From peri-implant distance ranges to spatial niches: a resolution-aware analytical framework for the intracortical microelectrode–brain interface

1 Introduction

IMEs form biohybrid interfaces at which an engineered conductor, insulator, and device geometry remain in intimate contact with living brain tissue. Their value depends on access to local neural activity, and chronic signal quality varies across channels, devices, animals, and time (Williams et al., 2025). Mechanical injury at insertion, persistent micromotion, vascular disruption, inflammatory signaling, glial and immune responses, and neuronal displacement or loss can contribute to the evolving tissue response, while material and device properties are associated with its magnitude and time course (Polikov et al., 2005; Salatino et al., 2017; Hermann and Capadona, 2018; Gulino et al., 2019; Perna et al., 2023). Around the interface, these interdependent processes produce a multicomponent response that includes heterogeneous cellular states rather than a single uniform reaction (Bouadi and Tay, 2021). Recording performance varies with this tissue context together with electrode design, electrochemistry, cortical depth, neuronal availability, and temporal history (Prasad and Sanchez, 2012; Nolta et al., 2015; Ereifej et al., 2020; Urdaneta et al., 2023; Williams et al., 2025). This combination motivates an analytical distinction that preserves the material-centered geometry of the interface while resolving the local biology organized around it.

IME studies already use a practical spatial language. Histology and molecular assays are commonly summarized in concentric rings, fixed radial bins, or near-versus-far regions centered on the track or device surface (Perna et al., 2023; Gupta et al., 2024). Recent spatial transcriptomics (ST) and regional profiling studies retain the same reference while measuring molecular variation continuously or in study-specific bands (Whitsitt et al., 2022; 2024; Gupta et al., 2026; Song et al., 2024; Druschel et al., 2024). This coordinate system states where a measurement lies relative to the engineered interface, supports proximal-to-distal comparison, and exposes gradients hidden by whole-section averages. A distance value or ring label precisely records location while leaving local tissue context open for measurement. Candidate sources of variation among locations at the same radius include vascular adjacency, lesion morphology, neuronal density, glial composition and state, cortical layer, and position along or across a planar probe (Bouadi and Tay, 2021; Gupta et al., 2024); spatial molecular assays can test which of these contexts accompany local molecular differences.

Spatially resolved molecular technologies make that within-coordinate heterogeneity testable, an opportunity also recognized in recent implant-host single-cell and spatial-omics reviews (Farbehi et al., 2025; Yang et al., 2025). Direct IME studies now include Visium whole-transcriptome measurements with multicellular capture spots, regional GeoMx RNA or protein profiles, registered histology and immunofluorescence (IF), and targeted Xenium transcript-coordinate maps with marker-based cell-type inference reported in a 2026 preprint (Whitsitt et al., 2024; Gupta et al., 2026; Song et al., 2024; Druschel et al., 2024; 2026; Thompson et al., 2026). Each assay contributes a distinct native unit: a Visium spot captures a local multicellular mixture; a Xenium readout provides transcript coordinates within a targeted panel, with study-specific segmentation and cell-type inference; a GeoMx area of illumination (AOI) represents a selected tissue aggregate; and an image feature reflects its staining, segmentation, and registration. Together these measurements support analysis of recurring local contexts within implant-centered coordinates.

Spatial-omics studies characterize cellular niches through the composition and molecular states of neighboring cells (Varrone et al., 2024; Haviv et al., 2025). Building on this usage, we use spatial niche to represent local cellular, molecular, and structural context and specify four minimum reporting elements for this review: the focal unit, a biologically and technically justified neighborhood rule and scale, the input features used to represent context, and a validation strategy. The validation strategy states the criterion used to evaluate the representation and supports reproducible comparison. A niche class is a recurrent discrete category assigned to similar niche representations; continuous niche representations need not be discretized. A distance ring, capture spot, or unsupervised spatial cluster alone is therefore not a niche unless it is characterized by local biological or structural features and evaluated according to the declared validation strategy.

Existing reviews have synthesized neural-implant tissue responses and device constraints (Salatino et al., 2017; Gulino et al., 2019; Perna et al., 2023; Williams et al., 2025), heterogeneous glial states (Bouadi and Tay, 2021), and structure–function–genetic changes around brain electrodes (Gupta et al., 2024). Adjacent biomaterial reviews extend that foundation to emerging cell types, metabolism, and implant-host single-cell and spatial analysis (Yang et al., 2023; Callaghan et al., 2024; Farbehi et al., 2025; Yang et al., 2025). Direct material studies further associate contrasts in coating chemistry, nanoscale topography, mechanics, drug elution, substrate, and device geometry with tissue or recording endpoints (Hernández-Reynoso et al., 2023; Sturgill et al., 2024; Duncan et al., 2024; Mueller et al., 2025; Haghighi et al., 2025; Ocoko et al., 2026). Because individual comparisons often couple material and geometry and report either tissue or recording endpoints, they enter the evidence chain at different levels. The organizing question is how this material-centered evidence relates to the native units of spatial assays and to downstream functional evidence. Regional aggregates, multicellular spots, segmented cells, inferred molecular relations, and recording units contribute different parts of that connection and support different levels of interpretation.

This methodological framework review organizes the evidence around three analytical directions: identifying and characterizing spatial niches; examining implant-referenced patterns, gradients, boundaries, and molecular-state ordering; and evaluating niche-conditioned transcript and protein relationships. It then considers how niche-derived features could be compared with recording variables at animal, implant, shank, or channel scales. Figure 1 summarizes this analytical framework.

2 From implant-centered coordinates to spatial niches

2.1 Established spatial coordinates at the IME–brain interface

In vivo, a planar intracortical shank extends along its insertion axis and leaves a three-dimensional (3D) probe tract in the surrounding tissue. The terminal spatial-omics image is a thin tissue section nominally transverse to that axis; it therefore approximates a two-dimensional (2D) probe-tract cross-section rather than a longitudinal view of the intact implant. Where this cross-section is visible, its center often appears as a small irregular void or scar. An implant-centered coordinate can therefore be defined from the observed or reconstructed boundary of that cross-section, the reconstructed material surface, or another registered device reference. Measurements may then be expressed as continuous distance, grouped into radial bands, or compared between proximal and distal regions. Reported coordinate representations vary with assay resolution, tissue integrity, registration confidence, and response scale (Perna et al., 2023; Gupta et al., 2024). Direct ST studies have summarized expression in bins around a tract cross-section (Whitsitt et al., 2022; 2024); an expanded dataset contrasted tissue within approximately 200 μm with tissue near 500 μm (Gupta et al., 2026). GeoMx studies have sampled selected AOIs (Song et al., 2024; Druschel et al., 2024), including RNA and protein profiles using corresponding 0–90, 90–180, and 180–270 μm radial-band definitions (Druschel et al., 2026). These values define study-specific sampling boundaries. Section angle, device cross-section, tissue deformation, cortical layer, and uncertainty in the reconstructed tract cross-section determine the biological tissue represented by a numerical distance.

