A record of post-accretion asteroid surface mixing preserved in the Aguas Zarcas meteorite

Particle ejection and redeposition events on the surface of asteroid 101955 Bennu, which led to transport, mixing and loss of material, have been observed frequently by NASA’s OSIRIS-REx mission. Besides large-scale impacts, this may be one of the most important post-accretional processes on small carbonaceous asteroids. Here we looked for relics of such activity in a Bennu analogue, the carbonaceous chondrite Aguas Zarcas. We discovered compact fragments that were strongly shocked, redistributed and deposited onto an unshocked lithology, consistent with surficial re-accretion on Aguas Zarcas’s parent body. Such re-accretion could be driven by large-scale impacts or by frequent pebble transport from endogenous asteroidal activity such as observed at Bennu. The latter hypothesis is supported by the matching size distribution of the Aguas Zarcas compact fragments with that of the Bennu ejecta. Such mixing has hitherto been unexplored in other regolith breccias, and further analysis will determine how common such processes are. Some fragments of the Aguas Zarcas carbonaceous meteorite have been shocked before being redeposited over an unshocked lithology. As their size distribution is similar to that of the ejecta observed at Bennu, they might be the signature of activity of the Aguas Zarcas parent body. Alternatively, they might be the result of a large-scale impact.


Results
We found a compact lithology by disaggregating ~79 g of AZ fragments from a large and freshly broken sample (1.894 kg) into fine powder with the freeze-thaw method. More than 10 sub-cm-sized compact fragments (3.2 wt%) with a dull, black, smooth lustre similar to the appearance of melt rock survived the disintegration (Methods). The compact lithology was more resistant to the mechanical breakdown and has a higher density (~2.7 g cm -3 ) and compactness compared with the regular host lithology (~2.4 g cm -3 ). Because of the low abundance of the compact lithology, we assumed that a randomly selected AZ fragment is not compact and belongs to the regular AZ lithology.
To better understand the petrological history of AZ, we μCT-scanned Murchison and Leoville meteorites as reference samples. Murchison is a CM2 chondrite that is similar to Aguas Zarcas in petrology and mineralogy 8 ; Leoville is a CV3 chondrite that has undergone strong deformation 9 . We observed prominent deformation and a preferred orientation of chondrules in the μCT and SEM data of compact AZ and Leoville but no such effect in regular AZ and Murchison (Fig. 1). Murchison is one of the best-studied meteorites with heterogenous deformation from which an undeformed fragment was studied here 10 . Leoville is known as one of the most deformed chondrites showing flattened chondrules aligned in parallel 10,11 . We utilized undeformed Murchison and strongly deformed Leoville as two extreme endmembers for our study.
To assess and quantify the type and strength of deformation, we outlined chondrules in the μCT dataset, fitted ellipsoids following an established method 12 and used axial ratios, fabric parameters and shape analysis with ternary diagrams 13,14 . Fabric is the geometric arrangement of components in a rock. In our case, it refers to the spatial arrangement of chondrules and their preferred orientation that manifests itself as elongated (rod-shaped, with lineation) or flattened (disc-shaped, with foliation) shapes. Fabric parameters (K and C) are defined by a set of direction vectors of axes of the best-fit ellipsoids of chondrules and can be used to distinguish deformation type (lineation or foliation) and quantify the deformation intensity 15 . Shape parameter K < 1 for the longest axes set and K > 1 for the shortest axes set demonstrates a foliation, and K > 1 for both axes sets indicates a lineation. For the strength parameter C, higher values indicate stronger fabrics.
