Multibeam swath bathymetry and sub-bottom profiler data are used to establish constraints on the flow and retreat history of a major palaeo-ice stream that carried the combined discharge from the parts of the West Antarctic Ice Sheet now occupied by the Pine Island and Thwaites glacier basins. Sets of highly elongated bedforms show that, at the last glacial maximum, the route of the Pine Island-Thwaites palaeo-ice stream arced north-northeast following a prominent cross-shelf trough. In this area, the grounding line advanced to within ∼68 km of, and probably reached, the shelf edge. Minimum ice thickness is estimated at 715 m on the outer shelf, and we estimate a minimum ice discharge of ∼108 km3 yr−1 assuming velocities similar to today's Pine Island glacier (∼2.5 km yr−1). Additional bed forms observed in a trough northwest of Pine Island Bay likely formed via diachronous ice flows across the outer shelf and demonstrate switching ice stream behavior. The “style” of ice retreat is also evident in five grounding zone wedges, which suggest episodic deglaciation characterized by halts in grounding line migration up-trough. Stillstands occurred in association with changes in ice bed gradient, and phases of inferred rapid retreat correlate to higher bed slopes, supporting theoretical studies that show bed geometry as a control on ice margin recession. However, estimates that individual wedges could have formed within several centuries still imply a relatively rapid overall retreat. Our findings show that the ice stream channeled a substantial fraction of West Antarctica's discharge in the past, just as the Pine Island and Thwaites glaciers do today.
 Pine Island and Thwaites glaciers account for most of the ice discharge from the Amundsen Sea drainage sector of the West Antarctic Ice Sheet (WAIS) (Figure 1). They presently exhibit the most rapidly decreasing surface elevation within the WAIS [Shepherd et al., 2001; Pritchard et al., 2009] and increasing discharge over the past 35 years through flow acceleration (Pine Island) and widening (Thwaites) indicates that this is dynamic thinning [Rignot, 2006, 2008; Scott et al., 2009]. Grounding line retreat associated with this thinning has also been demonstrated for Pine Island Glacier [Rignot, 1998]. The dynamic thinning has been widely attributed to intrusion of relatively warm circumpolar deep water onto the continental shelf, melting the bases of glacier tongues and ice shelves and reducing their “buttressing” effect [e.g., Jacobs et al., 1996; Rignot and Jacobs, 2002; Payne et al., 2004; Shepherd et al., 2004; Walker et al., 2007; Thoma et al., 2008]. Deep bathymetric cross-shelf troughs, which are thought to have been incised by glaciers through the Miocene-Quaternary (Figure 1) may be important for this process, channeling warm water toward the ice fronts. Enhanced geothermal heat flux beneath the Amundsen Sea glacier catchments might be another driver of flow changes [Rignot et al., 2002], enhancing melting at the glacier beds and thus increasing lubrication. Alternatively, the recent imbalance may simply reflect a short-term deviation from the ice sheet's long-term retreat trajectory. In order to test this hypothesis, it is necessary to establish the maximum extent of ice during the last glaciation and the timing and pattern of its subsequent retreat. An important question is whether or not there may have been previous periods of rapid thinning and retreat interrupted by intervals when the grounding line and flow stabilized.
 This history of flow stability is particularly pertinent to the Amundsen Sea sector today because the ice bed currently lies below sea level beneath its main glacier trunks and slopes inland monotonically from the grounding line in the Thwaites catchment [Holt et al., 2006; Vaughan et al., 2006]. According to some theoretical studies, this geometric configuration is one of the key criterion for a marine ice sheet collapse [Vaughan and Arthern, 2007; Schoof, 2007] and has contributed to the Amundsen Sea sector often being referred to as “the weak underbelly” of the WAIS [Hughes, 1981], with suggestions that the region may be prone to future rapid deglaciation [Vaughan, 2008].
