4D single-cell spatial transcriptomics reveals dynamic morphogenetic gradients and regenerative domains in planarians

Abstract

Background

Understanding how organisms reconstruct complex tissue architectures following injury requires precise mapping of gene expression and cellular responses across space and time. Although planarians serve as a classic model for whole-body regeneration, capturing the continuous spatiotemporal dynamics of positional information and cell fate decisions at the organismal scale remains a significant challenge.

Results

Using high-definition spatial transcriptomics, we generated a 4-dimensional atlas encompassing over 3.5 million cells from whole animals across 8 distinct regeneration timepoints. This comprehensive dataset enabled the definition of 36 spatial domains and the tracing of body axis restoration, revealing that positional control genes recover through self-organizing dynamics analogous to an underdamped control system. We identified an injury-induced spatial domain termed the anterior regenerative zone. This unique region is characterized by the convergence of epidermal, muscular, and neural lineages enriched with positional signals. Furthermore, we demonstrated that the transcriptional co-factor Mediator 8 is a critical regulator of this zone. Depletion of Mediator 8 impairs the formation of the anterior regenerative zone, disrupts polarity establishment, and prevents successful blastema formation.

Conclusions

Our study provides a holistic molecular and cellular reconstruction of whole-body regeneration, directly linking dynamic gene expression gradients to morphological restoration. The discovery of the Mediator 8-regulated anterior regenerative zone highlights the importance of spatial domains in coordinating tissue repair. The resulting interactive atlas serves as a foundational resource for deciphering the logic of spatiotemporal patterning in regeneration.

Background

Understanding the mechanisms of tissue and organ regeneration following injury is a fundamental biological question with profound implications for regenerative medicine, wound repair, and aging. Regeneration involves a complex series of events, including local and systemic responses to injury, the restoration of positional information, and the formation of new tissue structures [1, 2]. While progress has been made in understanding these processes, several critical questions remain [3, 4]. How do cells spatially respond to injury, and how do molecular gradients influence tissue patterning? What roles do specific cell types, morphogen gradients, and injury-responsive regions play in regeneration across complex organisms? Furthermore, how can we capture and quantify these processes at the molecular level across an entire organism?

To address these questions, profiling organisms, cells, and genes across multiple spatial and temporal scales is crucial [5, 6]. However, capturing continuous positional signals at molecular, cellular, and organismal levels remains a significant challenge, especially in 3-dimensional space and time [7]. The complexity of tissue heterogeneity, large body sizes, and the preservation of cellular and molecular organization in extensive tissue sections complicate spatial transcriptomic analyses. Moreover, the absence of quantitative assays and frameworks to capture continuous positional signals at the transcriptome level in 3-dimensional space and time presents an additional technical obstacle. This challenge is further compounded by the limited availability of classical model organisms that can fully regenerate their bodies. To date, a comprehensive 3-dimensional molecular reconstruction of cellular architecture and morphogen gradients across an entire organism has not been achieved, limiting our understanding of the dynamic changes in cellular and molecular identities during regeneration.

Planarians, renowned for their exceptional regenerative abilities, serve as an ideal model for studying the spatial and temporal dynamics of tissue regeneration [3]. These bilateral metazoans possess a complex anatomy [8], including a brain, nerve cords, peripheral nervous system, epidermis, intestine, muscles, excretory system, and a centrally located pharynx. Composed of a variety of cell types derived from 3 germ layers, planarians rely on pluripotent stem cells, or neoblasts, for constant tissue turnover and regeneration [9, 10]. They utilize precise positional cues to guide body axis establishment and tissue patterning [11, 12]. Numerous genes involved in signaling pathways for body plan patterning have been identified, expressed in a complex spatial map across the dorsoventral (D/V), mediolateral (M/L), and anteroposterior (A/P) axes [13–16]. These genes, known as position control genes (PCGs), are largely expressed in muscle tissue and play a critical role in regulating positional information during regeneration [17]. However, the precise phenotypic outcomes associated with many of these genes remain poorly understood, and it is still unclear whether positional information is confined exclusively to muscle tissue [17]. The regenerative process requires cells to establish, record, and interpret positional information to rebuild the body’s complex structure. This highlights the importance of profiling genes and cells across multiple spatial and temporal scales. Although advances in single-cell RNA sequencing (scRNA-seq) [18–24] and spatial transcriptomics (ST) [25, 26] have enabled the profiling of cell and gene expression patterns, these technologies still face limitations in achieving high spatial resolution at both the cellular and organismal levels. Furthermore, the mechanisms by which injury-induced local signals guide stem cells to reconstruct the body axis and regenerate fully functional 3-dimensional structures remain poorly understood [3, 27]. Thus, there is a need for comprehensive analytical frameworks that can capture the dynamic molecular and cellular events of regeneration across entire organisms.

In this study, we used Stereo-seq [28] and our custom framework [29] to create an extensive atlas of 3,508,004 segmented cells from 353 sections of 16 complete planarians, spanning 8 time points of whole-body regeneration (WBR). With a resolution of 715 nm, we generated detailed transcriptional and anatomical maps of the regeneration process, constructing positional and transcriptional gradients along the body axis. Our 4D transcriptomic atlas, annotated with 36 refined spatial domains, provides a comprehensive view of gene expression dynamics across cellular, tissue, and organismal scales. Our findings offer key insights into the regenerative process. First, we reveal complete spatial gene expression patterns that define positional gradients along the body axes, tracking their dynamic spatiotemporal fluctuations following amputation. Additionally, we identify an injury-induced anterior regenerative zone (ARZ), marked by ROD1, which exhibits enriched positional signals in epidermal, muscle, and neural cells. The ARZ is regulated by Mediator 8 (med8), which is crucial for polarity establishment, blastema formation, and overall regeneration. These results provide a comprehensive molecular and spatial map of regenerative processes, highlighting dynamic changes in regeneration-responsive cells, spatial domains, and key regulatory factors.

Data description

To comprehensively map the spatiotemporal dynamics of WBR, we employed high-resolution Stereo-seq (715 nm) on 353 cryosections derived from 16 Schmidtea mediterranea individuals across 8 regeneration timepoints (0 h to 14 days). This 4D reconstruction yielded 3,508,004 segmented cells, from which we annotated 36 distinct cell clusters and identified the injury-induced ARZ. This dataset enables genome-wide transcriptional imputation across body axes and provides a foundational resource for modeling morphogenetic gradients and regenerative patterning. Raw sequencing data are deposited in the CNGB Nucleotide Sequence Archive under accession STT0000028, while processed data, code, and 3D visualizations are accessible via the PRISTA4D interactive portal and GitHub for unrestricted community reuse.

Analyses

Reconstruction of planarian 4D spatiotemporal transcriptomes at single-cell resolution

To investigate the cellular and molecular dynamics of regeneration, we generated a comprehensive 4D atlas of gene expression and cellular changes during the regeneration process (the “Methods” section). We focused on pre-pharyngeal amputations, which regenerate the head, tail, and pharynx over a 2-week period (Fig. 1A). Using the Stereo-seq technique [28], which integrates tissue cryo-sectioning with in situ RNA sequencing at 715 nm resolution, we profiled gene expression at multiple stages of regeneration. Animals were collected at 8 distinct time points: 0, 12, and 36 hours post-amputation (hpa) and 3, 5, 7, 10, and 14 days post-amputation (dpa). For each time point, 2 animals were sampled, resulting in a total of 16 regenerating animals. These animals were sectioned along the dorsal–ventral axis to capture the entire organism (Supplementary Fig. S1A). The 16 animals were processed into 10-μm-thick sections, producing a total of 353 slices for ST analysis using the Stereo-seq platform (Fig. 1A and Supplementary Fig. S1A).

Figure 1.

3D molecular reconstruction of whole-body planarian regeneration using 4D ST. (A) Schematic representation of the amputation strategy and sampling design for planarian WBR. The dotted line indicates the amputated pre-pharyngeal fragments. The central diagram illustrates the arrangement of 17 embedded tissue samples per block, including 1 intact sample and 2 replicates at each of the 8 time points post-amputation. (B) 3D spatial visualization of 36 SPC clusters (top) and tissue meshes (bottom) in representative animals at 8 time points during WBR. SPC clusters are labeled in the bottom panel. Tissue meshes highlight the intestine (purple), pharynx (green), and central nervous system (gray). (C) Top: Bar plots showing changes in body volume size; line charts depicting variations in the length of the D/V, M/L, and A/P axes relative to the 0 hpa sample at 8 time points of regeneration. Bottom: Bar plots showing SPC cell counts and line charts illustrating the average cell volume across the 8 time points of regeneration. (D) Left: Spatial patterns of the pharynx (Pha.19) and CNS (Neu.25) clusters at 8 regenerative time points. Right: Line plots showing the volume proportions of the reconstructed organs (neural and pharynx) during regeneration. Error bars represent the standard deviation from 2 replicates. Linear fitting lines show nearly linear time-dependent changes. (E) FISH staining showing spatial patterns of pharynx (foxA) and neural (pc2) markers during regeneration. Scale bars: 500 µm. n ≥ 3. (F) Schematic diagrams illustrating the 3-step process for identifying blastema subdomains based on the watershed algorithm for pigment variations. The head blastema region is divided into HAR (head anterior region), HBOR (head border outer region), and HBIR (head border inner region), while the tail blastema region includes TPR (tail posterior region), TBOR (tail border outer region), and TBIR (tail border inner region). (G) Left: Conceptual schematic illustrating the spatial definition of blastema sub-regions based on relative distance to the wound. Right: UMAP visualization of transcriptomic profiles from the corresponding blastema regions. (H) Heatmap depicting the relative expression of genes enriched in different blastema regions across all time points. (I) Stacked bar plot illustrating the number of differentially expressed genes in each SPC cell type or blastema region under the indicated comparison conditions. Genes with a significance threshold of adjusted P-value < 0.05 and absolute log10(Fold Change) > 0.5 were included.

To facilitate the generation of a comprehensive 3D reconstruction, we aligned and stitched the individual tissue sections (the “Methods” section). This resulted in 3,508,004 segmented cells across the 16 reconstructed animals, with cell counts ranging from 58,450 to 432,197 per time point after quality control (QC) (UMIs per cell > 50) (Fig. 1B, Supplementary Fig. S1A, and Supplementary Table S1). This ST atlas spans all 8 time points, providing a high-resolution, 4D view of regenerating planarians at subcellular resolution (Fig. 1B and Supplementary Fig. S1B). The dataset allows for the tracking of the spatial dynamics of regeneration-responsive genes and cellular interactions at various stages of regeneration. To validate the quantitative accuracy of our spatial transcriptomic atlas, we performed a correlation analysis between the Stereo-seq data (aggregated as pseudo-bulk) and public bulk RNA-seq data [30]. The high correlation coefficient (R = 0.8, Supplementary Fig. S1C) demonstrates the robust reliability and technical reproducibility of our spatial dataset. The dataset is publicly available via our searchable browser, PRISTA4D (Supplementary Fig. S1D).