The average axial ratios of fitted ellipsoids of chondrules increase from regular AZ and Murchison to compact AZ and Leoville, which is consistent with the result of C parameters (Supplementary Table 1). The fabric strength varies from 'moderately weak' to 'moderately strong' with C of longest axes ranging from 0.75 to 2.53 and C of shortest axes ranging from 1.04 to 2.65. These parameters are highest for compact AZ and Leoville and are lowest for regular AZ and Murchison, indicating that the former two are strongly shocked and that the latter are weakly shocked. K parameters for the longest axes in compact AZ and Leoville range from 0.09 to 0.40 while those for the shortest axes range from 1.97 to 13.06, arguing that the fabrics in compact AZ and Leoville are both foliations. That is, the chondrules are more flattened than elongated, though the difference cannot be distinguished by two-dimensional (2D) analysis ( Fig. 1). Meanwhile, in the ternary diagram ( Fig. 2) that plots an object's shape as a function of a perfect sphere, elongated rod and platy disc shape, we note two patterns of chondrule shape distribution. One group, represented by regular AZ and Murchison, has the majority of their chondrule shapes located in the top 'equant shape' sub-triangle. The other group, represented by compact AZ and Leoville, has a remarkable number of points in the areas signifying more deformed shapes (44.0% for compact AZ, 64.6% for Leoville, 17.4% for regular AZ, 19.0% for Murchison). Combining the indices above, we infer that the deformation intensity sequence is regular AZ < Murchison < compact AZ < Leoville. An exception to this sequence is regular AZ fragment RF-3, which displays a higher axial ratio and C parameter compared with other regular AZ fragments and is closer to compact AZ (Extended Data Fig. 1). Therefore, we classify RF-3 as a compact AZ fragment. The classification is not binary. RF-3 is more deformed than regular fragments but may not be as deformed as the other compact fragments. This deformation sequence is also reflected by the fragments' average densities (2.72 g cm -3 for compact AZ, 2.43 g cm -3 for regular AZ and 2.55 g cm -3 for RF-3).
Besides deformation, fractures and veins are often used to investigate meteorite stress histories in shock events 3  section of compact fragment CF-10, we observed 6 major fractures (0.5-6 mm long, 7-20 μm wide) and many minor fractures (several hundred micrometres long, ~2 μm wide) in the matrix and some small fractures within chondrules, some of which are filled with metal sulfide veins oriented independent of the chondrule flattening direction (10-20 μm; Fig. 1). In contrast, the fractures in the matrix are all empty and mostly parallel or subparallel to the direction of the chondrule elongation. We also examined the μCT data of regular AZ and found several unfilled fractures (10-20 μm wide, 1-2 mm long) without any preferred orientation ( Supplementary Fig. 1). Fractures and deformation are common in chondrites 17 , and an impact origin is supported by an increasing amount of evidence such as the correlation between shock stages and aspect ratios of chondrules, noncoaxial strain and the abundance of unfilled fractures versus fractures filled with secondary minerals 10,12,16,18 . Meanwhile, fracturing in the compact AZ matrix is approximately in the same orientation as chondrule flattening. As shown in Fig. 1e, angles between the fractures and the direction of chondrule flattening vary within ±8° with a mean of 1.3° and 1 s.d. of 3.7°. Therefore, we propose that the same generation of impact events caused the shock effects observed in the compact AZ lithology and that no remarkable aqueous alteration occurred after the shock; otherwise, these fractures would be filled with secondary precipitates. Metal sulfide veins are not common in CM chondrites but are often seen in ordinary and some CV meteorites 18,19 that are likely to have experienced collisions and heating events. While veins from impacts are usually large and cross into the matrix, the veins in compact AZ are thin (10 to 20 μm wide) and only exist in chondrules. Based on cross-cutting relationships and the observation that fractures outside of chondrules are unfilled, we infer that the chondrule veins formed before the impact-induced matrix fractures. Otherwise, we would expect to see all fractures, including those in the matrix, to be filled.
We can exclude deformation by the burial processes: we modelled the lithostatic pressure for chondritic bodies with varying sizes ( Supplementary Fig. 2) and find that the maximum pressure in the centre of a Ceres-like asteroid is only about 0.5 GPa. The non-isotropic stress that may cause deformation is typically lower than the lithostatic pressure 11 . This is much deeper than a plausible burial depth and a much lower pressure than needed to explain the deformation of compact AZ (see the 'Shock pressure estimate' in the following paragraph).