Distance-based representations provide an explicit material reference. Continuous-distance models represent monotonic or nonlinear proximal-to-distal change, whereas study-defined rings summarize radial regions used in histological studies. Angular and axial coordinates represent additional dimensions of non-radial probe geometry. Within this Review, implant coordinates form the material-referenced layer, and multivariate local contexts form a second layer that captures biological organization within and across those coordinates. Their combination preserves the defining geometry of the biohybrid interface.

An implant-centered coordinate is most interpretable when it uses the shortest distance from each observation to a declared reconstructed tract or material-surface boundary rather than only the distance to a tract centroid. For a planar probe, face, edge, tip or shaft region and axial or angular position provide additional context when sectioning and registration permit. Registered device and tissue landmarks describe section angle and tissue deformation, while cortical layer and vessel proximity distinguish native tissue organization from the implant response.

2.2 Spatial domains and spatial niches

The framework separates three vocabularies that are often conflated: established implant-referenced coordinates and sampling regions, assay-native measurement units, and analytically derived spatial contexts. Table 1 defines the principal terms and their analytical scope across these vocabularies. This separation prevents a distance band, capture spot, broad domain, local niche, and niche-conditioned state from being treated as interchangeable observations. Recent cellular-neighborhood syntheses similarly distinguish recurring local cell compositions from broader spatial domains and functional ecosystems (Ma et al., 2026). Existing methods may cluster aggregated cell states, learn continuous niche representations, or test context-dependent expression; these operations address different levels of the framework (Varrone et al., 2024; Birk et al., 2025; Mason et al., 2024).

TABLE 1. Operational definitions of implant-referenced, assay-native, and analytically derived spatial terms used in this Review.

Within the analytically derived layer, niche representations are operational and tied to analytical function. A niche may be continuous or discrete and may be anchored to a spot, a selected cell type, all cells, or an image-defined structure. Neuron-centered focal units represent neuronal-support contexts, whereas broader focal units represent interface-remodeling contexts.

2.3 Common representations of a spatial niche

The simplest representation is a neighborhood composition vector: for each spot or cell, nearby cell types or inferred proportions are counted or weighted within a fixed radius, k nearest neighbors, or a graph. Such transparent summaries make neighborhood scale and cellular composition explicit. Existing methods extend this representation through soft composition programs, aggregated graph neighborhoods, multiview representations, or prior-informed continuous embeddings (Ma et al., 2026; Chen et al., 2020; Varrone et al., 2024; Hu et al., 2024; Qian et al., 2025; Birk et al., 2025).

Topological representations extend composition-based representations. A neighborhood graph can encode which niche classes contact, mix, segregate, or form repeated motifs, while a local boundary score can quantify transition sharpness. Continuous niche features can also enter expression or outcome models without discretization. This distinction separates niche representation from downstream context-conditioned testing, which is treated in Section 6. Continuous features preserve graded structure, whereas optional labels summarize recurring configurations.

2.4 Scale relationships among rings, spots, cells, and niches

As illustrated schematically in Figure 1A, geometric and analytical scales form nested, distinct levels: a distance ring can contain multiple niche classes, while one niche class can recur across several rings. A single Visium spot can in turn contain several cell types or microstructures. The 55-μm Visium capture diameter provides local whole-transcriptome information at a multicellular scale (Gupta et al., 2026). Cell-coordinate assays support focal-cell neighborhoods within the state space defined by the targeted panel. The current IME Xenium preprint used a 100-gene panel and electrically inactive surrogate probes, described as nonfunctional in the source study. These probes enabled comparisons of material and geometry but did not provide electrophysiological recordings. The study provided material- and size-associated transcript-density maps with marker-based cell-type inference at segmented nuclei (Thompson et al., 2026). GeoMx AOIs or segmented regions supply regional RNA or protein context. RNA and protein measured on different slides or through different aggregations provide complementary evidence at corresponding registered regions when sample lineage and alignment permit comparison (Song et al., 2024; Druschel et al., 2024; 2026).

Neighborhood scale is a property of the analytical representation in addition to an assay property. Cell-scale, spot-scale, and regional-ring neighborhoods describe different forms of local or regional context. Reported scales reflect cell size, vascular and tissue structures, platform resolution, registration error, and the physical dimensions of the implant. Exact implant distance, native measurement unit, and neighborhood definition are therefore descriptors of a reported niche.

3 Current assay-native spatial evidence at the IME–brain interface

Building on these scale distinctions, direct IME studies now provide four complementary spatial-data types: spot-resolved whole-transcriptome RNA, targeted Xenium transcript-coordinate RNA with marker-based cell-type inference, regional RNA or protein profiles, and registered morphology or IF (Whitsitt et al., 2024; Gupta et al., 2026; Song et al., 2024; Druschel et al., 2024; 2026; Thompson et al., 2026). Registered marker images include neuronal nuclei (NeuN), glial fibrillary acidic protein (GFAP), and oligodendrocyte transcription factor 2 (OLIG2), where available. The native observations are, respectively, multicellular capture spots, transcript coordinates with inferred cell types at segmented nuclei, selected tissue aggregates, and image-resolved structures or marker patterns. Figure 2 compares these assay-native units, and Table 2 provides an overview of the related publications before the individual platforms are considered in detail.

TABLE 2. Current spatial studies at the IME–brain interface.

Si, silicon; n, total animals reported per publication, including controls where applicable.

We identified relevant literature through a main search in OpenAlex, with Semantic Scholar used for record checks. Search terms combined spatial transcriptomics or spatial omics with biomaterials, implants, foreign body response, the brain electrode interface, spatial niches, spatial domains, integration across samples, and outcome prediction. Titles and abstracts were screened before available full texts were reviewed. Crossref and, where relevant, DataCite were used to verify citation metadata; PubMed Central, Europe PMC, and publisher records were used to locate accessible full text. Studies that preserved spatial molecular or imaging information at the IME interface formed the direct evidence base. Methods papers from other tissues were included when they addressed one of the analytical questions examined here. Preprints were identified as such.