 However, confidence in models that purport to simulate future changes in the ice sheet can only be established by verifying those models against past behavior; in particular, flow history and changes since the last glacial maximum (LGM) at ∼20 cal kyrs before present (B.P.), here referring to the last period of regional maximum ice sheet extent around Antarctica [Anderson et al., 2002] when grounded ice had advanced onto the adjacent continental shelf, and during the deglaciation between ∼20 and 10 cal kyrs B.P. [Anderson et al., 2002; Lowe and Anderson, 2002]. The most readily available and detailed information on the former footprint of these ice streams, and thus their flow history, comes from the morphological record preserved at the sea floor today [Shipp et al., 1999; Canals et al., 2000; Evans et al., 2005; Ó Cofaigh et al., 2005a, 2005b; Dowdeswell et al., 2008]. Previous studies have used the presence of geomorphic ice flow indicators such as preserved streamlined bedforms, to suggest that an ice stream flowed out of Pine Island Bay toward the continental shelf edge in the past. Geophysical data and sediment cores from the large cross-shelf trough that extends offshore from Pine Island and Thwaites glaciers were interpreted by Lowe and Anderson [2002, 2003] as showing that grounded ice had extended at least as far as the middle shelf (as a minimum possible extent), and probably to the shelf edge at the LGM (as a maximum scenario). On the basis of radiocarbon dates on calcareous microfossils extracted from cores, Lowe and Anderson  concluded that ice subsequently retreated from the middle shelf before ∼16 14C ka B.P. (uncorrected) and retreated to the inner shelf by 10 14C ka B.P. (uncorrected). Evans et al.  presented multibeam data showing subglacial bedforms that extend to the shelf edge in an outer shelf trough at 114°W and argued that their sea-floor position and the lack of overlying sediment drape indicate that the WAIS was grounded at the shelf edge during the LGM. However, coverage of multibeam swath bathymetry data remains sparse on large parts of the outer Amundsen Sea shelf, especially north of 72°30′S, where the path of the main Pine Island cross-shelf trough is less clear (Figure 1) [Nitsche et al., 2007]. In addition, only two radiocarbon dates constrain the “timing” of post-LGM deglaciation and even less information exists about the “style” of ice retreat from the eastern Amundsen Sea shelf [Lowe and Anderson, 2002], the latter of which constitutes a key parameter for validating ice sheet numerical simulations.
 In this paper we describe streamlined subglacial bedforms that record the flow and retreat style of a major palaeo-ice stream, the Pine Island-Thwaites palaeo-ice stream (PITIS), across the Amundsen Sea continental shelf. This ice stream received input from the ancestral Pine Island, Thwaites, and Smith glacier systems (Figure 1). The bedforms are located in a trough that extends to the shelf edge, at 106°W (Figure 1). Using new geophysical data sets, we test the following hypotheses: (1) the LGM WAIS extended to the outer shelf in the eastern Amundsen Sea; (2) the major drainage pathway and main outlet of the PITIS can be traced from the present grounding line through the trough in the eastern Amundsen Sea Embayment (ASE); and (3) sea-floor geomorphic evidence can be used to constrain the style and course of grounding line retreat to its present-day configuration.
 We used marine geophysical data acquired using a Kongsberg EM120 (191 beams at 11.25–12.75 kHz) and an Atlas Hydrosweep DS-2 (59 beams at 15.5 kHz) multibeam swath bathymetry system, together with TOPAS parametric sub-bottom profiles (“burst” pulse at secondary frequency of 2.8 kHz and “chirp” pulse with secondary frequencies of 1.5–5 kHz), to map the topography and distribution of relict subglacial bedforms on the continental shelf of the Amundsen Sea. Geophysical data were collected during cruises JR84, JR141, and JR179 of the RRS James Clark Ross (JCR; 2003, 2006, and 2008) and during cruise ANT-XXIII/4 of the R/V Polarstern (2006). These data were combined with existing swath bathymetric data from the Lamont-Doherty Earth Observatory, Marine Geosciences Data System (http://www.marine-geo.org/; Figure 1). Navigation data were acquired using GPS receivers. Ping-edited swath data were gridded at a 30 m cell size.
