Human trunk embryoids with patterned anterior-posterior and dorsal-ventral body axes: utility for understanding human development and disease

pluripotent stem cells, embryoid, bipotent NMPs, notochord, A-P and D-V 26 axes, human trunk development 27 28

29 Understanding the molecular and cellular aspects of peri- and post-implantation human 30 embryogenesis are fundamental for a better understanding of congenital disease, the 31 application of human pluripotent stem cell (hPSC) towards regenerative medicine and 32 potentially, drug validation. Developing hPSC-derived embryoid models offers 33 opportunities to address these issues.1-3 Although success has been achieved in 34 developing mouse embryoids that exhibit multi-axial and multi-tissue patterning along 35 the anterior-posterior (A-P), dorsal-ventral (D-V) and left-right (L-R) body axes,4-6 36 attempts to generate multi-axial human trunk models suffer from several limitations. 37 Initial human trunk models using bipotent neuromesoderm progenitors (bi-NMPs) are 38 uni-axial and comprised of somite-only7-9 or neural tube-only10 structures. Recently 39 developed coupled models11-13 are composed of a neural tube and somites but, are 40 often aberrantly structured (e.g. unilateral somites and twisted neural tube) and fail to 41 establish a D-V body axis. Notably, these models lack a notochord and are heavily 42 biased towards dorsal patterning.12,13 In contrast, trunk models that form a notochord-43 like structure are ventrally-biased and lack a neural tube and somites.14 These 44

highlight the need for a human model that supports co-development of the 45 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 3 notochord alongside bi-NMPs, together with the establishment of the A-P and D-V body 46 axes. 47 Temporal signaling by WNT, NODAL, BMP, retinoic acid (RA), FGF and SHH specifies 48 tailbud bi-NMPs and notochord progenitor cells along the A-P and D-V axes.14-18 A 49 major challenge is to establish a concerted signaling environment that supports these 50 co-developmental events. We iteratively optimized conditions to generate human trunk 51 embryoid models (hTEMs) from bi-NMPs and notochord progenitors. hTEMs self-52 organize into notochord, dorsal neural tube, floor plate, and bilateral somites. Validation 53 of this was based on morphological, cellular and molecular criteria using spatial and 54 single-cell transcriptomics, live imaging and scanning electron microscopy (SEM). 55 Comparative analyses showed that hTEMs recapitulate posterior trunk development 56 equivalent to that in human embryos at Carnegie stages (CS) 8-10. 57 The utility of hTEMs for modeling human development was investigated by perturbing 58 notochord identity and notochord-derived SHH activity. By this approach, the notochord 59 was shown to be a major driving force for D-V axis establishment in the human trunk. 60 Moreover, hTEMs faithfully reproduced human neural tube defects (NTDs), as shown by 61 loss-of-function mutations in the planar cell polarity (PCP) genes, VANGL1 and 62 VANGL2. These findings highlight the versatility of trunk embryoids for understanding 63 human development and congenital disease and points towards additional applications 64 for drug discovery and regenerative medicine. 65 66

and Discussion 67 Dorsally biased neural tube-somite coupled embryoids 68 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 4 We first aimed to generate topographically coupled embryoids with A-P segmented 69 somites flanking an elongating neural tube (hTEM.v1; Figures 1A and 1B). To achieve 70 this, size-controlled 3D human embryonic stem cell (hESC) spheroids received WNT 71 agonist CHIR99021 (CHIR), FGF2, and inhibitors for BMP (LDN-193189 (LDN)) and 72 NODAL (SB-431542 (SB)) pathways. Resultantly, hTEM.v1 embryoids became 73 asymmetric and by day4, underwent mediolateral narrowing and axial elongation 74 (Figures 1B and S1A). Matrigel and retinal (RAL) promoted axial elongation and somite 75 segmentations at days 4-7 (Figures 1C, S1B, and S1C).8 6-8 pairs of bilaterally 76 positioned somites progressively formed along with gradual extension and closure of 77 the neural tube (Figures 1D, S1D, and Video S1). By day 7, hTEM.v1 reached a mean 78 length of 1857 μm (Figure 1C). 79 Immunostaining of hTEM.v1 (days 2-4) revealed the following features indicative of A-80 P patterning; (i) polarized expression of TBXT and CDX2 at the posterior end, (ii) 81 mutually exclusive positioning of central neural (SOX2) and mediolateral pre-somitic 82 mesoderm (PSM; TBX6) cells; (iii) A-P symmetry breaking, indicated by anterior somitic 83 (SIX1) cells undergoing an epithelial-to-mesenchymal transition (EMT; N-cadherin) and 84 caudalized tailbud (CDX2) cells (Figures 1E, S1E-S1G). In total, these characteristics 85 are reminiscent of the highly mitotic (pH3-Ser10), A-P patterned human CS8 embryo 86 (Figure 1E).19,20 Upon further inspection of hTEM.v1 (days 4-7), SOX2+TBXT- neural 87 tubes were observed on both dorsal and ventral sides along the A-P axis (Figures 1F-88 1H and S1H). This is coincident with the absence of a SOX2-TBXT+ notochord 89 structure (Figures 1E and S1I). This suggests that although the A-P axis had formed, D-90 V patterning in hTEM.v1 was not established. 91 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 5 To understand D-V patterning defects in hTEM.v1, single-cell RNA sequencing 92 (scRNA-seq) analysis was performed (days 4/5/7). Clustering and hierarchical 93 transcriptomic profiling identified 12 major cell identities, including two major continuums 94 delineating somitogenesis and neural tube formation (Figures 1I, 1J, S1J, and S1K); (i) 95 the somitic lineage contains NMP-Meso (TBXT, MSGN1), posterior PSM (HES7, TBX6), 96 anterior PSM (MESP2, RIPPLY2) and pan-somite cells (PAX3, SIX1, comprised of 97 early-, middle- and late-somite (E-, M- and L-somite) subtypes along the time course; (ii) 98 the neural lineage includes NMP-Neural (SOX2, NKX1-2), caudal neural plate 99 progenitors (Caud. NP; MSX2, SOX2, NKX1-2) and neural tube cells (PAX6, HES5). 100 Overall, due to the absence of ventral somite compartments (PAX1, PAX9) or ventral 101 neural tube (NKX6-1) and floor plate (FOXA2), the spatiotemporal expression profile of 102 hTEM.v1 confirmed dorsally-biased cell identities in both the somitic and neural 103 lineages (Figure 1K). Immunostaining of transversely sectioned day-7 hTEM.v1 further 104 confirmed the unrestricted distribution of dorsal BMP4/7 signals and over-expansion of 105 dorsal neural (PAX6) cells, with few to no ventral neural (NKX6-1) cells in the neural 106 tube region (Figure 1L). 107 Conventionally, bi-NMPs are identified by the co-expression of SOX2 and TBXT,21-23 108 but questions about their molecular identities remain unresolved.24-26 Here, we identified 109 distinct NMP-Neural (SOX2high,TBXTlow) and NMP-Meso (SOX2low,TBXThigh) subtypes 110 that exhibited contrasting expression patterns (Figures 1J and 1M). Additionally, they 111 expressed differential levels of FGF3/4/8/17, WNT3A/5A/5B/8A, and BM2/4/7 (Figure 112 S1L), which are believed to be the driving force for bi-NMP fate bifurcations in human 113 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 6 trunk formation.27 This suggests hTEM.v1’s potential to resolve the plasticity of bi-NMP 114 bifurcation. 115 Then, we asked why notochord was absent from hTEM.v1. scRNA-seq combined with 116 in situ hybridization chain reaction (HCR) only detected few cells expressing notochord 117 markers (TBXT, NOTO, CHRD, SHH), possibly representing the ventral node (SHH) 118 surrounded by PSM (TBX6, HES7) (Figures 1J, S1M, and S1N).28 In human embryos, 119 correct D-V axis specification is initiated by SOX2high/TBXTlow in the dorsal neural plate 120 and SOX2low/TBXThigh in the ventral notochordal plate.20,27 However, in day-4 hTEM.v1, 121 TBXT protein was restricted to the tailbud and was undetectable by day 7, while SOX2 122 expression extended to both dorsal and ventral tube structures (Figures 1E, 1G, and 123 S1I). The SOX2 over-expansion and neural tube duplication phenotype in hTEM.v1 124 coincides with previously reported NTDs caused by mutations/misregulation of Tbxt in 125 mouse notochord progenitors.29,30 We, therefore, attributed the neural tube duplication 126 from hTEM.v1 to the misregulation of TBXT and consequently, the inability to specify 127 notochord cells. 128 Overall, bi-NMP-derived hTEM.v1 with spatially coupled neural tube structures and 129 flanking somite segments were generated. Notochord is well-known for its role in D-V 130 patterning of neural tube and somitic cells in the embryo.31-33 The absence of a 131 notochord and SHH signaling along the A-P axis in hTEM.v1 explains the heavy 132 dorsalization observed. Moreover, it is noteworthy that we observed a subset of somitic 133 cells resembling the endotome (EBF2) (Figure 1J), a poorly understood somite 134 compartment that contributes to the dorsal aorta, endothelium and hematopoietic stem 135 cells.13,34,35 The anteriorly positioned EBF2:mScarlet signal coincided with vascular 136 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 7 endothelial cells (SOX17, VE-cadherin) at the somite periphery (Figure S1O and Video 137 S1). These findings indicate that hTEM-based embryoids have utility for studying more 138 advanced events related to later stages of gastrulation in humans, such as 139 somitogenesis, neural tube morphogenesis and somite-derived vasculogenesis.27 140 141 Exogenous SHH activation induces ventral fates in trunk embryoids 142 As notochord-derived SHH signaling is key to ventral patterning,33 we postulated that 143 SHH activation could establish a D-V axis and rescue the dorsal-biased defects in 144 hTEM.v1. To test this, day-4 hTEM.v1 embryoids were exposed to varied durations and 145 concentrations of Smoothened agonist (SAG) (Figure S2A). SAG-treated hTEM.v1 146 exhibited elevated transcripts (e.g., ventral neural tube; NKX6-1. Sclerotome; PAX1/9) 147 of ventral identities by day 7, in a dose- and time-dependent manner (Figure S2B). A 148 pulse of 100 nM SAG was chosen to generate hTEM.v2, because this condition 149 established D-V patterning in neural and somitic cells without affecting morphology or 150 axial elongation (Figures 2A-2D and S2B). Immunostaining of day-7 hTEM.v2 confirmed 151 the emergence of ventral (NKX6-1) neural cells in both neural tubes along the A-P axis 152 (Figure 2E). Notably, neural tube duplication was not rescued in hTEM.v2 (Figure 2E). 153 Through scRNA-seq analysis, we observed distinct ventral cell identities in hTEM.v2, 154 including ventral neural tube (NKX6-1, NKX2-8), floor plate (FOXA2), ventral somite 155 (SNAI2, TWIST1), syndetome (SCX), and sclerotome (PAX1/9, SOX9), alongside 156 dorsal (PAX6) neural and dorsal (PAX3) somite cells (Figures 2F-2H, S2C, and S2D). 157 However, notochord (NOTO, CHRD) cells remained limited in number (Figure 2G). This 158 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 8 confirmed that dorsally-biased patterning in hTEM.v1 was switched to a D-V balanced 159 patterning in hTEM.v2, due to SAG-induced SHH activity. 160 Attention was then turned to evaluate bi-axial patterning in hTEM.v2. Using Visium HD 161 technology, transverse and longitudinal sections of day-7 hTEM.v2 spanning the A-P 162 axis were acquired (Figures 2I, 2J, and S2E-S2H). To validate A-P axis formation, 163 spatially resolved developmental events were assessed, including symmetry breaking, 164 somite segmentation and inter-tissue RA signalling crosstalk. As seen in human 165 embryos, the tailbud-to-hindbrain A-P patterning was evident by tailbud cells expressing 166 CDX2 and anterior neural cells expressing CRABP1 (Figure 2I).36 The somite 167 determination front was marked by co-expression of RIPPLY1 and LFNG in the PSM 168 region (Figure 2I).37 Mutually expressed TBX18 (rostral) and UNCX (caudal) were 169 observed (Figure S2I), indicative of somite segmentations.38,39 ALDH1A2 (RA synthesis) 170 expression in somite and RARB (RA effector) expression in neural tube cells were 171 suggestive of somite-neural tube crosstalk along the A-P axis (Figure 2I).40 The anterior 172 expression of ALDH1A2 was opposed to posterior CYP26A1 (RA degradation) (Figure 173 2I), consistent with “source and sink” RA signaling patterns that are fundamental to A-P 174 axis establishment in chick and mouse embryos.41,42 As seen in previous scRNA-seq 175

