4.1. Evidence of Periodically Enhanced Methane Flux
 The anomalously low δ13C and Δ14C measured in the planktic and benthic foraminifer samples are the result of mixing between foraminifer calcite and authigenic carbonates associated with anaerobic oxidation of methane (AOM). This two-end-member mixing has been observed in foraminifera from modern active seeps, including at Hydrate Ridge [Orphan et al., 2004] and Papua New Guinea [Schmidt et al., 2002]. Not all foraminifera with low δ13C have authigenic overgrowths as obvious as the examples in Figure 5. A small amount of authigenic carbonate on or within the test could account for the isotopic anomalies. Based on the average δ13C of N. pachyderma (s.) during the glacial period, the estimated δ13C of the authigenic carbonates, and the mass mixing model, samples that have δ13C less than −2‰ are at least 8% authigenic carbonate by mass. The isotopic anomalies are larger for N. pachyderma (s.) than U. peregrina (Figure 3), probably due to the high surface area of a planktic foraminifer test, which is more likely to accommodate contamination by small crystals of authigenic carbonate than the imperforate test of U. peregrina.
 Since the two-end-member mixing model is also consistent with the stable isotope composition of U. peregrina, this means that the calcite test has an isotopic composition distinct from the authigenic carbonate minerals, and that the foraminifera must have calcified its test before the authigenic minerals precipitated. This is similar to observations at Hydrate Ridge [Torres et al., 2003], where Rose Bengal–stained benthic foraminifera had δ13C much higher than the pore waters in which they were collected, implying that the benthic foraminifera calcified their tests during periods of lower methane flux, or grew before active methane oxidation began. In contrast at southern Hydrate Ridge [Hill et al., 2004], individual Rose Bengal–stained benthic foraminifera had variable δ13C, with values as low as −12.6‰, and whose shells, studied with scanning electron microscopy, had no visible evidence of diagenesis or authigenic carbonates, implying the calcite in the foraminifer tests incorporated methane-derived carbon. It is possible that there is some variability in the δ13C of the U. peregrina tests that we studied, which would change the foraminifer calcite end-member δ13C. The current data are insufficient to test for this possibility, and many of our samples clearly have authigenic carbonates.
 The δ18O we estimate for the authigenic carbonate is 1.5–1.6‰ higher than mean δ18O of U. peregrina measured between low δ13C events in samples 20–25 ka. This enrichment in 18O is not likely to be caused by a decrease in the bottom water temperature, since today, the temperatures at the two coring sites are 2.2°C at 51JPC and 3.2°C at 57JPC, and a benthic δ18O record from 2209 m at Bowers Ridge exhibits no remarkable changes during MIS3 [Brunelle et al., 2007]. It is also unlikely to be due to water from dissociated methane hydrates, which is ∼3‰ enriched in 18O [Matsumoto and Borowski, 2000], since it would require pore water to be 50% hydrate water. This site is deep enough that even large changes in temperature are unlikely to cause significant destabilization of methane hydrates [Xu and Lowell, 2001].
 The mineralogy of the authigenic minerals is probably the main influence on their δ18O. In modern cold seeps, the authigenic carbonates can be composed of high-Mg (magnesian) calcite, aragonite and/or dolomite [Peckmann and Thiel, 2004]. The fractionation of oxygen isotopes as a function of temperature was determined experimentally by Kim and O'Neil  for calcite (at 10–40°C), by Kim et al.  for aragonite (at 5–40°C), and by Fritz and Smith  for dolomite (at 25–79°C). At low temperatures, the δ18O of aragonite and calcite are very similar, but dolomite δ18O is on average 2.6‰ higher. In magnesian calcite there is greater incorporation of 18O with higher molar fraction of Mg in the mineral [Tarutani et al., 1969]. The δ18O and Mg/Ca (see auxiliary material) of the authigenic minerals is consistent with the presence of dolomite and/or magnesian calcite.
