Ultraslow spreading mid-ocean ridges have a low magma budget and melt is distributed unevenly along the ridge axis. There is little or no basaltic crust between isolated magmatic centers. The processes that focus melts to segments of robust magmatism are not yet understood. During a seismic survey of the ultraslow spreading Knipovich Ridge in the Norwegian-Greenland Sea with ocean bottom seismometers, we discovered a seismic gap in the upper mantle beneath Logachev Seamount, where micro-earthquakes clearly delineate a shallowing of the maximum depth of faulting. A topography of the lithosphere that allows melts to travel laterally along its base and rise in areas of thin lithosphere has been proposed as a possible mechanism to explain the focusing of melts at volcanic centers, but has never been confirmed observationally. Our results are the first geophysical evidence for an along-axis variation of the lithospheric thickness at an ultraslow spreading ridge.
 The morphology of ultraslow (<20 mm/yr full rate) spreading mid-ocean ridges strongly differs from that of faster spreading ridges [Dick et al., 2003]. Generally, ultraslow spreading ridges are characterized by a deep rift valley, reaching water depths in excess of 5000 m, bounded by steep fault-controlled rift flanks culminating 2000 m above the rift valley floor [Michael et al., 2003]. The typical segmentation pattern of faster spreading ridges, with volcanically active segments centers associated with Mantle Bouguer Anomaly lows limited by transform faults or nontransform discontinuities [Cannat et al., 1999], is absent at most ultraslow spreading ridges. Instead, ultraslow spreading ridges are characterized by segments of very different characteristics [Michael et al., 2003]: magmatically robust sections resembling slow spreading ridges alternate with sections of little or no volcanism [e. g. Sauter et al., 2004]. Isolated volcanic centers, spaced 50–160 km apart, punctuate these melt-poor segments. Increased crustal thickness with basalts at the seafloor [Jokat et al., 2003], pronounced central magnetic anomalies [Jokat et al., 2003], hydrothermal plumes [Edmonds et al., 2003] and earthquake swarm activity [Schlindwein, 2012] may go along with these volcanic centers. If the melt supply system feeding these volcanoes is stable in time and space, seamount chains form that extend in spreading direction away from the volcanic centers like at the ultraslow-spreading Gakkel and Knipovich ridges [Okino et al., 2002; Curewitz et al., 2010], both part of the Arctic Ridge System.
 The other representative of an ultraslow spreading ridge, the Southwest Indian Ridge (SWIR), shows in its eastern magma-starved portion a more variable melt supply pattern in space and time [Cannat et al., 2003]. Away from the volcanic centers, a thin carapace of basalts covers the rift valley floor or, alternatively, mantle-derived peridotites are exposed directly at the seafloor, reflecting a highly variable melt delivery along the axis of ultraslow-spreading ridges [Cannat et al., 2006]. This pronounced variability in melt budget along-axis of an ultraslow-spreading ridge is one of the most important differences to all other spreading systems [Cannat et al., 2003]. This limited and variable melt supply has its expression in rough oceanic basement topography [Jokat and Micksch, 2004; Ehlers and Jokat, 2009], and a 2–4 km thin oceanic crust between the volcanic centers in the rift valley and in the adjacent basins [Urlaub et al., 2010].
 Explaining the melt generation and distribution system at ultraslow spreading ridges is therefore one of the major challenges of modern mid-ocean ridge research [Sauter and Cannat, 2010]. At the SWIR a wealth of observations including petrological sampling and seafloor mapping and imagery has enabled the construction of conceptual [Cannat et al., 2003; Standish et al., 2008] and numerical models of melt flow [Montési et al., 2011]. These authors postulate that a topography of the lithosphere-asthenosphere boundary (LAB) helps to guide melts to the volcanic centers. Once this topography is established, the process is self-sustained. This would explain the long-lived volcanic centers of Gakkel Ridge, Knipovich Ridge, or the western SWIR. At the eastern SWIR, however, with a less stable distribution of volcanoes in time, anomalies in the mantle composition or physical state are invoked to initiate a new topography of the boundary, which is then enhanced during the lifetime of a volcano [Cannat et al., 2003].
