Intricate Regolith Reworking Processes Revealed by Microstructures on Lunar Impact Glasses

Glasses cooled from impact melt and vapor are a common component in lunar regolith, carrying important information about protolith composition, regolith formation, and impact flux on the Moon. Interpretations, however, are frequently challenged due to widespread ambiguity in determining their provenances. Regolith samples returned by China's Chang’E‐5 mission provide a unique opportunity to study the microscopic mechanism of regolith reworking on the Moon, because as evidenced by the coherent radioisotope ages and petrographic characteristics of basaltic clasts in the regolith, the Chang’E‐5 regolith was mainly evolved from local mare materials, containing minor exotic components. Here, we report 153 glass particles larger than 20 μm in diameters that were screened from 500 mg of Chang’E‐5 regolith. Most glass particles have rotational shapes and contain structural and/or compositional heterogeneities in interiors, and geochemical analyses reveal a dominant origin as impact melt of local mare materials. Surfaces of the impact glasses are observed to have abundant protruded and dented microstructures, which are classified as different groups based on their morphology and geochemistry. Similar microstructures were observed on impact spherules collected by the Apollo and Luna missions, but those on the Chang’E‐5 impact glasses were formed without substantial involvement of exotic ejecta. Microstructures such as silicate melt pancakes that frequently exhibit flow spikes at margins, nano‐phase iron‐rich mounds that are arranged with semi‐equidistant spaces in curves and patches, spatially clustered microcraters that are indicative of secondary impacts, and blunt linear scratches with terminal particles all suggest that regolith reworking mainly occurred among local materials at low speeds.


Introduction
The surface of the Moon is almost entirely covered by regolith that is formed by impact gardening and space weathering (McKay et al., 1991). Average regolith thicknesses on the lunar mare and highland (5 and 10 m, respectively) are larger than detection depths of most lunar remote sensing techniques (Fa & Wieczorek, 2012). Samples returned by the Apollo, Luna, and Chang'E-5 missions were mostly collected from surface regolith. Therefore, knowledge of lunar geology and geochemistry was mainly obtained from the surface regolith. Lunar regolith is a reworked product of various protolith, for example, bedrocks, pyroclastic deposits, and ejecta deposits (Head & Wilson, 2020). Both macroscopic (e.g., bedrock fracturing, excavation, and mixing by large impact events) and microscopic (e.g., thermal weathering, sputtering by solar winds and cosmic rays, and micrometeorite bombardments) processes were involved in regolith reworking, which changed the physical and chemical characteristics of the protolith. Therefore, the formation process of lunar regolith can provide crucial mineralogical and chemical information for the interpretation of data obtained by lunar remote sensing and sample analyses.
Lunar regolith is mainly composed of local materials, and thus, regolith is a good mirror of its protolith in terms of compositions (McKay et al., 1991). Meanwhile, as evidenced by the high formation efficiency of far-flung impact rays (Speyerer et al., 2016) and thick distal ejecta deposits , lateral transportation of heterochthonous materials (also termed as foreign or exotic materials) is also an efficient process in the formation of lunar regolith at least at regional scales (Huang et al., 2018;Li & Mustard, 2000;Petro & Pieters, 2006). Occasional discovery of meteoritic components in regolith samples further suggests multiple material sources involved in regolith formation (Joy et al., 2012). The relative abundance of heterochthonous components in lunar regolith was frequently estimated using parameters-dependent ballistic sedimentation models (Huang et al., 2017;O'Brien & Byrne, 2021;Petro & Pieters, 2006;Xie et al., 2020). However, there are few constraints on the interaction mechanism between distal ejecta and local regolith particles based on sample analyses.
Glasses formed by volcanism and impact cratering are a main component of lunar regolith, making up to ∼20-30 vol% in fine-grained surface soils (Delano, 1986;Taylor et al., 2019;Zeigler et al., 2006). Glass particles quenched from impact melt and vapor are ubiquitous in regolith samples returned by the Apollo and Luna missions (Heiken, 1975 and references therein). Impact glass spherules on the Moon, similar to those on the Earth, have spherical and other rotational shapes (Bastin, 1979;Chernyak & Nussinov, 1975), for example, ellipsoidal, teardrop, and dumbbell. Typical impact glasses on the Moon exhibit similar compositions with the local bulk regolith, suggesting that small-scale impacts that melted local regolith materials were the dominant source (e.g., Delano et al., 2007;Korotev et al., 2010;Norman et al., 2012;Zeigler et al., 2006;Zellner et al., 2002). Some impact glasses (especially glass shards) on the Moon have homogeneous interior compositions that are distinctively different from the local regolith, and they were regarded as distal ejecta that were sourced from terranes featuring different compositions (Delano et al., 2007;Korotev et al., 2010;Zellner & Delano, 2015). Therefore, impact glasses are an important indicator for the provenance of lunar regolith (e.g., Zellner, 2019). Together with compositional data, formation ages of lunar impact glasses (including both spherules and shards) provide valuable information about impact fluxes on the Moon (Culler et al., 2000;Nemchin et al., 2022;Norman et al., 2019;Zellner, 2019;Zellner et al., 2009).
There is a major theoretical uncertainty, however, in applications that used impact glasses to probe protolith compositions and impact flux: the sizes and locations of parent craters that formed the impact glass are usually unknown. While impact glasses with exotic compositions may be formed by distal impact events (Delano et al., 2007;Korotev et al., 2010), in regolith that contains heterogeneous compositions, various-sized single and/ or multiple impacts could form glasses that have diverged compositions and/or sizes such as the wide size and compositionally ranges of microtektites in a same strewn field on Earth (Glass & Simonson, 2012). Furthermore, feldspar is preferentially concentrated in fine-grained regolith particles (Hörz & Cintala, 1984), and element partitioning could occur during impact melting and vapourization (Warren, 2008), so that impact glasses formed from homogeneous regolith by a single impact may feature different compositions as well. Therefore, large uncertainties exist when applying empirical scaling laws to estimate sizes of parent craters based on sizes of impact glass spherules (Melosh & Vickery, 1991). Uncertainties also exist when determining possible source craters of impact spherules by referring to the compositional similarities between individual glass spherules and impact craters  10.1029/2022JE007260 3 of 36 Microstructures formed on lunar impact glasses provide a wealth of information about the reworking process of regolith materials, which have not been extensively discussed since the end of the Apollo and Luna missions (e.g., Bloch et al., 1971;Carter & MacGregor, 1970;Clanton et al., 1978;Gault et al., 1972;Hörz et al., 1971;Keller & McKay, 1997;McKay et al., 1970;Neukum et al., 1970;Rode et al., 1979;Wentworth et al., 1999). In the first place, the formation of impact glasses is part of regolith formation because most impact glasses on the Moon were cooled from impact melt of local regolith (Korotev et al., 2010;Zeigler et al., 2006;Zellner et al., 2002). Therefore, microstructures formed on surfaces of lunar impact glasses record subsequent modification of the glasses that happened during regolith reworking. In Apollo and Luna samples, almost all regolith particles (e.g., breccias, lithic clasts, and mineral shards), including glass spherules, are featured by widespread microstructures, such as silicate melt pancakes (Keller & McKay, 1997;Wentworth et al., 1999), iron-rich mounds (Carter, 1973;Carter & MacGregor, 1970), microcraters Hartung et al., 1972;McKay et al., 1970), and ameboid-like iron-rich strips (Carter, 1971;Carter & McKay, 1972). Such features were separately interpreted as accretional or destructional products that were formed during regolith reworking. However, the relative contribution of heterochthonous ejecta that were launched from terranes with different compositions is basically unknown in regolith reworking. A critical reason is that both Apollo and Luna sampling sites were located on terrains older than 3.7 Ga (e.g., Hiesinger & Head, 2006;Stöffler & Ryder, 2001), and heterochthonous materials are common components in the samples (Petro & Pieters, 2006;Stöffler & Ryder, 2001).
The Chang'E-5 mission recently returned 1,731 g of lunar regolith to Earth from one of the youngest mare units on the Moon. The samples were both scooped from the surface and drilled from about 100 cm in the subsurface (C. L. Li, Hu, et al., 2022). Widespread impact rays and secondary craters are visible around the sampling site (Qiao et al., 2021). Ballistic sedimentation models Xie et al., 2020) predicted that up to 10%-40% of exotic materials might exist in the sampling unit. However, sample analyses revealed abundant locally derived basaltic fragments in the Chang'E-5 regolith that have a uniform crystallization age of ∼2.0 Ga (Che et al., 2021;Li et al., 2021) and coherent petrographic characteristics that can be ascribable to a single episode of lava flows (C. L. Li, Hu, et al., 2022;Tian et al., 2021Tian et al., , 2022. Structurally homogeneous impact glass spherules discovered in the regolith also exhibit compositions similar to that of the basalt clasts , further supporting the dominance of locally evolved materials in the regolith. Therefore, exotic materials delivered from the other geochemical terranes are a minor component in the Chang'E-5 regolith. This result is further supported by the well-matched compositions between the bulk regolith and local mare basalts (C. L. Li, Hu, et al., 2022;Tian et al., 2021). From the perspective of regolith formation processes, Chang'E-5 regolith samples provide a unique opportunity to study microscopic mechanism of regolith reworking that involved little contribution from distal ejecta.
Permitted by the China National Space Administration, we were allocated of 500 mg of Chang'E-5 regolith to study impact glasses. In this study, we report 153 glass particles that were separated from this regolith aliquot (Section 2). We recognized impact glasses formed from impact melting based on their interior structures (Section 3.1) and geochemistry (Section 3.2). We investigated the morphology and semi-quantitative compositions of the widespread microstructures (most of them are less than 1 μm in size) on the glasses (Section 3.3). The microstructures were classified as protrusive (Section 3.3.1) and dented (Section 3.3.2) features according to their appearances on the glass particles, and each class contained several subtypes. We interpreted the origin of these microstructures based on their morphology, compositions, crosscutting relationships, and comparisons with similar features observed on Apollo and Luna regolith grains (Sections 4.1,4.2 and 4.3). This study shows that although regolith developed on different-aged lunar surfaces may contain various proportions of exotic materials, microstructures on regolith particles that have different exposure ages are similar in morphology, indicating that regolith reworking mainly occurs among autochthonous particles via ejecta deposition during regolith gardening (Section 4.2). While materials involved in regolith reworking have phases spanning from vapor to solid, their encounter speeds are much lower than those of primary impacts by interplanetary materials (Section 4.3).