In many meteorite types, shock effects in olivine are used to determine the shock pressure, but they may not reflect the shock history of CMs well because the abundant matrix (~70 vol%) in CM meteorites can remarkably attenuate a shock wave to a low intensity such that it cannot affect olivine crystals 3 . Indicators of shock pressure in CMs include chondrule flattening and fractures. First, empirical relationships between chondrule aspect ratio and shock pressure were established in impact shock experiments 10,20 . Based on these relationships for CV and CM chondrites (Fig. 3), we determined shock pressures for Leoville and compact AZ as ~17 GPa and ~18 GPa, respectively. The published shock stage for Leoville is S3, corresponding to a shock pressure of 15-20 GPa in a single impact 18,21 . The consistency between the shock pressure determined from chondrule aspect ratio and published shock stage for Leoville demonstrates the suitability of this method. Second, the existence and density of fractures are qualitative indicators of shock pressure. In shock experiments with Murchison, the recovered sample showed that the fracture (<5 μm) density in the matrix increased slightly when the pressure increased up to 10 GPa. At 21 GPa, fractures became wider (20 μm) and more preferentially (r 1 -r 2 )/(r 1 -r 3 ) (r 1 -r 2 )/(r 1 -r 3 ) r 2 :r 1 r 2 :r 1 r 2 :r 1 r 3 :r 1 r 3 :r 1 r 3 :r 1

Fig. 2 | Chondrules fitted with ellipsoid shapes.
The ellipsoid axis lengths are used to plot each chondrule shape within the triangle whose apices represent the idealized particle shapes. r1, r2 and r3 represent the longest, intermediate and shortest axes of each fitted ellipsoid, respectively. The fraction of data points in the top sub-triangle (red) is 82.6% for regular AZ, 81.0% for Murchison, 56.0% for compact AZ and 35.4% for Leoville. n is the number of best-fitted ellipsoids. orientated and olivine grains showed undulatory extinction and planar fractures, consistent with shock stage S3 10 . The occurrence of 2-20-μm-wide, unfilled fractures in compact AZ matrix (some pass through the flattened chondrules) parallel or at a low angle to the direction of chondrule deformation is consistent with a shock pressure of 15-20 GPa.
Two types of impact collision have been considered for the origins of meteoritic breccias 4 . Accretionary impacts happened during the accretion of asteroids at relatively slow speeds (typically less than a few hundred m s -1 ). Hypervelocity impacts occurred after asteroidal orbits were dynamically excited when asteroids collided at speeds of a few km s -1 . An impactor with a speed <1 km s -1 cannot generate pressure greater than 10 GPa (ref. 10 ). Most of the meteorites that contain high-aspect-ratio chondrules provide independent evidence of hypervelocity impact in the form of shock fractures 16 that we also observed in compact AZ fragments. Therefore, we infer that the compact lithology must have experienced at least one hypervelocity impact.

Discussion
AZ is highly brecciated with multiple lithologies that were thought to be the result of different degrees of aqueous alteration and impact modification 7 . Brecciated carbonaceous chondrites are not unusual. However, the occurrence of strongly deformed fragments that include oblate chondrules next to undeformed rock fragments with spherical chondrules is striking. Aqueous alteration cannot deform rocks and explaining the observations requires another process 22 . Also, because of the cumulative nature of compaction and shock events, heterogeneous shock effects on the sub-millimetre scale can be expected 23 , but shock propagation is not likely to produce deformed and undeformed lithologies in such close proximity. Rather, the deformed compact fragments must have been transported into an undeformed lithology before final lithification. Previous studies on microstructure and misorientation of olivine in chondrules found a deformation difference between olivine grains in thin sections that formed before or during the agglomeration or originated from different parent bodies 24,25 . However, in our study, chondrule flattening is restricted to the compact fragments and parallel to fractures in the matrix, so the shock event must have occurred after the agglomeration. We compared the petrology and mineralogy of the compact AZ with that of the regular AZ and did not find any compositional discrepancy (Supplementary Text and Extended Data Figs. 2 and 3). Thus, the compact fragments are not xenoliths from a different body but rather deformed rock fragments of the same composition that were transported into an undeformed lithology before final lithification.