To enhance the identification of biologically relevant tissue domains, we combined gene expression data with spatial information using the spatial proximity-based clustering (SPC) method. This method groups cells based on both transcriptional similarity and spatial proximity (the “Methods” section). Using SPC, we identified 36 distinct spatial domains (Fig. 1B). Correlation analysis with published single-cell datasets revealed that the majority of these domains represent lineage-restricted populations (Supplementary Fig. S2A–C). Furthermore, comparison with published 10x Genomics Visium datasets confirmed consistent spatial transcriptional signatures across platforms (Supplementary Fig. S2D). Together, these systematic cross-platform validations confirmed the accuracy of our cell identities while highlighting the superior resolution of Stereo-seq in resolving fine-grained spatial heterogeneity within broader tissue domains [29]. For example, our analysis revealed significant spatial heterogeneity in the parenchyma, which was subdivided into 11 distinct subclusters (Supplementary Fig. S3A), consistent with previously identified heterogeneity [18, 23]. We also identified well-known tissue sub-populations, including epidermal progenitors, dorsal and ventral epidermal populations, and subpopulations of goblet cells and phagocytes within the intestine (Supplementary Fig. S3B). Furthermore, we discovered a new spatially localized domain, Clu.31, within the blastema. Initially emerging in both the head and tail blastemas at 36 h post-injury, this domain eventually became restricted to the head by the end of regeneration (Supplementary Fig. S3A), which we designated as the ARZ.

For better tissue contour characterization, we generated tissue meshes for the intestines, pharynx, and neuronal regions. The spatial distributions of these clusters and tissue meshes were reproducible across 2 animals at each time point (Fig. 1B and Supplementary Fig. S1B), confirming the robustness of our data at the organismal level. Together, this 4D atlas provides a valuable resource for studying the temporal and spatial dynamics of gene expression and cellular coordination during WBR.

Capturing tissue and organ remodeling and identifying genes responsive to regeneration

The comprehensive 3D reconstruction enabled precise measurements of tissue volume changes through the 4D dataset. Analyzing the length ratios along the D/V, M/L, and A/P axes revealed that the D/V and M/L axes shortened, while the A/P axis elongated during regeneration (Fig. 1C, top). Although the cell counts decreased, the average cell volume remained largely unchanged, suggesting that the observed volume changes were primarily due to a reduction in cell numbers (Fig. 1C, bottom). This finding aligns with the body-wide plasticity previously observed in planarians [31].

Using the 4D dataset, we tracked the regeneration of the pharynx, as well as the remodeling of the nervous system (Fig. 1D), with validation through fluorescent in situ hybridization (FISH) for pharyngeal (foxA) and neural markers (pc2) (Fig. 1E). Both the pharynx and central nervous system, particularly the cephalic ganglia, exhibited increased volume over time, while the intestine showed a decrease in size but underwent remodeling. The pharynx began to form between 3 and 5 dpa, while the central nervous system matured by 5 dpa (Fig. 1D and E), reflecting the gradual remodeling in these organs.

To investigate the dynamic cellular responses during regeneration, we analyzed the proportions of different domains over time. We classified cell cluster dynamics into 5 patterns: continuous increase, initial increase followed by decrease, continuous decrease, initial decrease followed by increase, and unchanged (Supplementary Fig. S2C). Notably, dorsal epidermal progenitors (Epi.30), neural progenitors (Neu.28), and pharyngeal lineages (Pha.19 and Pha.29) exhibited a gradual increase (Supplementary Fig. S3C–D), reflecting the expansion of these specific lineages required for tissue reconstruction. These results suggest that planarians may balance cell and organ proportions during regeneration, dynamically rescaling body proportions and restoring axial polarity.

Due to the fragility of blastema tissue, conventional FISH methods are challenging for capturing internal gene expression and spatial distribution in this region [32]. To overcome this limitation, we hypothesized that digital segmentation of the blastema region, based on pigmentation intensity, would provide a more effective means of analyzing gene expression and cellular composition. Using image recognition algorithms, we segmented the animals into head blastema, tail blastema, and pre-existing trunk regions (Supplementary Fig. S3E). Statistical analysis revealed positional heterogeneity in cellular responses. For instance, the pharynx (Pha.19) was initially localized to the tail blastema and later to the trunk, suggesting its centripetal migration toward the body center. Additionally, the ARZ (Clu.31) was induced in both blastemas but persisted only in the head region (Supplementary Fig. S3F). This segmentation highlights the potential of our virtual 4D data in identifying distinct cell subtypes that emerge at various stages and locations during regeneration.

To further investigate the molecular responses in finer regions, we divided the regenerating head and tail blastemas into 3 subdomains, proximal, middle, and distal, along with the trunk, and identified region-specific gene expression patterns through clustering (Fig. 1F and G). Both the head and tail blastemas exhibited similar wound-healing and remodeling gene expression, with enrichment in genes associated with Wnt and BMP signaling pathways. However, different subdomains within these regions showed distinct gene expression profiles (Fig. 1H, Supplementary Table S2). Additionally, comparison of differentially expressed genes (DEGs) within the same clusters or regions across various regeneration time points revealed temporal variations in gene expression patterns. For example, we observed that parenchymal domains and epidermal progenitors responded to injury within the first 12 hpa, while goblet cells and cathepsin⁺ cells were activated between 12 hpa and 3 dpa. Neuronal cells and the ARZ (Clu.31) domain showed a response after 3 dpa (Fig. 1I, Supplementary Table S2), highlighting distinct cellular and regional responses at different stages of regeneration.

In summary, the 4D atlas offers an in-depth view of tissue remodeling, dynamic changes in cellular localization, and the spatial distribution of cell populations throughout planarian regeneration.

Spatiotemporal dynamics of positional gradients during WBR

In an accompanying study, we characterized gene expression patterns along body axes and identified genes with regional expression patterns in 3D intact planarians, which we proposed as spatially biased genes (SBGs) [29] (Supplementary Table S3). Some of these genes, involved in patterning processes, were classified as positional control genes (PCGs) [11] (Supplementary Table S4). To investigate the dynamics of these SBGs during regeneration, we employed our 4D dataset to map gene expression across the entire organism. We hypothesized that injury would disrupt SBG expression, with recovery occurring gradually as regeneration progressed.

To test this, we analyzed the spatiotemporal patterns of SBGs by mapping their expression dynamics onto 16 canonical spatial clusters defined by normalized body length (100 bins) along the A/P axis using logistic regression (Fig. 2A, Supplementary Table S3). At 0 hpa, following injury, the physical loss of head and tail territories led to the disruption of anterior and posterior-specific clusters, whereas medial patterns remained stable. By 5 dpa, the overall gradient patterns had largely been restored to resemble those of uninjured individuals (Fig. 2A), highlighting the dynamic process of positional remodeling during regeneration. As regeneration proceeded, these spatial domains were progressively reconstructed, with gene expression patterns returning to their homeostatic states by 14 dpa. The complete gene composition for each spatial cluster is documented in Supplementary Table S3.

Figure 2.

Dynamics of SBGs during WBR. (A) Heatmaps showing changes in spatial expression patterns along the A/P axis during regeneration for the 16 A/P pattern cluster genes. These genes exhibit spatially biased expression along the A/P axis in intact animals. Hollow rectangles indicate the absence of specific gene expression clusters at particular time points. Animals are virtually divided into 100 sections along the A/P axis. Left margin annotations indicate cluster numbers. (B) PCA representation of pattern reconstruction for known and potential PCG candidates. PC1 corresponds to the head-tail gradient feature, while PC2 captures fluctuations in the pharynx region (convex and concave). Time-resolved trajectories in PCA space reveal universal self-organized dynamics during SBG pattern regeneration, resembling an underdamped mass-spring system (top). (C) Schematic diagrams illustrating the hierarchical reconstruction of SBG patterns, colored by cluster ID. SBGs are arranged by the timing of repatterning completion, as shown on the right. Hollow circles indicate the absence of expression at specific time points. Red arrows highlight newly identified SBGs. The color gradient represents the recovery of SBGs at different stages post-amputation. The box highlights the time points when repatterning is complete. (D) Spatiotemporal modeling of SBG patterns during regeneration. The interaction between SBGs and their upstream regulators generates Turing patterns in an autoregulatory activator–inhibitor system (top). Gene expression at any time point after amputation can be predicted. The lower line charts compare predicted repatterning (dashed line) with measured data (solid line) for both known and newly identified SBGs. r, Pearson’s correlation. (E) Bar plot showing the number of SBGs highly enriched in each SPC cluster at different time points during regeneration.

Leveraging the quantitative nature of ST, we further explored the spatiotemporal dynamics of SBGs. Principal component analysis (PCA) on known PCGs along the body axes of intact planarians revealed that genes expressed in the same regions clustered together (Supplementary Table S4). For example, ndl-4 and sfrp1 were specifically expressed in the head (Supplementary Fig. S4A), confirming their region-specific patterns [33, 34]. This analysis also allowed us to quantify relative expression patterns, such as ndk in both the head and pharynx, and wntA in the pharynx (Supplementary Fig. S4B), consistent with established spatial distributions [15, 35–37].

Next, we mapped SBGs across multiple time points in PCA space to track their dynamic recovery during regeneration. Notably, several known PCGs, including ndk, fz5/8–4, fz4-1, and wnt11-2 [15, 35], exhibited a reciprocal recovery pattern along the A/P axis. These genes initially showed higher expression compared to uninjured individuals, before gradually returning to baseline levels (Fig. 2B, bottom). Our analysis revealed that following the disruption caused by amputation, SBG expression did not simply ramp up or down linearly. Instead, these genes exhibited transient dynamic fluctuations along the body axis. We modeled this recovery as a perturbation in a dynamic system, analogous to an underdamped mass-spring system (Fig. 2B, top), where the system is displaced from equilibrium (homeostasis) by amputation and subsequently driven back by a restoring force representing the gene regulatory network (GRN) [38, 39]. This dynamic trajectory eventually restored the gene expression profile to its homeostatic state by 14 dpa (Fig. 2A). The PCA (Fig. 2B) further delineates this process into distinct biological phases. The early time points (0–36 hpa) cluster separately from later stages, representing an “Acute Injury Phase” characterized by a distinct wound response program that drives the system far from equilibrium. This is followed by a “Patterning Phase” (3–10 dpa), where the trajectory shifts direction and converges toward the homeostatic attractor [40]. This observation is consistent with the distinct, transient injury states recently identified in single-cell studies [18], confirming that early wound responses are transcriptionally distinct from the later morphogenetic programs that restore body plan fidelity. We hypothesize that injury disrupts regional expression, and that the GRN may act as a restoring force, guiding the recovery of disrupted PCG expression through feedback mechanisms (Fig. 2B, top). Consistent with this model, our analysis showed that PCG expression was elevated between 3 and 10 dpa, and returned to baseline by 14 dpa upon completion of regeneration (Supplementary Fig. S4C), further supporting this recovery model.

To visualize the dynamic recovery of spatial patterns, we color-coded the expression patterns of known PCGs at each time point (the “Methods” section). These patterns were restored by 12, 36 hpa, or 3 dpa (Fig. 2C), suggesting a temporal progression in their recovery. Notably, spatial patterns were restored earlier than corresponding detectable changes in gene expression. For instance, the spatial pattern of ndk was restored by 12 hpa (Fig. 2C), while its expression began to increase only at 36 hpa (Supplementary Fig. S4C and Supplementary Table S3). These temporal dynamics suggest that spatial patterning may influence the regulation of gene expression.