Chang'E-5 Regolith Sample and Glass Particles
The regional geological context of the Chang'E-5 sampling site has been deciphered extensively, and regolith thickness in the sampled region is ∼5 m as estimated by orbital remote sensing (e.g., Qian et al., 2021;Qiao et al., 2021;Yue et al., 2019). The sample studied in this work is 500 mg aliquot (CE5C0400YJFM00404; Figure 1a) of the fine-grained regolith (CE5C0400) that was scooped from the surface (C. L. Li, Hu, et al., 2022). The regolith is dark gray in color (Figure 1a; see also https://moon.bao.ac.cn). Abundant lithic clasts, melt-bearing breccias, and olivine and pyroxene shards are visible (Figure 1b). Few grains in the regolith are larger than 1 mm in the long axis, and the majority regolith grains are much smaller in size (Figure 1b; C. L. Li, Hu, et al., 2022). Seen through stereoscopes, glass in the regolith exhibits distinctive vitreous lusters and various colors.
From CE5C0400YJFM00404, 153 glass particles were handpicked and indexed in a super clean room at the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS). The particles are larger than 20 μm in diameters and most of them are smaller than 200 μm. Further screening of this regolith sample would yield more such particles, but we were not meant to handpick all particles larger than 20 μm, and 153 glass particles are representative and adequate for our research.

Analytical Methods
We employed the analytical procedure established by J.  in this study, and the analysis procedures followed two parallel criteria: (a) nondestructive measurements came first and (b) searching for comprehensive and adequate supporting data. After picking out the glass particles from the regolith, all of them were placed in a glass groove and repeatedly cleaned with alcohol to remove loose adhesions on the surface. We documented the optical properties of the glass particles under stereoscopes in the super clean room (Figure 1c), noticing that most of the particles contain structural heterogeneities, such as voids and surface adhesions. All the glass particles were then carbon-coated and examined under a scanning electron microscope (SEM) that was equipped with an energy-dispersive X-ray spectrometer (EDX). The general morphology of the particles and the detailed morphology and semi-quantitative compositions of all microstructures on the particles were investigated and classified. Afterward, we selected 10 representative particles (i.e., those appear as homogeneous glass spherules as seen in stereoscopes and SEM, and some were coated by surface adhesions) to investigate their interior structures using micro-CT, noticing that the apparently homogeneous spherules also contain widespread heterogeneous structures (e.g., bubble voids and/or unmelted mineral fragments) in the interior. Finally, we embedded 33 glass particles that have representative shapes, sizes, colors, and transparency in epoxy resin and analyzed their major element abundances on polished sections. Eleven of these particles exhibit homogeneous compositions in the polished sections, and we measured their trace element compositions to investigate whether or not the particles were derived from local materials.