Variably shocked materials within one sample are sometimes observed in impact melt rocks/breccias containing shock-melted clastic material and unmelted but substantially shocked material 26,27 . However, strongly shocked lithologies embedded in unshocked lithologies (not as xenoliths) have been rarely observed 17,28 . Previous studies on NWA 7298 (an H3.8 chondrite) and Mokoia (a weakly shocked CV3 chondrite) showed distinctly shocked lithologies within single hand samples 28,29 , which were attributed to relatively strong impacts that produced the different types of rock simultaneously but were spatially separated. A necessary and subsequent transport would be required but was not explained in detail. Conceivably, shock features would be more likely to be prevalent throughout the lithology if exposed to high pressures during a hypervelocity impact. In most meteorites showing deformation effects, all chondrules show the same degree of deformation, with Leoville being one of the most extreme cases. Often large-scale impacts are proposed to explain mass transport and regolith mixing on asteroids 30,31 . These compact AZ fragments could be distal ejecta from hypervelocity impacts on AZ's parent body or from re-accretion of early collided asteroids. However, the endogenous transport that was observed on asteroid Bennu provides a novel explanation for observations of AZ, NWA 7298 and Mokoia.
Particle ejection and re-accretion of millimetre-to centimetre-sized particles onto the regolith of Bennu 1 is an important but until recently undiscovered mass transport process on asteroids. According to the OSIRIS-REx observations, the larger events with more than 70 particles observed each time occurred every 2 weeks and smaller detected events with less than 25 particles observed each time happened every 1-2 days. During the time period of observation, no hypervelocity impact events were detected. This implies that hypervelocity impacts are much less frequent than the particle ejection and re-accretion events. Through the latter, about 10 4 to 10 5 particles may be launched per year, with 85% of them redeposited and the remainder exceeding the escape velocity 2,32 . Consequently, a large number of pebble-sized fragments  10 and Nakamura et al. 20 , respectively, acquired from shock experiments were used to create standard curves (red dashed lines). The last two Murchison data points in a with high shock pressures were not included in the linear regression because of the nearly constant aspect ratio at higher pressure >25 GPa. Triangles are used for our data. Error bars represent one s.d., not data uncertainties, so the mean value is still useful in determining shock pressures.
were relocated on Bennu's surface, leading to global and thorough mass transport and regolith mixing. Many mechanisms have been proposed to explain the ejection events, while low-energy dust impacts and thermal fracturing received the most attention 2,32-34 . Spectral data of Bennu suggest a CM composition 6 , similar to AZ, and therefore thermal stressing and mass transport may have also occurred on the AZ parent body and mixed the compact AZ lithology with the regular one. We compared the size distribution of compact AZ fragments with that of Bennu ejecta and found a good match (Fig. 4) that supports a similar breakup/transport mechanism. Both datasets are truncated at the lower end due to the observational detection limits of OSIRIS-REx for Bennu and sample processing in the laboratory for AZ. To better understand such activity on asteroids and its potential to transport mass globally on the AZ parent body, we conducted a Monte Carlo analysis. Tens of thousands of fragments were released from the surface of asteroids with 1-100 Bennu radii. The ejecta redeposited onto the surface after orbiting the asteroid up to several times or escaped the asteroid's gravity directly. Fragment trajectories were recorded and the efficiency of global transport was evaluated. To quantify the particle relocation, we used the concept of displacement angle as the central angle between the launching site and landing site. A large displacement angle represents a global transport, otherwise a local one, and we arbitrarily set a threshold equal to π/4 to distinguish between the two types of transport. Our model yields a pronounced equatorial excess of particle redeposition on a Bennu-sized body (Extended Data Fig. 4), consistent with the spacecraft observation of Bennu's shape 1,2 . The model also predicts that the particle ejection and redeposition process operates as an effective mixing process on asteroids with a radius of up to 50 Bennu radii (Fig. 5).
The model results support the hypothesis that such a process occurred on the AZ parent body. Hypervelocity impact deformation is a local phenomenon, and specimens from the same meteorite fall have the identical petrofabric in most cases 17,28 . Nonetheless, if the active pebble transport occurred on the AZ parent body, it was able to eject and reaccrete compact AZ fragments globally. After mass transport, compact AZ fragments were mixed into unshocked regular AZ lithology, and later impacts consolidated the breccia and ejected it to Earth. The consolidating impact may have resulted in the final ejection and delivery to Earth.