Motivated by the underdamped response of PCG trajectories (Fig. 2B) and the temporal hierarchy between spatial pattern recovery and gene expression (Fig. 2C), we hypothesized that SBG expression dynamics could be modeled mathematically. Extending the self-organizing model proposed for the Wnt pathway along the A/P axis [33], we applied the Gierer–Meinhardt model of a simple activator–inhibitor system [41] within the Turing system framework [42] to predict global changes in SBG expression during regeneration. While the model traditionally describes the diffusion and reaction kinetics of secreted morphogens, such as the Wnt/Notum pair, we applied it here to characterize the expression of downstream TFs, such as hox4b and EGR1. We propose a Readout hypothesis where these nuclear factors do not diffuse themselves but act as high-resolution spatial proxies that interpret the primary, diffusive morphogen gradients. By analyzing spatial gradients in exponential form and excluding the influence of the pharynx-enriched genes, we separated independent temporal and spatial components to simulate changes in activator and inhibitor expression at each time point during regeneration (Fig. 2D, top). Our findings demonstrate that the predictions of trajectories of these “readout” genes, such as ARNT [43], Ndk [15], EGR1, hox4b, Nfya, and glra1 (Fig. 2D, bottom, Supplementary Table S4) closely follow theoretical activator–inhibitor kinetics, reinforcing the conclusion that planarian regeneration is guided by scalable, self-organizing patterning systems.

We next explored whether SBGs were enriched in specific domains or regions. By quantifying the number of SBGs enriched in each SPC at different stages of regeneration, we observed that SBGs were expressed across multiple lineages, including muscle, epidermal, and neural lineages (Fig. 2E). Interestingly, the ARZ domain (Clu.31) displayed characteristics from several lineages and contained the highest number of SBGs (Fig. 2E). This observation suggests that the ARZ may play a role in remodeling and maintaining polarity. Overall, our 4D analysis provides a detailed view of the spatiotemporal dynamics of SBG expression during regeneration, offering support for a model based on self-organized reaction-diffusion patterns.

Characteristics of the injury-induced ARZ enriched in SBGs

Having demonstrated that the ARZ (Clu.31) exhibits injury-induced anterior localization and is enriched in SBGs, we hypothesized that the ARZ plays a crucial role in regulating PCGs and maintaining regenerative polarity during regeneration, similar to the proposed function of the organizer [44]. We further sought to characterize this region. At homeostasis, the ARZ is localized to the anterior side, with signatures of 3 distinct lineages (Fig. 3A and Supplementary Table S5). Gene expression analysis within this domain revealed enriched expression of SMED30003831 (smed03831 or Rod1), caveolin3, and SMED30001640 (smed01640) (Supplementary Fig. S5A and Supplementary Table S2). Notably, the ARZ spans both the peripheral epidermal and subepidermal areas of the head, distinguishing it from the Equinox-expressing wound epidermis [45] (Supplementary Fig. S5A). Gene Ontology analysis of ARZ-enriched genes identified processes related to epidermal differentiation, muscle contraction, and neural development (Supplementary Fig. S5B). Co-FISH experiments with the ARZ marker smed03831 and lineage markers confirmed that the ARZ encompasses epidermal (agat-1) (Fig. 3B), muscular (collagen) (Fig. 3C), and neural (pds) cells (Fig. 3D), solidifying its trilineage characteristics. These findings suggest that the ARZ is a co-regulated, regeneration-responsive region.

Figure 3.

Identification and characterization of the ARZ (Clu.31) domain in the blastema region. (A) Spatial visualization of the Clu.31 domain in a homeostatic worm. (i) 3D ST data showing Clu.31 (green) with the right panel displaying the fifth section from ventral to dorsal (z = 5). (ii) Top: Enlarged view of 3 lineage signatures within Clu.31 in a single slice. Bottom left: Enlarged view of the head region from (i). Bottom right: ssDNA staining highlights the presence of multiple cell layers coexisting in Clu.31, with the green dashed line indicating the location of Clu.31. (B–D) FISH staining for the Clu.31 marker (smed03831, magenta) with the epidermal marker agat-1 (green) (B), muscle marker collagen (green) (C), and neuron marker pds (green) (D). Co-expressed cells are indicated by yellow arrowheads. Scale bars: 500 µm (left); 50 µm (right). n ≥ 3. (E) Spatial distribution of the Clu.31 domain during regeneration, shown from top, front, and side views. Green dots represent SPC cells within Clu.31, pink dots represent SPC cells in the blastema region, and gray dots represent other SPCs. (F) The position and cell number of Clu.31 during regeneration. Top: Line plot (red) showing the decreased minimal distance (min dis) of Clu.31 to the wound surface, accompanied by an increase in the cell number (cell num) of Clu.31 during wound healing (yellow line). Min dis represents the minimal distance to the wound surface in UV spatial coordinates, while cell num refers to the number of Clu.31 cells. (G) FISH staining showing smed03831 expression in the head blastema during regeneration. Enlarged areas are shown to the right. n ≥ 3. Scale bars: 500 µm (left); 50 µm (right). (H) Pseudotime trajectory analysis of Clu.31 across 6 time points of WBR, from 36 hpa to 14 dpa. (I) Feature plot showing the expression of representative cell-type marker genes—epidermis (left), muscle (middle), and neuronal (right) lineages—along pseudotime trajectories from (H). (J) Bubble plot displaying the gene set module scores of markers for neural (Neu.), muscular (Mus.), and epidermal (Epi.) lineages within Clu.31 during regeneration.

To track the temporal dynamics of the ARZ, we analyzed its spatial location throughout regeneration. At 36 hpa, the ARZ was present as scattered clusters near the ventral wound sites in both head and tail fragments. By 3–5 dpa, these cells converged toward the midline and expanded to cover the wound area, coinciding with wound closure and blastema formation. By 10 dpa, ARZ cells diminished in the tail but persisted in the head region (Fig. 3E, Supplementary Fig. S5C). Measurements of the distance from the wound surface revealed that the ARZ gradually approached the amputation site during the first 5 days, with an increase in cell number within the zone (Fig. 3F). This was further confirmed by staining for the ARZ marker smed03831 in the regenerating head region (Fig. 3G).

To investigate the putative origin of ARZ cells, we traced their pseudotime trajectory during regeneration using Monocle [46]. This analysis revealed three distinct branches (Fig. 3H), each enriched for genes specific to epidermal, muscle, or neuronal lineages (Fig. 3I, Supplementary Fig. S5D). The earliest reappearance of epidermal signatures at 12 hpa marked the emergence of the ARZ. To validate this injury-induced differentiation dynamics, we profiled the temporal expression of the late epidermal progenitor marker agat-1 within the ARZ. We observed a progressive enrichment of agat-1 starting from 36 hpa and peaking at 3 dpa (Supplementary Fig. S5E), mirroring the kinetics of active blastema differentiation rather than static tissue maintenance [18, 47]. This was followed by the emergence of muscle and neuronal markers at 3 dpa, coinciding with blastema formation (Fig. 3J). By 14 dpa, the ARZ cellular composition had largely reverted to epidermal cells, resembling the homeostatic state (Fig. 3J). The expression of ARZ-enriched genes aligns with these cellular dynamics (Supplementary Table S5), further supporting the coordinated and timely assembly of the ARZ domain during regeneration.

Cellular composition and regulation of the polarity-enriched ARZ domain

Having demonstrated the coordination between ARZ formation and regeneration, we next sought to investigate the cellular components that control ARZ formation. Our focus was on the epidermis, as it constitutes the earliest cell type to emerge within the ARZ. To identify the regulatory factors involved in ARZ formation, we employed RNA velocity, a method that distinguishes between unspliced and spliced mRNAs [48], to predict the putative trajectory of the SPC clusters in epidermal lineages (Supplementary Fig. S6A). The velocity vectors indicated that epidermal cells in the ARZ primarily originate from the ventral epidermal lineage (Epi.1) (Supplementary Fig. S6dA and B). To visualize ARZ formation in a spatiotemporal context, we projected 3D spatial data onto 2D maps of the head blastema at various time points via unwrapping (Fig. 4A), a dimensionality reduction technique transforming a 3D model’s surface into a 2D plane (see the “Methods” section). Pseudotime trajectory analysis of the unwrapped 2D map revealed a cell-state transition from ventral epidermal cells toward the ARZ, while left and right dorsal–ventral boundary cells on both sides of the head blastema moved toward the anterior pole (Fig. 4B–D). These observations suggest that ARZ formation involves interactions between ventral and dorsal epidermal cells. Indeed, comparing spatial maps at 36 hpa and 3 dpa revealed the expansion of the ARZ domain from both ventral and dorsal sides, coinciding with wound closure (Fig. 4E).

Figure 4.

Cellular composition and regulation of the ARZ domain. (A) Schematic illustrating the workflow for unwrapping the 3D epidermal surface of the head blastema into 2D spatial coordinates. UV unwrapping of the 3D mesh was performed by manually marking a seam along the D/V boundary (Step 1) and unwrapping the surface by cutting along the seam (Steps 2 and 3). Epidermal cells were projected onto the 2D coordinates by minimizing the distance between each 3D cell center and the nearest subdivision vertices (Step 4). See the “Methods” section for details. (B) UV spatial mapping of the ventral epidermal transition forming the Clu.31 domain at 12 hpa. Top: Distribution of Clu.31 (green) and epidermal SPCs on the unwrapped UV map, with dot colors representing SPC types. Labels and dashed lines indicate key positions. Bottom: Predicted cell transition streams of the epidermis as modeled using Dynamo. DVB, dorsal–ventral boundary; A pole, anterior pole; P, posterior. (C) UV spatial map showing Clu.31 transition patterns at 36 hpa. Top: Distribution of Clu.31 and epidermal SPCs on the unwrapped UV spatial map. Bottom: Cell transition streams of the epidermis predicted using Dynamo. (D) UV spatial map showing Clu.31 transition patterns at 3 dpa (top) and 5 dpa (bottom). The distribution of Clu.31 (green) and epidermal SPCs on the unwrapped UV spatial map is shown. (E) Visualization of Clu.31 movement during wound healing (top, 36 hpa) and blastema formation (bottom, 3 dpa). The white arrow indicates the predicted trajectory of the epidermal transition, delineated based on the spatial dynamo shown in (C) and (D). Arrowheads highlight the location of either the wound surface (top) or blastema (bottom). (F) Pseudotime trajectory analysis of ventral epidermis and neoblast cells. Left: Distinct states of SPCs identified by pseudotime analysis, with cells colored by SPC clusters (top left) and pseudotime (bottom left). Right: Heatmap showing significantly altered genes discovered by Monocle 2 along the trajectory. (G) Expression and localization of med8 and smed03831 in the head blastema of regenerative fragments at the indicated time points. Scale bars, 500 µm. (H) Percentage of med8⁺, smed03831⁺, or co-expressing cells in the blastema shown in (G). ns, P > 0.05; **P < 0.01; 2-tailed unpaired t-test. (I) Representative phenotypes following med8 RNAi at 5 dpa. n = 10 animals for each condition. Scale bars: 500 µm. (J) Expression and localization of smed03831 and sfrp-1 in the blastema of control and med8 RNAi animals at 1.5 (top left) and 3 dpa (top right). Scale bars: 500 µm. The ratio of smed03831- or sfrp-1-expressing cells in the blastema was quantified (bottom). ns, P > 0.05; *P < 0.05; **P < 0.01; 2-tailed unpaired t-test. (K) FISH staining of smed03831 in control and med8 RNAi animals. n ≥ 3 animals per condition. Scale bars: 500 µm (top); 50 µm (bottom).