 The δ13C of DIC of pore water influenced by AOM is strongly dependent on methane flux which will ultimately determine methane oxidation rate and the depth of the SMTZ [Borowski et al., 1996]. Our estimates of the δ13C of the authigenic carbonates (−24.1 and −22.4‰ in 57JPC and 51JPC, respectively) could be consistent with high AOM rates and a very shallow SMTZ or more moderate AOM rates and deeper SMTZ [Zeebe, 2007]. For example, they are similar to the δ13C of carbonate crusts from mud volcanoes from the eastern Mediterranean (−28.9‰ and −24.8‰ at the sediment-water interface [Aloisi et al., 2002]), and in sediments from the Chilean Margin (−24.6‰ at 355 cm below seafloor [Treude et al., 2005]).
 The DIC that originates from fossil methane deep within the sediment column is radiocarbon free [Winckler et al., 2002], and with high methane flux toward the seafloor, could incorporate carbon from younger carbon reservoirs [Kessler et al., 2008]. Samples of N. pachyderma (s.) with authigenic carbonates appear 2000–3000 years older than would be expected from the age model (Table 1), similar to observations of authigenic carbonates from the Japan margin [Uchida et al., 2008]. The lower Δ14C in the authigenic carbonates is equivalent to −350‰ and −345‰ in the present day. This is not as low as radiocarbon measurements on authigenic carbonates at modern seeps in the Gulf of Mexico (−898‰ to −992‰ [Aharon et al., 1997]), Aleutian accretionary margin (−860‰ [Greinert et al., 2002]) or the Black Sea (−918‰ to −901‰ BP [Peckmann et al., 2001]). However, the δ13C of the carbonates at these sites were lower than in this study, implying that they had a higher proportion of methane-derived DIC, and would be expected to have a carbon pool more depleted in 14C as well.
 Deviation of the isotope measurements from the theoretical mixing lines (Figure 4) could be the result of variations in the isotopic composition of the foraminifer and authigenic carbonate end-members from event to event. However, since we can use a single mixing model to describe the N. pachyderma (s.) and U. peregrina δ18O, δ13C and Δ14C in each core, the isotopic composition of DIC during the recurring episodes of authigenic carbonate precipitation were similar, even given the simplified mixing model. This suggests that the multiple events through time and in the two study regions may have drawn from the same carbon pool, and that these episodes were not strictly local, transient phenomena.
 Lipid biomarkers can give clues as to the intensity and location of AOM. Assuming that the δ13Cmethane here is similar to that observed by Claypool et al.  (mean = −72‰, σ = 4‰, n = 13), δ13Carchaeol is up to 4‰ lower than δ13Cmethane (Figure 9). At settings with intense AOM, δ13C values of archaeaol are typically lower than those of methane (by up to −50‰) [e.g., Peckmann and Thiel, 2004]. The most plausible interpretation is that the values observed here represent a mixed origin of archaeol of methanogenic and methanotrophic archaea. Production of these respective fractions likely did not occur simultaneously. Rather, we suggest that the methanotrophic portion formed during a punctuated event when conditions where conducive of AOM, while methanogenic production of archaeol extended over longer periods after burial in the subsurface. This interpretation is consistent with the concentration profile that suggests methanogenic background production of archaeol throughout most of the sediment column, and concentration maxima that coincide with increased 13C depletion. A similar mixed origin was inferred in a late Quaternary record from the Santa Barbara Basin (SBB), where only during an event at around 44 ka, δ13Carchaeol showed a negative spike of −25‰ lower than background values [Hinrichs et al., 2003]. Minimum δ13C values of archaeol in SBB coincided with the sudden appearance of several 13C-depleted dialkylglycerolethers (DAGEs), which are likely derived from sulfate reducing bacteria [e.g., Hinrichs et al., 2000], as well as minima in the δ13C of both benthic and planktic foraminifera [Hinrichs et al., 2003; Kennett et al., 2000].