Table 2 groups related publications by the spatial assay and analysis reported in each study. Note that the datasets do not necessarily represent independent cohorts. For example, Gupta and colleagues expanded the Whitsitt dataset and included four previously reported specimens (Whitsitt et al., 2024; Gupta et al., 2026). The Song study and the two subsequent Druschel studies form a linked experimental series: Song and colleagues reported proteomic data at 4, 8, and 16 weeks, while the later studies extended the 4- and 8-week comparisons through cell and distance segmentation and multi-omic profiling (Song et al., 2024; Druschel et al., 2024; 2026).

3.1 Assay-native measurements

Visium provides the current whole-transcriptome, spot-resolved evidence at this interface. In a 2024 rat motor-cortex study, multicellular capture spots were analyzed with registered NeuN and GFAP images from the same tissue section, and selected expression features were related to multiunit activity, local field potentials, signal-to-noise ratio, neuronal density, and GFAP intensity (Whitsitt et al., 2024). An expanded dataset used 55-μm capture spots to compare spatial expression in near and farther tissue at acute and chronic stages and to describe inter-animal variation (Gupta et al., 2026). These studies establish device-referenced whole-transcriptome variation and exploratory molecular–recording associations at the multicellular-spot scale.

Targeted Xenium provides the current transcript-coordinate evidence. A 2026 preprint reported transcript coordinates around implanted probes and used marker-based scores to infer cell types at segmented nuclei, together with registered NeuN, OLIG2, and GFAP imaging (Thompson et al., 2026). The study compared electrically inactive surrogate probes differing in material and cross-sectional size and described local cell-type composition and targeted expression profiles. Its 100-gene panel defines the measured state space, and recording outcomes were outside the study design.

GeoMx studies provide region-resolved RNA and protein evidence. Song and colleagues and Druschel and colleagues measured protein abundance in selected peri-implant regions (Song et al., 2024; Druschel et al., 2024), and a later study compared whole-transcriptome RNA and targeted protein profiles across corresponding radial bands (Druschel et al., 2026). These studies establish regional molecular differences and provide protein-level context for transcript findings. The selected AOI or compartment is the measurement unit, and RNA–protein correspondence is interpreted at the regional scale defined by each study.

Registered histology and IF accompany the current Visium and Xenium measurements, providing probe-tract geometry and neuronal, astroglial, and oligodendroglial marker context alongside molecular data (Whitsitt et al., 2024; Thompson et al., 2026). Within the available spatial evidence, Whitsitt and colleagues also related selected transcript features and registered image measures to electrophysiological outcomes. Separate IME studies have related conventional histological measures to recording outcomes and electrode impedance to longitudinal function, establishing relevant tissue–function and device–function context without spatial-niche resolution (Prasad and Sanchez, 2012; Nolta et al., 2015; Ereifej et al., 2020).

3.2 Statistical practice in current IME studies

Animal-level replication and within-animal sampling are distinct in the current IME datasets. In the Xenium study, Thompson and colleagues averaged gene-expression estimates for each cell type within specified spatial regions of each animal before analysis of variance, with the animal as the biological replicate (Thompson et al., 2026). The GeoMx studies sampled multiple mice and several implant sites within each mouse, with regional molecular measurements around those sites (Song et al., 2024; Druschel et al., 2024; 2026). Cells, spots, sampled regions, and implant sites within the same animal are therefore nested observations; counting them as independent animal replicates would constitute pseudoreplication.

Batch effects were explicitly examined in the expanded Visium dataset (Gupta et al., 2026). The three correction approaches tested by Gupta and colleagues reduced slide-associated clustering, but their outputs did not meet the intended downstream analysis requirements. The authors retained standard aggregation, which showed little clustering by slide and apparent operator-associated clustering. Using these data, per-animal pseudobulk analysis of the near-implant region showed generally greater between-animal expression variability in the examined genes at 1 week than at 6 weeks, with individual animals contributing strongly to the high variance of several genes.

Spatial correspondence also depends on tissue integrity and alignment. Whitsitt excluded regions affected by tissue folds, tears, or staining artifacts, while Druschel manually outlined deformed tract boundaries for GeoMx sampling (Whitsitt et al., 2024; Druschel et al., 2026). Uncertainty in image registration and tract-boundary placement can shift distance assignments and the molecular or image features attributed to a local region.

Taken together, the current literature documents peri-implant molecular and image heterogeneity at spot, transcript-coordinate, inferred-cell-type, and regional scales, together with a same-study link between selected spatial features and recording measures. Subsequent niche analyses of current and future IME datasets retain the same distinction between nested observations and biological replicates, with batch information and registration uncertainty informing interpretation. Building on this evidence base, the following sections connect niche identification, spatial organization, niche-conditioned molecular relationships, and recording performance within a common analytical framework. None of the niche-analysis methods reviewed here has yet been applied to IME data. We therefore discuss their compatibility with current IME datasets and the biological insights their application could provide.

4 Niche identification and biological characterization

These assay-native observations define the focal objects for niche identification, while local context is constructed at the resolution each assay supports. Spot-resolved contexts can use adjacent spots, a metric radius, or a graph corrected for missing tissue; cell-resolved contexts can weight neighbors by distance, contact, or tissue connectivity within the measured panel. Regional AOIs remain aggregated contexts, whereas registered images can delineate morphology-aware contexts from probe-tract geometry, vessels, cellular density, marker-intensity patterns, and other image objects. Animal and section identity, implant distance, cortical layer, treatment of probe-tract voids or damaged tissue, and image-derived covariates all affect interpretation. Across these units, niche representations draw on the four feature classes summarized in Figure 1B: cell-type composition, within-cell molecular state, tissue structure, and local interaction or relational context. The available combination varies by assay resolution.

Niche-identification methods produce more than a colored cluster map. Available outputs include continuous niche features, optional niche labels, abundance or area fractions, and cell- or spot-state summaries. Topological outputs include adjacency, mixing, segregation, boundary length, and graph motifs. Together, these outputs describe local organization within a specimen, its distribution across implant coordinates, and patterns reported across materials, implantation stages, or device designs.