(Figure 2G), few node-like (NOTO, FOXJ1) cells were found adjacent to bi-176 NMPs in the tailbud (SOX2, TBXT) (Figure 2I). Moreover, caudal expression of HOXC9 177 and anterior HOXC4 were noted (Figure 2I), indicating the presence of an emerging 178 HOX code.43 179 Next, D-V patterning of neural tube structures in transversely sectioned hTEM.v2 was 180 confirmed by visualizing the spatially restricted expression of roof plate (WNT1, MSX1), 181 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 9 dorsal neural tube (PAX6, DBX2) and ventral neural tube (OLIG2, NKX6-1) markers 182 (Figure 2J). Likewise, dorsal somite (PAX3, RDH10)44 and ventral somite markers 183 (TWIST1, COL1A1) exhibited a dorsolateral-ventromedial pattern within the bilateral 184 somites (Figure 2J). Of note, endotome (EBF2) and endothelial cells (KDR) were found 185 in the lateral somite compartment, as observed in hTEM.v1 (Figures 2J and S1O). 186 These results confirmed the embryo-like cell-cell organization along the D-V axis for 187 neural and somitic lineages (Figures 2K-2N). Important signals, including WNTs 188 (WNT1/3/3A/4), BMPs (BMP4/7), PDGF (PDGFA) and heregulin (NRG1) displayed 189 dorsally enriched expression patterns in hTEM.v2 (Figure 2M), similar to that observed 190 in the neural tube in vivo.45-48 However, the floor plate (FOXA1/2, SHH) identity was 191 under-represented (Figures 2H and 2M), suggesting an incomplete D-V axis 192 establishment in hTEM.v2. Although WNTs were expressed in hTEM.v2 and are known 193 to induce myogenesis in vivo and in vitro,46,49 the excessive expression of FRZB (water-194 soluble WNT antagonist) from the ventral somite cells accounts for the absence of 195 dermomyotome and myogenic populations (Figures 2H, 2J, and 2N). 196 Altogether, in the absence of a notochord, exogenous SHH activation established only 197 a limited D-V axis in hTEM.v2 and failed to rescue neural tube duplication. These 198 observations emphasize the critical role of the notochord in trunk development and for 199 correct D-V patterning in somitic and neural lineages.50,51 To establish a bi-axial 200 embryoid comparable to the posterior trunk in human embryos, the next challenge was 201 to establish suitable culture conditions that support co-development of notochord 202 progenitor cells with bi-NMP descendants. 203 204 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 10 Co-development of notochord, neural tube and bilateral somites 205 During gastrulation, the coordinated emergence of notochord with bi-NMPs depends on 206 sustained WNT signaling and temporal modulation of NODAL and BMP signaling 207 activity.14,15,52 Following initial WNT activation in the anterior primitive streak (APS), 208 notochord induction coincides with the activation of a NODAL autoregulatory loop, 209 including CER1 and/or LEFTY2.15,53 This precise modulation was not achieved in 210 hTEM.v1/2, where broad and persistent NODAL inhibition by SB disrupted the APS-211 derived notochord process. Notably, CER1 expression precedes notochord formation 212 and localizes to notochord adjacent APS, definitive endoderm (DE), visceral endoderm 213 (VE) and axial progenitor populations in gastrulating mouse54 and human20 embryos 214 (Figures S2J-S2L). Furthermore, the co-development of notochord and bi-NMPs are 215 involved in initiation of D-V axis establishment, which depends on opposing signals of 216 SHH and BMP2/4/7 that are also active within these early populations (Figures S2J–217 S2L).55,56 218 Since these critical signals were not intact in hTEM.v1 (Figure S1L), we hypothesized 219 that its notochord deficiency resulted from; (i) disruption of APS-derived notochord 220 induction by early and persistent NODAL inhibition (SB); (ii) absence of opposing SHH 221 and BMP signals necessary for D-V patterning and notochord/bi-NMP specification. 222 To test this hypothesis, we replaced SB with recombinant human CER1 for the first 24 223 hours of bi-NMP induction (Figure 3A). By day 2, it was evident that temporal NODAL 224 modulation by CER1 followed by SB supported the balanced co-emergence of APS 225 (OTX2, EOMES), early notochord (NOTO, FOXA2), NMP-Meso (TBXT, TBX6) and 226 NMP-Neural (SOX2, NKX1-2) progenitors (Figures S3A and S3B). Further interrogation 227 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 11 of transcriptomic profiles in gastrulating mouse and human embryos revealed an array 228 of signaling pathways with synergistic and opposing activity including pathways of 229 SHH55,56, BMPs47,55-58, FGFs12,57,59-62 , and RA13,16,58,60 within notochord-adjacent 230 populations, such as APS, DE, VE and axial progenitors (Figures S2J–S2L). Through 231 empirical testing, we established a cocktail comprised of SHH, BMP2/4/7 at days 2-4, 232 plus temporal FGF2/3/4/8b/17 and RA at day 2-3 (Figures 3A and S3C), which 233 supported the coordinated progress of notochord (NOTO), somitic (TBX6) and neural 234 (NKX1-2) lineages to day 4 (Figure S3D). At day 7, markers (SHH, FOXA2) for 235 notochord and floor plate were highly expressed in these embryoids in contrast to 236 hTEM.v1/2 (Figure S3E). Day-7 embryoids elongated to ~2 mm in length and formed 6–237 8 pairs of somites flanking a midline neural tube (Figures 3B, 3C, and S3G). We refer to 238 these embryoids as hTEM.v3. 239 240 Morphological characteristics of hTEM.v3 241 Using a NOTO:mClover3 H9-hESC line, time-lapse imaging captured notochord 242 morphogenesis in hTEM.v3 at days 3-4. This began with a salt & pepper pattern of 243 NOTO:mClover3 expression followed by axial elongation of NOTO expressing cells, 244 near the caudal end (Figure S3H). This aligns with in vivo observations that Noto is 245 expressed in the node and nascent notochord in mice.63 Meanwhile, NKX1-2:mScarlet+ 246 cells were progressively enriched along the midline of hTEM.v3, indicative of caudal 247 neural plate morphogenesis (Figure S3I). 248 249 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 12 In day-4 hTEM.v3, distinct FOXA2+ notochord formed along the midline, flanked by 250 caudal PSM (TBX6) cells (Figure 3D). Immunostaining-HCR results further confirmed 251 the embryo-like arrangement of a dorsally-localized caudal neural plate (SOX2, NKX1-252 2) and a ventrally localized notochord (transcripts of NOTO, CHRD SHH and proteins of 253 NOTO:mClover3, TBXT, FOXJ1, FOXA2) in hTEM.v3 (Figures 3E, 3F, S3I, and S3J). 254 These features closely resemble the gastrulating human CS8-9 embryos.20,27 After day 255 4, hTEM.v3 exhibited embryo-like D-V arrangements of the neural tube, notochord and 256 bilateral somites. The emergence of a ventral notochord coincided with rescue of the 257 neural tube duplication defect seen in hTEM.v1/2 (Figure 3G). 258 In mice, the node is located at the anterior tip of APS and is composed of columnar 259 epithelial cells in the dorsal region and teardrop-shape, ciliated cells in the ventral part.64 260 SEM imaging of transversely fractured day-4 hTEM.v3 revealed a group of ciliated, 261 squamous cells on the ventral side near the posterior end, resembling the ventral node 262 structure in embryos (Figure S3K). Confocal imaging further confirmed ciliated cells 263 (FOXJ1, ARL13B) residing in the presumptive node (TBXT) region (Video S2). At day 264 5.5, SEM of hTEM.v3 revealed a dorsally positioned lumen representing the neural 265 tube, a vacuolated and ventrally localized notochord and bilateral rosette patterns 266 formed by cells in the somites (Figure 3H). Consistent with observations in frog, rabbit 267 and chick65-67, the inner canal of hTEM.v3 notochord contains lipid droplets of varying 268 sizes, while the outer layer is rich in extracellular matrix (ECM) fibers, making notochord 269 cells structurally distinguishable from neural tube cells. Notochord cells in hTEM.v3 270 were flattened, vacuolated and larger than neural tube cells, indicative of changes in the 271 nature of the cytoplasm during notochord maturation. This is reminiscent of 272 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 13 morphological features of the chicken notochord at Hamburger & Hamilton stage 14.66 273 The structural similarities between notochord structures in hTEM.v3 and frog, rabbit, 274 chick and mouse embryos, signifies the conservation of notochord development across 275 vertebrate species. It also validates the hTEM.v3 as a legitimate model for human trunk 276 development. 277 278 Cellular composition of hTEM.v3 279 scRNA-seq integration and clustering analysis of hTEM.v3 (days 3-7) identified 29 280 major cell types (Figures S3L and S3M). RNA velocity analysis further unveiled intricate 281 developmental trajectories of the tailbud (CDX2, CYP26A1) stemming from notochord 282 (TBXThighSOX2low, NOTO, SHH), NMP-Meso (TBXThighSOX2low, MSGN1) and NMP-283 Neural (TBXTlowSOX2high, NKX1-2) (Figure 3I and 3J).20,27 Bi-NMPs diverge into two 284 streams, including; (i) NMP-Meso → posterior PSM → anterior PSM → E-/M-Somite 285 (pan-somite) → L-Somite (somite compartments comprised of sclerotome, syndetome, 286 endotome, myogenic progenitors, and dermomyotome) and; (ii) NMP-Neural → Caud. 287 NP → E-Neural tube and E-Floor plate → Neural tube and Floor plate (Figures 3I and 288 3K). In contrast to hTEM.v2, distinct myogenic progenitors (PAX7), dermomyotome 289 (MYF5, MYF6) and floor plate (FOXA1/2, NKX2-8) cells were observed in hTEM.v3, 290 highlighting the importance of notochord for control of cell fate specification along the D-291 V axis. 292 To better understand human notochord development, sub-clustering, RNA velocity and 293 pseudotime analyses were conducted on the ‘notochord’ subset from hTEM.v3 (Figures 294 S3N-S3P). Four cell subtypes delineating notochord maturation were identified (Figures 295 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 14 S3Q and S3R). In line with APS-derived notochord processes,15,17 the first subtype 296 designated as ‘node or notochord progenitors’ was marked by NODAL, WNT3A, 297 CYP26A1 and CDX2. The second and third subtypes were both marked by NOTO, 298 equivalent to newly generated notochord cells in vivo.63 The second subtype was 299 enriched for ‘nascent notochord’ (FOXA2, CHRD, SHH) markers,20 while the third 300 subtype was signified by ‘ciliated notochord’ (FOXJ1, RFX2, TCTEX1D1) markers.68 In 301 the fourth ‘mature notochord’ subtype, NOTO transcript was decreased associated with 302 upregulation of FOXA1, SOX9, NOG and SEMA3C.14 hTEM.v3 is therefore a platform 303 on which the detailed processes of human notochord development and function can be 304 explored. 305 306 hTEM.v3 recapitulates key aspects of the trunk A-P axis 307 To assess whether hTEM.v3 could model aspects of in vivo A-P axis at the multi-tissue 308 level. Pseudotime analysis was individually performed on bi-NMP-derived somitic and 309 neural lineages and notochord cells. We ordered cells according to the rank of 310 pseudotime indices and inferred the A-P axis for each lineage (Figures S4A and S4B). 