 No direct verification of this hypothesis has yet been reported. Seismic refraction data and micro-earthquake data are sorely missing at ultraslow spreading ridges [Sauter and Cannat, 2010]. Their remote location and extreme weather conditions have so far prevented systematic seismic surveys. Only reconnaissance-type studies provide first insights at isolated locations [Muller et al., 1999; Minshull et al., 2006; Jokat and Schmidt-Aursch, 2007; Schlindwein et al., 2007]. In addition, the LAB is, in general, difficult to determine with sufficient accuracy with geophysical methods.
 In 2009 during cruise ARK XXIV/3 of R/V Polarstern, the Alfred Wegener Institute for Polar and Marine Research conducted a seismic refraction survey along and across the rift valley of Knipovich ridge with ocean bottom seismometers (OBS) spaced at 20 km distance [Jokat et al., 2012]. As part of this experiment, five additional OBS were deployed about 25 km off-axis for reconnaissance purposes. The cruise schedule allowed to deploy the entire set-up for a period of 10 days to record microseismicity especially along the rift axis (Figure 1). We recorded 919 local earthquakes (Ml 0.0–3.5) and calculated 525 hypocenters [Bergner, 2012]. Figures 1 and 2 show the earthquake foci for 313 events. For these events we used a stringent quality criterion to ensure reliable hypocenter results [see Bergner, 2012]. Each event had to be recorded by at least five stations, had to have a location error (semimajor axis of 95% the error ellipse and depth error) of less than 5 km and a root-mean square travel time residual of less than 0.4 s. A priori knowledge of the velocity structure of the crust and upper mantle from the seismic refraction profile [Jokat et al., 2012] and the use of an enhanced localization algorithm [Schweitzer, 2001], which is particularly well suited to locate events with few observations [Läderach et al., 2011], enabled a robust hypocenter determination.
3 Seismicity Along the Knipovich Ridge
 Figures 1 and 2 show the earthquake activity as recorded between 28 August and 6 September 2009. In this contribution, we focus on the earthquake distribution in the mantle, which is in two aspects very remarkable:
 1. Earthquakes occur as deep as 25 km below sea level, about 17 km below the seismic Moho. Compared to other seismicity surveys at mid-ocean ridges, these are exceptionally deep hypocenters. At the fast spreading East Pacific rise, earthquakes occur down to depths of 1.7 km [Bohnenstiehl et al., 2008] below seafloor (bsf); the slow-spreading Mid-Atlantic ridge has maximum hypocenter depths of about 8 km bsf [Tilmann et al., 2004]; the ultraslow spreading Lena Trough shows local earthquakes at 14 km depth bsf [Läderach et al., 2011].
 In general, global observations suggest that the maximum depth of earthquakes along mid-oceanic ridges increases with decreasing spreading rate. Extrapolating centroid depths of faster spreading ridges to full spreading rates below 10 mm/yr, a maximum depth of faulting of 15 km at the most has been estimated [Solomon et al., 1988]. Figure 2 shows more than 60 earthquakes that are located between 15 and 25 km depth below sea level (12–22 km bsf) exceeding these estimates by far. These earthquakes together with less numerous observations at Lena Trough [Läderach et al., 2011] give us the first reliable observation of the maximum depth of faulting at an ultraslow spreading ridge.
 Because earthquakes are generated by brittle failure of rocks, the maximum depth of earthquakes in a region reflects the boundary between brittle and ductile deformation behavior. At mid-ocean ridges, this transition typically occurs at temperatures of about 600°C [McKenzie et al., 2005; Bohnenstiehl and Dziak, 2008]. Applying this model to our observations, we are able to delineate the 600°C isotherm along-axis beneath an ultraslow spreading ridge and get a first quantitative estimate of its temperature regime, confirming the existence of a thick, cold, and brittle upper mantle at ultraslow spreading rates [Montési and Behn, 2007]. Based on a 2-D corner flow model, these authors predict a depth of the 600°C isotherm of about 13 km for a ridge spreading obliquely with a full rate of 14.5 mm/yr and an angle of 50° (representative values for Knipovich Ridge from DeMets et al. ). Our observations clearly point to an even colder temperature regime at Knipovich Ridge away from volcanic centers, suggesting that for example cooling by hydrothermal circulation may be more efficient or the assumption of two-dimensionality may not be valid for the pronounced along-axis variations in melt delivery at ultraslow spreading ridges.