Published approaches differ in the object they represent and in the way local context is encoded. The literature is organized here into five method families: composition-centered niches, state/program-centered niches, spatial-domain scaffolds, image-linked characterization, and serial-section/3D extension. Table 3 aligns these families with their representation outputs, characterization dimensions, spatial distinctions, and representative approaches. The five families describe complementary analytical content.

TABLE 3. Niche-identification method families, representations, and characterization outputs.

Composition-centered niches describe local context through cell-type proportions, phenotype co-occurrence, or soft composition programs. Cell-resolved implementations summarize phenotyped cells within a neighborhood, whereas spot-resolved implementations use deconvolved cell fractions or neighboring-spot summaries. Spatial-LDA represents soft mixtures of phenotype-defined microenvironments, CellCharter represents graph neighborhoods as niche classes, and CytoCommunity represents local phenotype communities (Chen et al., 2020; Varrone et al., 2024; Hu et al., 2024). Their outputs describe cellular makeup and heterogeneity among locations at the same implant distance.

State/program-centered niches add the molecular state of the focal cell or spot, neighboring states, or spatially patterned gene programs to local composition. scNiche combines focal-cell, neighboring-state, and neighboring-composition views in a joint embedding, whereas NicheCompass represents prior and de novo spatial programs in a graph embedding (Qian et al., 2025; Birk et al., 2025). Niche-DE associates focal-cell expression with an effective niche defined from local neighborhood composition, without necessarily producing recurrent niche classes (Mason et al., 2024).

Spatial-domain scaffolds describe broader organization through domain labels or shared spatial embeddings. BASS, PRECAST, SpaGCN, BayesSpace, BANKSY, and GraphST combine expression with spatial coordinates, neighborhood graphs, or registered histology in different forms (Li and Zhou, 2022; Liu W. et al., 2023; Hu et al., 2021; Zhao et al., 2021; Singhal et al., 2024; Long et al., 2023). These methods represent regional structure that can contain several local contexts; a broad spatial domain and a local niche therefore remain distinct analytical objects.

Published niche studies biologically characterize inferred contexts through enriched cell types, molecular states or programs, spatial adjacency, morphology, and associations with tissue remodeling or cell plasticity (Varrone et al., 2024; Qian et al., 2025; Birk et al., 2025). These features describe what a learned context contains and how its components are organized. Spatial localization is a separate descriptor: the same characterized context may be enriched proximally and also occur distally near a vessel or lesion.

4.1 Histology- and morphology-aware niche identification

Morphology supplies structural information that molecular coordinates alone leave unresolved. Relevant image objects include the probe-tract void, tissue fissures, vessels, neuronal layers, zones of dense immunoreactivity, and local distortion. Image-derived measurements include cell morphology, stain intensity, vessel distance, texture, lesion masks, and phenotype relationships. Image-linked methods expose complementary components of niche characterization: Starfysh combines spot expression with histology, histoCAT represents protein-defined neighborhoods and contacts, and SPIAT quantifies point patterns, mixing, distances, and tissue interfaces, as summarized in Table 3 (He et al., 2025; Schapiro et al., 2017; Feng et al., 2023).

Image-derived niche representations vary with tract and tissue annotation, segmentation near the implant void, autofluorescence, distorted morphology, and registration error relative to the neighborhood scale. Same-section registration relates molecular and image features at the same spatial observation, whereas adjacent-section imaging relates them across serial sections. These two configurations represent different forms of molecular–morphological correspondence.

4.2 Serial-section alignment and 3D reconstruction

2D sections sample only part of the device–tissue interface. A plane can intersect the 3D probe tract obliquely, miss an axial feature, or make a continuous structure appear as isolated islands. Serial-section methods extend these observations through full or partial coupling in PASTE and PASTE2, graph-derived cross-section anchors in STAligner, reference-assisted modeling in STitch3D, domain-guided coordinate alignment in SPACEL, and image-feature registration in STIM, as summarized in Table 3 (Zeira et al., 2022; Liu X. et al., 2023; Zhou et al., 2023; Wang et al., 2023; Xu et al., 2023; Pentimalli et al., 2025). Their outputs include slice correspondence, shared embeddings, aligned coordinates, integrated domains, and reconstructed 3D regions or neighborhoods.

3D reconstruction estimates continuity from measured sections rather than tracking the same cells through an intact volume. Adjacent sections contain different cells, damaged tissue can remove alignment landmarks, and molecular similarity can draw distinct regions together. Device geometry, histology, and section spacing contribute complementary spatial coordinates. The resulting representations range from repeated occurrence across sampled cross-sections to inferred volumetric continuity and axial organization.

4.3 Analytical value beyond distance-based summaries

Spatial-niche analysis adds four analytical distinctions to distance-based analysis. CellCharter reports niche recurrence and abundance across locations (Varrone et al., 2024); Niche-DE reports focal-cell expression associated with effective neighborhood composition (Mason et al., 2024); and point-pattern summaries add tissue topology to average-intensity measures (Feng et al., 2023). Together, these outputs distinguish heterogeneity among locations at the same radius, recurrence across radii, niche redistribution, molecular-state variation, and tissue topology. These distinctions preserve the separation between biological characterization and spatial localization: composition, state, structure, and local relations describe what a niche contains, whereas implant distance, vascular proximity, cortical layer, and lesion adjacency describe where it occurs.

The IME Xenium study reports transcript coordinates and cell-type scores at segmented nuclei. If transcripts are assigned to segmented cells, CellCharter and BANKSY can use the resulting cell-by-gene matrix and the cells’ non-grid spatial coordinates to characterize cell-resolved neighborhoods. For the regular hexagonal spot arrays in current IME Visium studies, both methods can characterize multicellular domains from spot expression and coordinates, while BayesSpace uses the known array adjacency. When several IME Visium sections are analyzed together, PRECAST takes an expression matrix and spot coordinates for each section, using shared genes to identify corresponding spatial domains while accounting for section-specific variation. GeoMx AOIs provide regional RNA or protein profiles and support comparisons at that scale. The following sections extend these descriptors to spatial patterns, gradients, boundaries, and state ordering, niche-conditioned molecular relationships, and recording-related outcomes.