Surface Morphology and Interior Structures
We investigated the surface morphology for each of the 153 glass particles using a Zeiss Auriga Compact focused ion beam scanning electron microscope (FIB-SEM) at IGGCAS. The cleaned glass particles were mounted on a copper conductive adhesive and then coated with carbon film. The SEM features a Schottky field emission Gemini electron column that was operated between 100 V and 30 kV. The SEM was operated at a 5 kV accelerating voltage with a working distance of 6.0-7.0 mm. This SEM can resolve features ∼1.0 nm in dimensions at 15 kV and ∼1.9 nm at 1 kV.
To study the three-dimensional interior structures of the glass particles, we investigated the No. 2 and 149 spherules (considering their homogeneous appearance seen under stereoscopes) using a Carl Zeiss Xradia 520 Versa High Resolution X-ray Tomographic Microscope (HR-XRTM) at the micro-CT lab of the Nanjing Institute of Geology and Palaeontology, Chinese Academy of Sciences (NIGP-CAS). The glass spherules were fixed in a plastic tube using latex to ensure a stable 360° rotational scan. The scanner was set up to 70 kV voltage and 6 W power. A total of 4,001 projections were collected over a rotation of 360°, producing a voxel dimension of ∼0.35 μm for the spherules.
Another eight particles (Nos. 4,14,16,31,44,140,145,147) that appear either as obvious homogeneous spherules (Nos. 4,16,31,44,147) or closer to agglutinates (Nos. 14, 140 and 145) were selected to investigate the interior structures using the FEI Heliscan Micro CT at IGGCAS. Five of these particles (Nos. 4,14,44,140 and 147), with representative morphologic characteristics and different interior structures, were selected for further geochemical analyses. The glass particles were mounted on the end of a polished glass rod using 3M tape, and the sample stage was rotated and moved following a space filling trajectory during the measurement. The source voltage and current were 60 kV and ∼100 μA, respectively. A total of 2,880 projections were taken for each particle, and the acquired volume voxel size was 0.9855 μm. The volume data were processed using the Vgstudio Max (3.0 version) software to reconstruct the 3D morphology and obtain interior slices.
In addition, we extracted two ultra-thin foils (10 μm × 3 μm × 100 nm) containing microstructures from the glass particles (e.g., iron-rich mounds on the No. 32 spherule; Section 4.2) using the Zeiss Auriga Compact FIB-SEM at IGGCAS. Before the FIB-cutting, platinum was deposited first to protect the microstructures from possible Ga + damage that may occur during cutting and milling. The ion beam conditions for milling and final polishing were 5-30 kV high voltage with various beam currents (20 pA-4 nA).

Semi-Quantitative Composition of the Surface Microstructures
Since most of the observed microstructures on the glass particles are less than several microns in sizes, chemical compositions of these microstructures were investigated using an Oxford X-MAX80 EDX spectrometer that was coupled on a Zeiss Auriga compact FIB-SEM at the IGGCAS. This SEM-EDX can conduct rapid elemental analyses for such small features without destruction based on semi-quantitative elemental mapping. To obtain a high-count spectrum for rigorous quantitative microanalyses, all the SEM-EDX experiments were carried out with an accelerating voltage of 15 kV, larger than the critical excitation energy of examined elements in the particles. The pixel density of the elemental map was 1,024 pixels by 1,024 pixels, and the dwell time per pixel was 150 μs. The main error of the estimated quantities of elements was the system error, which was caused by statistical fluctuation of X-ray photon counts with time. The system error was 1% when the X-ray count was 10 4 and 0.1% when the X-ray count was 10 6 . Most X-ray counts in this study were greater than 2.5 × 10 5 , and the system error was less than 0.2% (Li, 2019). Therefore, EDX semi-quantitative composition analyses of the surface microstructures on the glass particles can reflect the compositional similarities and differences between the surface microstructures and the host glass particles on the first order. To verify relative compositions sensed by EDX mapping, chemical compositions of the FIB sections were obtained using a JEM-2100HR TEM coupled with M-MAX (operating at 200 kV) in IGGCAS.

Analyses of Major Elements of Chang'E-5 Glasses
Major element composition is critical to discriminate the volcanic or impact origin of lunar glass and determine their provenances (Delano, 1986). We selected 33 representative glass particles in terms of color, shape, transparency, and size for major element analyses, that is, No. 1,[4][5][6][7]9,10,12,13,14,44,100,105,107,109,112,113,115,117,120,124,128,129,133,136,137,140,143,147,[150][151][152][153] (Table S1 in Supporting Information S1). Among them, 12 particles were structurally homogeneous and translucent spherules (i.e., No. 1,4,10,13,105,113,115,124,128,147,150,and 151), and they do not contain obvious surface adhesions. After polishing, the cross-sections were coated with carbon film. A JEOL JXA-8230 electron probe micro-analyzer (EPMA) coupled with a backscattered electron (BSE) detector at the Purple Mountain Observatory, Chinese Academy of Sciences, was employed for the major elements analyses. All the EPMA experiments were performed with an accelerating voltage of 15 kV and a beam current of 10 nA. The peak counting time and background counting time for S, Ni, Ca, Ti, Si, P, Fe, Mn, Cr, Mg and Al were 20 and 10 s, and for K and Na were 10 and 5 s, respectively. We collected at least five points on the polished section of each glass particle and took the average values as the composition.

Analyses of Trace Elements of Chang'E-5 Glasses
Abundances of trace elements of lunar glasses are also useful to investigate both their identity (e.g., volcanic or impact glasses) and provenance of precursor materials. We measured the trace elements for eleven of the 33 glass particles that were exposed for major element analyses (i.e., No. 1,4,5,10,13,100,124,128,147,151 and 153), considering that the 11 particles exhibited homogeneous compositions in the exposed sections. A laser ablation inductively coupled plasma-mass spectrometry (LA-ICP-MS) at IGGCAS was used for this experiment. This instrument is equipped with an Element XR high-resolution ICP-MS (HR-ICP-MS) that is coupled to a 193 nm ArF excimer laser system (Geolas HD). The laser diameter was 26 μm, the repetition rate was 3 Hz, and the laser energy density was ∼3.0 J/cm 2 in our measurements. The Element XR was equipped with a high-capacity interface pump (OnTool Booster 150) in combination with Jet sample and normal H-skimmer cones to achieve a detection efficiency in the range of 1.5% (based on U in a single spot ablation of NIST SRM 612). Helium was employed as the ablation gas to improve the transporting efficiency of ablated aerosols. The NIST SRM 610 (Jochum et al., 2011) reference glass was used for external calibration, and ARM-1 (Wu et al., 2019) and BCR-2G (Jochum et al., 2005) glasses were used for quality control monitoring.

General Morphology and Interior Structure of the Chang'E-5 Glass Particles
The interior structures of glass particles, especially structurally homogeneous glass spherules, are indicative of their origin. For example, impact spherules that contain voids and/or unmelted fragments were cooled from impact melt instead of condensed from impact vapor (Agrell et al., 1970;Glass & Simonson, 2012); volcanic glass beads on the Moon may contain minor bubble voids (Heiken & McKay, 1978) but schlieren and exotic inclusions (e.g., unmelted lithic and mineral fragments) are not visible in the interior (Delano, 1986;Delano & Livi, 1981), but such structures are commonly seen in spherical agglutinates in lunar regolith. We performed detailed SEM observations for each of the 153 glass particles, noticing that most of them contain large heterogeneous components, for example, irregularly shaped voids and incompletely melted clasts/minerals. Figure 2 shows several examples, and the completed data set for each of the 153 glass particles is included in the Extended Data Set S1. Large (micron-sized) chunks of structural heterogeneities are frequently visible in the glass particles ( Figure 2), suggesting that such particles are likely agglutinates that were formed by the adhesion of impact melt and other regolith particles.
Three-dimensional structural reconstruction ( Figure 3) further showed that except for a few spherules (Figures 3i and 3j), most of the glass particles contain widespread voids, unmelted fragments, and other structural and/or compositional heterogeneities in the interiors ( Figure 3). Therefore, we can conclude that most of the 153 glass particles are impact melt glasses instead of vapor condensed or volcanic spherules (e.g., Chao et al., 1970;Delano & Livi, 1981).
Interior structures of the 33 glass particles selected for EPMA analyses were also investigated based on the polished sections. Sixteen of them exhibit homogeneous compositions in the glass matrix, but bubble voids are also frequently visible, for example, white arrows in Figures 4a-4b. The melt matrix of the other 17 particles contains widespread compositional heterogeneities that exhibit different shapes and X-ray diffraction contrasts, indicating that the 17 particles are also impact melt products (Figures 4c-4d).