3. Subglacial Bed Forms and Features
 A recent regional bathymetric data compilation for the Amundsen Sea Embayment [Nitsche et al., 2007] shows a continuous cross-shelf trough that arcs north-westward across the shelf and which has two possible outlets, as depicted by bathymetric depressions: one to the NW [Evans et al., 2006] and another to the NE (Figure 1). Hereafter, we refer to these outlets as Pine Island Trough West (PITW) and Pine Island Trough East (PITE).
 On cruise JR179, we followed the axis of the largely unsurveyed PITE, from its mouth at the shelf edge to the main cross-shelf trough and then continued onward to the deepest part of the trough on the inner shelf, near the March 2008 fast ice edge (to within ∼100 km of the modern Pine Island ice shelf front). Water depths generally increase inshore along this 400 km long track, from ∼500 m at the shelf edge to >1600 m on the inner shelf (Figure 1).
Figure 2a shows grayscale shaded-relief swath bathymetry from the outer shelf in the PITE axis. A suite of crudely aligned to parallel lineations are observed on the sea floor. Bedforms have a general NNE alignment (25°–35°; north azimuth) and occasionally crosscut. They are up to 500 m wide, 16 km long, and have amplitudes of 2–6 m (Figure 2a, inset). The maximum elongation (length-width) ratio for lineations is ∼64:1, and most have ratios greater than 10:1. The main set of imaged lineations is formed on the landward flank of a bathymetric high and terminate at a prominent seaward facing (ice leeward) ramp (Figures 2a, 3a, and 4a). The ramp is up to 20 m high and imaged over a distance of 1700 m, producing a “wedge” transverse to the trough long axis at the seabed (GZW1) (Figures 2a, 3a, and 4a). Curvilinear furrows characterize the shallower portions of the surrounding bathymetry and, in places, cut across more parallel lineations (Figure 2a). A narrow corridor of multibeam data extending to the northeast of GZW1 shows that lineations are also present further north. A second ramp, with similar dimensions to that forming the seaward limit of GZW1, lies ∼20 km to the south of the first wedge (GZW2) (Figures 2a, 3a, and 4a), and there are subtle changes in lineation orientations across each of the wedge fronts (on the order of several degrees).
 Following the trough south, additional sets of highly parallel bedforms comprising undulating grooves and ridges are apparent at the sea floor, lying within the deepest parts of the trough (Figures 2b and 5b). Bedforms are 5–14 km in length, 250–490 m in width, and have amplitudes of 3–20 m; most ∼5 m (Figure 2b, inset). These dimensions give rise to high elongation ratios, up to 52:1 (minimum 18:1). The trend of the bedforms is again uniformly NNE–SSW (31°–33°), suggesting they may simply be the up-trough extension (albeit, a slightly younger generation) of the bedforms imaged in Figure 2a. The bedforms shown in Figures 2a and 2b both lie within the outer shelf region defined by Lowe and Anderson  as geomorphic zone 4, in which they suggested the only bedforms are randomly oriented iceberg furrows.
 Farther landward along the trough, additional sets of parallel and attenuated bedforms on the middle shelf are interrupted at the sea floor by continuous seabed ramps that form prominent escarpments, striking transverse to the trough long axis, and with gently dipping backslopes (Figure 2c). At least three separate ramps are imaged in the middle shelf area: one at 72°33′S (GZW3 front) (Figure 3b), another at 72°42′S (GZW4 front) (Figure 3c), and a third at 72°51′S (GZW5 front) (Figures 1, 2c, and 3c). In TOPAS profiles and from multibeam analysis, it is apparent that individual ramps are the seaward flanks of sea-floor wedges (Figures 2c, inset; 3 and 4b). GZW3 is a small ridge ∼15 m high, with a series of more subtle, low-amplitude ridges formed behind it (Figures 3b and 4b). GZW4 is ∼35 m high at its crest, has an along-trough extent of ∼16 km, and is imaged within a 2.5 km wide corridor of multibeam data. Small, lobate mounds characterize the lower slope of the sea-floor ramp in front of GZW4 (Figure 4b). GZW5 is the largest of the three midshelf “wedge” landforms with an estimated volume of at least 6 km3 (50 m average height, by 20 km along-trough extent, by a minimum of 6 km across the wedge front, Figure 3c; note also the similar slope angles of its northern and southern flanks, Figure 4b). All the wedges are likely formed of sediment, as shown by an existing seismic profile through the landforms [Lowe and Anderson, 2002, Figure 4b], and they form a stacked, back stepping complex on the middle shelf (Figure 4).