5 Spatial organization of niches: Patterns, gradients, boundaries, and state ordering

5.1 Spatially variable features and pattern discovery

Spatial-organization analysis connects generic spatial patterning within a tissue section to organization referenced to the implant. Discovery methods differ chiefly in the spatial structures they represent, their scalability, and their use of an explicit neighborhood graph; Table 4 compares representative approaches (Svensson et al., 2018; Sun et al., 2020; Zhu et al., 2021; Weber et al., 2023; Miller et al., 2021; Ospina et al., 2022). These methods identify spatially structured features, programs, or expression surfaces; an IME-centered interpretation then asks whether those patterns align with the registered implant geometry.

TABLE 4. Analytical directions and IME-focused interpretations of spatial organization. References for representative approaches are provided in the corresponding subsections.

The measured interface is also structured by cortical layer, vascular adjacency, implant-facing or tissue-edge effects, and local cell composition (Bouadi and Tay, 2021; Gupta et al., 2024). Spatial pattern discovery can therefore compare alignment with the device coordinate and with native tissue structures. Proximal enrichment may reflect a change in niche abundance, a state change within a recurring niche, or both; spatial-niche and context-dependent expression methods describe these complementary dimensions (Varrone et al., 2024; Mason et al., 2024).

Niche maps themselves become spatial features. GraphCompass summarizes neighborhood enrichment, graph topology, and cross-condition organization, whereas SPIAT quantifies mixing, segregation, contact, and tissue interfaces (Ali et al., 2024; Feng et al., 2023).

5.2 Reference-aware gradients and learned spatial boundaries

Ring-based summaries can be extended toward continuous gradients and localized transitions. This direction distinguishes a gradual near–far change, a localized peak, and a boundary-like shift and compares these spatial forms among molecular features, niches, materials, or implantation stages (Kueckelhaus et al., 2024; Chitra et al., 2025).

Applicable methods span reference-path gradients and learned topographies with discontinuities, as summarized in Table 4. At an IME, the registered material boundary provides an external coordinate for comparing implant distance with cortical layers, vascular structures, and other non-radial tissue organization. Planar shanks and multi-site devices extend this comparison to the device face, edge, tip, and longitudinal geometry (Perna et al., 2023; Gupta et al., 2024).

Spatial boundaries can themselves be treated as biological features: a transition may occur in gene programs, niche abundance, niche adjacency, or their relationship to registered morphology. Learned topographies represent continuous gradients together with expression discontinuities; graph topology describes organization across conditions; and point-pattern methods quantify mixing and interfaces (Chitra et al., 2025; Ali et al., 2024; Feng et al., 2023). These outputs describe transition-zone variation across materials, device designs, and implantation stages.

CellCharter reports niche recurrence and abundance, whereas Niche-DE reports expression associated with effective neighborhood composition (Varrone et al., 2024; Mason et al., 2024). These outputs separate niche redistribution from molecular-state variation within local contexts when interpreted along implant position.

5.3 Spatial state ordering

Spatial state-ordering methods use expression embeddings or neighborhood composition to place heterogeneous locations along a continuous tissue axis, as summarized in Table 4 (Ren et al., 2022; Wang et al., 2025b; a). At an IME, these methods offer a route to ask whether candidate neuron-rich, reactive, vascular-remodeling, and lesion-associated contexts occupy a continuous order and how that order relates to implant-centered position.

SpaceFlow has related spatial state order to cortical layering and developmental organization, whereas ONTraC has related niche-trajectory scores to cortical depth, developmental stage, and expression variation within spatial domains. These precedents motivate asking whether peri-implant state order follows implant distance, shifts across implantation stages, or captures continuous organization within conventional radial bands.

5.4 Assay compatibility

For current IME Visium data, whole-transcriptome spot expression and coordinates provide inputs for spatialGE to analyze spatial expression patterns and for GASTON to infer gradients and domain boundaries. With an annotated implant boundary, Spatial Gradient Screening can examine expression changes with distance from that boundary. Xenium-derived cell-type labels and nuclear coordinates provide inputs for SPIAT to quantify distances and local arrangements among identified cells. The IME Xenium panel measures 100 RNA genes, which define the available gene set for expression-based analyses. GeoMx RNA and protein profiles support comparisons among sampled regions.

6 Niche-conditioned cellular and molecular relationships

Once spatial niches have been identified and placed relative to implant geometry, the next question is which cellular and molecular relationships characterize each niche, occur across its boundary, or vary along a niche-associated state order. The analytical object is not communication between abstract niche labels, but relationships among focal cells, neighboring phenotypes, and molecular programs conditioned on local context. Transcript measurements reveal niche-associated receiver programs and candidate ligand–receptor relationships, whereas spatial protein measurements reveal phenotype adjacency, receptor abundance, and activation-state patterns. This section organizes these parallel routes around the niche context they resolve.

6.1 Transcript relationships conditioned on and mapped to spatial niches

Transcript analysis can enter the niche framework through two complementary routes. In a niche- or context-first analysis, a previously defined niche label, mask, continuous feature, boundary, or state order organizes the molecular comparison. Niche-DE relates focal-cell-type expression to effective niches derived from local neighborhood composition, and its downstream niche-LR analysis prioritizes candidate ligand–receptor explanations for the resulting receiver programs (Mason et al., 2024). MISTy can instead encode the focal location, a declared local niche, and a broader implant-referenced region as distinct spatial views (Tanevski et al., 2022). These approaches ask which receiver responses and molecular dependencies characterize a particular local context.

An interaction-first route begins with spatial molecular relationships rather than a pre-existing niche label. SpatialDM identifies spatially associated ligand–receptor pairs and local hotspots, while COMMOT generates distance-aware candidate sender–receiver fields (Li et al., 2023; Cang et al., 2023). LIANA + provides a complementary integration layer for comparing candidate relations across estimators, views, or conditions and can organize those comparisons by niche labels or boundaries (Dimitrov et al., 2024). Mapping these outputs to a prespecified niche map, boundary, or state order shows whether a candidate relationship is concentrated within one recurring niche, spans an interface between niche classes, or varies among states of the same local context. Table 5 distinguishes these routes according to when niche information enters the analysis.

TABLE 5. Transcript-based routes for resolving relationships within and across spatial niches. Methods are grouped by when niche context enters the analysis.

The two routes produce a common niche-resolved output. A niche-conditioned receiver program can nominate local ligands, receptors, or neighboring sender identities, while an interaction hotspot can reveal molecular structure within a defined niche or at its boundary. Cell-coordinate data assign these relationships to neighboring cells within or across cell-resolved niches; spot data relate composition-informed receiver programs to multicellular niches. The current IME Xenium dataset provides targeted transcript-coordinate localization and marker-inferred cell types within the interaction space covered by its prespecified panel (Thompson et al., 2026). In each case, the biological comparison is organized by niche membership, boundary position, or niche-associated state rather than by a generic tissue-wide interaction score.