To explore the role of the ARZ as a signaling-rich domain during regeneration, we conducted trajectory analysis to identify potential regulators involved in cell differentiation within the ARZ (Fig. 4F). While genes such as smed03831 and caveolin3 serve as definitive markers of the differentiated ARZ, we sought to identify the upstream drivers governing its formation. Notably, the mediator complex subunit 8 (med8) emerged as an early-expressed gene along the pseudotime trajectory (Fig. 4F), preceding the expression of structural markers. Med8 is an essential component of the mediator complex, playing a critical role in transcription regulation [49]. The planarian med8 homolog is evolutionarily conserved and shares high sequence identity with orthologs in other species (Supplementary Fig. S6C). In the homeostatic state, med8 is highly expressed in neoblasts (Supplementary Fig. S6D), with enrichment observed across multiple neoblast subpopulations (Supplementary Fig. S6E–F). Following injury, med8 expression increased in the wound area at 12 hpa, prior to the emergence of smed03831⁺ cells at 1.5 dpa (Fig. 4G). This suggests that med8 may regulate ARZ formation. The co-expression of med8 and smed03831 in a substantial portion of blastema cells after 1.5 dpa further supports its regulatory role in ARZ formation (Fig. 4G and H). To assess the functional role of med8, we performed RNA interference (RNAi) knockdown experiments (Supplementary Fig. S6G) and measured the expression of the ARZ marker smed03831. Knockdown of med8 resulted in impaired blastema regeneration by 5 dpa (Fig. 4I) and a significant reduction in the number of smed03831⁺ cells at 3 and 5 dpa (Fig. 4J), suggesting that med8 is required for ARZ formation.

Having shown that med8(RNAi) hinders ARZ reconstruction (Fig. 4J), we next investigated the specific stage at which med8 influences ARZ formation by examining gene expression at different regeneration time points. Given that the ARZ is enriched with various PCGs (Fig. 2E), we hypothesized that the failure in ARZ reconstruction would prevent the re-establishment of anterior polarity. To test this, we examined the expression of the anterior pole marker sfrp-1 [14, 50], which was significantly reduced upon med8 knockdown (Fig. 4J), suggesting a failure to generate the anterior pole identity during regeneration. In control animals, smed03831⁺ cells were enriched at the wound site at 1.5 dpa and fully covered the wound, followed by the appearance of sfrp-1⁺ cells at the anterior pole by 3 dpa (Fig. 4J). This temporal sequence suggests that ARZ formation precedes anterior pole formation. In contrast, med8(RNAi) animals exhibited impaired ARZ formation and a reduction in sfrp-1⁺ cells (Fig. 4J), indicating an inability to re-establish anterior identity. By 7 dpa, ARZ formation was completely disrupted in med8(RNAi) animals, and regeneration failed (Fig. 4K), linking ARZ formation to successful regeneration. Together, these findings support the idea that med8-mediated ARZ formation is essential for providing the cellular basis for pole formation during regeneration.

Med8 is required for ARZ lineage specification to support blastema development

To identify the transcriptional programs mediating changes in the ARZ, we conducted scRNA-seq on med8 and control RNAi animals with amputated tails undergoing head regeneration. By integrating the data from both groups, we identified known cell lineages, including stem cell populations and 8 distinct cell types (Fig. 5A), consistent with previous findings [9, 23]. Notably, med8 RNAi expanded the neoblast population while reducing the proportions of ARZ-associated lineages, specifically neural, muscle, and epidermal cells (Fig. 5B). This observation was further confirmed by pseudotime trajectory analysis using Monocle, which revealed similar reductions in the differentiation of these cell lineages (Supplementary Fig. S7A). Label transfer analysis matching ARZ cells across the scRNA-seq dataset confirmed a decrease in each ARZ cellular component in med8 RNAi animals (Fig. 5C and Supplementary Fig. S7B).

Figure 5.

Med8-dependent regulation of ARZ and blastema formation. (A) UMAP visualization of scRNA-seq analysis depicting cell lineages from control and med8 RNAi-treated tail fragments. (B) Bar plot showing the changes in cell populations following med8 RNAi knockdown. (C) Left: UMAP visualization of neural, muscular, and epidermal cells highlighting the ARZ signature (Clu.31-like) in scRNA-seq data. Right: Bar plot illustrating the decrease in the fraction of ARZ (Clu.31) cells in med8 RNAi-treated animals compared to controls. (D) Violin plot showing changes in the module score of A/P gradient genes in ARZ (Clu.31) cells. P-values are from the Wilcoxon test: ns, P > 0.05; ****, P < 0.0001. (E) Dot plot illustrating the relative expression of representative PCGs in anterior or posterior regions from scRNA-seq data. (F–G) WISH analysis showing the expression and localization of posterior markers wnt-1 (F) and wnt11-1 (G) in control and med8 knockdown animals. Scale bars: 500 µm (top), 50 µm (bottom). n = 6 animals with consistent results. (H–I) Expression and localization of muscle marker collagen and epidermal marker agat-1 in the blastema of control and med8 RNAi animals at 3 (H) and 5 dpa (I), respectively. Scale bars: 500 µm. n = 6 animals with consistent results. The percentage of positive cells in the blastema was quantified. ns, P > 0.05; *P < 0.05; **P < 0.01, 2-tailed unpaired t-test. (J) Expression and localization of the neural marker pc2 in the blastema of control and med8 RNAi animals at 3 and 5dpa. Scale bars: 500 µm. The percentage of positive cells in the blastema was quantified (right). *P < 0.05; **P < 0.01, 2-tailed unpaired t-test. All FISH images are maximum-intensity projections. Error bars represent SEM. n ≥ 3 biologically independent experiments.

Given that the ARZ is enriched for genes involved in polarity formation, we next examined whether the differentiation blockade caused by med8 RNAi led to a loss of cells expressing anterior and posterior pole markers. Analysis of A/P axis marker gene expression in the scRNA-seq data revealed that med8 RNAi led to a reduction in the polarity signature score within ARZ cells, but not in other cell types (Fig. 5D and Supplementary Fig. S7C). Further analysis showed a decrease in both anterior and posterior markers (Fig. 5E). Whole-mount in situ hybridization (WISH) confirmed reduced expression of posterior markers, including wnt1 and Wnt11-1 [37]. However, body-wide polarity was not completely disrupted in med8 RNAi animals (Fig. 5F and G), suggesting that the observed changes are linked to regenerative growth. These results indicate that med8 RNAi impairs ARZ formation, a prerequisite for establishing positional landmarks within the blastema during regeneration.

To investigate how med8 affects ARZ formation, we examined the expression of ARZ markers within the blastema. The spatial extent of major cell lineages in the ARZ, including epidermal (agat-1) and muscle (collagen) cells, was reduced in med8 RNAi animals (Fig. 5H). Additionally, the expression of transcription factors essential for the differentiation of neural (tcf/lef-1) [51], epidermal (p53) [52], and muscle (dmrt2) cells was diminished upon med8 knockdown (Supplementary Fig. S7D), indicating that med8 is crucial for maintaining the transcriptional programs associated with ARZ cell fate [53]. In line with this, gene expression analysis in neoblasts revealed downregulation of pathways related to stem cell division as well as neural and epidermal fate determination (Supplementary Fig. S7E). To quantify this defect at the cellular level, we calculated the differentiation efficiency across lineages in our scRNA-seq dataset. We found that despite an accumulation of undifferentiated neoblasts, the efficiency of generating lineage-committed progenitors was severely compromised for the neural, muscle, and epidermal cells (Supplementary Fig. S7F), confirming a broad differentiation blockade. Furthermore, FISH staining at 3 and 5 dpa demonstrated reduced expression of markers for epidermal and muscle cells in the blastema region (Fig. 5H and I), suggesting impaired cell fate decisions in ARZ cells. We also observed decreased expression of neural markers in med8 RNAi animals (Fig. 5J), further confirming the disruption of ARZ lineage specification and impaired head regeneration.

Finally, we assessed whether med8-mediated ARZ formation is required for homeostasis. While short-term med8 RNAi treatment caused minimal phenotypic changes in homeostatic animals, prolonged med8 knockdown led to head regression (Supplementary Fig. S7G), suggesting that sustained loss of ARZ function results in homeostatic defects. In summary, our data support the role of med8 in controlling blastema growth by enabling the specification of the ARZ-associated lineages, which is essential for proper tissue regeneration and the maintenance of homeostasis.

Discussion

Understanding the full spectrum of spatial information and the principles governing pattern formation during regeneration in tissues and organs remains a significant challenge. This complexity is driven by the intricate tissue geometry, the large size of multicellular organisms, and the limited number of model organisms capable of regenerating entire tissues. Additionally, the absence of techniques capable of capturing high-resolution spatial transcriptomic data across an entire organism in 3D over time further complicates this challenge. In this study, we applied high-resolution Stereo-seq (715 nm) to planarians to reconstruct the 4D spatiotemporal landscape of WBR. Our 4D dataset addresses several limitations of traditional techniques, including low-throughput FISH assays and 2D single-slice-level ST, providing a holistic and high-resolution view of genes and spatial domains before and during regeneration. While recent ST methods provide spatial context [25, 26, 54], they often lack full 3D or single-cell resolution, limiting the ability to comprehensively profile morphogen gradients and domains across entire organisms over time. In contrast, our 4D regeneration atlas offers a high-resolution, time-resolved framework for analyzing regenerative dynamics. Mining our dataset allowed for the recovery of single-cell transcriptomes at the spatiotemporal level, enabling the visualization of gene expression patterns, positional signals, and cell type distributions across multiple scales. This analysis revealed morphogen gradient gene dynamics, identified regenerative domains, and highlighted key regulatory factors. The full dataset is available through our online resource, PRISTA4D.