 2. Figure 2 shows a prominent zone devoid of earthquakes in the upper mantle beneath Logachev Seamount located in the central rift valley of Knipovich Ridge (Figure 1). This earthquake-free zone starts at a depth of 10 km below sea level and extends down to 25 km with an along-axis extent of 50 km. Its boundary is clearly delineated by roughly 50 events, which equal about 50% of the earthquakes located in the upper mantle. A second, less prominent seismic gap is located underneath OBS 228 at the southern limit of the survey area.
4 Thermal Structure of Knipovich Ridge
 The along-axis axis distribution of the deepest earthquakes then allows us to trace the 600°C isotherm, which clearly rises to depths of about 10 km underneath the volcanoes. Underneath mid-ocean ridges, the LAB may be located at an isotherm of about 1000°C for the typical stress, pressure, and chemical conditions there [Cannat, 1996]. If we assume that the 600°C isotherm and the 1000°C isotherm have largely the same along-axis shape, our seismicity study yields for the first time convincing geophysical evidence that a pronounced topography of the LAB may exist.
 In Figure 3 we combined a 3-D cross-section of bathymetry and earthquake hypocenters with a conceptual model [Standish et al., 2008], which illustrates how melts may be guided toward volcanic centers along a sloping horizon. We could quantitatively determine the position of the 600°C isotherm, boundary between brittle and ductile deformation. Warm areas with a shallow 600°C isotherm correlate with Mantle Bouguer Anomaly lows [Okino et al., 2002] and increased crustal thickness [Jokat et al., 2012] indicating a higher magma supply.
 Because temperature gradients are unknown, we cannot calculate the position and shape of the 1000°C and, hence, the LAB in detail. However, assuming that the temperature gradient is laterally largely invariant away from the volcanic centers and potentially higher underneath the magmatic centers, the topography of the LAB will parallel that of the 600°C isotherm with even steeper slopes underneath the magmatic centers. Therefore, we estimate that the LAB may be inclined by more than 20° underneath Logachev Seamount, providing a valuable constraint for models of melt flow, which require critical slopes of a permeability barrier to enable flow along its base [Montési et al., 2011].
 The current 10 day data set provides a first glimpse on processes below an ultraslow spreading ridge. Our seismicity study yielded two important results: (1) The lithosphere underneath ultraslow spreading ridges is cold and shows brittle behavior to depths of at least 25 km below sea level. (2) A pronounced thinning of this seismogenic layer underneath a volcanic center is the first geophysical evidence of an often postulated undulating plate base that guides melts toward volcanoes.
 Despite the short observation period, we are confident that a seismic gap beneath the Logachev Seamount exists, because the seismicity pattern in the upper mantle appeared in any arbitrary time slice of our 10 day data set (see Figure S1 in the auxiliary material). Furthermore, it appears unlikely that the earthquakes delineating the seismic gap belong to an accidentally captured seismic sequence on a deep reaching fault. Thus, we propose that the surrounding earthquakes are part of the background seismic activity and that the earthquake-free zones indicate areas of elevated temperatures, where brittle faulting is physically not possible. Our results provide first observational controls for theoretical models of temperature and melt regime underneath ultraslow spreading ridges, yielding evidence for even thicker lithosphere than expected.
 We thank the crew of R/V Polarstern for their excellent support during the expedition, and the DEPAS Pool for providing ocean bottom seismometers. D. Sauter and L. Montési are thanked for very helpful comments. V. S. is funded by the Deutsche Forschungsgemeinschaft grant SCHL853/1-1.