Current IME Visium expression matrices and spot coordinates provide inputs for MISTy, SpatialDM, and COMMOT at the multicellular-spot scale. Niche-DE additionally uses spot cell-type compositions and cell-type mean-expression profiles. Assigning Xenium transcripts to cells yields expression profiles that, together with cell coordinates, provide cell-level inputs for these methods. With cell-type labels and corresponding mean-expression profiles, Niche-DE relates cell-type-specific expression to neighboring cell composition, while niche-LR uses these expression changes to nominate candidate ligand–receptor relationships. In the current IME Xenium dataset, the 100-gene panel determines which ligands, receptors, and downstream target genes are available for analysis.

Spatial co-localization and ligand–receptor compatibility identify candidate interactions. Spatial controls compare observed associations with a background distribution to assess whether their localization is more specific than expected by chance (Li et al., 2023). Protein localization examines whether the corresponding ligands and receptors are present in compatible sender and receiver compartments, for example, through immunostaining in the relevant neighborhood. Receiver responses can be characterized by coordinated target-gene expression and inferred activity of the corresponding transcription factors (Dimitrov et al., 2024). Independent measurements of downstream activation or target-protein expression in the proposed receiver cells provide additional functional support for the candidate signal (Cang et al., 2023; Armingol et al., 2024).

6.2 Protein-native niche relationships and transcript–protein complementarity

Spatial protein measurements provide an independent route to niche relationships rather than serving only as confirmation of transcript-derived candidates. Building on the image-linked characterization described earlier, the relational features considered here include phenotype co-occurrence, enriched or depleted contacts, mixing or segregation, boundary-associated cell states, and the localization of phosphoprotein or downstream-effector signals. These features reveal how cellular organization and molecular state differ within a niche, across its boundary, and among niche classes.

histoCAT provides a framework for deriving cell phenotypes and quantifying their spatial interactions in multiplex image cytometry, while SPIAT measures cell distributions, distances, colocalization, neighborhood organization, and tissue interfaces (Schapiro et al., 2017; Feng et al., 2023). Once protein-defined neighborhoods have been identified, these outputs can compare phenotype adjacency, mixing, and activation patterns among IME niches and across their boundaries. Table 6 aligns transcript- and protein-resolved readouts across four niche-centered relationship dimensions.

TABLE 6. Transcript- and protein-resolved readouts across four dimensions of niche-centered relationships.

Transcript and protein measurements become complementary when they are assigned to the same niche, boundary, or registered local context. Transcript data contribute niche-conditioned receiver programs and candidate ligand–receptor relationships, while protein data contribute phenotype organization, receptor abundance, and activation-state patterns. Song et al. (2024) and Druschel et al. (2024) establish peri-implant protein profiling; Druschel et al. (2026) provide corresponding regional RNA–protein comparisons. When those regional measurements are registered to a declared molecular or morphology-aware context, they provide a region-centered view of the same niche relationship; cell-resolved multiplex protein data extend that view to phenotype adjacency and activation within a protein-defined niche. Transcript and protein approaches therefore provide parallel views of the cellular and molecular relationships that characterize spatial niches at the implant interface.

7 Spatial niches and recording-related outcomes

7.1 Existing approaches to recording-performance modeling

To connect these spatial features to function, recording performance has already been posed as an explicit supervised-learning target in longitudinal intracortical studies. In a non-peer-reviewed OpenReview submission, Urdaneta and colleagues combined temporal, electrochemical, electrode-design, cortical-depth, and prior-recording variables to predict whether a channel would yield sortable units in a target session; subject, time after implantation, previous spike metrics, and site depth were among the most influential variables (Urdaneta et al., 2023). The resulting task predicts target-session channel state from information available beforehand.

A parallel literature has established biological and device correlates of function. Longitudinal impedance has been related to single-unit detectability and used in an impedance–function model (Prasad and Sanchez, 2012); blood–brain-barrier leakage, astrogliosis, tissue loss, neuronal density, and inflammatory measures have been analyzed in relation to unit-based recording metrics (Nolta et al., 2015; Ereifej et al., 2020). Whitsitt and colleagues subsequently related individual genes and reduced Visium gene sets to multiunit activity, local-field-potential measures, signal-to-noise ratio, neuronal density, and GFAP intensity using regression-based analyses (Whitsitt et al., 2024). Together, these studies establish both the functional demand and the available modeling ingredients, while also distinguishing predictive modeling from tissue–recording association.

They also delimit three different questions. Contemporaneous association asks whether biological and recording measurements covary at a matched stage; retrospective explanation relates terminal tissue to an earlier or cumulative recording trajectory; prospective prediction requires an input, or a locked surrogate of that input, measured before the target session or failure event. The outcome unit may be a registered channel or site, a shank, an implant, or an animal; cells and spots provide biological features nested within that unit.

7.2 Candidate predictors and functional targets

Existing studies imply a hierarchical predictor space. Baseline features include implantation time, material and device geometry, cortical depth, impedance and other electrochemical variables, and prior recording history (Prasad and Sanchez, 2012; Urdaneta et al., 2023). Conventional biological features include neuronal density, vascular leakage, astrogliosis, inflammatory markers, and other terminal tissue measures (Nolta et al., 2015; Ereifej et al., 2020). Spatial molecular data add gene-level or program-level measurements, device-referenced expression profiles, and inter-animal variation (Whitsitt et al., 2024; Gupta et al., 2026); targeted Xenium measurements further add transcript-coordinate expression profiles and marker-based cell-type inference from tissue surrounding electrically inactive surrogate probes (Thompson et al., 2026).

Niche-derived predictors extend this hierarchy from individual measurements to recurrent local contexts. Candidate representations include the local composition of identified niches; focal-cell programs and pathway scores conditioned on niche identity; adjacency, mixing, and boundary organization between identified niche classes; radial, angular, and axial distributions of niche prevalence or state relative to the device; variation in niche composition or state across registered sites on the same device; and recurrent niche assemblies or ecotypes at the specimen level. For prospective prediction, a compact molecular or image surrogate is defined by a locked niche definition and surrogate-construction procedure.