 Additional information on anaerobic methanotrophic communities in our study area is from analysis of GDGT-derived biphytanes. Methane oxidizing archaea of the ANME-1 cluster are known to produce substantial amounts of GDGTs with zero to three cyclopentanes [e.g., Rossel et al., 2008; Wakeham et al., 2003]. In marine sediments, several groups of metabolically distinct archaea need to be considered as sources of GDGTs: planktic crenarchaea, which can be preserved well in marine sediments [e.g., Schouten et al., 1998; Sinninghe Damasté et al., 2002]; benthic archaea, including methanogens [e.g., Koga and Nakano, 2008], methanotrophs [e.g., Hinrichs et al., 2000]; and other heterotrophic archaea [e.g., Biddle et al., 2006]. GDGT-0 (caldarchaeol, no rings) is a common constituent in polar lipids of a wide range of methanogens, GDGT-1 and -2 (one and two cyclopentane rings) are rare among cultured methanogens [e.g., Koga and Nakano, 2008]. Both planktic and benthic archaea produce primarily GDGT-0 (caldarchaeol) and GDGT-5 (crenarchaeol, four cyclopentane rings and one cyclohexane ring) and smaller amounts of GDGT-1 and -2, with typical δ13C values for the constituent biphytanes ranging from −30‰ to −20‰ [Biddle et al., 2006; Hoefs et al., 1997].
 The relative distribution of GDGT-cleaved biphytanes in all our samples reflects strong contribution of crenarchaea, with the greatest abundance of acyclic biphytane (predominantly GDGT-0 derived) and similar abundances of bicyclic and tricyclic biphytane (predominantly GDGT-5 derived), and small amounts of monocyclic biphytane (derived from GDGT-1 and -2) (Figure 8). The δ13C values of bicyclic and tricyclic biphytanes fall within the typical range observed in planktic and benthic archaea. The low δ13C values of monocyclic biphytane in both sediment cores indicate the contribution of GDGT-1 and -2 from methanotrophic archaea. The maximum 13C-depletion in this compound is consistent with the high relative abundance of GDGT-1 and -2 in methanotrophic archaea. The lower degree of 13C-depletion in acyclic biphytane is because the greatest proportion of this compound is most likely derived from planktic and unknown heterotrophic benthic archaea [cf. Biddle et al., 2006; Lipp and Hinrichs, 2009; Schouten et al., 2002].
 The 13C-depleted bacterial lipids that cooccur with 13C-depleted archaeol and GDGT-derived biphytanes (Figure 8) are likely from syntrophic sulfate reducing bacteria (SRB). DAGEs from SRBs are typical for sediments hosting ANME consortia [Blumenberg et al., 2004; Hinrichs et al., 2000; Teske et al., 2002]. Background δ13C values of DAGE-C30 are −47 to −42‰. In samples where 13C-depleted archaeal lipids indicate the presence of methanotrophic biomass, the δ13C of the DAGE-C30 drops, reflecting the contribution of a methane-derived substrate to the SRB.
 Notably, we did not detect diplopterol or diploptene in any samples, which would indicate aerobic, bacterial-driven methanotrophy within the water column [Hinrichs, 2001; Summons et al., 1994]. Though the benthic foraminiferal abundances suggest release of methane into the water column, it is possible that any aerobic methanotrophic biomass was low enough to be undetectable, resulting in a lipid profile that represents exclusively the sedimentary anaerobic methanotrophic biomass.
 Even with methane release into the water column, a substantial flux of methane into the atmosphere would be required to trigger greenhouse warming. Bacterial methanotrophy in the oceanic water column is highly effective (see review by Reeburgh ), and in the modern ocean, methane flux to the atmosphere from shallow submarine seeps and mud volcanoes is considered small [Judd, 2000], with virtually no methane from deeper sites reaching the atmosphere [McGinnis et al., 2006]. However, with sudden, high methane flux, a rising bubble plume could advect a mixture of methane-saturated seawater and gas bubbles toward the surface, allowing methane flux from water depths greater than 250 m to reach the atmosphere [Leifer et al., 2006]. Recent measurements from the Gulf of Mexico suggest that current estimates of methane bubble dissolution may be too low, and flux from deepwater seeps may be larger than previously assumed [Solomon et al., 2009].