311 As expected, expression of marker genes for respective lineage progenitors were 312 enriched at the inferred posterior end, whereas marker genes for differentiated tissues 313 were highly expressed at the anterior end (Figure 3L). 314 Having built the inferred A-P axis, we then assessed the spatial distribution of RA, 315 FGF, WNT and NOTCH pathways that are crucial for trunk development as observed in 316 vertebrate embryos.16,42,69,70 RA signaling is important for the balanced bi-NMP 317 specification into somitic and neural lineages.71 It was therefore important to establish if 318 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 15 RA signaling was active in hTEM.v3. The expression of RA degradation enzyme 319 (CYP26A1) peaked in the posterior-most NMP-Neural and notochord progenitor cells, 320 with RA receptor gamma (RARG) exclusively expressed in the NMP-Neural cells 321 (Figure 3M). Expression of RA synthesis genes (RDH10, ALDH1A2) were restricted to 322 anterior somitic cells and the positionally parallel expression of RARB was exclusive to 323 the anterior neural cells, coinciding with the expression of PAX6 (Figures 3L and 3M).72 324 The A-P patterned RA circuit genes confirms the integrity of RA signaling in hTEM.v3 as 325 observed in mouse embryos (Figure 3N).16,70 326 How FGFs exert differential roles in coordinating multi-tissue co-patterning along the 327 A-P axis remains unknown.59,70 Distinct expression patterns of FGFs were noted among 328 different cell types in hTEM.v3. For example, FGF3/4/8/17/19 in posterior neural and 329 somitic cells, FGF13 in anterior neural cells, FGF13/18 in anterior somitic cells, while 330 FGF8/17 were expressed throughout the notochord without A-P polarity (Figure S4C). 331 Like FGFs, WNT ligands also showed A-P graded expression in somitic and neural 332 lineages. Canonical (WNT3A/8A) and noncanonical (WNT5A/5B) WNT ligands were 333 expressed in the posterior-most bi-NMPs (Figure S4D). In contrast, WNT3A/5B were 334 expressed throughout the notochord lineage. Gradients of CTNNB1 (WNT effector) in 335 each lineage was opposed to the posteriorly restricted WNT ligands, consistent with 336 polarized cell proliferation and movements during axial elongation. Anterior expression 337 of SFRP1/2 (WNT inhibitors) in both neural and somitic lineages were opposed to the 338 WNT ligands in bi-NMPs, consistent with a negative WNT feedback loop during trunk 339 formation in mice.1 How the interacting gradients of FGF and WNT signaling drive 340 notochord formation, segmentation clock and neurogenesis in humans is not well 341 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 16 understood hTEM.v3 will be useful for addressing such important developmental 342 questions. 343 Next, we investigated the patterning of NOTCH signaling in hTEM.v3 as it is critical for 344 axial elongation.64,73 How NOTCH signaling participates in D-V patterning is poorly 345 understood, but hTEM.v3 is likely to be a useful tool to address this. In the somitic 346 lineage, anteriorly expressed NOTCH modulator LFNG was opposed the posterior 347 NOTCH ligands (DLL3) and its effector (HES7) (Figure S4E), mirroring the anterior-to-348 posterior somitogenesis.42 Although NOTCH receptors (NOTCH1/2/3) were lowly 349 expressed in the neural and notochord lineages, NOTCH effectors (HES1/4) and their 350 target (CCND1) were highly expressed in anterior neural cells and throughout the 351 notochord, respectively. This could explain the rapid morphogenesis and axial 352 elongation of neural tube and notochord after day 4 (Figures 3B and 3C).74,75 353 Interestingly, we noted the expression of a NOTCH coactivator MAML2 throughout the 354 notochord (Figure S4E). The role of Maml2 is possibly involved in Sox9-dependent 355 inhibition of the WNT pathway in mouse sclerotome.76,77 hTEM.v3 offers a unique 356 opportunity to characterize this notochord-dependent NOTCH regulatory mechanism in 357 human ventral patterning. 358 Collectively, hTEM.v3 faithfully recapitulates A-P axis development at multi-tissue 359 levels, mirroring key signaling networks of early human embryogenesis. By achieving 360 high-fidelity reconstruction of essential signaling pathways, this model opens new 361 avenues to interrogate spatiotemporal signaling crosstalk and cell-fate decisions at a 362 multi-tissue level. 363 364 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 17 Patterning of neural and somitic cells along the D-V axis 365 To delineate D-V specifications in hTEM.v3, sub-clustering and RNA velocity analysis 366 was performed on neural and somitic cells, respectively (Figures 4A and 4B). The 367 resultant UMAP demonstrated clear D-V patterning, evident by distinct transcriptomic 368 profiles of dorsal/ventral neural cells-floor plate and somite compartments, respectively 369 (Figures 4C, 4D, S5A, and S5B). This high degree of complexity regarding D-V 370 specifications was not seen in hTEM.v1/2 (Figure 4E). Immunostaining of transversely 371 sectioned day-6.5 hTEM.v3 confirmed D-V patterned dorsal (PAX6) and ventral (NKX6-372 1, OLIG2) neural cells along the neural tube (Figure 4F). These midline-positioned 373 neural cells were flanked by distinct somite compartments including the dorsal somite 374 (PAX3), dorsolateral dermomyotome (MYF5:mClover3), myogenic progenitors (PAX7), 375 lateral endotome (EBF2:mScarlet), ventromedial sclerotome (PAX1) and surrounding 376 vascular endothelial cells (SOX17) (Figures 4F and 4G). 377 To understand the molecular mechanisms underlying D-V axis establishment, we 378 performed Gene Ontology (GO) and pathway enrichment analysis on scRNA-seq data 379 for hTEM.v3. First, SCENIC78 was used to generate a regulon module enrichment 380 heatmap illustrating representative transcription factor (TF) genes associated with all 381 cell types of hTEM.v3 (Figures 4H, 4I, and S5C). As expected, biological processes in 382 GO terms enriched for each regulon module were consistent with associated cell types 383 (Figure S3L, S5D, and S5E). For example, module M4 comprised of key TFs (PAX6, 384 IRX3, NKX6-2) important for regulating neural tube formation, was significantly enriched 385 in “GO:0001840 neural plate development”, was highly expressed in the “Neural tube”, 386 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 18 but not in “NMP-Meso” or “L-Somite” clusters in hTEM.v3 scRNA-seq data (Figures 4I 387 and 4J). 388 Regulon module-based gene regulatory network (GRN) analysis identified key GRNs 389 (NOTO, PAX6, NKX6-2, FOXA1/2) and their target networks that drive D-V axis 390 formation (Figures 4K and S5F). The NOTO GRN was enriched for SHH signaling 391 (GLIS1), TBXT, and cilia functions (TPPP3, SCG3, TCTEX1D1). Consistent with Noto’s 392 role in mice,79 NOTO target genes were expressed in human 'nascent' and 'ciliated' 393 notochord subtypes (Figure S3Q). The NKX6-2 GRN contained targets that balance 394 ventral neural tube patterning, including interneuron (HES5)80 and motoneuron 395 specifiers (OLIG1/2)81 (Figure S5F). Finally, FOXA1/2 GRNs 396 exhibited differential functions. FOXA1 targets were associated with floor plate (ARX)82 397 patterning and notochordal fluid trafficking (CFTR)83, whereas the FOXA2 targets 398 governed node formation (PPIL6)84, cilia function (CFAP43)85, and maintaining 399 notochord structure (KRT8, EPCAM, FN1)63 (Figure S5F). The GRN analysis revealed 400 multifaceted networks in ventral patterning, demonstrating the establishment of D-V axis 401 in hTEM.v3. 402 Overall, hTEM.v3 self-organizes into notochord, neural tube and somitic tissues with 403 proper A-P and D-V patternings (Figures 3E-3H and 4A-4G). A key advance in hTEM.v3 404 is its ventral neural specification. Unlike hTEM.v2, which displayed elevated HES5 and 405 low NKX6-2 expression accompanied with a floor plate deficiency (Figure 2H), hTEM.v3 406 exhibited distinct floor plate (ARX) and ventral neural (OLIG2) markers alongside 407 reduction in HES5 and elevated NKX6-2 (Figure 3K). Our GRN analysis suggests that 408 NXK6-2 acts as a pivotal node, connecting SHH signaling and Notch-mediated HES5 409 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 19 activity to balance motoneuron (OLIG2)/interneuron (HES5) specification (Figures S5A 410 and S5F).86,87 These data suggest a previously undescribed SHH-NKX6-2-HES5 411 regulatory mechanism in human neural tube patterning. 412 413 Spatially transcriptomic profile reveals embryo-like signatures of hTEM.v3 414 To map spatial allocation of cells in hTEM.v3, we performed Visium HD on longitudinal 415 sections (days 4/6.5). Key cell types including notochord, NMP-Meso, NMP-Neural and 416 D-V fated neural and somitic cells were all identified (Figure S6A). However, full 417 recovery of D-V fated cells from Visium HD data, especially in somite compartments, 418 was difficult due to limited sections. We therefore performed cell type deconvolution 419 (RCTD)88 by leveraging the reference annotations from scRNA-seq data of hTEM.v3 420 (days 3-7) (Figure S6B). The resulting spatial UMAP exhibited embryo-like spatial 421 patterns that correspond to anatomical regions of the posterior trunk in human CS8-10 422 embryos (Figures 4L, S6C, and S6D).20,27,36 For example, by day 4, the Caud. NP 423 (NKX1-2, GBX2) extending anteriorly out of the tailbud was flanked by anterior PSM 424 (RIPPLY2, CER1)89 (Figure S6E). By day 6.5, neural tube and floor plate were specified 425 in the midline along the A-P axis and flanked by compartmentalized somites (Figures 4L 426 and S6D). Remarkably, NMP-Neural and NMP-Meso cell populations occupied distinct 427 regions in the day-4 tailbud (Figure 4L, S6C and S6E), representing an unprecedented 428 bi-layered structure of bi-NMPs in vitro. This bi-layered structure signifies the onset of 429 D-V patterning as observed in a spatially resolved human CS9 embryo.27 430 Next, we verified the inferred A-P signaling gradients from spatially resolved ligand-431 receptor gene expression. For example, the RA (RARG-CYP26A1), FGF (FGF8-432 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 20 FGFR1), WNT (WNT3A-FZD7) and NOTCH (DLL3-NOTCH1) ligand-receptor pairs 433 were spatially enriched in the tailbud of hTEM.v3 (Figures 4L and 4M), consistent with 434 their inferred A-P distribution in Figures 3M and S4C-S4E. In addition, CellChat90 435 predicted tissue-tissue communications from scRNA-seq data that were also validated 436 by Visium HD data (Figure 4O). For example, PAX1 (target of SHH pathway) 437 expression in the sclerotome region was shown to be in close proximity to SHH 438 expression in the floor plate (Figure S6F), reflecting the SHH signaling range. The 439 anterior expression of ALDH1A2 in somites and posterior CYP26A1 in tailbud (CDX2) 440 were evident of the RA signalling along A-P axis (Figures 3M and 4N). The detection of 441 RARB and CRABP1 (RA target) in anterior neural cells confirms the somite-to-neural 442 RA crosstalk has been established as in the spinal cord of human CS10 embryos 443 (Figure S6F).36 Moreover, CellChat-predicted PDGF (PDGFA-PDGFRA) signaling 444 between neural and somitic cells was revealed on the spatial UMAP (Figures 4N and 445 4O). This suggests a neural-to-somite regulatory mechanism in patterning the 446 sclerotome.91 These spatial data demonstrate hTEM.v3’s utility in studying advanced 447 organogenesis co-patterning. 448 The emergent HOX expression is key to axial elongation and spatially aligned with 449 body axes formations,92 we therefore evaluated the fidelity of HOX coding in hTEM.v3. 