 Individual streamlined bedform sets are associated with each of the imaged wedges, with lineations terminating either at the crest of the seabed ramp (Figures 2c, inset; 3c) or further landward of the wedge backslope, as a result of ploughing by iceberg keels near the crest (Figures 2c, inset; 3c). The orientation of lineations changes progressively up-trough, from a NNE–SSW orientation (28°–32°) to an NNW–SSE orientation (353°) either side of GZW4 (Figures 2c, inset; 3b). On the up-trough flank of GZW5, further large and elongated lineations are well developed at the sea bed, and these trend NNW–SSE (353°) (Figures 2c and 3c). These bedforms extend southward along the trough floor to the transition of the inner to the middle shelf (Figure 2d) [cf. Lowe and Anderson, 2002]. The lineations have dimensions of 4–12 km long, ∼250–500 m wide, and amplitudes of 2–10 m (average ∼5 m) with length-width ratios up to ∼48:1 (Figure 2d), and thus are similar to those in Figures 2a and 2b.
 On the inner shelf, more than 280 km from the shelf edge, elongated drumlins, less elongated streamlined grooves, meltwater channels, and tunnel valleys were reported by Lowe and Anderson [2002, 2003]. The authors showed that these features were ubiquitously formed as erosional bedforms on substrate that they identified as acoustic basement in seismic profiles (which they interpreted as crystalline bedrock). By contrast, all the bedforms and GZWs described north of ∼73°30′S are formed over a seaward thickening sedimentary substratum, as shown in seismic records [Lowe and Anderson, 2002, 2003].
4. Palaeoglaciological Interpretations
 Large, straight to curvilinear lineations on the outermost shelf in the ASE, at the crests of the two sea-floor wedges closest to the shelf edge (GZW1 and GZW2; Figures 2a, 3a, and 4a), are interpreted as a subglacial or grounding line proximal bedform signature. We interpret their formation at, or close to, a laterally restricted, grounded, and calving ice sheet margin (such as an ice stream front) based on their (1) high elongation ratios and long lengths; (2) few crosscuts and grouping into general parallel to subparallel alignments that both suggest local restriction to movement of grounded ice, or icebergs which are just afloat; (3) distinct, rounded ridge-crest surfaces in crossing TOPAS profiles, reminiscent of subglacial lineations rather than typically “v”-shaped iceberg ploughmarks; and (4) larger, straighter, and more elongate morphology than neighboring iceberg furrows. The varying types of bedform geometry indicate that the sea floor in this location probably records a mixture of subglacial and iceberg-keel ploughed morphologies. Similar mixed morphologies have been observed on multibeam data in the Vega Trough, and in the Robertson Trough, offshore of the Antarctic Peninsula [Heroy and Anderson, 2005, Figure 4].
 To the south on the outer shelf, highly parallel, pervasive sets of ridge-groove streamlined bedforms, measuring several to >15 km in length and with elongation ratios between 18 and >60:1, are interpreted, more typically, as subglacial mega-scale glacial lineations (MSGLs) [Clark, 1993; Stokes and Clark, 1999]. MSGLs, analogous in both geometry and form to those imaged in this study, are found as a relict expression of fast flow on many palaeo-ice stream beds [e.g., Wellner et al., 2001; Stokes and Clark, 2002; Shipp et al., 1999, 2002; Evans et al., 2005; Ó Cofaigh et al., 2005a, 2005b] and have recently been imaged at the bed of the Rutford Ice Stream in a downstream zone where it is fast flowing (>300 m yr−1) confirming, through direct observation, an ice stream-MSGL connection [King et al., 2009]. Like flutings on glacial forelands, the MSGLs are aligned parallel to former glacier flow [Clark, 1993]. Therefore, the MSGLs in the PITE indicate a dominant flow direction arcing to the north and NNE.
