Pervasive detachment faults within the slow spreading oceanic crust at the poorly coupled Antilles subduction zone

Oceanic crust formed at slow-spreading ridges is currently subducted in only a few places on Earth and the tectonic and seismogenic imprint of the slow-spreading process is poorly understood. Here we present seismic and bathymetric data from the Northeastern Lesser Antilles Subduction Zone where thick sediments enable seismic imaging to greater depths than in the ocean basins. This dataset highlights a pervasive tectonic fabric characterized by closely spaced sequences of convex-up Ridgeward-Dipping Reflectors, which extend down to about 15 km depth with a 15-to-40° angle. We interpret these reflectors as discrete shear planes formed during the early stages of exhumation of magma-poor mantle rocks at an inside corner of a Mid-Atlantic Ridge fracture zone. Closer to the trench, plate bending could have reactivated this tectonic fabric and enabled deep fluid circulation and serpentinization of the basement rocks. This weak serpentinized basement likely explains the very low interplate seismic activity associated with the Barbuda-Anegada margin segment above. Pervasive detachment faults in the oceanic basement at the NE Antilles Subduction Zone could enable deep fluid circulation and serpentinization and partly control the reduced interplate seismicity, according to seismic reflection and bathymetric data

O ceanic basement formed at slow-spreading mid-ocean ridges (MORs), exhibits remarkable variations in crustal thickness, seismic velocity and tectonic fabric, as previously inferred from bathymetric data [1][2][3] , paleomagnetic studies 4,5 sampling and drilling of outcrops of deep-seated rocks 6,7 and numerical modelling 8,9 . In contrast, few seismic data image this fabric variability at depth [10][11][12][13] . Poorly sedimented seafloor near MORs causes severe scattering of seismic waves during mapping expeditions, which impedes accurate intracrustal imaging at depth 14 . As a result, deciphering the complex variability of oceanic tectonic fabric in seismic images remains challenging.
This variability, from magmatically robust to tectonically dominated segments, depends on the spreading rate and the relative contribution of tectonic extension and magmatic diking to oceanic spreading 8,9 . At magmatically robust segments, oceanic spreading is mainly taken up by vigorous melt delivery, which leads to typical velocity-depth "Penrose" structure of extrusive basalts overlying intrusive gabbros 15 . In contrast, tectonically dominated spreading favors stretched and thinned crust possibly hosting widely spaced, long-lived, low-angle, ridgeward-dipping detachment faults 16 , exhuming serpentinized peridotites with a varying amount of gabbro bodies. This heterogeneous crustal composition is referred to as the "plum-pudding model" 17,18 . The velocity-depth structure then usually consists of one layer with a rather constant velocity gradient [19][20][21] depending on a decreasing serpentinization degree with depth 22 . Reston et al. 11 argue that more closely spaced faulting may result from a tectonic sequence where a detachment fault forms, slips, flexes and becomes inactive when a new detachment fault develops nearby. However, in the absence of convincing deep seismic images for such tectonic sequences, the model of widely spaced detachment faults frequently prevails.
New bathymetric and multichannel seismic (MCS) data, collected in the NLA trench during cruises ANTITHESIS 1 and 3 23,24 call into question this generic model for the first time. The sedimentary layer reduces the scattering, allowing up-to-6-s-two-waytraveltime (stwt) deep seismic imaging, which is unprecedented for slow-spreading oceanic basement. These data reveal impressive along-strike variations in oceanic fabric, showing an unexpected deep and pervasive tectonic pattern within the basement created at a segment end of the Mid-Atlantic Ridge. These images challenge the long-lived detachment model, highlight that the tectonic imprint of slow-spreading onto the oceanic basement has possibly been underestimated, and raise questions about the seismic consequences of subduction of tectonically dominated, hydrated, serpentinized, and weak oceanic basement patches.

Results and discussion
Oceanic tectonic pattern near the Jacksonville Fracture zone. The~N120°-trending Jacksonville Fracture Zone extends from the Northwestern Atlantic to the NLA Subduction Zone 25 (Fig. 1). To the south, the 15-20 Fracture Zone subducts beneath the margin at a convergence rate of 20 mm/yr in the N254°E direction 26 . Between these fracture zones, the Cretaceous oceanic basement in the trench (Fig. 1A) ages westward 27 . Based on the 300-330 km distance between chron C34 (83 Ma) and C32 (71.6 Ma) 28 , the mean half-spreading rate was low, 26-29 mm/yr, at the spreading center.