Functional targets occur at different spatial and temporal scales. A registered channel or site carries local recording outcomes such as sortable-unit presence, unit yield, spike amplitude, signal-to-noise ratio, noise floor, impedance, or active-channel state. A shank, implant, or animal carries outcomes aggregated across sites or over time, including the fraction of functional channels, a recording-trajectory summary, recording longevity, or time to persistent functional failure. Both are recording-performance targets at different aggregation scales; cell- and spot-level molecular measurements instead describe nested tissue features.

Across the peri-implant studies reviewed here, recording-performance models and spatial molecular studies form complementary evidence streams. The former use electrical, temporal, device, and recording-history variables, whereas the latter provide within-cohort gene–recording associations or tissue-state maps. ST studies in other settings show how niche-conditioned molecular features and other niche representations have been pooled to an outcome unit (Mason et al., 2024; Fischer et al., 2023; Lee et al., 2026; Zuo et al., 2025; Zhang et al., 2026). These established applications provide the methodological bridge to recording-performance modeling.

7.3 ST methods for niche-feature construction and outcome-aligned pooling

Outcome-oriented spatial analyses distinguish local molecular modeling from modeling of the functional endpoint. A local method may quantify how neighborhood composition or spatial organization is associated with focal-cell expression, pathway activity, or latent state. These local outputs serve as niche features after reduction to a registered channel, implant, or animal and are analyzed alongside device, temporal, electrochemical, or recording-history variables. Representative methods across these stages are summarized in Table 7.

TABLE 7. Representative ST methods for niche-feature construction and outcome-aligned modeling.

GMM, Gaussian mixture model; ICI, immune-checkpoint inhibitor; NMF, non-negative matrix factorization; ROC, receiver operating characteristic; VAE, variational autoencoder; WTA, Whole Transcriptome Atlas.

Peri-implant molecular-state analysis addresses biological and material-response questions: existing spatial studies have related local genes to recording and histological measures, characterized spatial and inter-animal expression variation, and mapped targeted transcript profiles and inferred cell types around different probe materials and sizes. Those biological questions are treated in the preceding sections. In an outcome-oriented context, local ST models contribute candidate features. Niche-DE estimates niche-conditioned receiver-gene effects (Mason et al., 2024); NCEM reconstructs expression from receiver identity and a spatial graph (Fischer et al., 2023); MISTy partitions intrinsic, near, and broader spatial contributions (Tanevski et al., 2022); and NiCo models niche-associated partner identity and state covariation (Agrawal et al., 2024). COVET represents local gene–gene covariance directly, whereas ENVI incorporates COVET-derived context into a conditional latent embedding (Haviv et al., 2025). These outputs supply genes, pathways, latent states, covariance patterns, and view-specific importance values for endpoint-level modeling.

Outcome-aligned pooling conceptually matches these local features to the unit of the functional target. Within tissue registered to a channel or site, local niche composition, niche-conditioned state, and niche-interface organization can be summarized as predictors of the corresponding channel’s function. Across sites, device-relative niche distributions and across-site niche heterogeneity can characterize a shank or implant, while recurrent niche assemblies or ecotypes can characterize an implant or animal. Ecotype methods compress recurrent combinations of states and niches into a lower-dimensional mixture, whereas graph and multiple-instance methods learn local embeddings and apply mean, attention, or hierarchical pooling before an outcome head. Surrogate-transfer designs approximate a locked ST-defined feature with a compact molecular or image measurement available in the target cohort.

Outcome-linked ST studies illustrate increasingly structured pooling strategies. A melanoma GeoMx study derived a compartment-specific gene signature by averaging expression within tumor, leukocyte, and macrophage compartments before least absolute shrinkage and selection operator (LASSO) modeling (Aung et al., 2024). A single-cell CosMx study reduced nearest-neighbor niches to the patient-level fraction of favorable niches (Lee et al., 2026). stClinic instead learns within- and cross-slice graph embeddings, pools them with attention, and applies a classification or Cox head (Zuo et al., 2025).

Cross-modal transfer provides a second route from an ST-defined feature to a larger outcome cohort. Oral-cancer Visium programs were reduced to gene sets for bulk-cohort analysis (Arora et al., 2023); glioblastoma ST states were transferred to histology using a glioblastoma convolutional neural network (GBM-CNN) and related to survival with a Cox model (Zheng et al., 2023); and Spatial EcoTyper recovered recurrent ecotypes from bulk RNA, whereas Liquid EcoTyper recovered them from cell-free-DNA methylation data (Zhang et al., 2026). Related work transferred spatial protein-defined states to compact spatial RNA signatures (Aung et al., 2025).

For current IME data, regional GeoMx RNA profiles provide inputs for adapting the compartment-specific gene-signature approach to sampled peri-implant regions. If Xenium transcripts are assigned to cells, their expression profiles and coordinates support the neighborhood classification and niche-fraction representation illustrated by the CosMx study. Visium expression matrices and spot coordinates from multiple sections provide inputs for stClinic’s graph representation. Relating these candidate features to recording performance also depends on paired functional measurements. Whitsitt’s Visium data include such measurements, while future paired datasets would allow the corresponding GeoMx and Xenium representations to be examined.

7.4 Niche-based relationships with recording performance

Established recording comparators include implant-centered distance, overall cellular composition, impedance, and prior recording history (Prasad and Sanchez, 2012; Urdaneta et al., 2023; Whitsitt et al., 2024). Table 8 maps six candidate niche-derived feature families to recording outcomes at the corresponding functional scale.

TABLE 8. Candidate niche-derived feature families and recording-performance outcomes at matched functional scales. Each row presents a candidate relationship for testing in future IME studies; the final column summarizes related IME observations or hypotheses.

Histological studies relate neuronal, glial, and vascular measures to unit yield and recording performance (Nolta et al., 2015; Ereifej et al., 2020). These tissue profiles motivate representing the balance of local niche classes, while distance-resolved measurements motivate comparing niche distributions around the device. Whitsitt and colleagues reported regional gene associations with recorded activity and signal quality; niche-conditioned state extends this question to whether the recording association of a molecular program varies with its local context. Their discussion of reactive glia as a potential barrier to signal transmission provides a biological rationale for considering niche boundaries and mixing.