450 hTEM.v3 exhibited HOX collinearity in somitic lineage as observed in axialoids8 and 451 extended it to neural and notochord lineages, with over 50% of HOX genes displaying 452 coordinated expression (Figure S4F). Tissue-specific expression patterns of HOX genes 453 were also noted. For example, HOXA5 and HOXB5 were expressed in anterior somitic 454 and notochord lineages but not in the neural lineage. HOXA9 was restricted to the 455 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 21 Caud. NP and posterior PSM, but not expressed in notochord cells (Figure S4F), 456 consistent with its initial expression in caudal neural plate region in mouse gastrulating 457 embryos.93 We next exemplified the spatiotemporal HOX coding with scaled expression 458 of HOXC genes from days 4-6.5 (Figures S6G and S6H). During axial elongation, a 459 clear anterior expansion of HOXC6 and HOXC8 expression accompanied the caudal 460 downregulation of HOXC10, as observed in the human developing spinal cord.94 461 Collectively, our spatial and scRNA-seq transcriptomic data establishes hTEM.v3 as a 462 high-fidelity model of human posterior axial development. This system recapitulates the 463 spatially patterned signaling pathways and tissue-tissue crosstalk of human posterior 464 trunk. Importantly, hTEM.v3 demonstrated a spatiotemporal HOX coding alongside axial 465 elongation, thereby capturing the core signaling and transcriptional machinery for future 466 developmental studies. 467 468 Developmental staging of hTEM.v3 resembles primate CS8-10 embryos 469 Next, we sought to allocate the developmental lineages of hTEM.v3 to those involved in 470 trunk formation in primates and mice. First, a embryogenesis scRNA-seq reference, 471 consist of public datasets of human (CS7/8/10)20,27,84 and cynomolgus monkey 472 (CS8/9/11)19 embryos, was created based on primate orthologues (Figure S7A). Consist 473 with hTEM.v3, the notochord, NMP-Neural and NMP-Meso in human-monkey (H-M) 474

dataset showed well-matched expression profiles (Figure 3K and 5A). Next, 475 sub-clustering of caudal trajectories (caudal epiblast, caudal mesoderm, somitic 476 mesoderm and NMP) in mouse embryo (E7.0-E8.5) datasets19,54 identified previously 477 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 22 unresolved mouse NMP-Meso and NMP-Neural populations (Figure 5B), highlighting 478 the evolutionary conservation in trunk formation between mice and primates. 479 Next, we examined cellular composition similarities between hTEM.v3 (days 3-7) and 480 H-M (CS7-11) datasets using Seurat. Comparative projection of cell types showed that 481 hTEM.v3 accurately recapitulates H-M posterior trunk development including the 482 notochord, bi-NMPs, somite and spinal cord (Figures S7B and S7C). Likewise, hTEM.v3 483 was mapped to ortholog mouse trajectories of axial progenitors (notochord, caudal 484 epiblast, caudal mesoderm, somitic mesoderm, NMP), paraxial mesoderm and spinal 485 cord (Figures S7D and S7E). Collectively, an unsupervised cluster similarity analysis 486 summarized the major trunk cell types conserved across hTEM.v3, H-M, and mouse 487 embryos (Figures 5C and S7F). Further comparative analysis aligned the 488 developmental stages between hTEM.v3 and H-M, and hTEM.v3 and mouse, 489 respectively (Figures 5D and S7G). We noted the correspondences of advanced cell 490 types, such as neurons, intermediate mesoderm and endothelial cells at post-491 gastrulation stages between datasets (Figure S7G). Moreover, genes important for 492 notochord, bi-NMP specification and pathways for RA, FGF and WNT signaling 493 exhibited broad similarities across species (Figures 5E and S7H). One exception is the 494 absence of Fgf17 expression in mouse notochord. In hTEM.v3 and H-M notochord 495 populations, FGF17 expression is highly expressed, suggesting a divergence in 496 mechanisms of notochord progress and D-V establishment between mice and primates 497 (Figures 5A and 5E). These findings establish that hTEM.v3 is suitable for identifying 498 new mechanisms of human development. 499 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 23 In summary, combined with the embryo-like morphological features of hTEM.v3 500 (Figures 3D and S3I), cross-species comparative analyses indicates that hTEM.v3 501 accurately models cellular and molecular aspects of trunk development in CS8-10 H-M 502 embryos. hTEM.v3 (days 3-7) equates approximately to E19-23 in humans, E20-24 in 503 cynomolgus monkeys and E7.5-8.5 in mice (Figure 5F). 504 505 Changes in notochord identity and SHH signaling switch D-V fates 506 Next, the utility of hTEM.v3 as a testbed for investigating human D-V patterning at the 507 genetic level was explored. Focus was directed towards NOTO which is important for 508 notochord function in mice by serving as a key regulator of Shh signaling in mice.20,63 509 Loss-of-function mutation of NOTO (NOTO-LOF) derived day-6 hTEM.v3 exhibited 510 irregular posterior morphology and shorter axial length (Figure S8A-S8C), consistent 511 with the shortened tail phenotype in Noto-null mice.79 Comparative scRNA-seq analysis 512 between NOTO-LOF and wild-type (WT) revealed an expanded population of ‘mature 513 notochord’ (FOXA1, NOG) cells in day-6 hTEM.v3 (Figures 6A–6C). This shift in 514 notochord identity correlated with a pronounced upregulation of SHH expression and 515 ventrally-biased fates (Figures 6C-6E). For example, cellular proportion of ventral 516 lineages (floor plate; FOXA2, NKX6-1. sclerotome; PAX1, PAX9) were upregulated at 517 the expense of dorsal identities in neural tube (PAX6, MSX1) and somites (PAX3) 518 (Figures 6D, 6E, and S8D). Focusing on the neural tube, elevated SHH activity (SHH, 519 GLI1) in NOTO-LOF sharply suppressed PAX6 while inducing NKX6-1 (Figures 6E and 520 6F). Moreover, NOTO-LOF downregulated cilia-related genes (FOXJ1, RFX3, 521 TCTEX1D1) (Figure S8E), consistent with Noto’s conserved role in regulating 522 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 24 ciliogenesis.79 These data demonstrate hTEM.v3 as a valuable testbed for genetic 523 dissection of human development and establish the role of NOTO in restraining SHH 524 activity and safeguard nascent notochord identities. 525 In contrast to the NOTO-LOF, exposure to SANT195 (smoothened antagonist, 250 nM 526 at days 4-6) led to a downregulation of ventral (PAX1, NKX6-1, FOXA2) markers 527 without negatively impacting dorsal (PAX3, PAX6) markers or axial elongation (Figures 528 S8B-S8D). scRNA-seq and immunostaining results confirmed that SANT1 treated 529 hTEM.v3 were dorsally-biased similar to hTEM.v1 (Figures 6E and 6F). This was 530 supported by reduced ventral populations (e.g., floor plate, sclerotome) and a 531 concordant reduction in the expression of ventral markers (NKX6-1/6-2, OLIG1/2, 532 FOXA1/2) (Figures 6D, 6E, and S8D). It is noteworthy that the proportion of notochord 533 cells and endogenous levels of SHH were not negatively affected in SANT1 (Figures 6C 534 and 6D), indicating that the dorsally-biased fates in SANT1 can be attributed to the 535 blockade of SHH signaling between notochord and its targets. Pharmacological 536 inhibition of the SHH pathway downregulates BMPs, VEGFs and KDR which control 537 dorsal aorta development and vasculogenesis.96 In support of this observation, 538 reductions in the proportions of endotome, endothelial cells and related gene 539 expression (MEF2C, CD34, KDR) are noted in hTLE.v3 (Figures 6B, S8F, and S8G). 540 Collectively, these observations show that by modulating notochord identity and 541 derived SHH signaling, hTEM.v3 embryoids can be patterned into ventrally-biased 542 (NOTO-LOF) or dorsally-biased (SANT1) fates. This responsiveness allows timed 543 perturbations in SHH to be assessed and enables human bi-axial development to be 544 explored with temporal precision and quantifiable measurements. 545 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 25 546 Disease modeling of axial truncation by knockout of VANGL1/2 547 We next evaluated hTEM.v3 as a platform for the study of human congenital disease. 548 For proof-of-concept, PCP signaling core members VANGL1 and VANGL2 were 549 selected, as mutations in VANGL1/2 lead to NTDs including spina bifida and congenital 550 vertebral malformations.97-99 As expected, VANGL1/2 expression was widespread 551 throughout day-6.5 hTEM.v3, with VANGL2 enriched in neural tube and VANGL1/2 552 slightly enriched in caudal neural plate and PSM (Figures 6G and S8H). To model 553 VANGL-related defects in these tissues, loss-of-function mutations of VANGL1/2 H9-554 hESC lines were made, including VANGL1 -/- (VANGL1-LOF), VANGL2 -/- (VANGL2-555 LOF), and VANGL1 +/-; VANGL2 -/- (VANGL1/2-LOF) (Figure S8A). 556 As in mice,99,100 VANGL-mutant derived hTEM.v3 exhibited axial truncations with 557 different degrees of penetrance (Figures 6H and S8I). VANGL1/2-LOF displayed an 558 early failure in mediolateral narrowing on day 4 and subsequently, severe dysregulation 559 of somitogenesis (no segmentation) and neural tube genesis (loss of a neural tube) 560 (Figure 6I). These observations are consistent with the well-established role of Vangl in 561 C&E movements and PCP signaling during axial elongation in zebrafish and mouse 562 models.99,100 Contrarily, VANGL1-LOF and VANGL2-LOF showed milder axial 563 truncation defects compared to VANGL1/2-LOF, with VANGL2-LOF reproducing the no-564 segmentation defect presented in VANGL1/2-LOF (Figure 6I). 565 To further characterize the role of VANGL2 in trunk development, we performed a 566 comparative scRNA-seq analysis of VANGL2-LOF and WT embryoids (Figure 6A). In 567 VANGL2-LOF, a sharp decrease in the proportion of neural tube cells was accompanied 568 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 26 by a marked increase in PSM, while the somite population remained unchanged (Figure 569 6J). The expanded PSM population and the caudal accumulation of TBXT protein in 570 VANGL2-LOF mirrored the observations of widened PSM and caudally restricted 571 expression of Tbxt and Tbx6 in Vangl2 mutant mice (Figure 6I).99 Focusing on the loss 572 of somite segmentations in VANGL2-LOF, differential gene expression analysis 573 revealed that ITGA5, a gene required for fibronectin (FN1) accumulation during somite 574 boundary formation and neural tube closure (Figure 6G),101,102 was downregulated in 575 pan-somite cells (Figure S8J). This indicates the importance of stable interactions 576 between VANGL2 and integrins in somite segmentation and points towards an unknown 577 VANGL2-dependent mechanism that regulates ITGA5 expression.103 578 Turning to VANGL2-related PCP components, PRICKLE1, DVL1/3, PTK7, CELSR1/3 579 and FZD6 were broadly expressed throughout the somite and neural tube trajectories in 580 day-6.5 hTEM.v3 (Figure 6G). Although VANGL2 protein is typically negatively 581 correlated with PRICKLE1/2 protein levels,104 VANGL2-LOF resulted in increased 582 PRICKLE1/2 transcript levels in PSM in VANGL2-LOF (Figures 6K, 6L and S8J). This 583 finding points towards the existence of compensatory transcriptional mechanism within 584 the PCP network in PSM cells. Additional PCP genes including CELSR3, DVL2/3, PTK7 585 were concordantly upregulated with over-expanded UNCX (caudal polarity) expression 586 in PSM and pan-somite cell populations (Figures 6L and S8J). In contrast, VANGL2-587 LOF caused downregulation of PRICKLE2, CELSR3, DVL2 and PTK2 in neural cells, 588 accompanied by a shortened neural tube (Figures 6I and S8J), consistent with their 589 reported implication in human NTDs.105 This contrasting expression pattern in VANGL2-590 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 27 LOF indicates a differential role of VANGL2 in coordinating PCP components in; (i) 591 establishing A-P polarity somite segmentations and, (ii) neural tube elongation. 592 Current VANGL-related early congenital disease models are based solely on 593 observations made in zebrafish and mice.99,106 Using VANGL1/2 knockouts, we 594 demonstrated the potential for our multi-tissue, bi-axial embryoids to serve as a NTD 595 model and as a platform for developing genetic screens, therapeutic strategies, toxicity 596 and drug tests in a more human relevant context. 597 Overall, by leveraging the co-development of notochord and bi-NMPs, embryo-like 598 structures were generated that, at the morphologic and molecular level, recapitulating 599 A-P and D-V axes formations in the human posterior trunk. These embryoids are self-600 organized and non-integrated. This provides a valuable tool that meets the criteria2 for 601 benchmarking early human embryogenesis and the development of human disease 602 models. This can also be a platform for future development of next-generation 603 embryoids that recapitulate additional aspects of human development. 604 605