Geochemistry of the Chang'E-5 Glass Particles
For the 16 compositionally homogeneous glass spherules (Section 3.1), our EPMA measurements of major elements abundances (Table 1) are consistent with those of Chang'E-5 basaltic clasts (Tian et al., 2021) and lunar soil (C. L. Li, Hu, et al., 2022), except for the slightly low content of Na 2 O and K 2 O of several particles ( Figure 5). We noticed that the green-yellowish transparent glass spherule No. 4 ( Figure 1c) has a different composition compared to the rest of the homogeneous glass spherules. This is evidenced by its higher contents of SiO 2 , K 2 O, and Na 2 O, but lower contents of FeO, TiO 2 , and MnO (Table 1 and Figure 5).
The 16 glass spherules are classified into three types according to their major element abundances and the classification criteria of lunar mare basalts (Neal & Taylor, 1992): low-Ti/low-K (No. 1,5,10,13,100,115,124,128,147,150,and 151),113,152,and 153), and high-Al/high-K (No. 4). Figure 6 shows that except for the No. 4 high-silica spherule, source materials of the other 15 glass spherules are consistent with low-Ti/low-Al/low-K basalts, which are the same with compositions of basalt clasts (Tian et al., 2021) and bulk fine-grained soil (C. L. Li, Hu, et al., 2022) in the Chang'E-5 regolith ( Figures 5 and 6).
EPMA data also confirm that the glass matrix of heterogeneous glass particles has identical compositions with the homogeneous glass spherules (Figure 7 and Table 1), that is, glass matrix in the 33 particles has the same composition regardless of the structural homogeneity. Unmelted fragments in the heterogeneous glass particles include pyroxene, plagioclase, and ilmenite (Figures 4c-4d; Table 1). The pyroxene and plagioclase fragments have similar compositions with those observed in basaltic clasts in the Chang'E-5 regolith (Tian et al., 2021), indicating that the mineral clasts engulfed in the particles were sourced from the local mare unit (Figure 8). Therefore, the heterogeneous glass, which is the dominant type in CE5C0400YJFM00404, has the same material source as the homogeneous glass.
For the trace element compositions of the eleven homogeneous glass spherules (Table S2 in Supporting Information S1), the rare-earth element (REE) patterns of the low-Ti/low-K and mid-Ti/low-K glass spherules are consistent with that of the bulk fine-grained soil (C. L. Li, Hu, et al., 2022) and basaltic clasts (Tian et al., 2021), although basaltic clasts in the Chang'E-5 regolith exhibit a larger depletion of Eu (Figure 9). The REE pattern of the No. 4 high-Al/high-K spherule is similar to that of the Apollo 15 KREEP but different from the other Apollo and Luna samples (Figure 9).

Morphology and Composition of Microstructures on Surfaces of Glass Particles
We performed high-resolution SEM observation for each of the 153 particles and recognized various-shaped microstructures on the surfaces of the glass particles. The microstructures are classified as protruded and dented features based on their occurrences with respect to the surface of the host particle. Subtypes were defined for each class of the microstructures based on their morphology and SEM-EDX elemental mapping.

Protruded Microstructures
Protruded microstructures are high-relief features on the glass particles. Sorted by the occurrence frequency, the most frequently observed protruded microstructures include vesicular silicate-melt foams, silicate-melt pancakes, iron-rich mounds, and iron-rich pancakes. Other types of protruded microfeatures (e.g., ameboid-like melt strips and high-Si chips) are comparatively fewer. The 3D tomography (panels a and e) was obtained by the HR-XRTM at NIGP-CAS, and broken bubbles are shown in an aerial view in the insets (scale bar is 50 μm). Interior bubbles (green) and unmelted fragments (purple) that have different compositions than those of the glass matrix are annotated in panels (b and f) and (c and g), respectively. Panels (d and h) are two slices along the long axis of the spherules. (i-p) Slices of three-dimensional reconstructions for the other eight glass particles (data were obtained using the micro-CT at IGGCAS). Note that the No. 4 and 147 spherules have homogeneous interior structures.

Vesicular Silicate-Melt Foams
Vesicular silicate-melt foams are the largest protruded microstructures in terms of sizes on the glass particles. Melt foams host high abundances of degassing pits that are usually connected to form honeycomb patterns ( Figure 10). The foams can more-or-less continuously cover the entire (Figures 10a and 10c) or part of the particles (Figures 10b and 10d). SEM-EDX elemental mapping revealed similar compositions between melt foams and their underlying glass particles ( Figure S1 in the appendices). The polished section of the No. 5 spherule is circumferentially coated by silicate-melt foam (Figure 4b), and EPMA measurements reveal similar compositions between the glass matrix in the foam and the interior of the homogeneous glass spherules (Figure 7).

Silicate-Melt Pancakes
Circular and elliptical silicate-melt pancakes are widespread on the Chang'E-5 glasses ( Figure 11). Such pancakes are thin carpeting features, and they are usually micrometers in size. In some cases, large circular pancakes exhibit gentle depressions in the center and slightly thickened rims (yellow arrow, Figure 11a). On a given spherule, a group of elliptical silicate-melt pancakes frequently exhibit identical orientations according to their long axis (white arrows, Figure 11b). Tongue-shaped pancakes ( Figure 11c) and protruded branches at the periphery ( Figure 11d) are also abundant, suggesting subsequent flows after the initial contact between the melt droplets and the solid spherules. EDX elemental mapping (Figures 11e−11h) revealed similar compositions between melt pancakes and the host spherule within the detection limit (Section 2.2.2), suggesting that their precursor materials feature similar compositions. Overlapping relationships between the silicate-melt pancakes and other microstructures on the surface are also frequently visible (red arrows, Figure 11). Note that a small circular void is exposed at the lower-right of the cross-section, and materials with different contrast are adhered to the surface of the spherule. (b) Melt matrix of the No. 5 spherule is homogeneous in composition and the spherule surface is almost entirely coated by adhesives. (c-d) The No. 12 and 14 particles contain widespread structural heterogeneities in the interiors, and mineral fragments, including pyroxene, plagioclase, and ilmenite, are common (major element compositions shown in Table 1). The data were obtained by SEM background scattering electron (BSE) imaging.