The Jacksonville Patch, originated from the ridge segment end close to the Jacksonville Fracture Zone, is currently located in the trench between 18 and 19°N. The bathymetry and deep structure within this patch drastically differs from that of the incoming oceanic plate in neighboring zones.
To the southeast and the northwest of this patch, the oceanic fabric of the incoming plate corresponds to~N20°-trending elongated topographic highs sub-parallel to the magnetic anomalies (Fig. 1B, Supplementary Fig. 1). In addition, in every seismic line perpendicular to the trench (Ant01 07, 10, 11, 12, 14, and 50) reflectors of the seafloor, oceanic sediments, and the basement top step down westward along steep fault planes that dominantly dip toward the margin ( Fig. 2A). These normal faults penetrate the basement down to gently southwestward-dipping discontinuous reflections M, 1.9 to 2.2 stwt beneath the top of the basement, interpreted as the Moho. These faults crop out at bathymetric scarps, directed N100-120°E, sub-parallel or slightly oblique to the deformation front.
This margin sub-parallel faulting in the outer trench wall has long been described as the result of the incoming plate bending into the subduction zone 29,30 . According to these studies, plate bending mainly reactivates inherited tectonic structures of the oceanic fabric when favorably oriented (sub-parallel to the trench) and produces new faults when the fabric is highly oblique with respect to the trench. Offshore of the NLA, the oceanic fabric trends at more than 70°angle to the trench. The southwestward-dipping faults, sub-parallel to the deformation front, are thus likely to be newly formed plate bending faults.
Within the Jacksonville Patch, the bathymetric map and associated dip and strike seismic lines (Ant06, 43, 44, 45, 52, 53, and 54) show a drastically different tectonic pattern in the trench (Fig. 1B, Supplementary Fig. 1B). The smoother seafloor is neither spiked with the N20°-trending ridges of the oceanic fabric, nor deformed by the margin-subparallel scarps of platebending normal faults. In contrast, short, shallow and steep faults dipping toward east and west bound~4-6 km wide grabens in the oceanic basement. These grabens define~N100-110°-directed seafloor undulations trending at 40°angle to the margin front. The seismic lines do not show organized reflections at typical Moho depths. The most striking features are 5-10-km-spaced, convex-up, high-amplitude reflector sequences, which dip from the top of the oceanic basement down to 5 stwt below the seafloor (Fig. 2B). The sequences are 0.1-0.2-stwt-thick (200 to 500 m) and locally up-to-0.5-stwt-thick (1 to 1.5 km).

Ridgeward-Dipping oceanic-basement Reflectors (RDRs).
In order to estimate the true dip direction and geometry of these reflector sequences, we performed depth-conversion of MCS lines Ant45 and 53 interpretations (Fig. 3) as well as a pre-stack depth migration of line Ant45. We used a combined MCS / wide-angle seismic (WAS) velocity model based on nearby WAS line Ant06 31 ( Supplementary Fig. 3). In this model, the basement corresponds to a 5.6-6.5-km-thick single layer with a 5.5-7.4 km/s velocity range from top to bottom and a constant velocity gradient of 0.27 s −1 . Moreover, we used seismic attributes derived from seismic data in order to confirm the reflector sequences geometry. Computing RMS amplitude provides information about reflection physical properties and particularly fluid content 32 . This analysis suggests that the reflector sequences, compared to other intracrustal reflections, show physical properties consistent with fluidrich and/or serpentinized rocks within the upper 6 km of the oceanic basement. At greater depths, the dimming of reflections suggests a decreasing fluid content and/or serpentinization degree ( Supplementary Fig. 4). The mean apparent dip angle increases from 17°to 25°in N125°E-trending line Ant53 and from 15°to 35°in N40°E-trending line Ant45 (Fig. 3A, Supplementary  Fig. 5). These lines intersect each other (Fig. 3B) and reveal that the reflectors dip in a N60-90°E direction, towards the Mid-Atlantic Ridge, with a dip angle that increases from 20-30°in the upper 3 km, up to 45°between 3 and 8 km depth (Fig. 3C). We refer to these sequences as Ridgeward-Dipping oceanic-basement Reflectors (RDRs) (Fig. 4).