of this study 606 Few to no neural crest (SOX10), endoderm (HNF4A), cardiac mesoderm (HAND1) or 607 anterior neural ectoderm (OTX2) cells were detected in hTEMs. The neural tube 608 structures in embryoids are believed to represent secondary neurulation, which 609 accounts for the caudal portion of the A-P axis.107 More advanced embryo models are 610 needed to reconstitute the fully body axes of A-P, D-V and L-R. 611 612 Resource availability 613 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 28 The raw and processed sequencing data in this study are available in the GEO 614 repository GSE314260. Source code for analyzing the sequencing data is available at 615 https://github.com/tianmingwucuhk/hTEM. 616 617

618 We thank the Core Laboratories of School of Biomedical Sciences at the Chinese 619 University of Hong Kong for the service of single cell sorting, histology sections, 620 confocal imaging and SEM. This study was supported by grants to S.D. from the 621 Research Grants Council of Hong Kong (General Research Fund) and the Hong Kong 622 Jockey Club Charities Trust. S.D. is a Global Stem Scholar and Director of the JC 623 STEM Lab of Stem Cells and Regenerative Medicine. 624 625 Author contributions 626 Conceptualization, T.-M.W. and S.D.; Sequencing data generation, T.-M.W, H.Y., and 627 J.V.; Sequencing data analysis, T.-M.W, H.Y., B.S.-H.W., L.X., and S.K.-W.T.; 628 Experiments, T.-M.W., K.-X.T., W.-M.X., E.S.-K.N., A.Y.-F.K., and Jianan Z.; 629 Interpretation and discussion, T.-M.W., S.D., Jiannan Z., B. G.; Supervision, T.-M.W. 630 and S.D.; Manuscript writing, T.-M.W. and S.D.; Funding acquisition, S.D. 631 632 Declaration of interests 633 S.D. and T.-M.W. are the applicants and inventors on a patent filed by the Chinese 634 University of Hong Kong under reference number 25/MED/1610. The rest authors 635 declare no conflicts of interests. 636 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 29 637 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 30 638 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 31 Figure 1 Generation of dorsally biased fates hTEM.v1 639 (A) Schematic of the hTEM.v1 generation. 640 (B) Representative images of hTEM.v1 from day 0 to day 7. Scale bars, 200 μm. 641 (C) Box plot of embryoid length over time from three independent biological replicates 642 (rep). Each dot represents an individual hTEM.v1 length measurement (n = 9-124 for 643 each time point). 644 (D) SEM snapshots of day-7 hTEM.v1 samples (n = 3) in longitudinal orientation from 645 ventral (left), dorsolateral (middle) and transversely fractured views. nt., neural tube. 646 sm., somite. 647 (E) 3D projections of representative day-4 hTEM.v1 stained for NMPs (TBXT, SOX2), 648 PSM (TBX6), early unsegmented somite (SIX1), and mitotic cells (pH3-Ser10). Scale 649 bars, 100 μm. 650 (F) Immunofluorescence image of longitudinally sectioned day-6 hTEM.v1 showing the 651 neural tube (SOX2) and flanking pairs of somites (PAX3). Scale bar, 100 μm. 652 (G) Immunofluorescence images of a transversely sectioned day-7 hTEM.v1 showing 653 the spatial organization of duplicated, epithelialized (N-cadherin) neural tube structures 654 (SOX2) relative to paired somites (PAX3). 655 (H) 3D projections of day-7 hTEM.v1 from the dorsal (left) and ventral (right) displaying 656 the duplicated neural tubes (SOX2) with flanking somites (SIX1). Caudally positioned 657 NMPs (TBXT) were barely detectable on day 7. Scale bars, 100 μm. 658 (I) UMAP of integrated hTEM.v1 integration from day 4/5/7 (total of 47,908 cells). Cell 659 type annotations are indicated below. 660 (J) UMAP showing the expression of selected cell types from hTEM.v1. 661 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 32 (K) Dot plot displaying the expression levels (color) and proportion (dot size) of marker 662 genes for dorsal and ventral fates. Proportions below 5% were omitted. 663 (L) (Top) Both neural tube (SOX2, PAX3/6, BMP4/7) and somite cell (PAX3, BMP4/7) 664 identities were dorsal-biased in day-7 hTEM.v1, (bottom) with few detectable ventral 665 neural tube (NKX6-1, arrowhead) cells. Scale bars, 100 μm. 666 (M) Density plot showing the progress of NMP-neural, NMP-Meso and Node-like cells 667 over time. 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 33 683 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 34 Figure S1 Characterization of hTEM.v1, related to Figure 1 684 (A) SEM image showing the size contrast between a representative day0-hESC 685 spheroid (top) and day-4 hTEM.v1 (below). 686 (B) Representative images of day-6 hTEM.v1 with or without 4% Matrigel. Scale bars, 687 200 μm. 688 (C) Box plot showing length measurements of day-6 hTEM.v1 with (n = 15) or without (n 689 = 6) 4% Matrigel. Data are presented as mean ±standard deviation. 690 (D) Number of somite pairs in hTEM.v1 over days 5-7. Each dot represents an 691 individual hTEM.v1 randomly chosen from 3 biological replicates. Vertical lines 692 represent mean values. 693 (E) Top row, 3D projection of day-2 hTEM.v1 showing polarized patterns of bi-NMPs 694 (SOX2, TBXT, CDX2). Bottom row, complementary TBX6+ PSM and SOX2- regions. 695 (F) Immunofluorescence images of longitudinally sectioned day-4 hTEM.v1 showing A-696 P patterned bi-NMPs (TBXT, SOX2), neural plate (SOX2), PSM (TBX6), and flanking 697 somitic (SIX1) cells. 698 (G) 3D projection of A-P symmetry breaking of day-4 hTEM.v1 with posterior tail bud 699 (CDX2) and anterior somitic cells (N-cadherin). Scale bar, 100 μm. 700 (H) Immunofluorescence image showing the duplication of the neural tubes with flanking 701 somites in transversely sectioned day-7 hTEM.v1. 702 (I) 3D projections of day-4 hTEM.v1 showing the duplication of neural plate (SOX2) 703 structures along A-P axis. TBXT was absent from the presumed ventral side and 704 restricted to the posterior-most region. nt., neural tube. sm., somite. 705 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 35 (J) Heatmap of scRNA-seq showing clusters of transcripts associated with cell identities 706 of hTEM.v1 (days 4/5/7) as seen in Figure 1I. 707 (K) Changes in cell type composition in hTEM.v1 over days 4-7. Colors for cell types are 708 in consistent with (J). 709 (L) UMAP showing transcript levels for FGF, BMP and WNT family members in 710 hTEM.v1 (days 4/5/7). 711 (M) HCR-IF images of day-4 hTEM.v1 showing the node-like (SHH, arrowhead) 712 structure adjacent to PSM (HES7, TBX6). 713 (N) Counts of the node-like cells from hTEM.v1 scRNA-seq data, based on detected 714 marker genes at listed days. 715 (O) Left, 3D projection showing anteriorly positioned vascular endothelial cells (SOX17). 716 Right, Immunofluorescence of vascular endothelial cells (SOX17, VE-cahderin) at the 717 outer layer of epithelialized somites (N-cadherin). 718 All scale bars in the immunofluorescence images are 100 μm. 719 720 721 722 723 724 725 726 727 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 36 728 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 37 Figure 2 SHH activation by SAG caused D-V specifications in neural tube and 729 somite without a notochord 730 (A) Schematic of hTEM.v2 generation. 731 (B) Representative images of hTEM.v2 over days 5-7. Scale bars, 200 μm. 732 (C) Bar plot showing length measurements comparing hTEM.v1 to hTEM.v2 over days 733 5-7. Each dot represents a length measurement of individual hTEM.v1/2 embryoids at 734 shown time point (n = 14-57 for each group from same batch). n.s., no statistical 735 significance. 736 (D) qPCR results showing dorsal and ventral marker gene expression in hTEM.v2. Data 737 are presented as mean ±standard deviation. Data were reproduced twice. 738 (E) 3D projection of day-7 hTEM.v2 showing dorsal and ventral neural tubes (SOX2) 739 flanked by paired somites (SIX1). Arrowhead, somites. Dashed line, duplicated neural 740 tubes. Scale bar, 100 μm. 741 (F) UMAP showing identified cell types from hTEM.v2 (days 5/7, total of 16,535 cells). 742 Day-4 hTEM.v1 scRNA-seq data was integrated for accurate trajectory annotations. 743 (G) UMAP showing the expression of indicated lineage markers in hTEM.v1. 744 (H) Dot plot showing the expression profile reflecting somite compartments and neural 745 tube D-V patterning, induced by SAG treatment. Proportions below 5% were omitted. 746 (I-J) Spatial gene expression patterns in longitudinal (I) and transverse (J) day-7 747 hTEM.v2 sections. The major tissue trajectories were consistent with those shown in 748 Figures S2E and S2G. 749 (K-L) Schematic diagram of D-V patterning of the neural tube (K) and adjacent 750 compartmentalization of the somite (L). 751 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 38 (M-N) Dot plots showing the expression profiles of D-V patterns of neural tube (M) and 752 somite (N) cells from transversely sectioned day-7 hTEM.v2 Visium HD data. The listed 753 clusters were consistent with those in Figure S2G. 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 39 773 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 40 Figure S2 Characterization of hTEM.v2, related to Figure 2 774 (A) Experimental design for timely and varied SAG concentrations, based on the 775 hTEM.v1 protocol. 776 (B) qPCR showing expression levels of marker genes for indicated cell types. Sample 777 indices are the same as those in (A). Data are presented as mean ±standard deviation. 778 Data were reproduced twice. 779 (C-D) Numbers of D-V patterned cells from somitic (C) and neural (D) clusters in 780 hTEM.v2 scRNA-seq data. Dorsal somite marker; PAX3. Ventral somite markers; 781 TWIST1, SCX, PAX1. Dorsal neural tube marker; PAX6. Ventral neural tube markers; 782 NKX6-1, NKX2-8, FOXA2. 783 (E-F) UMAP (E) and spatial UMAP (F) showing identified cell types from transversely 784 sectioned day-7 hTEM.v2 (n = 4). Transverse samples 1-4 are sections from posterior 785 to anterior positions. 786 (G-H) UMAP (G) and spatial UMAP (H) showing identified cell types from longitudinally 787 sectioned day-7 hTEM.v2 (n = 4). 788 (I) Spatially expressed TBX18 (anterior) and UNCX (posterior) indicating somite 789 segmentation in longitudinal section of day- 7hTEM.v2 (Visium HD sample 1). Signals of 790 TBX18 and UNCX were quantified and scaled along the A-P axis. 791 (J) UMAP trajectory showing notochord emergence in E7.25-7.5 mouse embryos (E-792 MTAB-6967). 793 (K) Dot plot showing the expression of signals emanating from indicated tissues during 794 early mouse embryogenesis (E-MTAB-6967). 795 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 41 (L) Dot plot showing the expression of signals emanating from the indicated tissues in 796 the human CS8 embryo (HRA005567). Plot was generated using the online tool in 797 http://cs8.3dembryo.com/. 