By leveraging the complete repertoire of SBGs across 4 dimensions, we examined gene expression across anatomical regions and scales, particularly within the delicate blastema region, allowing for quantitative analyses across multiple body regions. We confirmed known PCGs and identified potential new PCG candidates, facilitating the modeling of morphogenetic gradients using the Turing reaction-diffusion model. While the Turing system has been proposed for fission and regeneration [33, 42, 55], its applicability to modeling planarian regeneration remains unclear [56]. The temporal dynamics of SBGs suggest that the restoration of positional identity functions as a self-organizing system. Rather than simple oscillations, the recovery follows an underdamped trajectory where the initial injury response (0–36 hpa) creates a significant displacement from the transcriptomic baseline, followed by a directed convergence during the patterning phase (3–10 dpa). This “overshoot and return” dynamic supports a model where the GRN acts as a restoring force, progressively dampening the perturbation introduced by amputation until the homeostatic equilibrium is regained. Our real dataset supports simulation predictions, with specific genes exhibiting consistency between modeled and observed data. This rich dataset and quantitative approach provide a foundation for studying scalable, self-organized pattern formation in more detail, offering a framework for understanding how positional information is re-established during regeneration. Integrating high-resolution ST with single-cell analysis, our study offers a valuable resource for investigating how positional information is maintained, disrupted, and interpreted across different tissues and domains following injury. This dataset bridges the gap between molecular, cellular, and morphological aspects of regeneration, offering a comprehensive multimodal view of WBR dynamics in organisms. This supports a hierarchical model where global self-organizing gradients are interpreted by local GRNs to restore tissue identity. These findings align with the perspective that regeneration relies on the re-establishment of organizers that coordinate tissue repair through these dynamic molecular gradients [44, 57].

Muscle cells have long been recognized as the primary conveyors of positional cues in adult planarians [17]. However, our 4D dataset expands this view, identifying non-muscular lineages, such as neuronal and epidermal cells, as contributors to the regenerative positional landscape. This suggests that pattern remodeling involves a multi-lineage process, with multiple cell types participating in encoding, reading, and interpreting positional information [58, 59]. We comprehensively profiled SBGs, which extend beyond PCGs, to define a systemic patterning system involving genome-wide spatial regulation. Since regenerative patterning involves both global polarity and local fate specification, most SBGs may not show defects in polarity but could alter cell fate [34, 60, 61]. The temporal dynamics of different SBG classes suggest that encoding and interpreting positional gradients is a distributed and hierarchical process, coordinated among various domains and exhibiting self-organizing properties within the organism.

The distinct temporal lag between the restoration of spatial patterning, such as ndk asymmetry at 12 hpa, and the subsequent transcriptional surge at 36 hpa implies a hierarchical “Pattern-then-Amplify” regulatory logic. We propose that the immediate post-injury phase involves the unmasking of positional identity within pre-existing tissues. As muscle cells serve as the primary source of positional instructions, surviving tissue at the wound boundary utilizes stable landmarks to rapidly reset the anatomical coordinate system [17]. This process is likely orchestrated by early signaling modulators, such as Ptpn11, that fine-tune wound response pathways and Wnt dynamics prior to significant transcriptional activation [62]. Consequently, a precise spatial blueprint is established at low transcriptional levels by 12 hpa to spatially constrain the subsequent proliferative burst. The later peak in expression at 36 hpa thus reflects the amplification of this established pattern to support the massive cellular demands of blastema formation, effectively decoupling the definition of “where to regenerate” from the execution of “how much to grow” [63].

Analysis of SBG distribution revealed a specific spatiotemporal domain, Clu.31, which we define as the ARZ. While this domain contains mechanosensory neurons and epithelial cells characteristic of the anterior peripheral nervous system, our data suggest it represents a dynamic cellular neighborhood [18, 54] where the intercalating peripheral nervous system integrates with the dorsal–ventral boundary epithelium and body wall muscle to orchestrate regeneration. Importantly, given its baseline presence during homeostasis, the ARZ likely does not represent an entirely de novo structure exclusive to regeneration. Instead, it constitutes a specialized, resident cellular neighborhood within the uninjured anterior tissue. Upon injury, the constituent cells within this resident niche become hyper-responsive, undergoing profound transcriptional alterations to serve as a regenerative organizing center. The remarkable plasticity of this neighborhood is evidenced by its transient ectopic induction even in the posterior wound (tail blastema) (Fig. 3E) and its dynamic compositional shifts, transitioning from an initial injury-induced epidermal state to a multi-lineage hub before resolving to its homeostatic baseline (Fig. 3H–J). The temporal dynamics of agat-1 expression within this domain further mirror transient regeneration-activated cell states described in other contexts [18, 47], supporting the hypothesis that the ARZ functions as a highly responsive regenerative organizing center rather than solely representing the regenerating nervous system. From an evolutionary perspective, the presence of this spatiotemporally coordinated domain is functionally reminiscent of the apical epithelial cap (AEC) observed in vertebrate appendage regeneration [64]. Although anatomical differences exist, particularly the integral role of neural components in the planarian ARZ, the principle of a transient signaling niche established by the convergence of epithelial and sub-epithelial tissues [65, 66] appears conserved. Similar to the AEC, the ARZ exhibits conserved molecular marker gene expression involving the Wnt/β-catenin and FGF pathways [14, 15, 50]. Further comparative analyses will be required to determine if the specific molecular circuits governing these cellular neighborhoods are homologous across metazoans [44].

Our findings using med8(RNAi) as a proxy suggest that epidermal, muscular, and neural cells within the ARZ likely contribute to positional information for blastema induction. The persistence of this domain in adult planarians may help explain their homeostatic maintenance, providing insights into the regulation of regeneration [44, 63]. We further demonstrate that med8, a subunit of the Mediator complex acting as a bridge between transcription factors and RNA polymerase II [49], modulates key genes related to epidermal, muscle, and neural specification within the regenerative domain. The Mediator complex is known for its role in maintaining stem cells, as well as in lineage-specific differentiation [53]. While the loss of Smed-med14 specifically affects stem cell populations, the loss of med8 does not [67], suggesting that distinct Mediator components have different requirements for stem cell function in planarians. Our findings build upon previous studies and demonstrate that med8 is important for the differentiation of stem cells into neural, muscle, and epidermal lineages within the ARZ region, thereby ensuring the production of proper regenerative patterning signals. This suggests that Mediator, in conjunction with transcription factors [68–73], may be involved in establishing the epigenetic landscape necessary for lineage commitment and cell fate transitions.

The observation that med8 knockdown markedly reduces the expression of polarity markers without causing gross disruption of the whole-body AP axis (e.g., double-head formation) warrants further discussion. Our single-cell and lineage analyses suggest that this discrepancy likely stems from the hierarchical role of med8 in cell fate specification rather than direct gradient scaling. Since med8 is required for the differentiation of specific ARZ lineages (epidermal, muscle, and neural), the observed reduction in polarity markers (e.g., sfrp-1, wnt1) reflects a loss of the signal-producing cells themselves, rather than a simple downregulation of gene expression within an intact tissue. Furthermore, med8 knockdown leads to a symmetric reduction in both anterior (sfrp-1) and posterior (wnt1, wnt11-1) signals. This balanced reduction likely preserves the relative antagonism between the anterior and posterior poles, preventing the dominance of one pole over the other that typically drives ectopic structure formation. Finally, while gross morphology is maintained in the short term, we noted that prolonged med8 RNAi eventually leads to head regression, confirming that med8-mediated cellular turnover is indeed essential for the long-term maintenance of cell populations that harbor positional landmarks of the global body axes.

Potential implications

Despite the advantages of our 4D approach, specific limitations remain. First, the sequencing depth is lower compared to scRNA-seq, which may impact the detection of rare cell populations. Second, biological replicates are limited due to the technical challenges of generating whole-organism 4D data. Additional replicates and validation will be required to further assess the robustness of morphogenetic gradients across individuals. Future efforts should focus on increasing sequencing depth, expanding the number of biological replicates, and incorporating complementary approaches to validate the dynamics of gene regulation. Nevertheless, integrating these data with gene perturbation and longitudinal imaging studies will enable us to directly assess the functional contributions of specific positional signals in guiding pattern remodeling and regeneration.

Our study establishes a framework for a 4D high-resolution atlas of gene expression dynamics throughout WBR. By combining spatial and temporal transcriptomic data, we provide a novel framework for understanding the principles governing regenerative patterning, advancing both regenerative biology and ST methodologies. This comprehensive dataset serves as a valuable resource for future studies, enabling researchers to explore positional information, tissue remodeling, and the regulation of regeneration in biological systems.

Methods

Animal culture

Asexual S. mediterranea (strain CIW4) was maintained at 20°C in a recirculating 1× Montjuic salts solution without antibiotics, following a previously described protocol [74]. The animals were routinely fed beef liver. For experimental procedures, the animals were transferred to static culture and starved for at least 7 days.

Gene cloning and RNAi feeding

Genes of interest were cloned from a CIW4 cDNA library into the pPR-T4P vector as previously described [75]. The resulting plasmids were used to produce dsRNA for RNAi. RNAi was performed following established protocols for gene knockdown [76]. Briefly, bacterial pellets expressing the dsRNA were mixed with fresh beef liver paste in a 4:1 ratio. EGFP dsRNA was used as a control. Animals were fed every 3 days for a total of 4 or 6 RNAi feedings. After the final RNAi feeding, animals were amputated 3 days later to collect samples at various stages of regeneration. Sequences of all RNAi constructs and target regions for each construct are included in the Supplementary Materials.

In situ hybridizations

WISHs were conducted following previously established protocols [32, 77]. In short, the mucus from the animals was removed using 5% NAC in PBS, and then fixed for 1 h in 4% formaldehyde (FA) in PBSTx (0.5%). The animals were bleached with formamide and incubated with proteinase K (2 μg/ml, AM2546, Thermo Fisher) for 10 min. After a 2-h pre-hybridization step, the hybridization was performed at 56°C for over 16 h. Following extensive washes, the antibody signal was amplified using the tyramide signal amplification system. Tissue clearing was achieved using ScaleA2 to reduce background noise [32]. Antibody and probe sequences used in the study are listed in the Supplementary Materials.

Sample fixation and section preparation for Stereo-seq

Sample fixation was carried out using a modified version of a previously described protocol [78]. In short, animals were relaxed in 0.66 M MgCl₂ for 1 min, followed by fixation in methacarn solution (6 ml methanol, 3 ml chloroform, 1 ml glacial acetic acid) for 10 min. After fixation, the animals were rinsed in methanol 3 times, rehydrated in 50% methanol in PBS for 5 min, and then cryoprotected in 20% sucrose in PBS for 2 cycles. The cryoprotected tissues were embedded in pre-cooled OCT, frozen with dry ice, and stored at −80°C until cryosectioning. Following embedding, the specimens were photographed under a stereomicroscope to acquire brightfield microscopy images. These images documented the morphological features and macroscopic pigmentation patterns of each specific animal, serving as the anatomical reference for downstream spatial region segmentation and data alignment. Tissues were equilibrated in a −20°C freezing microtome for 30 min prior to sectioning. RNA quality from cryosections was assessed using an Agilent 2100 Bioanalyzer. The cryosections of S. mediterranea were cut serially at 10 μm intervals using a Leica CM1950 cryostat. Each section was placed onto a Stereo-seq chip, incubated for 3 min at 37°C on a Thermocycler Adaptor, and then fixed in methanol at −20°C for 40 min.

ssDNA staining and imaging of Stereo-seq slides

Prior to tissue permeabilization, sections on the Stereo-seq chip were stained with a nucleic acid dye (Thermo Fisher, Q10212) to visualize single-stranded DNA (ssDNA). The stained sections were then imaged using a Leica DM6M microscope. The images were stitched together and processed using the Leica Application Suite X software.