4.2. Location of AOM in Sediment Column
 Only the events associated with the strongest 13C depletion in foraminifera are associated with unambiguous molecular signals indicative of AOM (304 cm in 57JPC, 544 cm in 51JPC). We suggest that only some of the authigenic carbonate layers were associated with rates of methane oxidation high enough and for long enough to emplace sufficient methanotrophic biomass for detection. The carbon conversion efficiency of methane into biomass for methanotrophs is quite low [Wegener et al., 2008]. As a consequence, molecular signals of AOM communities remained elusive for decades, especially in settings where low diffusive flux is linked to relatively low rates of AOM [cf. Bian et al., 2001]. In an example from the North Sea, Niemann et al.  measured elevated sulfate reduction and methane oxidation rates in the SMTZ of an active methane seep, but no lipid biomarkers from the methane oxidizing microbial community were found.
 If the SMTZ is deep in the seafloor, the authigenic carbonates form in sediment that is much older than the episode of methane flux, and consequently all of the methane is oxidized within the sediment column [Treude et al., 2005; Ussler and Paull, 2008]. If this is the case in our study region, the age of the sediments containing authigenic carbonates would not be the same as the time the methane flux occurred. Therefore, it is important to establish where in the sediment column the SMTZ was when the authigenic carbonates precipitated.
 Faunal change has been used to identify episodes of methane seepage in the past where relatively high abundance of endobenthic foraminifera and high absolute abundance of benthic foraminifera were interpreted as indicating methane seepage [Bhaumik and Gupta, 2007; Wiedicke and Weiss, 2006]. However, in the modern system, benthic foraminifer densities are highly variable within seeps [Heinz et al., 2005; Rathburn et al., 2000] and outside of seeps [Bernstein et al., 1978]. There does not appear to be a unique assemblage of seep fauna or an endemic species of benthic foraminifera in seeps, but dominant species frequently include members of the genera Uvigerina, Bolivina, Epistominella and Nonionella (see references in the work by Heinz et al. ).
 There is the added uncertainty that all of the surveys of the abundance and species of benthic foraminifera living in seep areas referenced above rely on Rose Bengal, a protein marking stain, to identify “living” specimens. However, Rose Bengal will stain tests that contain any cytoplasm, even if the individual is dead [Bernhard, 1988]. Bernhard et al.  studied fauna in an active seep in Monterey Bay using a metabolic fluorescent tag that only labels adenosine triphosphate, which is only present in living cells, and observed that living individuals of Epistominella, Fursenkoina, and Spiroplectammina were more common in seep sediments than nonseep sediments. The seep assemblage also included Uvigerina, Bulimina, and Cassidulina, consistent with Rose Bengal–based studied at other seep sites.
 Though there is no diagnostic fauna present in modern seeps, a large geochemical change in the habitat of benthic foraminifera occurs with the initiation of a methane seep, and a change in the benthic fauna that is coincident with other geochemical tracers of high methane flux would suggest methane seeping from the seafloor. In both our study areas, absolute abundance of benthic foraminifera, N. labradorica and E. cf. batialis increases in several of the horizons with authigenic carbonates (Table 4 and Figure 7). In 51JPC, the absolute abundance of G. pacifica increases during each horizon of low δ13C of N. pachyderma (s.) and changes in the absolute abundance of N. labradorica at both sites during several events. These correlations are highly statistically significant (Table 4), and are consistent with the interpretation that each layer of authigenic carbonates is associated with periods of strongly elevated methane flux from the subsurface to shallow sediments.
 This relationship between the occurrence of authigenic carbonates and changes in benthic foraminiferal abundance is weaker in 57JPC than in 51JPC. However, site 57JPC is within the oxygen minimum zone (OMZ), and the position and intensity of the OMZ strongly influences benthic fauna (see review by Bernhard and Sen Gupta ). Both OMZ and seep fauna are quite variable, but both include species that are adapted to low-oxygen and sulfidic environments. Site 51JPC is well below the OMZ, so we would not expect the benthic faunal abundance to be overprinted by the OMZ as in 57JPC.