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 42 819 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 43 Figure 3 Co-development of notochord, NMP-Neural, NMP-Meso and subsequent 820 neural and somitic lineages in hTEM.v3 821 (A) Schematic of hTEM.v3 generation. 822 (B) Representative images of hTEM.v3 at day 4 (left) and day 7 (right). 823 (C) Boxplot showing length measurements of hTEM.v3 from days 0-7 (n = 21-84 for 824 each time point) from 9 independent biological replicates (rep). Each dot represents an 825 individual hTEM.v3 at shown time point. 826 (D) Immunofluorescence image showing the A-P elongating notochord (FOXA2), caudal 827 and bilateral PSM (TBX6) and apical junctions (Phalloidin) within anterior cells. Scale 828 bar, 100 μm. 829 (E) HCR 3D projection showing the D-V layers of caudal neural plate cells (NKX1-2) 830 and notochord cells (CHRD) in day-4 hTEM.v3. Scale bar, 100 μm. 831 (F) HCR image showing transcripts of key genes (NOTO, SHH, FOXA2) expressed in 832 notochord cells in day-4 hTEM.v3. The posterior node area was zoomed. Yellow arrow, 833 triple positive. White arrow, double positive. Scale bar, 100 μm. 834 (G) Top, Immunofluorescence images of transversely sectioned day-5.5 hTEM.v3 835 showing dorsally positioned neural tube (SOX2) and ventrally positioned notochord 836 (NOTO:mClover3). Arrowhead, SOX2+NKX6-1+ ventral neural cells. Bottom, D-V 837 patterned neural tube (SOX2) and ventral notochord (TBXT) and bilateral somites (ZO-838 1). nt., neural tube. sm., somite. noto., notochord. Scale bars, 100 μm. 839 (H) SEM images of day-5.5 hTEM.v3 (n = 3). Wholemount view (top left), transversely 840 fractured viewpoint (top-middle). Zoomed views of the transversely fractured sample 841 showing; somite (top-right), neural tube (bottom-left), notochord with lipids (yellow 842 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 44 arrows) and vacuoles (blue arrows) (bottom middle) and ECM fibers (white arrow, 843 bottom-right). 844 (I) RNA velocity analysis based on the UMAP of hTEM.v3 (days 3-7) scRNA-seq 845 integration (total of 100,370 cells). Black arrows indicate calculated differentiation 846 directions. Yellow arrows indicate the two major developmental streams stemming from 847 bi-NMPs. 848 (J) Differential expression of key genes to distinguish tailbud (CDX2, CYP26A1)-derived 849 NMP-Neural (SOX2highTBXTlow, NKX1-2), NMP-Meso (SOX2lowTBXThigh, MSGN1) and 850 notochord (NOTO, SHH). 851 (K) Dot plot showing the expression of identified cell type markers in hTEM.v3. 852 Proportions below 5% were omitted. 853 (L) Normalized expression of respective lineage marker genes along the pseudotime 854 inferred A-P axis. Data are shown as coloured smooth spline with standard deviation in 855 grey shade. Vertical black lines at the bottom of the plots show cells co-expressing 856 SOX2 and TBXT. 857 (M) Heatmap showing scaled expression of genes in the RA pathway along the 858 pseudotime inferred A-P axis. 859 (N) Schematic diagram of RA signaling along the A-P axis of hTEM.v3. 860 861 862 863 864 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 45 865 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 46 Figure S3 Experimental design, morphological and cellular composition of 866 hTEM.v3, related to Figure 3 867 (A) Experimental design to test the conditions of timely BMP and NODAL inhibition for 868 notochord cells and bi-NMP induction. 869 (B) qPCR results showing marker levels of bi-NMPs, notochord and anterior primitive 870 streak under test conditions shown in (A). Data are presented as mean ±standard 871 deviation. Results were reproduced more than three times. 872 (C) Experimental design to test combinations of BMPs, FGFs and SHH for concordant 873 maintenance and differentiation of notochord, NMP-Neural and NMP-Meso. 874 (D) qPCR showing expression of indicated lineage markers for notochord (NOTO), 875 NMP-Neural (NKX1-2), NMP-Meso (TBX6) at day 4, from conditions listed in (C). Data 876 are presented as mean ±standard deviation. Results were reproduced more than three 877 times. 878 (E) qPCR results comparing expression levels of dorsal (PAX6) and ventral (NKX6-1, 879 SHH, FOXA2) markers in neural and notochord cells in day-6 hTEM.v3. Test conditions 880 are shown in (C). Data are presented as mean ±standard deviation. Results were 881 reproduced more than three times. 882 (G) Dot plot showing numbers of segmented somite pairs in hTEM.v3 at days 5-7. 883 (H) Representative images showing the caudal enrichment of NOTO:mClover3+ 884 notochord cells in the midline of hTEM.v3 at days 3-4. 885 (I) Representative images showing the gradual enrichment of NKX1-2:mScarlet+ caudal 886 neural plate progenitor cells in the midline of hTEM.v3 at days 3-4. 887 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 47 (I-J) Immunofluorescence images of transversely sectioned day-5.5 hTEM.v3 exhibiting 888 the bi-layered organization of dorsal neural plate (SOX2) and ventral notochord (TBXT, 889 NOTO, FOXJ1, FOXA2). 890 (K) SEM images of day-4 hTEM.v3 (n = 2); dorsal view. Regions of interest (R1/2/3) 891 were dash lined. Zoomed views of R1+R2 and R3 are displayed, highlighting the neural 892 plate and the ciliated ventral node. 893 (L) UMAP showing identified cell types from hTEM.v3 scRNA-seq integration (days 3-7, 894 total of 100,370 cells). 895 (M) Changes in cell type composition of hTEM.v3 from day 3-7. Colors of cell types 896 correspond to those in (L). 897 (N-P) RNA velocity (N), sub-clustering (O), and pseudotime (P) analysis on notochord 898 cells (n = 888) subset from hTEM.v3 scRNA-seq dataset in (L). 899 (Q) UMAP showing expression of markers for indicated notochord subtypes. 900 (R) Relative proportions of notochord subtypes identified from (N). 901 902 903 904 905 906 907 908 909 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 48 910 911 912 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 49 Figure S4 A-P organization of cell types and gene expression profiles inferred 913 from scRNA-seq integration of hTEM.v3 (days 3-7), related to Figure 3. 914 (A-B) Pseudotime analysis showing the progression of somitic (A) and neural lineages 915 (B). Cells analyzed here were subset from the hTEM.v3 (days 3-7) scRNA-seq dataset 916 in Figure S3L. 917 (C-F) Heatmap showing scaled expression of FGF (C), WNT (D), NOTCH (E) pathways 918 and the HOX (F) genes along pseudotime inferred A-P axis. 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 50 936 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 51 Figure 4 Molecular profiling of D-V organization and spatial patterning for 937 hTEM.v3 938 (A-B) RNA velocity analysis of neural (A) and somitic (B) lineage subclusters from 939 hTEM.v3 (day 3-7) scRNA-seq integration in Figure S3L. To distinguish somite 940 compartments, regression was performed on TBX18 and UNCX to remove somite A-P 941 patterns before re-clustering. 942 (C-D) Density plot showing trajectories of D-V patterned neural (C) and somitic (D) 943 lineages in hTEM.v3. Dorsal neural markers; PAX6, DBX2, and IRX5. Ventral neural 944 and floor plate markers; NKX6-1, FOXA2 and SHH. pan- and dorsal somite; PAX3. 945 Dermomyotome; MYF5. Endotome; EBF2. Syndetome; SCX. Sclerotome; PAX1. 946 Myogenic progenitors; PAX7. 947 (E) Comparison of hTEM.v1/2/3 cell types to contrast the presence or absence of D-V 948 fates. 949 (F) Immunofluorescence images of transverse (left) and longitudinal (right) sections of 950 day-6.5 hTEM.v3 showing the presence of D-V patterning in the neural tube structure 951 along D-V and A-P axes. Dorsal neural marker; PAX6. Ventral neural and floor plate 952 markers; OLIG2 and NKX6-1. Scale bars, 100 μm. 953 (G) Immunofluorescence images of transverse sections of day-6.5 hTEM.v3, showing 954 somite compartments along D-V axis. pan- and dorsal somite, PAX3. Endotome, 955 EBF2:mScarlet. Dermomyotome, MYF5:mClover3. Myogenic progenitors, PAX7. 956 Endothelial cells, SOX17. Scale bars, 100 μm. 957 (H) Circular heatmap showing the regulon representing TF genes in the indicated cell 958 types from hTEM.v3. 959 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 52 (I) Heatmap (left) of regulon modules based on cell type associated regulon activities in 960 Figure S5C, with the corresponding GO terms related to each regulon module 961 highlighted (right). 962 (J) Heatmap showing the enrichment of regulon modules across listed cell types. L- 963 Somite, late somitie, including sclerotome, syndetome, dermomyotome, myogenic 964 progenitors and endotome cells from hTEM.v3 scRNA-seq data in Figure S3L. 965 (K) NOTO gene regulatory networks in hTEM.v3. The top 15 interacting genes are 966 displayed. 967 (L) Spatial UMAP of RCTD cell type annotations in day-4 and day-6.5 hTEM.v3 (n = 8 968 each). Colors for annotation and cell types displayed are consistent with those in Figure 969 S6B-S6D. 970 (M-N) Spatial gene expression patterns in longitudinally sectioned day-4 (M) and day-971 6.5 (N) hTEM.v3. Dashed lines, areas for neural plate (day 4) and neural tube (day 6.5). 972 (O) Bubble plot showing the selected ligand-receptor interactions between indicated cell 973 types from hTEM.v3. Dot color represents the normalized sum of expression level of 974 ligand in source cells and interacting receptor in targeting cells. 975 976 977 978 979 980 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 53 981 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 54 Figure S5 Molecular profiles of hTEM.v3, related to Figure 4 982 (A-B) Normalized gene expression profiles of listed neural (A) and somitic (B) lineages 983 from hTEM.v3, reflecting the D-V axis formation in hTEM.v3. 984 (C) Heatmap showing regulon activities represented by selected transcription factors 985 (TFs) in 500 randomly sampled cells in hTEM.v3. 986 (D) UMAP showing the regulon module activity over trajectories for hTEM.v3. 987 (E) Gene Ontology (GO) enrichment analysis based on top ranking genes enriched in 988 regulon modules from Figure 4I. 989 (F) Gene regulatory networks of PAX6, NKX6-2, FOXA1 and FOXA2 in hTEM.v3. Top 990 15 interacted genes were displayed. 991 (G) Cell-cell communication patterns grouped by cell types from hTEM.v3 (days 3-7; 992 left). Significantly enriched signaling pathways belonging to each pattern (right). 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 55 1005 1006 1007 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 56 Figure S6 Spatially resolved hTEM.v3 cell type organization using Visium HD, 1008 related to Figure 4. 1009 (A) UMAP showing the primary cell types identified from Visium HD results of day-4 and 1010 day-6.5 hTEM.v3. 1011 (B) UMAP showing the RCTD cell types using reference annotations from hTEM.v3 1012 scRNA-seq data in Figure S3L. To reduce annotation complexity and to discern 1013 annotation colors in RCTD, ‘E-Somite’ and ‘M-Somite’ from scRNA-seq reference were 1014 combined as ‘pan-Somite’ in RCTD. ‘E-Neural tube’ and ‘Neural tube’ from scRNA-seq 1015