Library construction and sequencing of Stereo-seq data

The library construction and sequencing protocols for Stereo-seq have been previously described [28]. In summary, tissue sections were first washed with 100 μl of 0.1× saline-sodium citrate buffer (Thermo, AM9770) containing 0.05 U/μl RNase inhibitor (NEB, M0314L) to remove any remaining staining solution. Sections were then permeabilized using 0.1% pepsin (Sigma, P7000) in 0.01 M HCl buffer (pH 2.0) and incubated at 37°C for 18 min. Released mRNAs were captured on the Stereo-seq chip and reverse transcribed overnight at 42°C using SuperScript II reverse transcription (RT) mix (Invitrogen, 18064-014), containing 10 U/μl reverse transcriptase, 1 mM dNTPs, 1 M betaine solution, 7.5 mM MgCl₂, 5 mM DTT, 2 U/μl RNase inhibitor, 2.5 μM Stereo-seq template switch oligo, and 1× First-Strand buffer.

After in situ RT, tissue sections were treated with a removal buffer (10 mM Tris–HCl, 25 mM EDTA, 100 mM NaCl, 0.5% SDS) at 37°C for 30 min. The remaining RT products were then collected and amplified using KAPA HiFi Hotstart ReadyMix (Roche, KK2602) and 0.8 μM cDNA-PCR primers. PCR products were used to prepare sequencing libraries, with the following steps: quantification of concentration using the Qubit™ dsDNA Assay Kit (Thermo, Q32854), DNA fragmentation with in-house Tn5 transposase at 55°C for 10 min, PCR amplification (KAPA HiFi Hotstart ReadyMix, Roche, KK2602) with 0.8 μM cDNA-PCR primers, and purification using Vazyme (N411-03). The purified PCR products were used to construct DNB libraries and sequenced on an MGI DNBSEQ-T1 sequencer (35 bp for Read1,100 bp for Read2). The sequencing data were processed to generate a quantified spatial gene expression matrix at the subcellular level.

ST data processing

Spatially resolved single-cell RNA-seq data obtained through Stereo-seq were pre-processed for further analysis. The first read (Read1) of the sequencing library contained coordinate identifiers (CIDs), molecular identifiers (MIDs), and poly-T sequences, while the second read (Read2) provided the captured cDNA sequences. Spatial x–y coordinates of cDNA at 715 nm resolution were determined based on the CID sequences with a 1-bp mismatch tolerance. cDNA sequences were aligned to the S. mediterranea genome (dd_Smes_G4), and only mapped reads were used to identify exon transcripts. The MID sequences served to provide unique molecular identifiers (UMIs) for transcript quantification, with PCR duplicates removed using handleBam. Read pairs with a MID quality score below 10 were excluded. Finally, gene expression matrices incorporating spatial information were generated using quality-controlled exonic data [79].

3D reconstruction, clustering, and cell type annotation of regenerating animals

Regenerating planarians were reconstructed using methods outlined in an accompanying manuscript, where we developed a 3D ST framework [29]. First, the MIRROR algorithm was applied to align the spatial gene expression heatmap with the ssDNA staining image. Cell segmentation was then performed utilizing using CellProfiler and Fiji. Gene expression data were mapped to each individual cell, creating a spatial transcriptome map at single-cell resolution [80]. Next, after performing dimensionality reduction and clustering, cell clusters were annotated based on known lineage markers. The SEAM algorithm was employed to align the sections along the z-axis, thus determining the x–y–z coordinates of each cell. Morphological distortions induced by experimental procedures were corrected based on established polarity gene patterns, and the 3D reconstructions were created using a combination of 3DSlicer and MeshLab. Finally, SPC cells from different stages of regeneration were integrated using the FindIntegration and IntegrateData functions in Seurat (v4.0.2) [81]. Dimensionality reduction and clustering were then conducted in Seurat following standard procedures. To facilitate a comprehensive understanding of the analytical strategies employed in this study, we provide a schematic overview of the entire computational workflow (Supplementary Fig. S8). This diagram illustrates the sequential processing pipeline, beginning with raw data input (h5ad and imaging files) and preprocessing via the GEM3D toolkit, followed by parallel analytical modules including WACCA for 3D reconstruction, SPC analysis for spatial clustering, and polarity analysis for modeling morphogenetic gradients.

Data QC and validation

To ensure the technical reliability, sensitivity, and reproducibility of our spatial transcriptomic atlas, we systematically evaluated key QC metrics across all biological samples, tissue sections, and annotated spatial domains (Supplementary Table S1). Following single-cell segmentation, we quantified the segmented cell area, the number of effective spatial spots (nDNB), total UMIs, and the number of detected genes per cell. Analysis of all 35 tissue sections demonstrated high technical consistency across the dataset. The median number of detected genes per cell across sections ranged from ~150 to 230, with median UMI counts ranging from 190 to 370.

To rule out potential technical biases that might favor highly transcriptionally active cells over cells with lower RNA content, we statistically evaluated the capture efficiency across all 36 identified spatial domains. Our analysis revealed no meaningful systematic bias toward specific lineages. The distributions of cell area, detected genes, and UMI counts remained broadly consistent and biologically appropriate across the diverse cellular populations. Robust transcript detection was achieved globally, ranging from large differentiated lineages to smaller undifferentiated neoblasts and progenitor states. This uniform data quality confirms that our Stereo-seq approach provides sufficient resolution and sensitivity to capture transcription factors and resolve heterogeneous cell states without significant transcript dropout for low-abundance populations.

Furthermore, to validate the quantitative accuracy of our Stereo-seq dataset at the global tissue level, we assessed its concordance with traditional bulk RNA-seq data. Single-cell spatial expression profiles from Stereo-seq sections were aggregated to generate pseudo-bulk transcriptomes. We then calculated the Pearson correlation coefficients between these pseudo-bulk profiles and corresponding bulk RNA-seq datasets of regenerating planarians. The high correlation observed (Supplementary Fig. S1C) verified the technical reproducibility of our platform and confirmed the absence of significant transcript dropout or amplification bias during in situ capturing and library preparation.

For the visualization of spatial gene expression patterns in Fig. 2A, a representative sample was selected for each time point from the biological replicates (n = 2). This selection was determined by calculating the mean gene expression vector (centroid) for each time point and identifying the replicate with the highest Pearson correlation to this centroid. All quantitative analyses and statistical tests were performed using the full integrated dataset across all replicates.

Correlation analysis with single-cell and spatial atlases

To characterize the cellular composition of the 36 identified spatial domains, we performed a Pearson correlation analysis comparing the expression profiles of our clusters against annotated cell types from 3 independent single-cell RNA-seq atlases [23, 24, 82] and spatial domains from a 10x Genomics Visium dataset [25]. This analysis categorized domains into High Fidelity (1-to-1 mapping), Lineage Restricted (mapping to a single tissue class), or Mixed Domains (containing signatures from multiple cell types, such as the neoblast-parenchyma niche).

Blastema region detection in 3D ST data

The blastema regions were identified based on pigmentation patterns. The boundary between the unpigmented blastema and pigmented trunk was defined using the Threshold function in ImageJ and refined via quadratic polynomial regression [83]. To spatially resolve the interface between pre-existing and newly formed tissue, we defined a border zone extending 20 µm on either side of this regression line. This subdivision created distinct spatial domains at the anterior wound: the head anterior region (HAR), head border outer region (HBOR), and head border inner region (HBIR). An analogous approach applied to the posterior wound defined the TPR, TBOR, and TBIR domains. For regional analyses, the 3 anterior (HAR, HBOR, HBIR) and posterior (TPR, TBOR, TBIR) compartments were collectively designated as the head and tail blastema regions, corresponding to their distal, middle, and proximal subdivisions, respectively. Cells were assigned to these discrete regions based on their spatial coordinates subsequent to the alignment of the transcriptomic data with the microscopy images utilizing TrakEM2.

Identification of temporally DEGs

Temporal alterations in gene expression were analyzed by comparing adjacent time points for each defined SPC cluster and anatomical region (e.g., blastema and trunk regions) using the DEsingle algorithm. Specifically, expression profiles corresponding to a specific cluster or region were extracted from the integrated dataset. For each pair of adjacent time points (e.g., 0 hpa vs. 12 hpa, 12 hpa vs. 36 hpa, etc.), we employed the DEsingle R package [84] to detect DEGs. Following the primary analysis, we utilized the DEtype function within the DEsingle package to classify the identified DEGs into distinct categories based on variations in gene expression abundance and distribution. To ensure statistical rigor, raw P-values were adjusted using the Benjamini–Hochberg procedure, and genes with a false discovery rate < 0.05 and absolute log10(Fold change) > 0.5 were retained as significant DEGs (Fig. 1I, Supplementary Table S2).

Identification of SBGs

To quantitatively examine spatial gene expression patterns, we established a molecular coordinate system by dividing the straightened planarian body into bins along the A/P (100 bins), M/L (40 bins), and D/V (14 bins) axes. SCT-transformed expression values for highly variable genes (HVGs) and known position control genes (PCGs) were averaged per bin, normalized by cell density, scaled, and smoothed using a Gaussian filter (sigma = 3). Genes expressed in fewer than 5 consecutive bins were excluded from downstream analysis. We then applied a hierarchical density-based clustering algorithm to the homeostatic dataset to aggregate genes with similar spatial profiles along the body axes. To refine the clustering, parameters were optimized for spatial distinctness, and unassigned genes were assigned to the most probable groups using linear regression. The resulting clusters and their constituent genes are listed in Supplementary Table S3.

Spatial pattern analysis of regenerating animals

To investigate the spatiotemporal dynamics of positional information during regeneration, we applied the aforementioned coordinate binning strategy to the regenerating samples at each time point. Specifically, the expression profiles of the identified SBGs were mapped onto the A/P, M/L, and D/V axes of the regenerating fragments. The regenerated animals were then divided into 100 bins along the A/P axis, in line with the homeostatic reference, and gene expression was categorized for each sample separately. These cluster IDs function as spatial coordinates rather than static gene lists, allowing us to quantify the physical restoration of morphogenetic gradients over time. By tracking the spatial distribution of these gene clusters over time and comparing them to their homeostatic baselines, we visualized and quantified the restoration of axial polarity and regional patterning across the regeneration process.

Application of Turing pattern models to SBGs

We hypothesize that the interactions between SBGs and their upstream regulators adhere to Turing patterns within an autoregulatory activator–inhibitor framework. Following the removal of the influence of spatial gradients, temporal gene expression data across 8 regenerative time points were normalized, interpolated, and smoothed. Reaction, degradation, and diffusion parameters were fine-tuned using linear regression. These optimized parameters allowed us to predict gene expression levels at any post-amputation time point for both established and candidate PCGs. Pearson’s correlation coefficients were then calculated to assess the accuracy of these predictions.

PCA of SBGs and PCGs

To integrate temporal changes in gene expression with spatial variations, PCA was applied to the binned expression data of selected PCGs along the 3 axes in homeostatic animals [29]. For the A/P axis, the training set comprised 25 known PCGs. The first principal component (PC1), which accounted for 64.7% of the variance, corresponded to the head-tail gradient, while the second principal component (PC2), which explained 24.7% of the variance, captured fluctuations in the pharyngeal region (convex and concave). Based on their locations in the reduced-dimensional space, genes were manually grouped into 5 categories: head, head-pharynx, trunk-pharynx, pharynx-tail, and tail domains.