Iron-Rich Mounds
Hemispherical iron-rich mounds, with diameters from tens of nanometers to several micrometers, are the most prominent and common protruded features on the Chang'E-5 glasses (Figure 12). The mounds exhibit a high contrast with the glasses seen in SEM SE images, indicating heavier elements dominating the mounds compared to the host glass particles (Figures 12a-12d).
SEM-EDX elemental mapping shows that the iron-rich mounds are mainly composed of Fe, and minor enrichment of Ni, P, and S are also detected (Figures 12e−12k).
Iron-rich mounds usually occur in clusters, and the mounds are arranged in swarms and curves (Figures 12a and 12b). In each cluster, adjacent iron-rich mounds usually exhibit semi-equidistant spaces, and smaller mounds are much more common than large ones (Figures 12a and 12b). Such mists of iron-rich mounds have an uneven distribution on a given glass particle. When resolved with sufficient resolution by SEM, large iron-rich mounds exhibit exquisite surface textures, such as circular cracks (Figure 12d), small depressions, and protrusions ( Figure 12e), but these textures are not correlated with characteristic element abundances (Figures 12f-12i), possibly due to the relatively limited spatial resolution of EDX elemental mapping.

Iron-Rich Pancakes
Besides hemispherical iron-rich mounds, iron-rich pancakes are also visible on the Chang'E-5 glasses ( Figure 13). These microstructures have an intermediate morphology between silicate-melt pancakes ( Figure 11) and iron-rich mounds (Figure 12), as their apparent height/diameter ratios are larger than those of similarly sized silicate-melt pancakes and they exhibit flatter surfaces than typical iron-rich mounds. Interestingly, iron-rich pancakes may either occur individually among a chain/cluster of iron-rich mounds or as a monomorphism chain/cluster (Figures 13a-13d). Linear scratches are occasionally visible on some of the iron-rich pancakes (Figures 13e−13f). SEM-EDX elemental mapping shows that most iron-rich pancakes are mainly composed of Fe and Ni with minor enrichment of minor P and S (Figures 14a-14d) similar to iron-rich mounds (Figures 12f-12k). Meanwhile, a minor portion of iron-rich pancakes are mainly composed by a combination of Fe and S (Figures 14e−14h).

Other Forms of Protruded Microstructures
Micron-sized angular chips of silicate fragments (Figure 15a) that exhibit higher contents of Si than the host (Figure 15c) are visible on some glass particles. These fragments are relatively depleted in Al, Ca, Mg, and Fe (Figures 15d-15g). Similar fragments were reported on Apollo 11 regolith grains (McKay et al., 1970).
A handful of cases of ameboid-like melt strips are observed on the Chang'E-5 glasses ( Figure 16). They are observed on the surfaces of conchoidal fractures and walls of degassing pits in the particles, suggesting they were deposited after the solidification of the host glass particles. Such melt materials feature steeper topography than silicate-melt foams ( Figure 10) and silicate-melt pancakes (Figure 11), suggesting that the melt had a low flow velocity and/or high viscosity when being attached to the solid particles, limiting lateral divergent flows. EDX elemental mapping of the ameboid-like melts has not been performed yet, but the features exhibit similar contrast to the glass substrates seen in SEM SE images, indicating that the ameboid-like melt strips have similar atomic numbers with the glass substrate. Microstructures with similar morphology observed on Apollo 16 regolith grains are mainly composed of iron (Carter & McKay, 1972). For comparison, the ameboid-like melt strips on the Chang'E-5 glass particles were unlikely enriched in iron, and their high viscosity on emplacement was likely caused by the possibly lower temperature of the deposited glass melt.
Some silicate mounds are more voluminous than silicate-melt pancakes (Figure 17a), and they exhibit larger apparent height-to-diameter ratios. EDX elemental mapping reveals no obvious compositional differences compared to the host glass spherules (Figures 17b-17e). The overlapping relationship between the silicate mound and silicate pancakes suggests multiple accretional events among the microstructures (Figure 17a). Moreover, micron-sized smaller glass spherules that have spherical and other rotational forms are frequently adhered on the Chang'E-5 glasses, and some of the small spherules are featured by chains of iron-rich mounds (Figure 17f). The smaller spherules exhibit various compositions, which can be either similar or different from the larger host glass particle (Figures 17g-17j). For example, an elliptical-shaped small spherule shown in Figure 17f (white arrow) exhibits an obvious enrichment of Mg (Figure 17j).

Dented Microstructures
Dented structures are incised into the glass particles, and four types of dented microstructures were discovered on the Chang'E-5 glass particles, including microcraters, linear fractures, degassing pits, and grain-filled depressions.

Microcraters
Microcraters on surfaces of the Chang'E-5 glass particles are much scarcer than those observed on Apollo and Luna glass spherules (e.g., Carter, 1973;Carter & MacGregor, 1970;Rode et al., 1979). In total, only about 20 typical microcraters that are larger than 1 μm in diameters are recognized on the 153 Chang'E-5 glass particles. The microcraters have various morphologies (Figure 18 and Figure S2 in Supporting Information S1). Some microcraters are developed with a completed flower-petal spallation zone and a central pit (Figure 18a), while others exhibit a central mound and lack obviously raised rims (Figure 18b). We also observed the profound effect of different target properties on the formation of microcraters. For example, impacts on hollowed glass particles that have thin walls did not form obvious microcraters, but they developed circumferential tensile fractures (Figure 18c), which may be due to incipient spallation . Impacts on porous breccias that were adhered on the glass particles formed deep pits without obviously raised rims (Figure 18d). Only a few microcraters possess melt-like materials on the rims (Figure 18e) and floors (Figure 18f).

Linear Fractures
Linear fractures with various morphologies are widespread on the Chang'E-5 glass particles (Figure 19). Narrow and long fissures (Figure 19a) may be incipient spalls formed during microscopic impacts . Highly elliptical depressions that have pronounced raised rims (Figure 19b) are cataloged here as linear fractures, considering that their morphologies are different from those of typical microcraters, but such fractures may alternatively be microcraters formed by highly oblique impacts. Widespread blunt fractures that are composed of multiple parallel curved grooves are abundant on the glass particles, and various-sized terminal fragments are visible at one end of the grooves, indicating that the grooves were likely scratched by the fragments (Figures 19c-19g). On a same particle, scratches with different orientations (Figure 19e), sometimes opposite  orientations (Figure 19f), are visible on the glass particles. They were formed either at different times and/or by fragments that collided from different directions.

Degassing Pits
Degassing pits formed by the escape of volatiles in impact melt are common microstructures seen in Luna, Apollo (e.g., Carter & MacGregor, 1970;Carter & Mckay, 1972;Rode et al., 1979), and Chang'E-5 glasses. With  various sizes, degassing pits are different from microcraters due to the absence of raised rims or spallation zones (Figures 20a and 20b). Many structurally heterogeneous glass particles contain a high volumetric percentage of degassing pits in the interior (Figure 20c), similar to those developed in melt foams on the surfaces of the glass particles (Figures 10 and 20d).

Grain-Filled Circular Depressions
An enigmatic type of grain-filled circular depressions can also be observed on Chang'E-5 glass particles. The grain-filled circular depressions have sharp boundaries and no raised rims, and the grain-filled depressions have similar or larger surface elevations compared to the surrounding glass ( Figure 21). Figure 21c shows a grain-filled circular depression that contains a high-elevation and near-circular dome in the center. The grains have various shapes (spherical to angular), and most of them are less than 50 nm in sizes (Figures 21a-21c). SEM-EDX elemental mapping shows that the grains are dominantly enriched in Fe, and finer grains around the central iron-mound are additionally enriched in S (Figures 21d and 21e). Compared with the host glass particle, the smaller filled grains in the circular depressions also have slightly higher contents of P and Ni (Figures 21f-21i).