were merged as ‘Neural tube’ in RCTD, ‘E-Floor plate’ and ‘Floor plate’ from 1016 scRNA-seq reference were merged as ‘Floor plate’ in RCTD. Colors and annotations 1017 are listed at right side. 1018 (C-D) Spatial UMAP showing the RCTD annotated cell types from Visium HD resolved 1019 from day-4 (C) and day-6.5 (D) hTEM.v3. Colors and annotations are from (B). 1020 (E-F) UMAP and spatial expression patterns of indicated genes in longitudinally 1021 sectioned day-4 (E) and day-6.5 (F) hTEM.v3. Dashed lines, the neural plate (day 4) 1022 and neural tube (day 6.5) structures. 1023 (G-H) Spatial expression of HOXC6 and HOXC8 at day 4 (G) and day 6.5 (H) in 1024 hTEM.v3. Signals of HOXC family genes were quantified and scaled along the direction 1025 of P-A at the right side. 1026 1027 1028 1029 1030 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 57 1031 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 58 Figure 5 Developmental staging of hTEM.v3 resembles primate CS8-10 embryos 1032 (A) Top, sub-clustering UMAP revealing the notochord, NMP-Neural and NMP-Meso 1033 populations from human (CS7/8/10) and monkey (CS8/9/11) embryo integration 1034 dataset. Bottom, differential gene expression patterns delineating these cell types. 1035 (B) Sub-(top-right) and re-(top-right) clustering UMAPs revealing the notochord, NMP-1036 Neural and NMP-Meso from mouse embryo dataset (E7.0 to E8.5). Bottom, differential 1037 gene expression patterns delineating these cell types. 1038 (C) Heatmap of correlation co-efficient among cell types from human, monkey, mouse 1039 and hTEM.v3. The correlation was calculated and summarized from Figure S7F. 1040 (D) Sankey diagram showing relationships between selected cell types in hTEM.v3, 1041 human (CS7/8/10), monkey (CS8/9/11) and mouse embryos (E7.0 to E8.5). 1042 (E) Gene expression profiles for key transcription factors and key signaling pathways 1043 (SHH, WNT, BMP, NODAL, and FGF), in cells forming the posterior trunk in hTEM.v3, 1044 human, monkey and mouse embryo datasets. 1045 (F) Cartoon summary of similarities across human, monkey, mouse embryos with 1046 hTEM.v3 in developmental stages. 1047 1048 1049 1050 1051 1052 1053 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 59 1054 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 60 Figure S7 Cross-species comparison between hTEM.v3 and embryos from 1055 human, monkey, and mice, related to Figure 5 1056 (A) Ortholog UMAP trajectory of organogenesis in early human (CS7/8/10) and monkey 1057 (CS8/9/11) embryos. See ‘Methods’ for published human and monkey embryo scRNA-1058 seq datasets. 1059 (B) Left, hTEM.v3 projection onto public human-monkey (H-M) embryo integration 1060 dataset. Middle, H-M cell types overlaid by hTEM.v3. Colors and annotations for H-M 1061 are consistent with those in (A). Right, hTEM.v3 cell types overlapping with H-M. Colors 1062 and annotations for hTEM.v3 are consistent with Figure S3L. 1063 (C) Sankey plot showing a detailed cell type overlay between hTEM.v3 and H-M 1064 embryos dataset. Colors and annotations for H-M are consistent with those in (A). 1065 (D) Top-left, UMAP showing cell types in mouse embryos (E7.0-E8.5). Top-right, UMAP 1066 showing hTEM.v3 projection on mouse embryo data (E7.0-E8.5). Bottom-left, mouse 1067 cell types overlaid by hTEM.v3. Bottom-right, hTEM.v3 cell types overlapping with 1068 mouse embryos. Colors and annotations for hTEM.v3 are consistent with Figure S3L. 1069 (E) Sankey plot showing the detailed cell type overlay between hTEM.v3 and mouse 1070 embryo dataset. Colors and annotations for mouse embryos are consistent with those in 1071 (D). 1072 (F) Heatmap showing cluster similarities in listed cell types between hTEM.v3, H-M, and 1073 mouse embryos, respectively. The highest correlations are highlighted by black boxes. 1074 The second highest is in red box. 1075 (G) Sankey diagram showing relationships between selected cell types in hTEM.v3 with 1076 human (CS7/8/10), monkey (CS8/9/11) and mouse embryos (E7.0-E8.5). 1077 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 61 (H) Heatmap showing the scaled expression pattern of genes involved in RA, FGF and 1078 WNT signaling pathways in indicated cell types across datasets of hTEM.v3 and 1079 embryos. 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 62 1102 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 63 Figure 6 SHH signaling modulation and disease modeling using hTEM.v3 1103 (A) UMAP of integration of respective day-6 hTEM.v3 scRNA-seq data from WT (14,462 1104 cells), NOTO-LOF (16,081 cells), SANT1 (250 nM, days 4-6, 14,502 cells), and 1105 VANGL2-LOF (16,504 cells). 1106 (B) UMAP showing the re-analysis of notochord cells from WT, NOTO-LOF, and SANT1 1107 in (A). Pseudotime scores reflect the developmental direction indicating notochord 1108 maturation. 1109 (C) UMAP showing expression of marker genes indicating notochord maturation. 1110 (D) Changes in cellular composition from NOTO-LOF (top) and SANT1 treatment 1111 (bottom) in contrast to WT. Dots represents log2(NOTO-LOF/WT or SANT1/WT) fold 1112 changes for listed cell types. 1113 (E) Heatmap showing changes in the expression of key genes in regulation of 1114 notochord development, neural tube D-V patterning, FGF and NOTCH signaling upon 1115 NOTO-LOF and SANT1 treatment. 1116 (F) Whole-mount immunostaining of dorsal (PAX6) and ventral (NKX6-1) fate changes 1117 in neural tube (SOX2) in WT, NOTO-LOF and SANT1 treated day-6 hTEM.v3. nt., 1118 neural tube. Scale bars, 100 μm. 1119 (G) Spatial expression patterns of representative PCP genes in day-6.5 hTEM.v3 1120 (sample 2 in Figure 4L). Dashed lines, areas of neural tube. Black arrows, somite 1121 segments. 1122 (H) Comparison of length measurements using hTEM.v3 between WT, VANGL1-LOF, 1123 VANGL2-LOF and VANGL1/2-LOF (n ≥ 21 in each group at different times). ****p 1124 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 64 values < 0.0001 were calculated using students’ t-test. n.s., no statistical significance. 1125

were reproduced in more than 3 biological replicates. 1126 (I) Whole-mount Immunostaining of day-6 hTEM.v3 showing defects in neural tube 1127 elongation (SOX2) and somite (SIX1) segmentation using WT, VANGL1-LOF, VANGL2-1128 LOF and VANGL1/2-LOF lines. Arrow heads, somite segment. nt., neural tube. Scale 1129 bars, 100 μm. 1130 (J) Changes in cellular composition in VANGL2-LOF vs WT. Dots represents 1131 log2(VANGL2-LOF/WT) fold changes for listed cell types. 1132 (K) Sub-clustering of PSM, pan-Somite, Caud.NP and Neural tube from VANGL2-LOF 1133 and WT samples shown in (A). 1134 (L) UMAP showing PCP gene expressions involved VANGL2-related human neural tube 1135 defects. 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 65 1148 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 66 Figure S8 Morphological and molecular changes upon genetic and chemical 1149 perturbations in hTEM.v3, related to Figure 6 1150 (A) Schematic showing the CRISPR/Cas9 design and genotyping validations of NOTO-1151 LOF, VANGL1-LOF, VANGL2-LOF, and VANGL1/2-LOF hESC lines. 1152 (B) Representative images of day-6 hTEM.v3 in NOTO-LOF or, SANT1 treatment (250 1153 nM, days 4-6). Scale bars, 200 μm. 1154 (C) Comparison of length measurements of day-6 hTEM.v3 between WT, NOTO-LOF 1155 and SANT1 treatment (250 nM, days 4-6) (n = 6-8 each group). *p value < 0.05 and 1156 ****p value < 0.0001 was calculated by Student’s t test. n.s., no significance. The results 1157 were reproduced in more than 3 biological replicates. 1158 (D) qPCR showing changes in D-V patterning in WT, NOTO-LOF and SANT1 1159 treatments (100 nM or 250 nM at days 4-6) in day-6 hTEM.v3. Data are presented as 1160 mean ±standard deviation. Results were reproduced twice. 1161 (E-G) Violin plot showing changes in expression of genes related to notochord, 1162 endotome and endothelial cells in NOTO-LOF or, SANT1 treatment (250 nM, days 4-6). 1163 (H) Violin plot showing expression of VANGL1 and VANGL2 across listed cell types in 1164 hTEM.v3 dataset from Figure S3L. 1165 (I) Representative bright field images of day-6 hTEM.v3 in VANGL1-LOF, VANGL2-1166 LOF, or VANGL1/2-LOF. Scale bars, 200 μm. 1167 (J) Scaled expression levels of key genes involved in axial elongation, PCP and 1168 NOTCH signaling in day-6 hTEM.v3 dataset from Figure 6A. 1169 1170 1171 1172 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 67

1173 Culture of human ES and iPS cells 1174 The following human ESC and iPSC lines were used: wild type H9-hESC (Sex: female, 1175 WiCell, WAe009-A), K3-iPSC (Sex: male)108 generated from human neonatal foreskin 1176 fibroblasts (ATCC, PCS-201-010), H9-hESC line carrying IRES-mClover3 allele in the 3’ 1177 UTR of PAX3 and IRES-mScarlet allele in the 3’ UTR of EBF2 (PAX3:mClover3; 1178 EBF2:mScarlet dual reporter), H9-hESC line carrying IRES-mScarlet allele in the 3’ 1179 UTR of NKX1-2 (NKX1-2:mScarlet reporter), H9-hESC line carrying IRES-mClover3 1180 allele in the 3’ UTR of PAX3 and IRES-mScarlet allele in the 3’ UTR of NOTO 1181 (NOTO:mClover3 reporter), H9-hESC line carrying IRES-mClover3 allele in the 3’ UTR 1182 of MYF5 (MYF5:mClover3 reporter), and H9-hESC derived NOTO-LOF, VANGL1-LOF, 1183 VANGL2-LOF, and VANGL1/2-LOF (Figure S8A). H9-hESC lines and K3-iPSCs were 1184 routinely cultured and passaged as described previously.109,110 Briefly, 5 × 104/cm2 1185 cells were seeded onto culture-treated petri dishes coated with 1:200 diluted Geltrex 1186 (Thermo, A1413302). The basic culture media is in-house prepared using DMEM/F-12 1187 w/o glutamine (Thermo, 21331020 or Servicebio, G4514), supplemented with 0.5% 1188 Probumin (Sigma, 821001), 1x Antibiotic-Antimycotic (Thermo, 15240062), 1x MEM 1189 NEAA (Thermo, 11140050), 1x Trace Elements A/B/C (Corning), 64 μg/mL Ascorbic 1190 acid magnesium (TCI, A2521), 10 μg/mL Transferrin (Athens Research and 1191 Technology), and 1 x GlutaMax (Thermo, 35050061). To maintain pluripotency of 1192 hESCs and hiPSCs, the basic culture media is completed by addition of 10 ng/mL 1193 Heregulin β1 (Qkine, QK045), 10 ng/mL Activin A (Qkine, QK001), 8 ng/mL FGF2-G3 1194 (145aa) (Qkine, QK052) and 200 ng/mL IGF-1 LR3 (Qkine, QK041). The complete 1195 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 68 pluripotency maintenance medium is abbreviated as HAIF. hESC lines and K3-iPSCs 1196 were cultured in HAIF with media changes every 24 hours at 37 °C in 5% CO2 and 1197 passaged at 90% confluency using Accutase (Thermo, 00-4555-56). 1198 Generation of CRISPR/Cas9 knock-in fluorescent reporter H9-hESC lines 1199 To generate knock-in fluorescent reporter lines from H9-hESCs, we utilized the 1200 homology directed repair approach based on Cas9/gRNA introduced double strand 1201 break. The px330 vector expressing human codon-optimized Cas9 and sgRNA was 1202 obtained from Addgene. Oligos for sgRNA targets were individually cloned into px330 1203 using restriction enzyme BbsI. Homology directed repair (HDR) donor arms flanking the 1204 internal sequences of ribosome entry site (IRES) and fluorescent protein coding 1205 sequence (mClover3 or mScarlet) was synthesized and cloned into a pUC57 vector by 1206 BGI. The px330 with gRNAs targeting the gene of interest, homology donor arm vector 1207 and corresponding surrogate reporter vector (pRGS, PNA Bio) (vector ratio of 2:3:1, 1208 total of 9 μg vectors per 3 x 106 cells) were electroporated into H9-hESCs using a Neon 1209 electroporation system (voltage = 1050 V, width = 30 ms, pulse = 2 cycles) (Thermo, 1210 MPK10025). To enhance cell viability, 1x CEPT111 (in-house preparation) was used 1211 during electroporation and FACS sorting. Two days after electroporation, single cells 1212 were isolated by FACS sorted and plated on pMEF (Sigma, PMEF-NL-P1) coated 96-1213 well plates based on pRGS surrogate reporter signals. Single cell derived clones were 1214 screened by genotyping PCR, expanded in HAIF and subject to hTEM protocols. 1215 To generate CRISPR/Cas9 knock-out H9-hESC lines, gRNAs targeting upstream and 1216 downstream regions of exon(s) were used together with a surrogate reporter vector. 1217 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 69 Electroporation and single cell clone screening procedures were the same as 1218 generating knock-in cell lines. All oligos for sgRNA targets are listed in the Table S1. 