As paired spatial and functional datasets grow, the same niche-to-outcome question can be examined through increasingly rich representations. Direct summaries encode local niche composition, conditioned state, interfaces, and device-relative distributions; graph and multiple-instance representations retain higher-order relations among niche classes; ecotype models capture recurrent niche assemblies at the specimen level; and cross-modal surrogates can transfer an ST-defined pattern into cohorts measured by compact molecular panels or histology. These approaches are complementary routes for representing peri-implant organization, rather than a hierarchy determined by current cohort size.

Across this evidence, temporal alignment determines interpretation. Terminal tissue supports biological explanation of a preceding recording trajectory, whereas prospective prediction uses a spatial representation, or a transferable surrogate of that representation, available before the functional outcome. The biological question is which aspects of peri-implant spatial organization are reproducible across animals, devices, materials, and time points and are associated with recording performance at the corresponding temporal scale.

8 Discussion

The spatial-niche perspective addresses a central conceptual gap at the IME–brain interface. Device-centered coordinates specify where tissue is measured, and niches specify the local multivariate context encountered at that coordinate. Combining these representations preserves geometric specificity while exposing biological organization within a distance band, recurrence across bands, and alignment with anisotropic device and tissue structures. This review organizes the relevant literature around three complementary analytical dimensions: characterization of local context, description of its spatial organization, and evaluation of molecular relationships conditional on that context. Recording-related IME studies provide the functional endpoints, while outcome-linked ST studies in other settings provide methodological precedents for pooling niche features at compatible units.

Direct IME evidence now spans whole-transcriptome spots, targeted transcript coordinates with marker-inferred cell types, regional RNA and protein profiling, registered morphology, and electrophysiology (Whitsitt et al., 2024; Gupta et al., 2026; Song et al., 2024; Druschel et al., 2024; 2026; Thompson et al., 2026). These modalities contribute complementary, resolution-specific evidence: Visium captures broad programs at a multicellular spot scale, Xenium localizes selected transcripts within a targeted panel and supports study-specific cell-type inference, GeoMx supplies regional molecular profiles, morphology provides structural context, and available recordings supply functional measures. The review builds on these complementary evidence types by organizing candidate niche analyses within a shared implant-centered coordinate system and relating niche-derived features to recording outcomes at matched functional scales.

The material-intervention literature supplies relevant design contrasts. Coating chemistry, mechanics, feature size, substrate structure, and device architecture have been associated with tissue or recording endpoints (Hernández-Reynoso et al., 2023; Sturgill et al., 2024; Duncan et al., 2024; Mueller et al., 2025; Haghighi et al., 2025; Ocoko et al., 2026). Because some comparisons couple material and geometry and endpoint types vary, these studies define material-to-tissue and material-to-function contrasts at different levels. The material-centered synthesis asks whether radial tissue changes and reorganization of vascular, neuronal, inflammatory, or lesion-associated local contexts represent distinct responses to device design.

A central application gap is that none of the niche-identification, spatial-organization, or niche-conditioned relationship methods summarized in Tables 3–5 has yet been applied as such to IME spatial data. Current IME studies have instead used distance- or region-based comparisons, regression, marker-based scoring, and regional profiling. Direct transfer requires care because animal cohorts are small relative to feature dimensionality, implant-tract voids interrupt spatial graphs, targeted panels constrain state and ligand–receptor discovery, and sectioning, tissue deformation, and registration can alter the coordinate system.

The current IME evidence base also includes overlapping experimental lineages across some publications, a targeted Xenium dataset analyzed from transcript coordinates with marker-based cell-type inference currently reported as a preprint, and multimodal measurements collected across different sections or spatial units. Interpreting findings by experimental lineage and measurement scale is therefore more informative than publication counts alone.

Each platform also contributes a distinct observation process. Visium aggregates transcripts across multiple cells and is sensitive to capture and tissue quality around the implant void. Targeted in situ assays restrict discovery to the selected panel and depend on segmentation and molecule assignment. GeoMx profiles selected regions whose composition contributes to pathway estimates. Imaging introduces staining, segmentation, and registration variability, while 3D integration adds alignment and missing-section uncertainty. Cross-platform comparisons therefore interpret niche labels and molecular relationships at their native measurement units.

Interface organization varies with cortical layer, local vasculature, insertion damage, material, device geometry, micromotion, sex, animal, and time. When niche models are applied, their learned labels may encode these variables together, so differences among inferred niches may reflect biological context, device effects, implantation history, and specimen-specific technical variation.

Niche-conditioned expression, spatial ligand–receptor compatibility, regional protein support, and functional association form distinct links in the evidence chain. Transcript-based relationships nominate associations, regional or registered protein localizes them at its measured scale, and spatial RNA–recording analyses connect selected features with function.

The spatial-niche perspective positions local context as an intermediate analytical level between functional materials and device performance. Distance preserves the implant-centered geometry, niche features describe organized local context, and recording evidence defines the functional variables to which those features can be compared. The resulting material–niche–function question provides a common language for comparing spatial omics, morphology, interface engineering, and chronic recording studies.

9 Conclusion

Distance is an essential coordinate for IME research, and spatial niches add the cellular, molecular, and structural context present within and across implant-centered coordinates. Reporting the focal unit, neighborhood rule and scale, input features, and validation strategy makes that context explicit and comparable.

Three complementary analytical dimensions organize the literature: identifying and characterizing niches; examining their spatial patterns, gradients, boundaries, and state ordering; and evaluating niche-conditioned transcript and protein relationships. Visium, Xenium, GeoMx, registered morphology, and electrophysiology contribute distinct evidence layers. Niche composition, conditioned state, interface topology, device-relative distribution, across-site heterogeneity, and recurrent assemblies provide comparable feature families across these layers; studies in other settings have extended comparable molecular or image surrogates to outcome cohorts.

The evidence chain is strengthened by recurrence across independent animals, resolution-matched multimodal support, and evaluation of spatial-feature associations with recording performance at matched spatial and temporal scales. These elements clarify how material properties, device geometry, and spatial tissue organization relate to interface function. Distance locates the interface; niche analysis describes the local context in which the interface functions.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The authors used OpenAI Codex (GPT-5, OpenAI, https://openai.com/codex) for language editing, reference checking, and manuscript formatting. The authors reviewed and verified all AI-assisted content and take responsibility for the final manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Enbo Zhu, University of California, Los Angeles, United States

Reviewed by: Siu Hong Dexter Wong, Ocean University of China, China

Pengxiang Chen, University of California, Los Angeles, United States