Given the paucity of previously identified PCGs with clear spatial patterns along the M/L and D/V axes in our Stereo-seq data, we expanded the training sets to include 42 and 67 newly inferred PCGs, respectively. Potential M/L PCGs candidates were selected based on Spearman’s rank correlation coefficients greater than 0.7 or less than −0.7 for binned SCT-transformed gene expression data, showing patterns similar to known PCGs. For potential D/V PCGs, a fold change greater than 1.5 between dorsal and ventral regions in binned SCT-transformed values served as a selection criterion.

PCA was performed using the Scikit-learn package with default parameters. The eigenvectors derived from this analysis were used to map both established and potential PCGs back into the reduced PCA space, trained on the homeostatic data. The regenerative trajectories of PCGs were visualized, resembling the behavior of an underdamped mass-spring system: the initial amputation stretched the “spring,” disrupting the PCG expression profile, while the GRN provided restorative feedback to re-establish homeostasis [85].

UV unwrapping for the head blastema epidermal region

To convert the 3D planarian body shell into a 2D plane, we employed the UV unwrapping technique, a common procedure in the field of computer graphics, using the open-source 3D creation suite Blender. UV unwrapping is a process in which the surface of a 3D model is mathematically unwrapped and mapped onto a 2D plane, enabling the precise application of textures and structures onto a flat surface, which is essential for accurate visualization and analysis. The process of UV unwrapping for the head blastema epidermis involved segmenting the planarian body, marking seams for accurate texture alignment, unwrapping the mesh, and mapping the epidermal cells onto a 2D plane. This approach enabled a high-precision representation of the head blastema’s epidermal region, which will be useful for further texture analysis and studies related to planarian regeneration [86].

Briefly, the planarian body mesh was first segmented into 2 sections along a defined plane located near the boundary of the head blastema. This segmentation was performed using the Bisect Tool within Blender, which allowed for a clean division of the mesh without distorting the geometry. The separation enabled us to isolate the head blastema area, which we intended to unwrap for detailed analysis.

Next, to prepare for the unwrapping process, seams were strategically marked to guide the unfolding of the 3D mesh. The edges connecting the blastema cutting plane to the anterior pole of the planarian were designated as a seam, specifically placed along the D/V boundary. This step ensured that the unwrapping process adheres to natural anatomical divisions, preventing distortion of the texture in the subsequent 2D plane. Seam marking was carried out using the Blender’s UV editor tool, where the mesh’s geometry was manipulated to set boundaries for the unwrapping operation.

Once the seams were defined, the head blastema mesh was subjected to the UV unwrapping operation. The mesh was unfolded into a 2D plane, with particular attention paid to the correct alignment of the marked seam. The seam was clipped and adjusted to create a smooth, curved incision, representing the head blastema epidermal cells in a flat space. This incision helped ensure that the texture mapping would preserve the anatomical integrity of the original 3D model.

Finally, we focused on the outer epidermal cells of the head blastema, which were part of the original 3D point cloud data. These epidermal cells were mapped onto the unwrapped 2D plane using a process that minimized the distance between the original 3D coordinates of each epidermal cell and the vertices of the subdivided mesh surface. This step ensured a precise mapping of cellular structures to the 2D plane, allowing for high fidelity in representing the epidermal region’s texture and topology. The minimization of these distances preserved spatial relationships and accurately represented the cellular organization in a flattened format.

Monocle3 analysis

To investigate the state transitions of the ARZ (Clu.31 cluster) during regeneration, we analyzed the trajectory dynamics using raw transcriptome counts from 36 hpa to 14 dpa. The raw counts were first normalized using SCTransform in Seurat (v4.0.2). Dimensionality reduction and clustering were then performed in Monocle3 (v1.3.1) [46]. The trajectory graph was constructed by fitting the principal graph with the learn_graph function, and pseudotime was calculated with the neoblast cell type set as the root. Marker genes for the ARZ (Clu.31 cluster) were identified using Seurat, and their expression was projected onto the trajectory branches. Cells were clustered in a 15-NN graph, which allowed us to divide the trajectory into distinct branches enriched for epidermal, muscle, and neuronal signatures.

RNA velocity analysis

RNA velocity analysis, based on Waddington’s epigenetic landscape and differential geometry, was used to make continuous, time-resolved predictions of cell state transitions. Cellular genes were aligned to the reference genome to identify exon and intron sequences. The relative abundance of spliced (mature) and unspliced (nascent) mRNAs was calculated to estimate splicing and degradation rates using Velocyto [48]. Each DNB was assigned to its corresponding cell based on its x and y coordinates. Spliced and unspliced count matrices for different domains were processed using the recipe_monocle function in Dynamo [87] to identify highly expressed genes. Following dimensionality reduction, the continuous velocity vector field was reconstructed in UMAP space to predict future cell fates.

Monocle2 analysis

Monocle2 (v2.18.0) was used to analyze the ventral epidermal trajectory across different regeneration time points, following the tutorial [88]. We focused on extracting SPC clusters during the putative ventral epidermal transition from the blastema region. DEGs for each cell type were identified and used to order the cells. Dimensionality reduction was carried out using the DDRTree method, and the plot_cell_trajectory function was used for visualization. Marker genes identified by Seurat’s FindAllMarkers function were projected along the estimated pseudotime to assess their potential role in the transition.

Visual representation of the PRISTA4D interactive spatiotemporal transcriptomic atlas database

To enhance the accessibility and utility of our PRISTA4D (Planarian Regenerative Interactive Spatiotemporal Transcriptomic Atlas in Four Dimensions) for researchers in the field of regeneration, we developed an open-source, interactive database, PRISTA4D. This platform enables users to explore the spatial distribution and dynamic changes of various genes and cells at different stages of regeneration.

The PRISTA4D database provides several functions, including browsing capabilities and access to experimental procedures, data analysis pipelines, and the ability to download the original dataset. It serves as a resource for studying cell differentiation and spatiotemporal cell interactions within the regeneration research community.

The website includes a homepage and 5 key functional modules:

3D model: visualizes different domains in 3 dimensions, allowing users to view cellular organization across regeneration stages.

Spatial clustering module: illustrates the distribution of genes and domains, offering insights into spatial gene expression patterns.

Stereo-seq module: provides detailed experimental protocols used to generate the transcriptomic data, ensuring transparency and reproducibility.

Sampling design module: offers information on sample design and the data analysis pipelines used, enabling users to understand how the data were processed and analyzed.

Download module: grants access to the complete original dataset, allowing users to download the raw data for further analysis and research.

Cell sorting and library construction for scRNA-seq

To prepare the cell suspension for scRNA-seq, CMFB buffer (CMF + 1% FBS) was placed on a cold plate (4°C), and the animals were incubated in this solution before their tissues were manually chopped to release cells as described previously [9]. After dissociation, the cells were pelleted by centrifugation at 290 × g for 5 min at 4°C. The resulting suspension was passed through a 40 μm filter and stained with DAPI (1:1000; Beyotime) and DRAQ5 (1:1000; BioLegend). The cells were washed and resuspended in CMFB buffer. Flow cytometry and sorting were performed using a Sony MA900 cell sorter, with the temperature maintained at 4°C to preserve cell integrity. Approximately 20,000 viable cells (DAPI−; DRAQ5+) were loaded onto the SeekGene platform using the Single Cell 3′ Transcriptome kit to generate scRNA-seq libraries. Library preparation followed the manufacturer’s guidelines to ensure optimal coverage, and sequencing was conducted on an Illumina NovaSeq platform with paired-end 150 base pair (150 PE) reads for detailed transcriptomic profiling.

Analysis of scRNA-seq data after RNAi knockdown

Raw scRNA-seq data were processed and aligned with the S. mediterranea reference transcriptome (smed_20140614). Cells containing fewer than 200 detected features or genes expressed in fewer than 3 cells were excluded. Cells exhibiting unusually high mitochondrial or ribosomal RNA content (where mitochondrial percentage was more than twice the median) were also filtered out based on the annotated reads (Supplementary Table S6). After QC, the remaining UMIs were quantified and analyzed using the Seurat package in R, which enabled normalization, scaling, and dimensionality reduction of the data. PCA was performed, and the top 30 principal components were used for 2D UMAP generation and clustering within Seurat. Cell lineages were annotated based on the expression of known marker genes, while SPC clusters were identified using markers derived from 3D ST. Batch effects due to technical variations across samples were corrected using Seurat’s integration functions, ensuring that the observed differences were biologically relevant [89]. For analyzing cellular differentiation trajectories and lineage relationships, Monocle 2 was applied. Additionally, MiloR was used to examine and compare the abundance of cells in specific neighborhoods or microenvironments between the control and RNAi-treated groups [90]. Briefly, PCA was used for dimensionality reduction, and a KNN graph was constructed with the buildGraph function (k = 30, d = 30) based on the top 30 PCA dimensions. Neighborhoods were defined using the makeNhoods function (prop = 0.1, k = 30, d = 30), and differential abundance testing was conducted with default parameters using the distinct function. Differentially abundant cell populations between control and knockdown groups were identified, and DEGs were determined using DESingle (v1.9.2) with standard settings [84]. Genes with an adjusted P-value of less than 0.05 were considered significantly different. This comparative analysis revealed how RNAi interventions altered the cellular composition within the tissue.

Availability of source code and requirements

Project name: 4D-BioReconX

Project home page: https://github.com/BGI-Qingdao/4D-BioReconX

Resource portal: https://db.cngb.org/stomics/prista4d

License: MIT License

Operating system(s): Linux or MacOS

Programming language: Jupyter Notebook, Python, R, Shell

Other requirements: anndata≥0.7.5, matplotlib≥3.6.2, numpy≥1.22.4, opencv-python≥4.6.0.66, pandas≥1.4.3, scikit-image≥0.19.2, scipy≥1.9.0, seaborn≥0.11.2

RRID:SCR_027919

bio.tools ID: 4d-bioreconx

WorkflowHub: 10.48546/workflowhub.workflow.2045.1

Additional files

Supplementary Figure S1: High-resolution 4-dimensional spatial and molecular characterization of planarian regeneration. (A) Consecutive Stereo-seq sections were utilized for 3-dimensional reconstruction across 8 regenerative time points. Two animals were sampled per time point, with each row representing an individual animal. Sections display the spatial distribution of cell types along the ventral-to-dorsal axis from left to right. Cells are color-coded according to lineage annotations as detailed in Fig. 1B. hpa, hours post-amputation; dpa, days post-amputation. (B) Three-dimensional spatial visualization of 36 spatial transcriptomic clusters (top row) and tissue meshes (bottom row) across identical regenerative time points. Cells are color-coded based on cluster annotations. Organisms are oriented with the anterior upwards in a dorsal view. (C) Correlation analysis between Stereo-seq pseudo-bulk data during homeostasis and 3 independent bulk RNA-seq replicates [30]. Pearson correlation coefficients are indicated for each replicate. (D) Overview of the interactive Planarian Regenerative Interactive Spatiotemporal Transcriptomic Atlas in Four Dimensions (PRISTA4D) database (https://db.cngb.org/stomics/prista4d/). The homepage and 5 principal functional modules are highlighted. The 3-dimensional model module visualizes cell types, while the spatial clustering module illustrates gene and cell type distributions. The Stereo-seq module details experimental protocols, the sampling design module provides information on experimental design and analytical pipelines, and the download module ensures access to the complete original dataset.