Previous seismic data depicted distant convex-up ridgewarddipping reflectors with similar dipping angle at segment ends of slow-spreading ridges, interpreted as large-offset long-lived detachment faults, for instance in the Cretaceous-aged Eastern Central Atlantic 11,13 and at the South West Indian Ridge (SWIR) 10 . At the SWIR, the faults are associated with similar 0.5-stwt-thick sub-parallel bright discontinuous reflectors interpreted as damage zones. The RDRs are also partly consistent with closely spaced,~1-km-thick sequences of convex-up LCRs (Lower-Crust ridgeward-dipping Reflectors) in the Northwestern Atlantic 12,33 as well as in lower crust generated at the faster Mid-Pacific spreading ridge offshore of the Middle America Trench 34 , Japan 35-37 , Alaska 38 , and Hawaii 39 . These LCRs have been interpreted as lithological layering resulting from magma flow in the Atlantic 33 and the Pacific 40,41 . However, discrete spacing of reflectors rather than pervasive layering more readily supports ductile shear zones 37,42 due to spreading-related deep tectonic events 12 and/or anomaly in melt delivery at the mid-ocean ridge 42 .
The RDRs size and geometry partly differ from these analogues. These reflectors extend from the top of the oceanic basement to, at least, 6 km below (Fig. 3, Supplementary Fig. 5) while the LCRs are restricted to the lower crust and sole out downward onto the Moho. Discretely spaced thin sequences of subparallel RDRs are poorly consistent with pervasive and massive fan-shaped layering at Seaward Dipping Reflectors (SDRs) and lava flows. At last, the RDRs are closely spaced and most of them do not deform or fracture the top of the oceanic basement and the sediment layer, contrasting with the classical image of distant detachment faults 11 with topographic expressions 3 , in the Northeastern Atlantic. However, this fault spacing at a slow to intermediate spreading axis depends on the fraction of the plate separation rate that is accommodated by magmatic ridge-axis dyke intrusion 8 . According to these authors, a tectonically dominated slow-spreading ridge segment with moderate magmatic activity can generate closely spaced detachment-type deformation zones during early stages of basement exhumation.
Based on this discussion, we propose that the Jacksonville Patch lithosphere consists of serpentinized mantle rocks, possibly hosting gabbro bodies exhumed along low-angle detachment systems 16 or by serpentine diapirism up high-angle faults 43 . Although we cannot rule out serpentine diapirism, inside corners of fracture zones are known to be prone to detachment faulting 44 , the RDRs more readily image pervasive proto-detachment shear zones related to early tectonic extension at a magma-poor inside corner of the segmented MAR.. In this interpretation, the RDRs more readily image pervasive proto-detachment shear zones related to early tectonic extension at a magma-poor inside corner of the segmented MAR. Approaching the trench, the plate bending, reactivates extensional strain along the RDRs, favoring fluid percolation, rock alteration and serpentinization, increasing acoustic impedance contrast and reflection amplitude. The RMS analysis supports this interpretation showing high RMS amplitude along the RDRs within the upper 6 km and at the top of the oceanic crust above the RDRs (Supplementary Fig. 4).