 The bedforms in the eastern ASE therefore record the former presence of a fast-flowing grounded ice stream (the PITIS) that drained through Pine Island Bay and its seaward extension, Pine Island Trough East. The bedforms clearly lie in the center and deepest parts of the main cross-shelf trough as depicted by bathymetric profiles and as expected for the locale of a major palaeo-ice stream (Figure 5a and 5b). The pristine preservation of the bedforms, combined with the published deglacial chronology [Lowe and Anderson, 2002], indicates that this ice stream existed at and shortly after the LGM and that ice must have grounded to within 68 km of the shelf edge. Beyond the most northeasterly streamlined bedforms, any further grounded ice indicators have been removed by ice-keel ploughmarks. Thus, the PITIS must have occupied a minimum of ∼87% of the currently ice-free trough and probably covered extensive parts of the shelf outside of it at the LGM.
 Sets of bedforms on the middle shelf, aligned at different orientations to the flowset on the outer shelf, record the back stepping subglacial imprint of the ice stream during its retreat across the continental shelf. Bedform sets are commonly associated with pronounced sedimentary wedges, which we interpret as grounding zone wedges (GZWs 1–5) (Figures 3 and 4): depositional accumulations formed at ice stream mouths during stillstands of the grounding line where deformable till sheets at the ice bed act as sediment conveyors [Alley et al., 1989; Anderson, 2007; Anandakrishnan et al., 2007]. The fact that bedforms terminate at wedge fronts and are well preserved in “soft” sedimentary strata [Lowe and Anderson, 2002] indicate that they are features formed during recent ice recession rather than older lineations that have been subsequently overridden [cf. Ó Cofaigh et al., 2005a; Dowdeswell et al., 2008]. Smaller mounds and flow-transverse ridges are interpreted as proglacial debris flows or minor recessional moraines, both associated with ice-marginal sediment delivery (Figure 3b); the scale and corrugated, subparallel morphology of the latter compare well to “corrugation moraines” that formed during the last deglaciation and were mapped in cross-shelf troughs in the Ross Sea [Shipp et al., 2002; Dowdeswell et al., 2008]. The presence of GZWs interspersed by distinct MSGL flowsets identifies an episodic style of deglaciation [e.g., Larter and Vanneste, 1995; Dowdeswell et al., 2008; Ó Cofaigh et al., 2008], with stepped grounding line retreat across the continental shelf. These GZWs afford a preliminary reconstruction of the retreat pattern of the PITIS, where former stationary or readvanced positions of the grounding line can be traced (Figure 1, marked by purple lines).
 On the inner shelf, bathymetric sills were likely to provide additional pinning points for the grounding line in a zone where sediments are sparse and the topography is highly pronounced (Figure 5c). Bedforms scoured into acoustic basement elsewhere on the inner shelf are interpreted as the upstream signature of palaeo-ice flow in the former Pine Island and Thwaites glacier systems [Lowe and Anderson, 2002, 2003], though they may not all relate to ice streaming at the LGM [cf. Graham et al., 2009].
5. Flow Outlet of the Pine Island-Thwaites Ice Stream
 Using an earlier, more limited multibeam data set, Evans et al.  suggested that the main outlet from Pine Island Trough at the LGM drained to the NW via a trough that reaches the shelf edge at 114°W (PITW) (Figure 6). The presence of subdued streamlined bedforms within this NW trough does support the idea that ice drained through this outlet at some point in the past. However, MSGLs on the outer shelf along the PITE at 106°W are more pristine and have greater elongation ratios than those in the PITW, which suggests that major discharge and outflow of the PITIS at its last maximum extent was directed to the northeast along the PITE (compare Figures 2b and 6). We rule out variation in postglacial sediment coverage to explain differences in bedform preservation between PITE and PITW. Evans et al.  showed only a thin postglacial sediment drape in the PITW from acoustic profiles, and thin postglacial sediments were also recovered in cores from the PITE [Lowe and Anderson, 2002]. Line A-A′ is the only continuous west-east bathymetric profile to cross the PITE on the outer shelf, near the NE Amundsen Sea shelf break (Figure 5a). In conjunction with our multibeam data sets, it shows, for the first time, that the PITE actually continues to the shelf edge supporting a main flow outlet in this location (Figure 1, shown by the regions of bathymetry shaded in orange; Figure 5a).