Another possible explanation is that the compact fragments are distal ejecta from a large-scale impact. Large hypervelocity impacts are less frequent, and none of them were observed by OSIRIX-REx. An additional strong impact occurring in the impact site to redistribute the shocked lithology into an unshocked one is rarer.
Based on our observations, we propose the following scenario for the formation of AZ (Extended Data Fig. 5). (1) A hypervelocity impact caused deformation of chondrules and formed cracks in matrix (2-20 μm wide). (2) The compact lithology of the AZ parent body was fragmented by a combination of meteoroid impacts and thermal fracturing, and a pebble transport process such as observed on Bennu ejected compact fragments that reaccreted into unshocked regolith later. (3) The absence of precipitates in cracks implies that no detectable aqueous alteration occurred after that, and that the AZ breccia was lithified by one or multiple later impacts. (4) A meteoroid containing regular and compact AZ lithologies was ejected from the parent body by an impact and delivered to Earth.
The high frequency of pebble transport on Bennu and AZ-like asteroids seems at odds with the low frequency of occurrence of compact fragments in unshocked lithology seen in most carbonaceous chondrites. There are several possible reasons for this. First, the abundance of compact material is relatively low (3.2 wt% in this study), and most studies do not usually survey sample volumes as large as in this study; meanwhile, more common surveys of polished sections only provide information from a small sample of the whole rock. Second, if undeformed pebbles were transported in this way into a similar host lithology, we cannot identify them. Third, the ejection process may be more complex than expected and not common on most carbonaceous asteroids. In fact, different shock stages of Murchison have been reported 3,18,20 , and the average three-dimensional (3D) aspect ratio of chondrules in Murchison ranges from 1.75 ± 0.39 (ref. 16 ) to 1.54 ± 0.22 (ref. 12 ) and 1.30 ± 0.15 (this study), whereas the chondrule aspect ratio in 2D sections ranges from less than 1.2 (ref. 10 ) to 1.67 ± 0.51 (ref. 16 ). All the evidence is consistent with similar ejection and redeposition processes on Murchison's parent body. The OSIRIS-REx observations of pebble transport that redistributes material on the surface of Bennu are undeniable and were frequent during OSIRIS-REx's residence in Bennu orbit. Thus, each volume of rock on Bennu's surface should contain some fraction that was delivered by the pebble transport from a different region. This process breaks the tacit assumption that mixing and brecciation is solely by large-scale impacts and advances our understanding of post-accretional processes. Documenting such activity with meteorites is challenging because of the need to demonstrate that the meteorite fragments experienced relocation on the parent body via the Bennu-type transport, not distal ejecta from hypervelocity impacts. While we cannot exclude the latter, we argue that pebble transport analogous to Bennu more likely explains our observations in AZ. The main arguments for this include the much higher frequency of the pebble transport process, the matching size distributions of observed ejected pebbles from Bennu and compact fragments and the predicted global redistribution from our Monte Carlo model. Instead of the conventional impact-mixing hypothesis that is usually offered as the sole explanation, active pebble transport is an important process that now needs to be considered, in addition to impacts, to explain mixed lithology in CM chondrites such as in AZ. We predict that other carbonaceous chondrite breccias, in particular CM chondrites, as well as the mission-returned samples from the asteroid Bennu, may contain compact fragments embedded in an unshocked lithology. Studying other carbonaceous breccias will provide new insights into the diversity and relative importance of this and other surface processes on active asteroids.

Methods
Sample preparation. A large 1.894 kg fragment of Aguas Zarcas was recovered rapidly after its fall before rain, purchased by Terry Boudreaux and donated to the Field Museum of Natural History. This specimen, FMHN ME 6112, is stored at the Field Museum in a stainless-steel cabinet in an inert nitrogen atmosphere at room temperature. A total of 79 g of fragments were separated from the large sample of AZ, FMHN ME 6112 with cleaned stainless-steel tools in a nitrogen-filled glove bag. We used freeze-thaw disintegration as the first step of an effort to separate objects of interest including refractory minerals, isolated olivine grains and presolar grains from the fine-grained matrix and organic matter of AZ. The selected pieces were roughly divided into 10 ~8 g chunks in ultrapure water (18.2 MΩ × cm electrical resistivity; Milli-Q) and each was disintegrated using alternating cycles of liquid nitrogen and 50 °C water. Typically, 30 cycles can break down the matrix of a CM2 chondrite such as Murchison into powder. With AZ, most fragments were disaggregated within 50 cycles, however, more than 10 sub-centimetre-sized fragments (3.2 wt%) remained intact and showed no signs of mechanical breakdown after 112 cycles. We call these intact fragments 'compact AZ' .