Supplementary Figure S2: Consistency assessment of cell type annotations across independent datasets. Heatmaps illustrate the correlation-based correspondence between cell type clusters identified in the current study using Stereo-seq (x-axis) and 4 previously published reference datasets (y-axis). Comparisons are detailed between the present dataset and those generated by (A) Emili et al. (GSE246681), (B) Plass et al. (GSE103633), (C) Fincher et al. (GSE111764), and (D) Cui et al. (10x Visium). The color gradient reflects the correlation coefficient, with red denoting high correlation and blue denoting low correlation. Robust diagonal signals validate the reliability of the cell type nomenclature and functional annotations applied in this study against established planarian spatial atlases.

Supplementary Figure S3: Characterization and spatial visualization of cellular populations across regenerative time points. (A) Spatial visualization of 18 spatial transcriptomic clusters across 8 regenerative time points, encompassing neoblasts, the Clu.31 regenerative domain, muscle lineages (2 subtypes), parenchymal cells (11 subtypes), and pharyngeal cells (3 subtypes). (B) Spatial visualization of 18 spatial transcriptomic clusters across 8 regenerative time points, comprising epidermal cells (6 subtypes), neural lineages (4 subtypes), intestinal cells (5 subtypes), cathepsin-expressing cells (2 subtypes), and protonephridial lineages. (C) Bubble plots delineating proportional changes in cluster cell ratios between uninjured states and the 8 regenerative time points. Clusters are categorized by population dynamics, with overall trends denoted by arrows. Corresponding cluster identifiers are provided on the right. (D) Three-dimensional projections of spatially categorized clusters across the 8 regenerative time points. Cluster groupings correspond to the respective bubble plots. Scale bar, 500 µm. (E) Schematic representation of the head blastema (purple), trunk (brown), and tail blastema (white) domains in regenerating specimens. Blastema boundaries are defined based on the absence of pigmentation. Scale bar, 500 µm. (F) Bubble plot quantifying the temporal shifts in cluster proportions across 8 regenerative time points, specifically within the trunk, head blastema, and tail blastema compartments.

Supplementary Figure S4: Self-organized morphogenetic gradient formation during polarity regeneration. (A) Principal component analysis of gene expression patterns for anterior–posterior, medio–lateral, and dorso–ventral position control genes or candidates in intact animals. The proportion of variance explained by the principal components is detailed for each respective anatomical axis. (B) Exponential function fitting plots of spatial expression patterns for anterior-enriched, pharynx-enriched (incorporating an optional Gaussian fit), and posterior-enriched position control genes in intact planarians. (C) Heatmaps detailing the dynamic expression levels of specified genes localized to the anterior or posterior body extremities (defined as one-tenth of the total body length) throughout regeneration. Regionally enriched genes exhibiting analogous temporal patterns are organized via hierarchical clustering. (D) Spatiotemporal expression dynamics of the posterior position control gene wnt1 in regenerating planarians. The left panel demonstrates the relative expression distribution of wnt1 along the anterior–posterior axis at 36 hpa, 3 dpa, and 5 dpa. Biological replicates were partitioned into 10 longitudinal bins, with average expression levels calculated per bin. The emerging peaks at the posterior terminus confirm the correct polarization of wnt1 expression during tail regeneration. The right panel provides spatial visualization of wnt1 expression mapped onto the morphological contour of representative individuals at 3 and 5 dpa. The expression remains strictly confined to the posterior midline and tail tip, corroborating the established regulatory role of wnt1 in posterior polarity determination.

Supplementary Figure S5: Composition and molecular signatures of the Clu.31 domain. (A) Molecular characterization of the Clu.31 regenerative domain. The panel features a bubble plot illustrating the expression of the established wound epidermis marker equinox alongside the top 3 Clu.31-enriched marker genes across all transcriptomic clusters. Additionally, whole-mount in situ hybridization images localize 3 Clu.31 markers in homeostatic animals, accompanied by double fluorescence in situ hybridization of a representative Clu.31 marker (smed03831) mapped against equinox. Scale bars, 500 μm. n ≥ 3. (B) Functional enrichment analysis of Clu.31-enriched transcripts in homeostatic animals. (C) Spatial visualization of 3 constituent lineages residing within the Clu.31 domain at the head blastema across distinct regenerative time points. (D) Heatmap profiling enriched gene expression along the inferred Clu.31 developmental trajectory. Functional enrichment terms corresponding to genes associated with each pseudotime branch are detailed on the right. (E) Dot plot tracking the temporal expression profile of agat-1 throughout the regenerative process.

Supplementary Figure S6: Transcriptomic characterization of the Clu.31 domain and expression profiling of med8. (A) RNA velocity streamline plot depicting the predicted cellular transition trajectories at 36 hours post-amputation, coinciding with the initial emergence of the Clu.31 domain. Directional arrows indicate that Clu.31 predominantly originates from the ventral epidermal progenitor population (Cluster 1). The inset displays the inferred pseudotime progression, represented by a sequential color scale. (B) Heatmap comparing the relative expression of epidermal signature genes alongside dorsal (bmp4) and ventral (admp) positional markers within Clu.31 and selected epidermal clusters. (C) Phylogenetic tree of Mediator 8 family proteins, utilizing PIWI-1 as the defined outgroup. (D) Heatmap quantifying med8 expression across distinct cell populations, utilizing reference data from Benham-Pyle et al. (E) Violin plot detailing med8 expression variance across diverse neoblast subpopulations, referencing data from Zeng et al. (F) Whole-mount in situ hybridization demonstrating broad med8 expression and its spatial co-expression with the neoblast markers piwi-1 and tgs-1. Consistent expression patterns were observed across all biological replicates (n ≥ 6). Scale bar, 500 µm. (G) Quantitative real-time PCR assessment demonstrating med8 knockdown efficiency following RNA interference.

Supplementary Figure S7: The med8-regulated Clu.31 domain modulates cellular production during blastema formation. (A) Pseudotime trajectory analysis of single-cell RNA sequencing data revealing impaired differentiation of Nb2 neoblasts into subsequent cell lineages in med8 RNAi animals. Altered lineage trajectories are highlighted by dashed lines. (B) Bar plot comparing the proportional composition of lineage components within the Clu.31 domain between control and med8 RNAi specimens. (C) Violin plots quantifying the signature expression scores of anterior and posterior polarity genes. Statistical significance was determined using the Wilcoxon rank-sum test (ns, not significant; *, P < 0.05; ****, P < 0.0001). (D) Comprehensive differentiation analysis incorporating Monocle2 pseudotime trajectories and miloR differential cell abundance mapping for predicted transitions from Nb2 neoblasts toward neural, epidermal, and muscle lineages. Corresponding scatter plots depict the relative expression dynamics of tcf/lef-1, p53, and dmrt2 along these inferred developmental paths. (E) Volcano plot identifying differentially expressed genes within the Nb2 neoblast population following med8 knockdown, accompanied by representative Gene Ontology enrichment terms for the significantly downregulated gene set. (F) Bar plot quantifying the reduction in differentiation efficiency across epidermal, muscle, and neural progenitor lineages. (G) Representative morphological phenotypes of med8 RNAi animals following 4 or 6 consecutive feedings, assessed 3 days post-final feeding. Scale bars, 500 µm.

Supplementary Figure S8: Schematic overview of the computational workflow. The diagram delineates the sequential data processing pipeline, commencing with raw data input from h5ad and imaging files. This is followed by spatial preprocessing via the GEM3D toolkit and branches into parallel analytical modules. These subsequent modules include the WACCA algorithm for robust 3-dimensional reconstruction, spatial transcriptomic analysis for cellular clustering, and positional analysis for modeling dynamic morphogenetic gradients.

data S1-20,260,422.xlsx

data S2.xlsx

data S3.xlsx

data S4.xlsx

data S5.xlsx

data S6.xlsx

List of abbreviations

A/P: anterior/posterior; ARZ: anterior regenerative zone; D/V: dorsal/ventral; DEG: differentially expressed gene; dpa: days post-amputation; FISH: fluorescent in situ hybridization; GRN: gene regulatory network; hpa: hours post-amputation; HVG: highly variable gene; M/L: medial/lateral; PCA: principal component analysis; PCG: positional control gene; RNAi: RNA interference; SBG: spatially biased gene; scRNA-seq: single-cell RNA sequencing; SPC: spatial proximity-based clustering; ST: spatial transcriptomics; TF: transcription factor; UMI: unique molecular identifier; WISH: whole-mount in situ hybridization.

Supplementary Material

giag064_Reviewer_1_Report_original_submission

Reviewer 1 -- 11/20/2025

giag064_Reviewer_2_Report_original_submission

Reviewer 2 -- 11/26/2025

giag064_Reviewer_2_Report_revision_1

Reviewer 2 -- 3/23/2026

giag064_Reviewer_3_Report_original_submission

Reviewer 3 -- 11/30/2025

giag064_Reviewer_3_Report_revision_1

Reviewer 3 -- 3/18/2026

Author contributions

A.Z., X.X., G.F., H.L., K.H., and M.X. conceived and directed the study. A.Z., X.X., M.X., and H.L. supervised the work. Y.C. and Y.R.L. performed animal experiments and RNAi. X.W.L., W.G., J.Q.W., W.W., and H.P. performed the Stereo-seq experiments. K.H., M.X., L.G., Y.L., Y.W., and Z.H. analyzed the data. L.G., Y.L., and T.Y. performed database construction. Q.L., L.Z., and X.M. assisted in the data analysis. R.Z., L.L., X.W., H.Z., X.S., S.L., W.Z., S.T.C, J.F., X.L., Y.G., J.W., G.P., and H.Y. performed investigations. A.Z., G.F., H.L., M.X., and K.H. wrote the manuscript with input from all authors.

Funding

This research was supported by the National Key R&D Program of China (2022YFC3400400 to X.X.), the National Natural Science Foundation of China (32570997 and 32070828 to A.Z.), Shenzhen Science and Technology Program (JCYJ20250604191305008 to M.X., SYSPG20241211173844005 to G.F., and RCJC20221008092804002 to Y.G.), the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB1030000 to A.Z.), and the Shanghai Science and Technology Committee (STCSM) (YDZX20243100004003 to A.Z.).

Data availability

All data generated in this study were deposited in the CNGB Nucleotide Sequence Archive (accession code: STT0000028). The accession number for Rod1 (SMED30003831) is OR211556. Processed data and 3D models can be interactively explored via our PRISTA4D database (https://db.cngb.org/stomics/prista4d). All original code supporting the current study is hosted on GitHub [91] and WorkflowHub [92].

Competing interests

Prof. Xun Xu serves as an Editor-in-Chief for GigaScience. He was not involved in the review or decision-making process for this manuscript. The article was handled in accordance with the journal’s standard procedures, with peer review conducted independently of Prof. Xu.