 In the Pine Island cross-shelf trough, Lowe and Anderson [2002, 2003] also defined a number of geomorphic zones forming a down-flow progressive model across the inner and middle shelf. Much of the sea-bed on the Amundsen Sea outer shelf has been pervasively scoured by icebergs so our discovery of subglacial bedforms in the PITE is significant for showing ice extent beyond the middle shelf, where streamlined features were previously thought to terminate [Lowe and Anderson, 2002, 2003]. These bedform patterns show that there was streaming ice flow, at various time intervals, over an along-trough distance of at least ∼200 km on the middle-to-outer shelf, demonstrating fast flow at the scale of or even larger than other major ice stream systems in West Antarctica during the LGM, such as those in Marguerite Trough [Ó Cofaigh et al., 2005b], Belgica Trough [Ó Cofaigh et al., 2005a], Boyd Strait [Canals et al., 2000], Anvers Trough [Pudsey et al., 1994; Domack et al., 2005], and the Ross Sea [Shipp et al., 1999; Mosola and Anderson, 2006]. Our study indicates an extensive palaeo-ice stream in the eastern ASE and strongly supports previous interpretations that the WAIS in this sector reached the shelf edge at the LGM [Lowe and Anderson, 2002; Evans et al., 2006].
6. Outer Shelf Ice Dynamics
 While we consider it likely that the PITE was the main outlet for the PITIS, the presence of two well-preserved subglacial bedform sets, in separate outer shelf troughs, points to complex ice dynamics. Figure 7 illustrates three possible interpretations, which could be used to reconcile the two data sets.
 Our favored interpretation is that phases of ice streaming within the PITE and PITW were asynchronous (Figure 7a). Switching ice-stream flow between the two adjacent troughs probably produced the bedforms at different times [cf. Dowdeswell et al., 2006a], and we suggest that flow through PITW preceded the flow through PITE. This flow-switching scenario is supported by (1) the better preservation and exceptional clarity of the bedform signature in the PITE compared with that in the PITW, suggesting a younger age of the MSGLs to the northeast; (2) the presence of only a single major trough crossing the inner and middle shelf, but two well-defined outer shelf outlets; (3) the deeper incision of the PITE outlet into the shelf when compared to the PITW outlet (i.e., incision to different base levels; Figure 5b), which increases the likelihood of earlier ice flow in PITW being captured by more dominant ice stream flow draining through the PITE; and (4) the existence of slope gullies at the mouth of both the PITE and PITW that indicate that slope processes have probably been diachronous across the Amundsen Sea margin. Although there is contention over their mode of formation [Noormets et al., 2009], gullies at the mouth of the PITW were interpreted as derived from sediment-laden meltwater or slope instability flows [Dowdeswell et al., 2006b; Noormets et al., 2009], whereas slope debris flows more typical of ice marginal sediment delivery are found near the mouth of the PITE [Dowdeswell et al., 2006b]. This disparity would appear to support a younger age for the PITE.
 Clear evidence of flow switching of this kind has been found in 3-D seismic data from the North Norwegian margin, linked to sedimentary infill of accommodation space within cross-shelf troughs [Dowdeswell et al., 2006a]. Flow switching in ice streams has not been directly observed but is suspected to be possible driven by changing subglacial water drainage [Vaughan et al., 2008] and has almost certainly occurred in the recent past [Conway et al., 2002].
 An alternative interpretation is that two separate ice flow limbs were synchronous features within the Late Quaternary WAIS and existed as two bifurcating ice streams on the outer shelf (Figure 7b). However, very few bifurcating analogues are observed in contemporary ice sheets, and most reconstructions of Quaternary palaeo-ice sheets in the Northern Hemisphere lack evidence for ice stream bifurcation in their downstream regions [Ottesen et al., 2005, 2008; Stokes et al., 2009].