 The correlation between the authigenic-carbonate-rich layers and abundance of benthic foraminifera is weaker in 57JPC at 368–392 cm, 464–480 cm. Since we did not study the biomarkers at high resolution for these episodes, we cannot rule out the possibility that in this core, these horizons could be related to strong vertical methane flux but not to methane seepage at the seafloor. However, considering the persistent relationship between the benthic foraminiferal abundance and authigenic carbonates in 51JPC, a site far from the influence of the OMZ, it is possible that in 57JPC the methane flux also reached the seafloor, but was poorly recorded in the benthic foraminiferal abundances. Therefore, the abundance and species distribution of benthic foraminifera provide strong evidence that the authigenic carbonates are a record of methane flux to the shallow sediments, or to the seafloor, rather than deep in the subsurface.
4.3. Climate Implications
 The radiocarbon-based age model allows us to date the three most recent authigenic-carbonate-rich horizons (Figure 6). These horizons are 2 m shallower in 57JPC than in 51JPC, but within the uncertainty of the age model, they are contemporaneous in the two sediment cores. The uncertainty in the calibrated radiocarbon dates at 25 ka is 340–410 years (1σ), so it is not possible to analyze phasing of the events between the two cores. However, assuming the authigenic-carbonate-rich layers were each formed at approximately the same depth range below the seafloor, we can calculate that the three episodes of low δ13C were spaced slightly greater than 1000 years apart. This is similar to timing of D-O events, and we observe no events in either core between 22 and 15 ka, a time when D-O events did not occur either. Based on the timing of events between the two cores and the similarity in the isotopic signature of the excursions, we hypothesize that these horizons each formed during region-wide episodes of high methane flux, and that these episodes are linked to the climate and ocean circulation changes during D-O events.
 Each 13C-depleted authigenic-carbonate-rich horizon unambiguously represents episodes of enhanced vertical methane flux during the last glacial period at times of known millennial-scale climate changes. This supports the idea [Kennett et al., 2003], that sedimentary methane interacts with the climate system on these timescales. However, both coring sites are cold and deep, and destabilization of methane hydrates would require an unrealistic warming (6–16°C) or sea level drop (300–1000 m) (Figure S1).
 Simulations of the response of methane hydrates to temperature changes shows that with as little as 1°C warming, shallow deposits can undergo rapid dissociation and produce significant methane fluxes within decades [Reagan and Morodis, 2007]. For deep deposits (>1000 m water depth), temperature change of up to 5°C adjusts the top and bottom boundaries of methane hydrate deposits, but does not result in significant methane flux from the seafloor [Reagan and Morodis, 2007; Xu and Lowell, 2001]. However, even at temperatures and depths where methane hydrates would be thermodynamically stable, methane can be mobilized. For example, (1) the concentration is not high enough to form hydrate, (2) there is a kinetic barrier to hydrate formation [Brewer et al., 1998], or (3) pore water chemistry is altered to prevent hydrate formation [Liu and Flemings, 2006]. Examples of active seeps within the methane hydrate stability zone include the Congo Basin and Black Sea [Charlou et al., 2004; Heeschen et al., 2003; Klaucke et al., 2005] and at areas of high fluid flow like the Oregon and Aleutian subduction zones [Kulm et al., 1986; Suess et al., 1998], and at mud volcanoes [Kopf, 2002].
 In the Bering Sea, fluid flow is strongly influenced by silica diagenesis. Opaline sediments in the Bering Sea reflect the high productivity in the surface waters [Talley and Joyce, 1992]. Productivity is particularly high in a belt along the Bering Slope and along the Aleutian Islands [Springer et al., 1996] and the Umnak Plateau is at the juncture of these two zones. When diatom-rich sediments are buried and undergo compaction, the opal-A (biogenic silica) turns into opal-CT (cristobalite and tridymite). This diagenetic dehydration reaction causes a decrease in porosity of the sediment and expulsion of water from the crystal matrix that result in elevated pore fluid pressure and upward fluid flow over large areas [Davies et al., 2008] or focused fluid flow in sediment [Davies and Clark, 2006].