An additional 11 compact AZ fragments and 3 randomly chosen AZ (FMNH ME6110.1) fragments that were not processed by freeze-thaw were μCT-scanned. The fragments of compact AZ are named CF-1 to CF-11 and the randomly selected AZ fragments are named RF-1 to RF-3.
X-ray microtomography. We scanned all AZ samples (CF-1 to CF-11 and RF-1 to RF-3) at the University of Chicago's PaleoCT Lab, on a GE v|tome|x S model micro-CT scanner using a 240 kV microfocus tube. The fragments were mounted in a 15 ml tube and scanned at a spatial resolution (or voxel size) of 17.028 μm. An 80 kV voltage and 220 μA current were used with an image acquisition time of 500 ms per frame. Three frames were captured and averaged for each position to reduce noise and a 0.2 mm Cu filter was used to reduce beam hardening. The total scan time for the tube of specimens was 1 h 50 min.
Two of the larger pieces of AZ (CF-10, 0.591 g, and RF-1, 0.730 g) as well as specimens of Murchison (FMNH ME2644; 1.171 g) and Leoville (FMNH ME2628.2; 1.706 g) for comparison, were scanned at the University of Texas High-Resolution X-ray Computed Tomography Facility (UTCT) at higher resolution. These 4 samples were scanned on a Zeiss Versa 620 at 80 kV and 125 μA with varying acquisition time per frame (40-70 ms) with 1 frame per position for all samples except Leoville, which had 2 frames per position. The LE3 filter was used for all scans and a beam-hardening correction was applied during reconstruction with the scanner software and the final voxel size of each sample scan was 6.77 µm (both AZ fragments; scan time 59 min each), 8.47 µm (Murchison; scan time 53 min) and 11.01 µm (Leoville; scan time 74 min).
Scanning electron microscopy. After μCT scanning, the compact fragment CF-10 was embedded in Buehler EpoxiCure 2 epoxy and cross-sectioned parallel to the long axis of the flattened chondrules with a Buehler IsoMet low-speed diamond wafering saw. The section was coarsely polished with diamond Allied High Tech Products Inc. lapping film followed by a final polish with Allied 1 μm diamond slurry. The polished mount was imaged and mapped with a field-emission TESCAN LYRA3 SEM/Focused Ion Beam (FIB) equipped with two Oxford XMax Silicon Drift Detectors (SDD) 80 mm 2 energy dispersive X-ray spectroscopy (EDS) detectors at the University of Chicago. An EDS map, a backscattered electron (BSE) map and a secondary electron map were acquired with an acceleration voltage of 15 kV and a beam current of 470 pA and a typical pixel dwell time of 25 μs at a nominal spatial resolution of 0.369 and 0.185 µm per pixel, respectively. EDS spectra were used to determine the mineral chemistry qualitatively at an accuracy of <5 wt%. Cross-sectional morphology and texture, including fractures, were examined with the EDS, secondary electron and BSE maps using Oxford AZTec software.