 A third hypothesis is that the two outlets and their bedform signatures result from one large fast-flow sector across the outer shelf, with ice flow in the form of a divergent “sheet” (Figure 7c). To test this idea, we estimated a potential discharge for sheet flow across the area of outer shelf from the PITW to the PITE, assuming a minimum cross-sectional flow area of ∼400 km × 550 m. The PITIS basin, including the now-submerged shelf areas and modern Pine Island-Thwaites catchment, would cover an accumulation area of ∼500,000 km2 at the LGM, approximately three times the size of the modern Pine Island Glacier catchment. By scaling the modern net surface accumulation over Pine Island Glacier catchment (∼69–71 km3 ice yr−1) [Rignot, 2006] to this catchment area (∼207–213 km3 ice yr−1) and assuming significantly lower precipitation rates during the LGM (approximately half; ∼105 km3 ice yr−1), we show it would be possible to sustain sheet discharge through this gate at a velocity of ∼480 m yr−1, a fast-flow rate comparable to some contemporary Antarctic ice streams. However, whereas some models of ice sheets predict “sheet”-style flows at marine margins [Hubbard et al., 2009], modern glacial analogues remain absent, at least at similar scales (but see Malaspina Glacier, Alaska, for a possible scaled-down comparison) [Sharpe, 1958]. Therefore, while conceivable, we do not favor this interpretation.
 Regardless of the specific scenario, our data document a major discharge of the ancient Pine Island-Thwaites ice stream along the PITE. Taking into account a lowered sea level at the LGM of ∼120 m, a minimum ice thickness of 715 m is estimated if ice grounded in the trough at 72°30′S. This estimate neglects any glacio-isostatic depression of the shelf by the ice that extended onto it, which would make the minimum ice thickness even greater. This cross section would, in turn, accommodate a minimum flux of ∼108 km3 ice yr−1 (based on an assigned flow velocity of 2.5 km yr−1) exceeding the modern grounding line flux of 75–80 km3 ice yr−1 from Pine Island Glacier today (where velocity is measured at ∼2.5 km yr−1). Thus, we deduce the PITIS to have been a key drainage feature of the WAIS, since at least the last glacial period.
7. Retreat of the Pine Island-Thwaites Ice Stream
 Our data appear to show that multiple GZWs formed at the grounding line of the PITIS as it retreated from the outer and middle shelf. Previously, only a single wedge had been identified in the PITE [Lowe and Anderson, 2002] with a second tentatively suggested for the PITW at ∼108°W by Evans et al. . Our geophysical data show that the midshelf wedge of Lowe and Anderson  is in fact a complex of back stepping wedges, formed at a number of ice-marginal positions (Figure 4). This geometry suggests that the ice margin periodically stabilized during retreat, when stillstands of the grounding line may have occurred. The presence of large GZWs (minimum volume of ∼6 km3 for GZW5) indicates significant subglacial erosion, beneath the upstream portion of the ice stream trunk and the Antarctic interior, and shows that the ice stream was also a major conduit for basal sediment transport, at least during the early deglacial phase. The lack of comparatively thick sediment cover on parts of the inner shelf may be due to the PITIS stripping sediment from its bed during ice retreat [Smith and Murray, 2009].
 Despite new constraints on palaeo-ice margin location, the deglacial history of the PITIS since the LGM remains poorly constrained. The chronology of PITIS retreat is restricted to a radiocarbon age of ∼15.8 ± 3.9 14C ka B.P. obtained from deglacial transitional sediments on the midshelf near Burke Island and a minimum age of ∼10.2 ± 0.4 14C ka B.P. for grounding line retreat from the inner shelf (at 73°55′S and 106°39′W) of the PITE [Lowe and Anderson, 2002]. However, these ages give little insight into the duration of stillstands during grounding line migration up the trough. If we adopt sediment flux for an ice stream of similar size to the PITIS (e.g., 8 × 103 m3 yr−1 per meter wide, Norwegian Channel Palaeo-Ice Stream [Nygård et al., 2007]), we can calculate an equivalent total sediment flux from its grounding line, across a GZW of minimum 6 km wide (and 6 km3 volume, similar to GZW5), of ∼0.05 km3 yr−1. Under these constant flux conditions, it is feasible for GZWs in the eastern Amundsen Sea to have formed in ∼120 years (i.e., on centennial timescales). This rough estimate differs with previous interpretations, which suggest that GZW formation may stabilize an ice margin for thousands of years or more [Anderson, 2007]. Consequently, landforms of “episodic” retreat in the ASE may not necessarily indicate prolonged ice margin recession but may instead be manifestations of effective sediment delivery during a punctuated but otherwise rapid deglaciation [Larter and Vanneste, 1995; Dowdeswell et al., 2008].