Impact Origin of the Chang'E-5 Glasses
Most of the 153 glass particles studied here contain structural and compositional heterogeneities in the interior, suggesting that they are impact melt products instead of being vapor condensed or volcanic glass particles. Our geochemical analyses for the representative particles further suggested that the remaining homogeneous glass spherules were also derived from impact melt. Previous investigations of lunar glass in Apollo and Luna regolith samples show that impact glasses have a low MgO/Al 2 O 3 ratio (<1.25) that is distinctively different from volcanic glass (e.g., Delano, 1986). The 16 homogeneous glass spherules studied here exhibit low MgO/Al 2 O 3 ratios (0.18-0.71), which are consistent with an impact origin ( Figure 22).
In addition, CaO/Al 2 O 3 ratio of homogeneous impact glass is a useful parameter to differentiate highland and mare precursor materials of lunar impact glass (Naney et al., 1976), as CaO/Al 2 O 3 > 0.75 indicating the mare source and CaO/Al 2 O 3 < 0.75 indicating the highland source. For the 16 compositionally homogenous glass spherules studied in this work, except the high-Al/high-K No. 4 spherule, their CaO/Al 2 O 3 ratios are consistent with being originated from the mare (Figure 22a). The No. 4 spherule has a CaO/Al 2 O 3 ratio of 0.64, indicating a highland provenance. Compositions of the No. 4 spherule are similar with those of lunar igneous rocks that are enriched in KREEP elements (Figures 22b-22d). As revealed by the compositional comparison between the Chang'E-5 glass particles and those collected by the Apollo and Luna missions (Figures 22b-22d and Figure 9), most of the Chang'E-5 glass particles were sourced from the local mare materials, and the No. 4 spherule is an exotic impact spherule that was likely sourced from a KREEP-rich highland terrane.

Formation Mechanism of the Protruded Microstructures
SEM-EDX and EPMA analyses suggest that the vesicular melt foams are impact melt of the local mare materials (Figure 7 and Figure S1 in Supporting Information S1) that were sputtered on surfaces of solid glass particles, and the expansion of gas bubbles in the melt formed the widespread degassing pits (Figure 10).The vesicular melt foams contain high volumes of degassing pits, and such features are also widespread on surfaces of Apollo and Luna regolith grains (e.g., Carter & MacGregor, 1970;Rode et al., 1979). Basaltic clasts in the Chang'E-5 regolith contain low abundance of water (Hu et al., 2021), which is measured from the main OH-bearing phases, such as apatite (1921 ± 910 μg/g) and ilmenite-hosted melt inclusions (6 ± 2 μg/g to 370 ± 21 μg/g). The high abundance of degassing pits in the melt foams suggests that the released volatiles were not likely originated from the local basaltic clasts but from implanted hydrogen or other volatile species in the regolith (Housley et al., 1973). The glass particles were coated by melt foams either when they were excavated by impact events, or the melt foams were deposited on the glass particles from adjacent impact events.  Silicate-melt pancakes are droplets of impact melt adhered to the impact glass particles. Some large circular pancakes appear to host gentle depressions in the center and slightly thickened rims (yellow arrow, Figure 11a), indicating that melt accretion occurred at relative speeds much lower than the primary impact velocities, because no obvious melt splashes or microcraters are visible. The same orientation of the long axis for groups of elliptical silicate-melt pancakes (white arrows, Figure 11b) suggests that swarms of melt droplets were accreted from the same direction. The similar composition between the melt pancakes and the host glass particles (Figures 11e−11h) suggests that the melt pancakes might be impact melt ejecta sourced from the local regolith, and they were deposited on solid glass particles that had landed on the lunar surface. Alternatively, it is also possible that both the impact melt ejecta and solid glass particles belonged to ejecta that were launched from a single impact event, and they encountered one another in the excavation flow and/or during subsequent ballistic trajectory. The overlapping relationship between the silicate-melt pancakes and other microstructures on the particles (e.g., red arrows in Figure 11) suggests that the melt pancakes were accreted at different times, some perhaps significantly later than the formation of the host particles.
Iron-rich mounds are widespread on surfaces of Chang'E-5, Apollo, and Luna regolith grains, and earlier studies suggested that the mounds were composed of metallic iron that was reduced by different processes (e.g., Carter, 1973;Carter & MacGregor, 1970;McKay et al., 1970;Rode et al., 1979). Small irregularly shaped and rimless shallow depressions (<1 μm in diameters) are occasionally visible among mists of iron-rich mounds (white arrows in Figures 12a and 12c). Termed as dimples, such pits on Apollo regolith grains were interpreted as voids left after iron-rich mounds were lost from the host glass (Carter & MacGregor, 1970). In another words, the hemispherical iron-rich mounds may have extended into the glass particle. We prepared an FIB section for such iron-rich mounds, confirming that a mirrored-hemispherical iron-rich body exists in the host silicate glass (Figure 23). The lower boundary between the spherical iron-rich mounds and the silicate glass is sharp. This observation supports the earlier interpretation for the formation of similar iron-rich mounds on Apollo regolith  grains (Carter & MacGregor, 1970): liquid or vapor iron (i.e., the precursor of iron-rich mounds) was immiscible with silicate melt (i.e., the precursor of glass particle), surface tension between different liquid phases caused contraction and then formed the iron-rich mounds, and some mounds were lost due to more intense contraction or subsequent settlement to the lunar surface (Carter & MacGregor, 1970). This interpretation could also explain the regular distribution patterns (e.g., mist, straight, and curved strings) of the iron-rich mounds, which are expected during the cooling of an iron-rich liquid film that partly covered a liquid silicate glass spherule, and/or during the condensation of iron-rich vapor on a still molten spherule, and subsequent contraction of the silicate glass formed the hemispherical iron-rich mounds (Carter & MacGregor, 1970;Rode et al., 1979). The source material of the iron-rich mounds, that is, iron vapor/melt, could be reduced from precursor regolith materials due to shock heating, and/or they might also be cooled from melting and/or vapourization of an iron-bearing impactor (Hamann et al., 2018). The regular spatial distributions and semi-equidistant spacing between adjacent mounds in a given cluster (Figures 12a and 12b) indicate that if they were formed by condensation, the speed difference between the silicate glass and the iron vapor/melt was small when condensation occurred, so the iron mounds were able to be attached on the spherules instead of occurring as individual iron spherules in regolith.  (Delano, 1986), Apollo and Luna soils (Lucey et al., 2006), and bulk fine-grained soil (C. L. Li, Hu, et al., 2022) and basaltic clasts (Tian et al., 2021) in the regolith sample CE5C0400 are shown here for comparison.
Alternatively, Housley et al. (1973) and Pillinger et al. (1976) interpreted that such iron-rich mounds on Apollo regolith grains were reduced iron from the silicate melt that formed the glass particles, and the reducing agent may be solar-wind-implanted hydrogen and/or carbon. Recently, it has also been proposed that without melting of regolith materials (maximum heating temperature ∼1,000°C), flash heating caused by micrometeorite bombardments could cause local reduction of regolith particles (minerals, glass spherules, etc.) that have experienced space weathering (i.e., with implanted hydrogen and/or carbon), so that iron mounds could form and grow on surfaces of regolith particles by single and repeated heating events (Thompson et al., 2017). String and mist of nanophase iron (np-Fe 0 ) were formed in laser-irradiated pyroxene and basaltic volcanic glasses (Sorokin et al., 2021(Sorokin et al., , 2022, and they were interpreted to be caused by in situ thermal reduction of impact melt that involved no vapor condensation. The string distribution was interpreted to be caused by heterogeneous heat deposited by the laser and/or inhomogeneities in the target material (Sorokin et al., 2021(Sorokin et al., , 2022. Furthermore, the observed iron-rich mounds on the glass particles here have similar morphology with some np-Fe 0 observed on surfaces of the other Chang'E-5 regolith grains, for example, olivine and pyroxene (Mo et al., 2022). A comparison of these features is presented in Figure 23 and Figure S3 in Supporting Information S1. Micro-probe analyses for such np-Fe 0 features formed on Chang'E-5 soil mineral grains lend support to the interpretation that they were formed due to impact-induced thermal alterations (i.e., disproportionation reaction and/or decomposition) in different Fe-bearing minerals R. Li, Guo, Li, Xia, et al., 2022;Mo et al., 2022). However, the The yellow-dashed rectangle shows the focused ion beam (FIB) site. (b) SEM image of the FIB foil for the dome-like grain-filled depressions (white arrows) and iron-rich mound (yellow arrow), and iron-rich particles with different sizes are heterogeneously distributed inside the section (e.g., green arrow). (c and d) EDX spectra of the iron-rich mounds on the surface and the largest iron-rich particle inside the FIB section, respectively. morphology and size-frequency distribution of the iron-rich mound on Chang'E-5 glass particles are both different from iron-rich mounds produced by flash heating, suggesting that the iron-rich mounds on the Chang'E-5 glass particles were not obviously reduced from the solid glass by single or repeated impact heating. Specifically, laboratory simulations of single flash heating produced flat-topped iron mounds, which are different from the hemispherical iron-rich mounds observed in this study (Figures 12a and 12b) and reduced from a single flash heating (Thompson et al., 2017). A swarm of iron-rich mounds on the Chang'E-5 glass particle shares a similar right-skewed distribution (Figures 24a and 24b) with that of iron-rich mounds produced by single thermal heating in high vacuum, but dramatically different from the normal distribution of iron-rich mounds grown through repeated flash heating (Thompson et al., 2017).
Iron-rich pancakes ( Figure 13) have not been reported particularly on Apollo or Luna regolith samples before, but they appear similar to some faceted iron-rich features that were produced by laser radiation of analog regolith materials (Sorokin et al., 2021). The similar distribution and composition between iron-rich pancakes and iron-rich mounds (Figures 12 and 13) indicate these two morphological types of microstructures might share a similar identity as iron particles. However, it is currently unknown why the two subtypes, even those in a same chain/cluster, exhibit different morphology (Figures 13a-13d). The origin of the Fe-and S-rich pancakes is unknown, and they might be modified circular depressions that were filled by Fe-S grains (Section 3.3.2.4), which are evidenced by extensive linear scratches on some of the iron-rich pancakes (Figures 13e and 13f).
Angular high-Si chips, similar to those shown in Figure 15a, were reported on Apollo 11 regolith grains before (McKay et al., 1970). They may be impact melt delivered from exotic geochemical terranes that featured higher Si than that of the local mare materials. Alternatively, such high-Si chips may be impact melt of local materials that have undergone elemental differentiation during cooling of impact melt and/or vapor (Korotev et al., 2010;Warren, 2008). Likewise, the smaller dumbbell-shaped spherule shown in Figure 17f might share similar origin(s). The voluminous silicate mounds reported in Figure 17a are interpreted here as large droplets of impact melt that may have a lower temperature than that of ejecta melt forming the silicate-melt pancakes when they were accreted on the glass particles. The overlapping relationship between large silicate mound and silicate pancake (Figure 17a) suggests multiple accretional events during regolith reworking, which occurred at speeds much lower than primary impact velocities on the Moon.