Chondrule segmentation and deformation analysis. First, we determined the μCT components in compact AZ by calibrating the CT data by comparing BSE and EDS maps of the polished cross section of compact AZ fragment CF-10 with a matching μCT slice. There are three types of object in AZ that we identified based on their grayscale values within the μCT data ( Supplementary Fig. 3): small bright objects without well-defined shapes, light-toned objects and dark-toned objects. Here we mainly discuss chondrules and neglect irregular clasts. Earlier μCT studies 12,16 of Murchison have shown that the brightest components are metal and sulfides such as pentlandite and that light-toned and dark-toned objects are mostly Fe-bearing chondrules/calcium-aluminum-rich inclusions and Fe-poor/ Mg-rich chondrules, respectively. SEM data of the polished AZ CF-10 confirm the same μCT components as in Murchison. In the μCT data of Leoville, we only observed dark chondrules and bright metal/sulfides. µCT data of regular AZ show the same three object types as compact AZ. According to previous research 12, 16 , dark-toned objects (that is, Mg-rich chondrules) are typically more deformed and display a stronger fabric compared with bright (metal and sulfide) and light-toned (Fe-bearing chondrules and calcium-aluminum-rich inclusions) objects. Regardless of the reason for this observation, we only delineated and segmented the dark-toned objects in μCT data and calculated the fabric strength to avoid any potential observational bias with objects of different X-ray contrast.
Second, we outlined components of interest (here dark-toned chondrules) from the tomographic dataset into distinct volumes of interest. We used manual segmentation in 3D Slicer software (http://slicer.org) where we used the 'draw' tool to mark chondrules in individual 2D slices, then filled between slices to obtain a 3D visualization 35 . This method is labour-intensive and time-consuming if performed for every chondrule in the dataset. Therefore, we only applied it to small fragments and used a more efficient alternative, the partial segmentation method, for large ones. For partial segmentation, one or more representative cross-sections in each chondrule's orthogonal plane are chosen for segmentation excluding the ambiguous chondrules such as those that are in contact with each other. The effectiveness of this method to accurately calculate the orientation and degree of anisotropy of objects in rocks relative to the full segmentation has been examined and confirmed 12 . In this study, we used the whole segmentation method for samples scanned at University of Chicago, as these datasets are small due to their lower resolution, as well as for Leoville, where our scanned volume contains only a few chondrules due to their relatively large size. For the remaining datasets, we used the partial segmentation method.
Third, after segmentation with 3D Slicer, we exported all the segments to DICOM (Digital Imaging and Communications in Medicine) files, loaded them into Fiji and converted them to TIFF files. For each chondrule, we used Blob3D 36,37 (http://www.ctlab.geo.utexas.edu/software/blob3d/) to determine the size, location and orientation information via best-fit ellipsoids to either the full segmentation or partial segmentation via a set of orthogonal planes. Orientation biasing can occur when an object covers only a few voxels. To avoid that, we removed objects with a short axis of less than three voxels 12 . To make the data volume manageable, we divided each large tomographic dataset into several subvolumes and segmented chondrules within each individual subvolume. This enabled faster processing of the data and a reduction in file size. We segmented 825 dark-toned objects in total. Parameters of best-fit ellipsoids for each object are shown in Supplementary Table 1. Fabric analysis of the tomographic data in this work follows an established method 12 , and further details regarding parameter calculations reported in Supplementary Table 1 can be found in that work. Here we briefly introduce the quantitative analysis of deformation. We take the direction vectors of a set of axes of the fitted ellipsoids as an example. These directions are plotted on stereonets in a lower hemisphere projection, and the forming pattern is used to test whether the orientations are non-random. Meanwhile, Woodcock and Naylor 15 defined K and C parameters to describe the shape and strength, respectively, of a fabric. An orientation tensor (3 × 3 matrix) is calculated from the above direction vectors, and three eigenvectors of the tensor are defined as S 1 , S 2 , S 3 . K is defined as K = ln(S 1 /S 2 )/ (S 2 /S 3 ) and C is defined as C = ln(S 1 /S 3 ). K ranges from zero (girdle or 'great circle' distribution on a stereonet for K < 1) to infinite (cluster distribution for K > 1) 38 . C ranges from zero (no fabric) to four or above (strong fabric) and is manifested as the concentration of data points on a stereonet 38 . Supplementary Fig. 4 illustrates K and C parameters and the chondrule orientations for four types of rock in this study. Shock pressure determination. Previous studies used Murchison and Allende in shock-recovery experiments to build up empirical relationships between aspect ratio and shock pressure in a single impact shock event 10,20 . In Murchison, 10 GPa was a threshold over which the aspect ratio started transferring from ~1.2 to ~1.5 (Fig. 3). Meanwhile, 25 GPa was another threshold over which the aspect ratio kept approximately constant. Specifically, the aspect ratios of chondrules in an impacted sample had a large range, but the distribution of the aspect ratios moved clearly with an increasing shock pressure. Accordingly, the mean values of those ratios rose. Also, the aspect ratios of unshocked Murchison's and Allende's chondrules and the data that were acquired under extremely high pressure that did not cause further deformation were not counted for the linear regression. In the shock-recovery experiments, the recovered samples were cut along the shock compacting axis, such that the mean 2D aspect ratio of chondrules in the section was most comparable to the mean ratio of the longest axis length to the shortest axis length in our 3D model (called 3D aspect ratio).