 The observed landform configuration (presence of GZWs and the lack of recessional moraines) on the outer shelf generally supports our interpretations of a rapid, though episodic PITIS retreat [cf. Dowdeswell et al., 2008; Ó Cofaigh et al., 2008]. Runaway collapse of the modern ice sheet has been suggested where the ice bed slopes continuously into the ice sheet interior [Vaughan and Arthern, 2007]. For the PITE and PITIS, the palaeo-ice bed from the outer shelf to the inner 50 km of the shelf consists of a broadly continual reverse slope toward the coast (Figure 5c). Notably, the only breaks in this gradient are locations where the sea-floor levels out to form “benches.” These benches occur in close association with grounding zone wedges (Figure 5c), and the dimensions of the GZWs themselves are too small to account for the changes in sea-floor slope. Thus, periods of relative stability in ice retreat appear to have coincided with ice grounding on lower gradient beds (average slope measured at 0.015°), while well-preserved bedforms occur on beds with higher gradients (average slope measured at 0.149°), with no evidence for grounding line stabilization in these areas. This relationship suggests phases of rapid retreat over beds with greater slopes. Therefore, our observations lend support to model experiments indicating that stepped patterns of ice stream retreat are highly sensitive to the gradient of the bed on which they are grounded [Weertman, 1974; Schoof, 2007].
 Between the wedges, the presence of MSGLs indicates that either (1) ice streamed consistently during the retreat or (2) renewed onset of streaming occurred before each phase of ice margin recession. The feasibility of these dynamic patterns of ice stream retreat may be tested by future coupled glaciological-sedimentary models, which attempt to reconstruct the retreat history of major West Antarctic ice stream systems. Such models should seek to replicate periodic stillstands of the PITIS grounding line.
 We have identified a major LGM outlet of the Pine Island and Thwaites ice streams in the ASE, based on observations from new sea-floor and sub-bottom imagery. Addressing the hypotheses introduced at the start of the paper, we conclude the following:
 1. Streamlined glacial bedforms show that the West Antarctic Ice Sheet extended to within 68 km of, and probably right up to, the continental shelf edge in the northeastern Amundsen Sea Embayment at the LGM, supporting previous studies farther west in the embayment, which suggested extensive LGM ice sheet limits [Evans et al., 2006; Larter et al., 2009].
 2. The major drainage pathway of the PITIS can be traced along a cross-shelf trough that connects to a trough mouth on the outer shelf at ∼106°W. This mouth was likely the main outlet for the PITIS. Earlier phases of ice flow may have diverted along an outer shelf trough to the west, with an outlet at ∼114°W. Flow switching is our favored explanation for the presence of the two outlets and the two sets of bedform preserved within each outer shelf trough.
 3. Sea-floor geomorphic evidence constrains a stepped, episodic style of deglaciation for the PITIS. The course of ice retreat from the outer and middle shelf in the trough of the eastern ASE is shown by five grounding zone wedges, which formed during retreat. Stillstands occurred in association with changes in subglacial bed gradient, and we suggest that more rapid phases of retreat be correlated to higher bed slopes. Prominent bedrock pinning points on the inner shelf probably served to halt this retreat even further, as the grounding line receded to its present-day configuration. Further chronological data from sediment cores are urgently required to better verify the timing of stepwise retreat during the last deglaciation.
 We thank the captains and crews of the RRS James Clark Ross and R/V Polarstern, and shipboard parties of numerous cruises, for collecting data. This work forms a contribution to the Ice Sheets component of the British Antarctic Survey Polar Science for Planet Earth Programme. It was funded by The Natural Environment Research Council.