Formation Mechanism of the Dented Microstructures
Microcraters are one of the most frequently observed dented microstructures on surfaces of Apollo and Luna glass spheres, but they are much fewer on Chang'E-5 glass particles (Section 3.3.2.1). Microcraters on Apollo and Luna glass spheres are frequently lined by impact melt on the walls and floors (e.g., Carter, 1971Carter, , 1973McKay et al., 1970), but such melt-lined microcraters are rather rare on the Chang'E-5 glass particles (Figure 18 and  Figure S2 in Supporting Information S1). It is generally believed that microcraters on lunar regolith grains were formed by micrometeorite bombardments (e.g., Hartung et al., 1972;Hörz et al., 1971;Neukum et al., 1970), but others suggested that at least some microcraters were carved by impact debris instead of micrometeorites, that is, secondary craters (e.g., Carter & MacGregor, 1970;Frondel et al., 1970;McKay et al., 1971). Physical simulations of micrometeorite bombardments on silicate glass suggest that melt-lined microcraters were formed at velocities larger than 7 km/s and/or an elevated target temperature as high as 700°C (Carter & McKay, 1971). Therefore, most of the microcraters on the Chang'E-5 glasses were not obviously formed by impacts of micrometeorites. The sharp contrast of impact velocity in the formation of microcraters on Chang'E-5 versus Luna and Apollo glass particles can be partly attributed to the younger exposure age of the Chang'E-5 regolith. Although top centimeters of lunar regolith are overturned efficiently by bombardments of both micrometeors and impact ejecta (on the order of <100,000 years; Speyerer et al., 2016), regolith on older lunar surfaces is overturned more severely due to the larger cumulative exposure ages (Xie et al., 2020). Therefore, fewer micrometeorite impacts occurred on Chang'E-5 regolith when compared to the Apollo and Luna samples.
Interestingly, we noticed that on some Chang'E-5 glass particles, microcraters exhibit highly heterogeneous spatial distributions ( Figure 25). In such cases, individual microcraters have raised rims but without obvious melt veneers (Figure 25b). Similar distribution pattern of microcraters was observed on Apollo glass spherules and mineral particles in Chang'E-5 regolith, and they were interpreted to be secondary microcraters (e.g., Carter & MacGregor, 1970;Frondel et al., 1970;McKay et al., 1970). However, this observation and interpretation were frequently ignored in works that employed the size-frequency distribution of microcraters to quantify the exposure time of regolith grains. The spatially clustered microcraters observed on Chang'E-5 glass particles are obvious secondary craters that were formed by a swarm of debris ejected from nearby impacts. Likewise, spatially clustered microcraters were also observed on surfaces of microtektites in the Australasian strewn field on Earth, and they were interpreted to be caused by mutual collisions between solidified glass spherules and other particles during flight, possibly within a hypothesized impact plume (Prasad et al., 2010;Prasad & Sudhakar, 1998). Therefore, the clusters of microcraters on the Chang'E-5 glass particles are consistent with being secondary craters that were formed in an impact plume and/or ejecta curtain, as the partially or entirely solidified glass particles were collided with other debris at adequate speeds. Based on the above interpretations, we further stress here that regolith reworking at the Chang'E-5 landing site mainly occurred among grains sourced from local materials, so impacts by exotic ejecta on the glass particles are less abundant, forming fewer microcraters that might exhibit signs of high speeds.
Linear fractures, similar to those observed on Chang'E-5 glass particles (Figure 19), were reported on Apollo glass samples (Carter & MacGregor, 1970). We were once concerned that such scratches might be due to tweezer scraping during handpicking and cleaning of the glass particles. However, the common existence of terminal fragments at one end of the blunt scratches  and the overlapping relationships between the scratches and other microstructures, such as iron-rich mound (Figure 19h), suggest that the scratches were formed during regolith reworking on the Moon. In addition, terminal fragments at ends of the parallel fractures (Figures 19c-19g) further indicate that the fractures are surface scratches formed by the low-velocity collisions  (Carter & MacGregor, 1970), and the impact velocity was not fast enough to form microcraters. Adjacent scratches in different directions on the same glass particle (Figures 19e and 19f) suggest that they were formed either at different times and/or by fragments that were incident from different directions. Tumbling of the glass and other particles caused by regolith gardening could explain the formation of these linear fractures. Alternatively, debris impacts onto already landed glass particles by low-velocity ejecta could also explain the origin of the linear fractures.
Grain-filled circular depressions, such as those shown in Figure 21, were also observed in Apollo and Luna glass spheres, and they exhibit similar morphology and compositions, that is, a central metallic iron core surrounded by a waist of incompletely wetted iron sulfide (e.g., Carter, 1971;Carter, 1973;Carter & MacGregor, 1970;McKay et al., 1970;Stone et al., 1982). Based on their surface morphology and composition, the iron core and iron-sulfide waist were interpreted to be formed due to reduction of impact melt (Carter & McKay, 1972), and the sharp boundary between circular depression and surrounding glass was interpreted to be caused by cooling shrinkage (Carter, 1973).