Lithostatic pressure model. At depth within a spherical asteroid, the force balance is as follows: GM/r 2 × ρ4πr 2 dr = −4πr 2 dP, where r is the radial distance from the centre of the parent body, G is the gravitational constant, M is the mass of the material below r, ρ is the density and P is the pressure. The left side is the gravitational force of a shell with a width of dr at a radial distance r from the centre and the right side is the supporting force provided by the pressure gradient. Due to M = 4/3πr 3 ρ, the simplified equation of force balance is dP = −4/3πGρ 2 rdr. The solution is P = 2/3πGρ 2 (R 2 − r 2 ), where R is the radius of the parent body. When r = 0, P reaches the maximum, that is, the pressure at the centre. Because most stony meteorites have densities 39,40 on the order of 3-4 g cm -3 , we take ρ = 3.5 g cm -3 in the model. We consider two cases to visualize the pressure profiles in meteoritic parent bodies. One is the maximum pressure (centre pressure) for spherical objects with different sizes; the other is the depth-pressure profile for a 100-km-sized body ( Supplementary Fig. 2). The calculated maximum pressure for the Moon is 5.2 GPa, and most petrological experiments and seismic detections all support a ~5 GPa pressure at the lunar core or core-mantle boundary [41][42][43] . The maximum pressure for a 100-km-sized body is <0.02 GPa.
Monte Carlo model. The movement of ejecta on Bennu is controlled by multiple forces such as Bennu's gravity, solar radiation, reflected pressure, Poynting-Robertson effect and so on. The gravitational force in the vicinity of Bennu is 2-6 orders of magnitude higher than the other forces 2 ; therefore, and for simplicity, we consider it as the only driving force in our model. Besides this, we set up initial conditions that include initial velocities of particles, launch position and rotation of the central body. The observed velocity of Bennu ejecta ranges from 0.05 m s -1 to >3.3 m s -1 , and the observations may not include all particles, especially fast-moving ones. Thus, we take 0.05-5 m s -1 as the initial velocity range. The particles can be ejected from anywhere on the surface, but were more frequently observed from low latitudes. We adopt the distribution of ejection sites from Chesley et al. 2 to our model. Generally, the spin period of asteroids decreases with their size and clusters between 2 and 12 hours (ref. 44 ). The rotation period for Bennu is 4.3 hours, and we apply this to all the simulations in this study. First, we modelled particle movement on a spherical body whose mass, bulk density and radius are the same as that of Bennu. We released 50,000 particles and only those with low velocities (<0.35 m s -1 ) fell back on the surface. Extended Data Fig. 4 depicts the distribution of the sine of latitude for ejecta deposition. We also ran our model with larger asteroids with 10 to 100 Bennu radii. Bennu's bulk density is low, ~1.26 g cm -3 , because it is a rubble pile asteroid with a high porosity 45 . Nevertheless, the fragment density should be close to that of its meteorite analogue AZ (~2.4 g cm -3 ). Here we argue that 2.4 g cm -3 is the approximate upper limit for such carbonaceous chondrites, because it ignores the pore space in the parent body. The mean densities of C, S and M class asteroids are 1.38, 2.71 and 5.32 g cm -3 , respectively, from calculations 46 . We adopted 1.26 and 2.4 g cm -3 separately in our model.

Data availability
All data needed to evaluate the conclusions in the paper are present in the paper and the Supplementary

Code availability
Monte Carlo simulations with MATLAB code are deposited in Knowledge@ UChicago 47 , a repository hosted by the University of Chicago.