Seismogenic behavior of subducting serpentine-bearing rocks. The NLA Subduction Zone has hosted only 39 thrust-faulting earthquakes, detected teleseismically (Mw > 5), with focal mechanisms compatible with interplate co-seismic slips since 1973 (Fig. 5). Most of these subduction-type events occurred to the North of Guadeloupe where they are aggregated in two seismicity clusters: from Montserrat to Barbuda and from the Anegada Passage to Virgin Island 45 . Very few of them occurred along the~110-km-wide margin segment in between and to the South of Guadeloupe. In the Southern and Central Lesser Antilles, numerous fracture zones in the subducting South American Plate (Fig. 1A) likely trigger deep crustal hydration and mantle serpentinization [46][47][48] . This high water budget is prone to impede large interplate coseismic rupture, rather favoring alternate slip behavior (SSE, VLFE, EETS) 49 and/or numerous low-magnitude events 46,48 . In contrast, in the NLA, the only fracture zone (the 15-20 FZ) of the subducting North American Plate to interact with the subduction zone ( Fig. 1) has not subducted deep enough to favor dehydration of the subducting serpentinized mantle 47 . This fracture zone, located at less than 30 km depth beneath the forearc 31 , could trigger shallow dewatering and margin tectonic deformation, weakening the interplate contact, reducing the seismic coupling and affecting the megathrust seismogenic behavior. However, the fracture zone underthrusts similarly the two clusters of subduction-type teleseisms (Mw > 5), and the gap in between (Fig. 5) suggesting a low influence onto the interplate seismicity in the NLA.
We propose that the reduced strength of the subducting plate basement at least partly made of serpentinized mantle rocks strongly contributes to the megathrust weakness and the interplate seismicity reduction. Low-temperature species of serpentine minerals, chrysotile, and lizardite have a low coefficient of internal friction, low fracture strength, and a nominally non-dilatant mode of brittle deformation, which favor localized slip on discrete surfaces, cataclastic flow by shear microcracking [50][51][52] , and plastic flow within individual grains 53 . This substantial weakening of serpentine-bearing rocks is not a linear function of the degree of serpentinization but is similar in slightly hydrated peridotites and pure serpentinites 22 . The subduction of an heterogeneously faulted, hydrated and serpentinized basement is likely to generate an interplate patchiness of contrasting frictional properties, which may impede full interplate coupling 53 , instead favoring a mix of stable and unstable behaviors prone to triggering small-Mw, slow-slip, and/or very-low frequency earthquakes 54 . Similar conditions are suspected in anomalous non-seismic regions in locally hydrated forearc mantle within Northeast Japan 53 .
The Lesser Antilles is an end-member subduction zone, which undergoes the subduction of highly hydrated fracture zones and unsuspected large-scale tectonically dominated oceanic patches. Our data depict for the first time pervasive and closely spaced proto-detachment shear planes, reactivated by the plate bending in the trench within the oceanic basement, at least partly made of serpentinized mantle rocks exhumed at a former inside corner of the MAR. The landward extent of this patch is unclear beyond 40 km from the deformation front, because of seismic amplitude loss at great depth. However, downdip, tectonic interaction and fluid circulation between the patch and the 15-20 Fracture Zone possibly alter the forearc strength. Thus, the reduced strength and fluid circulation related to the Jacksonville serpentinized basement, its pervasive tectonic fabric, and the proximity of the hydrated 15-20 Fracture Zone are likely to account for the heterogeneous distribution of subduction earthquakes in the NLA.

Methods
Data acquisition. Our results are based on recent multichannel seismic (MCS) wide-angle (WAS) seismic and bathymetric data collected during cruises ANTITHESIS 1 23 and ANTITHESIS 3 24 . Multibeam swath bathymetry data were recorded using a Kongsberg EM122 and a RESON Seabat7150 (432 -880 beams echosounders) during ANTITHESIS 1 and 3, respectively. We recorded MCS lines Ant01, 10.2 and 12 during Antithesis 1 23 , using a 7699 cu in 18-elements airgun seismic source towed at 17-m-depth and a 3-km-long streamer composed of 288 channels spaced at 12.5 m and towed at 20-m-depth. We acquired lines Ant45, 50, 53, and 54 during Antithesis 3 24 , using a 6500 cu in, 16-elements airgun seismic source towed at 14-m-depth and a 4.5-km-long streamer composed of 720 channels spaced at 6.25 m and towed at 15-m-depth. Shots were fired every 75 m providing a 30-fold coverage.
Data processing. Swath processing consists in spikes and excessive slopes removal by automatic procedure and manual ping editing using Caraïbes ® and Globe ® softwares (IFREMER). Digital terrain models were produced with a grid spacing of 75 m. Vertical accuracy is between a few meters and tens of meters depending on depth. Bathymetric and slope maps were calculated and processed using QGis. Maps reveal reliefs in the order of tens of meters high and few hundred meters apart.