Implications for Regolith Reworking on the Moon
The interpreted formation scenarios for the various microstructures on Chang'E-5 glasses can be summarized in Figure 26. Our comprehensive comparison of the semi-quantitative compositions between the microstructures and the host glass particles suggests that regolith reworking at the Chang'E-5 sampling site mainly occurred by local instead of exotic materials. This interpretation is consistent with local mare materials being the dominated protolith of the Chang'E-5 regolith (C. L. Li. Hu, et al., 2022;Tian et al., 2021). SEM-EDX elemental mapping reveals that protruded microstructures, such as silicate-melt foams (Figure 10), melt pancakes (Figure 11), and voluminous melt mounds (Figures 17a-17e), have similar compositions to the host glass particles. The same is true for materials that formed the dented microstructures, such as clusters of secondary microcraters (Figures 18, 25 and Figure S2 in Supporting Information S1) and linear scratches ( Figure 19). Therefore, the microstructures on the Chang'E-5 glass particles were also formed by local materials. In general, in terms of the diversity of material provenances that were involved in regolith reworking, the Chang'E-5 regolith is an end-member case compared to those sampled by the Luna and Apollo missions. Interestingly, microstructures formed during regolith reworking have similar morphology on the Chang'E-5 regolith and older Apollo and Luna regolith samples, although the latter ones contain more exotic materials and longer exposure times. We conclude that the direct contributions of distal ejecta that were derived from other geochemical terranes are not significant in the reworking of lunar regolith.
Only a limited number of microcraters on the Chang'E-5 glasses might be formed by micrometeorite impacts, as evidenced by the minor occurrence of possible impact melt veneered on the crater walls ( Figure 18). Most microstructures on the surfaces of the Chang'E-5 glass particles were adhered and incised at low relative speeds during regolith reworking. This interpretation is supported by the spatial distribution of silicate-melt pancakes (Figure 11), iron-rich mounds (Figure 12), blunt scratches (Figure 19), and secondary microcraters (Figures 18, 25 and Figure S2 in Supporting Information S1). Impact bombardment plays a dominant role in the reworking of lunar regolith (e.g., McKay et al., 1971McKay et al., , 1991Quaide et al., 1971), and most of the microstructures are interpreted to be formed by ejecta deposits that were sourced from the local regolith. We noticed that although the observed microstructures were formed by materials with different phases, including melt (e.g., melt foam, silicate-melt pancakes, and silicate-melt mounds), solid (e.g., materials that formed linear scratches and secondary microcraters), and even possibly vapor (e.g., iron-rich mounds), their encounter speeds with the host glasses were frequently quite small, less than that necessary to form secondary microcraters (Figures 18, 25 and Figure  S2 in Supporting Information S1). This observation indicates that during regolith reworking, ejecta formed by small impacts can have similar velocities but different thermophysical states.

Conclusion
We observed the morphology of 153 glass particles with diameters larger than 20 μm that were handpicked from 500 mg aliquot of Chang'E-5 lunar regolith. Interior structures, petrographic, and geochemical analyses show that the glass particles were cooled from impact melt, and most of them sourced from the local mare regolith. One homogeneous, yellow-greenish glass spherule possesses high contents of Al 2 O 3 and K 2 O, and its geochemical characteristics are consistent with a provenance from highland materials that were enriched in KREEP.
Abundant microstructures are observed on the Chang'E-5 glass particles, and they were classified as protrusive and dented features that each contain different subtypes. Although they share similar morphology with those observed on regolith glass spheres collected by the Apollo and Luna missions, geochemical analyses by EDX elemental mapping revealed that most microstructures on Chang'E-5 glass particles were formed by local regolith materials with minor involvement of exotic materials, such as micrometeorites and ejecta launched from the other geochemical terranes. The microstructures were formed during regolith gardening by various states of impact ejecta (vapor, melt, and solid) that were sourced from local materials. They were formed by low relative speeds between the glass particles and various impact ejecta, much less than primary impact velocities on the Moon. Therefore, direct contribution of distal ejecta in regolith reworking on the Moon is not significant, and regolith reworking mainly occurs among autochthonous regolith particles during impact gardening and ejecta deposition. Interpreted formation scenarios of microstructures on Chang'E-5 glass particles. Circled numbers denote typical microstructures observed in this study and their possible formation mechanisms. The inset figure shows the various space weathering processes on the Moon, and our described regolith reworking processes are intertwined with the other space weathering processes (modified after Pieters and Noble (2016), Thompson et al. (2021), Gu et al. (2022)).

Data Availability Statement
The SEM micrographs of the Chang'E-5 glass particles, geochemical data, and three-dimensional reconstruction of glass particles measured in this work can be found online at the zenodo repository . Geochemical composition of lunar pristine volcanic glasses, Apollo and Luna soil, basaltic clasts, and bulk fine-grain soil in Chang'E-5 returned samples used in this work is available through Delano (1986), Lucey et al. (2006), Tian et al. (2021), and C. L. Li, Hu, et al. (2022), respectively.