 The opal-A to opal-CT diagenetic front is seen as a bottom simulating reflector in the Umnak Plateau region at 670 m below the seafloor at the transition between terrigenous-rich, diatomaceous sediments (average 65% diatoms) to siltstone [Scholl et al., 1973]. Silica diagenesis could result in a buildup of overpressure that is accompanied by a decrease in the shear strength of sediment, priming it for failure. The exfoliation of submarine canyons driven by high pore fluid pressures is an important mechanism in shaping the canyons on the Bering Slope [McHugh et al., 1993], including Bristol and Bering canyons, between which our study area lies.
 In this study, we see a similar pattern of methane seepage at two different sites. This is consistent with large-scale regional fluid flow. Since the isotopic composition of the authigenic carbonates appears similar not only between each episode but also between the two coring areas (Figure 4), it suggests that the methane flux could have been operating over the entire region, drawing from a common reservoir of methane-laden fluids, resulting in similar conditions during episodes. Since the microbial biomass is dominated by sedimentary anaerobic methanotrophic communities, and there is no sign of aerobic methanotrophs, it is unlikely that catastrophic release of methane into the water column was the dominant mechanism. However, elevated methane flux on a regional scale could have resulted in methane release into the water column.
 What would have caused episodic methane flux in the past? One way to relieve overpressure in pore fluids is through seismic shaking. The Umnak Plateau lies between the Aleutian trench, an active subduction zone, and the Bering Slope, the position of the subduction zone in the Tertiary [Cooper et al., 1976], and which is associated with many relict faults. Seismic shaking could produce widespread, elevated fluid flow, as was observed following slow earthquakes on the Costa Rican margin [Brown et al., 2005]. This is consistent with the moderate AOM rates implied by the inferred ANME-1 consortium and the stable isotope and radiocarbon estimates of the authigenic carbonates. Gas and methane-rich fluids could also travel rapidly upward along fault planes or laterally along bedding planes that intersect a fault, forming seeps at the seafloor. These locations could release methane rapidly to the water column, and with high enough flux could potentially influence the atmospheric methane budget.
 If seismic shaking is the main way that overpressured pore fluids were released in the past, what controls the frequency of the episodes of fluid and gas release? We speculate that changes in the water column have the potential to have caused methane mobilization in the southeast Bering Sea during the last glacial period. The slip associated with slow earthquakes in subduction zones around the rim of the Pacific are observed to be periodic, and have been linked to the 14 month pole tide [Lowry, 2006; Shen et al., 2005]. A small hydrostatic pressure change of ∼160 Pa occurs on the subduction zones when the pole tide passes overhead, resulting in Coulomb failure, slow slip, and low-frequency shaking [Shen et al., 2005]. During the last glacial period, changes in ocean circulation and sea level [Siddall et al., 2003] occurred with the Dansgaard-Oeschger climate events. Sea level increased by up to 35 m at the beginning of the warm climate events, with rates of rise of up to 2 cm per year [Siddall et al., 2003].
 We speculate that the sediments of the southeast Bering Sea could be primed for rapid methane and fluid release by the overpressure of pore fluids by silica diagenesis. With sea level rise at the beginning of a Dansgaard-Oeschger event, the hydrostatic pressure change on the Aleutian subduction zone could have triggered seismic shaking. The shaking could have caused region-wide upward fluid flow and rapid expulsion of methane and methane bearing pore fluids along faults and bedding planes, inducing near-surface AOM with associated precipitation of authigenic carbonate minerals. During the stadials, the continuing silica diagenesis would then repressurize the pore fluids.
 To test this mechanism would require hydrogeological modeling, including methanogenesis, compaction, silica diagenesis (to determine the magnitude of overpressure that could be generated in 1 to 4 kyr), and estimation of the amount of methane that could be mobilized by a fluid expulsion episode. In addition, modeling of the stress field of the Aleutian subduction zone could test whether hydrostatic pressure changes of the magnitude that occurred during MIS3 could have caused earthquakes.