MCS data processing includes quality control, binning, band-pass filtering, fK filtering, external and internal mutes, noise attenuation, predictive deconvolution, multiple suppression, velocity analysis, normal move out and dip move out corrections, stacking and pre-stack time migration, using Solid-QC ® (Ifremer) and Geovation ® (CGG-Veritas) Softwares 55 . We performed iterative Prestack Kirchhoff time migration (PSTM) to yield optimal migration velocities and form the final prestack migrated images. PSTM results in focusing correctly seismic energy from genuine basement reflections but not from out-of-planes arrivals from seafloor or basement propagating through regions of lower root-mean-square velocity. Thus, any intra-basement event observed on presented PSTM images can be interpreted as true reflection.
Depth-converted interpretation and depth-migrated seismic data. Converting and/or migrating to depth the seismic images is a mandatory condition to address the questions of the geometry, dipping angle, and orientation of the RDRs. The depth of investigation (11 to 18 km) is much larger than the streamer length (4 and 4.5 km for Antithesis 1 and 3, respectively). At great depth, this relative shortness of the streamers results in high uncertainties in interval velocity model strictly inferred from Normal-Move-Out (NMO) velocities. In order to reduce this uncertainty, we build composite velocity models based on MCS lines Ant45 and 53 and WAS line Ant06 31 ( Supplementary Fig. 3), located within the Jacksonville Patch 70 km to the northwest. These models consists in: 1/ NMO velocities converted to interval velocities using the Dix formula at shallow depth (i.e., from the seafloor to the topmost hundreds of milliseconds in the subducting basement) and  Fig. 2B and Supplementary Fig. 2, respectively. Figure 4 shows an interpreted perspective view of the RDRs. 2/ velocities inferred from first-arrival traveltime tomography for line Ant06 at greater depth in the basement and the mantle. We then base our investigations on two complementary methodological approaches. We convert to depth the interpretation for seismic lines An45 and 53 (Fig. 3) using these combined MCS/WAS velocity models. In order to confirm this conversion, MCS line Ant45 is migrated to depth ( Supplementary Fig. 5) with a preserved amplitude Pre-Stack Depth Migration (PSDM) approach 56-60 performed in the angle domain. The velocity macro-model is iteratively corrected during migration, using the "migrationvelocity-analysis" approach 61 until Common Image Gathers (CIG) show flat reflections. When this condition is satisfied, the CIG are stacked, providing an increased accuracy for the migrated image. In this methodological approach, it is noteworthy that parallel lines Ant45 and 06 are located 70 km from each other. As a result, the PSDM of line Ant45 should be considered as an additional constraint for the RDRs geometry complementing the rougher depth conversion, more than as a robust image for the deep structure of the subduction zone. Despite this uncertainty, both methods result in similar location, depth, geometry, and dipping angles for the RDRs. These reflectors are slightly deeper and steeper in the depthconverted image than in the interpreted PSDM MCS line.
RMS amplitude analysis. Seismic amplitude attributes analysis is commonly used in basin and oil exploration in order to identify and delineate structural and stratigraphic features associated with fluid-rich intervals 32 . The Root Mean Square (RMS), based on reflection coefficient, independently from the reflection polarity, is particularly suited for fluid content analysis. The RMS amplitude A RMS is calculated from original signal amplitudes a i (t) over a time window of N samples indexed with i, using the ≪ Petrel ≫ software (Schlumberger): As a result, this analysis estimates the signal overall amplitude and describes the signal average amplitude within a time window.

Data availability
Every geophysical data of the ANTITHESIS cruises are available on the internet site of the French Oceanographic Fleet (https://campagnes.flotteoceanographique.fr/search). Interested readers write/select "Antithesis" in field "search campaign" and the desired data set in field "Data Managed by SISMER." Once every needed data set is selected, the readers can download it from "My basket" page.

Code availability
Seismic data processing used softwares SolidQC (Ifremer) and Geovation (CGG), bathymetric data were processed with softwares Globe and Claritas (Ifremer). We used Kingdom Suite for the RMS analysis. Maps and 3D bathymetric views were drafted using GMT, Qgis, and Adobe Illustrator. The ray-Born PSDM code derives from an original private version by P. Thierry