the RDOC-enriched North Atlantic Deep Water and globally distributed but localized sites of RDOC removal. Further analyses, including compositional and isotopic analytical techniques, need to be conducted on the global scale to determine if RDOC exported with NADW is essentially conserved until it reaches localized sinks.
 The discrepancy between observed and modeled DOC in the deep ocean noted above stems from erroneous model assumptions used by Hansell et al. . The correlation between DOC and water mass ages determined in that work assumed single end-member mixing; it appeared from that result that RDOC was removed slowly (0.003 µmol kg-1 yr-1) but steadily, likely everywhere in the deep ocean. The removal rate was then applied in the model, leading to the discrepancies. But the results reported here inform us that, in fact, RDOC is conserved over great distances of the deep Pacific and that mixing into localized sinks creates the concentration gradients.
 The relevance of these RDOC sinks to the carbon that forced variability in paleoclimate remains unknown until two important determinations are made. First, the controls on the sinks need to be identified: what changes in the ocean system would cause acceleration of RDOC loss required by the hyperthermals of the Paleocene and Eocene, where atmospheric CO2 concentrations changed rapidly (over ~5–10 kyr), and what changes would reverse this loss in order to rebuild a large RDOC inventory (over ~30 kyr), setting the stage for the next massive release and associated climate hyperthermal. Second, the fate of the DOC removed from the water column must be identified. If the fate is mineralization, then RDOC is indeed an exchangeable pool of CO2 for the atmosphere. However, if particle scavenging of RDOC is the primary sink, such that carbon is sequestered in the sediments, then the carbon may remain buried over geologic timescales.
 Statistical analyses benefited from discussion with C. Nelson. This work was supported by the U.S. National Science Foundation grants OCE-OCE0752972 and OCE-1153930.