1219 1220 Generation of human trunk embryoid models (hTEMs) 1221 Routinely 2D passaged hPSCs (H9-hESC lines and K3-iPSC) at 80% confluency were 1222 dissociated using Accutase for 5 min at 37 °C and counted by a hemocytometer. To 1223 start 3D spheroid formation, a total of 1-2 million cells in 5 mL of HAIF medium with 10 1224 μM Y-27632 (Aladdin) were seeded per well of an ultra-low attachment 6-well plate 1225 (Thermo, 174929). The plate was then placed on an orbital shaker (Thermo, 88881104) 1226 at 110-120 rpm at 37 °C in 5% CO2. The next day, a media change consisting of HAIF 1227 with 10 μM Y-27632 was applied. hPSC spheroids were allowed to form at size of 200-1228 220 μm in diameter within 40 hours of shaking. To generate hTEMs, hPSC spheroids 1229 were collected with a wide-bore 1 ml tip and transferred into a 1.5 mL Eppendorf tube, 1230 then washed with N2B27 basal medium. N2B27 basal medium comprises a 1:1 mix of 1231 DMEM/F12 and Neurobasal A (Thermo, 21103049) supplemented with 1×B27 (Thermo, 1232 17504001), 1× N2 (Thermo, 17502048), 1.5× GlutaMAX, 1×MEM NEAA, 1x Sodium 1233 Pyruvate (Thermo, 11360039), 1x Antibiotic-Antimycotic, and 64 μg/mL Ascorbic Acid 1234 Magnesium. The HAIF primed hPSC spheroids (day 0 of hTEM) were then individually 1235 transferred to a well of an ultra-low attachment 96 well plates (Thermo, 174929), subject 1236 to hTEM protocols. 1237 For hTEM.v1, hPSC spheroids (day 0) were induced by culture in bi-NMP induction 1238 media composed of N2B27 basal media and supplements of 10 μM CHIR-99021 1239 (CHIR) (MCE, HY10182), 500 nM LDN-193189 (LDN) (Tocris, 1517128), 10 μM SB-1240 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 70 431542 (SB) (Aladdin, S125924), and 20 ng/mL FGF2-G3. 48 hours later, N2B27 basal 1241 media was replenished. At day 4, media is replaced by N2B27 basal media 1242 supplemented with 4% Geltrex (Thermo, A1413302) of Matrigel (Mogengel, 827775) 1243 and 1 μM all-trans retinal (RAL) (Aladdin, A122355) to support somitogenesis and 1244 neural tube elongation. 1245 For hTEM.v2, 100 nM SAG (Aladdin, S872455) was added at day 4-5 based on the 1246 hTEM.v1 protocol, followed by replenishment of N2B27 basal medium containing 4% 1247 Matrigel and 1 μM RAL at day 5. 1248 For hTEM.v3, the hTEM.v1 day 0-2 protocol is modified by replacing SB with 20 ng/ml 1249 human recombinant CER1 (MCE, HY-P7822) for the first 24 hours. At day 2-3, media 1250 was replenished with N2B27 basal medium supplemented with 0.2 ng/mL SHH (R&D, 1251 8908-SH), 2 ng/mL BMP4 (R&D, 314-BP), 1 ng/ml BMP2 (Qkine, QK007), 0.5 ng/ml 1252 BMP7 (R&D, 354-BP), 8ng/ml FGF2-G3, 4 ng/ml FGF3/4/8b/17 (MCE, HY-1253 P700065/HY-P7014/HY-P70533/HY-P700060), 1 nM all-trans retinoic acid (RA) 1254 (Aladdin, R106320) and 0.2% Matrigel. At day 3-4, medium was changed to N2B27 1255 basal medium supplemented with 0.4 ng/ml SHH, 0.5 ng/ml BMP2/4/7, 10 ng/ml 1256 WNT5A (R&D, 645-WN) and 0.4% Matrigel. At days 4-7, medium was replaced with 1257 N2B27 medium containing 4% Matrigel and 200 nM RAL. 10 ng/ml WNT3A (R&D, 1258 5036-WN) and 0.2 ng/ml Heregulin β1 were included at days 4-7 to enhance neural 1259 tube genesis. 1260 All hTEM cultures were limited to 7 days due to accumulated cell death in the anterior 1261 region and ceased axial elongation. 1262 1263 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 71 Inclusion criteria of hTEM embryoids 1264 All hTEM embryoids were collected using wide-bore tips from ultra-low attachment 96 1265 well plates within 7 days of culture. Embryoids were inspected under a phase contrast 1266 microscope, based on morphometric features resembling human CS8-10 embryos. At 1267 day 3-4 (equivalent to CS8), embryoids undergo uniaxial symmetry breaking and 1268 become oval shaped. Embryoids with cylindrical morphology and visible mediolateral 1269 narrowing near the caudal end were selected for further use. The neural plate is to be 1270 visible as a groove along the midline extending from the posterior end and flanked by a 1271 shaded area anterior to the tailbud, reflecting the PSM cells (Figures 1B and 3B). If 1272 using the NOTO:mClover3 or NKX1-2:mScarlet reporter line, polarized and caudal 1273 accumulation of NOTO+ or NKX1-2+ cells at day 3 and axial distribution of these cells in 1274 the midline at days 4-5 (Figures S3F and S3G) were expected. This is equivalent to the 1275 C&E movements of caudal trunk progenitors in natural CS8 human embryos. Embryoids 1276 with a correct body plan require culture in media supplemented with 4% Matrigel over 1277 days 4-7 to allow somitogenesis and neural tube morphogenesis. At days 4-7 1278 (equivalent to CS9 to CS10) the following criteria were applied: (1) somites were to be 1279 clearly segmented after day 5; (2) presence of a lumen representing the closed neural 1280 tube was observed along the midline from hTEMs (Figures 1B, 2B, and 3B); (3) a single 1281 body axis with less-dense tailbud cell populations on the posterior end and high-dense 1282 unsegmented somite cells surrounding the neural tube tip on the anterior end. At day 4, 1283 ~70% of the embryoids with a correct body plan were collected for analysis or subject to 1284 further Matrigel culture. At days 5-7, ~30% of the hTEM.v1/2 and ~20% of hTEM.v3 1285 satisfied the criteria specified above. 1286 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 72 1287 Morphometric feature measurements 1288 The length of hTEMs were analyzed using ImageJ. To calibrate the length in pixels, a 1289 standard ruler is utilized to set the scale of images with the same magnification. The 1290 anterior-most and posterior-most points of the embryoids were set as start and end 1291 points for measurements. The length was measured along a custom defined midline 1292 along the direction of axial elongation. The counting of somite pairs was based on their 1293 order from the posterior end. The presence of two rows of somite segments flanking a 1294 neural tube structure was confirmed before counting all somite pairs along the A-P axis. 1295 1296 RNA extraction & RT-qPCR analysis 1297 hTEMs RNA was extracted using E.Z.N.A. MicroElute Total RNA Kit (OMEGA, R6831-1298 02). 500 ng of total RNA was reverse transcribed into cDNA using iScript Reverse 1299 Transcription Supermix (Bio-rad, 1708841) according to manufacturer’s instructions. 1300 Quantitative real-time PCR was performed using Taq Pro HS Universal Probe Master 1301 Mix (Vazyme, QN113-01) on QuantStudio 7 Pro Real-Time PCR System (Applied 1302 Biosystems, A43183). Expression levels for each gene re normalized to 18S rRNA 1303 check italics as an endogenous control using the ΔΔCt method. Taqman probes used in 1304 this study are listed in the Table S1. 1305 1306 Scanning electron microscopy (SEM) 1307 Embryoids were washed with DPBS to remove Matrigel, then fixed with 4% PFA for 30 1308 minutes at room temperature. Next, samples were treated with 1% osmium tetroxide in 1309 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 73 distilled water for 1 hour and then washed three times with distilled water for 10 minutes 1310 each. Samples were then dehydrated by serial washes in 70% to 100% ethanol. Critical 1311 point drying (CPD) was carried out using a Tourimis Samdri (PVT30) Critical Point 1312 Dryer. To prepare samples for imaging, an 80:20 platinum/palladium sputtering process 1313 was conducted with rotation, depositing a 3 nm conductive layer using the Quorum 1314 Q150T Automatic Coating System. Finally, SEM images were acquired using a Hitachi 1315 SU8010 cold-field emission scanning electron microscope at 9.8 kV. 1316 Whole-mount hTEM immunostaining and imaging 1317 hTEM embryoids were transferred to 1.5 ml Eppendorf tubes, fixed in 4% 1318 paraformaldehyde (PFA)/DPBS at RT for 1 hr and then washed with 1x DPBS at 10 min 1319 intervals for 30 min to remove residual PFA. For antigen retrieval, samples were 1320 immersed in warm 0.5% SDS/DPBS (preheated at 55 °C) for 15 mins, followed by 1321 incubation with 0.5% Trition X-100 in DPBS for 30 min. After blocking samples with 1322 Duolink blocking solution (Sigma) for 30 min at room temperature, samples were 1323 incubated with primary antibodies, diluted in MAXbind staining medium (Active Motif) 1324 overnight at 4 °C. Next, samples were washed three times with 0.04% Tween-20/DPBS 1325 (5 min each) and then incubated with DAPI (TCI, 1ug/ml) and Alexa Fluor secondary 1326 antibodies (Thermo, 1:500 dilution) in MAXbind staining medium for 2 h at room 1327 temperature. This was followed by one wash in 1 ml MAXwash washing medium (Active 1328 Motif) and two washes in 0.04 % Tween-20/DPBS (5 mins each) at room temperature. 1329 Optionally, a clearing procedure with Optimus Clearing Solution was used.112 Finally, 1330 samples were individually transferred into chamber slides (iBidi, 81811), cured by 1331 ProLong Gold antifade mountant (Thermo, P36934) overnight before imaging. All 1332 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 74 samples were imaged and analysed using a BZ-X All-in-one inverted fluorescent 1333 microscope (Keyence) or SP8 inverted confocal microscope (Leica). Antibodies used in 1334 this study are listed in the Key Resources Table. 1335 Cryosectioning of hTEMs 1336 hTEM embryoids were washed twice with ice-cold DPBS to remove Matrigel and fixed 1337 by 4% PFA, then dehydrated in 30% sucrose (Aladdin, S112234) overnight at 4°C. 1338 Samples were then transferred into a 1cm x 1cm cryomold, positioned and embedded in 1339 OCT compound (Tissue-tek, 4583). OCT embedded molds were snap frozen in dry ice 1340 and 95% isopentane (Macklin, I813377) and transferred to -80°C overnight before 1341 sectioning. The frozen and fixed samples were sectioned using a cryostat microtome 1342 (Leica, CM1950) at 10 μm. Sections were kept at -20°C before further analysis. All 1343 buffers in contact with hTEM embryoids were pre-treated with RNaseOUT (Thermo, 1344 10777019) to preserve RNA integrity for Visium HD process. 1345 Time-lapse imaging of hTEMs 1346 Bright-field and fluorescent images of live hTEMs were taken with a BZ-X All-in-one 1347 inverted fluorescent microscope (Keyence) in the ‘Time lapse capture’ mode using a 1348 10× plan objective. Images were captured using the z-stack mode at a step depth of 6-1349 10 μm, spanning a total of ~100 μm. For time-lapse imaging, the incubator module was 1350 set at 37 °C and 5% CO2 and images taken every 30  min or 60 min. To generate the 1351 time-lapsed video, stacked snapshots within the best focus range from each time point 1352 were used. 1353 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 75 1354 In situ hybridization chain reaction (HCR) 1355 hTEM embryoids were fixed in 4% PFA for 30 min at room temperature (RT), followed 1356 by one wash with ice-cold DPBS. Before HCR, fixed samples were pre-treated with 1357 0.5% SDS/DPBS for 15 min at RT, washed twice with 0.5% Triton X-100/DPBS and 1358 washed three times in 0.1% Tween-20/DPBS for 5 min each wash. HCR was performed 1359 following manual instructions (Molecular Instruments). In brief, samples were incubated 1360 in hybridisation buffer (HB) for 5 min at RT, then for 30 min at 37 °C. Probes were 1361 prepared at 8 nM in HB and incubated for 30 min at 37 °C before use. Samples were 1362 then incubated with probes for 12-16h on a thermal cycler (Bio-rad) at 37 °C. The next 1363 day, samples were washed with probe wash buffer (WB) three times for 15 min each at 1364 37 °C, then with 5x SSCT (5x SSC and 0.1% Tween-20 diluted in UltraPure Water) 1365 three times for 15 min each at RT. Samples were then pre-amplified in probe 1366 amplification buffer for >30 min at RT. Amplifier hairpins (h1 and h2) were prepared by 1367 heating at 95 °C for 90 sec separately followed by cooling down for 30 min at RT in 1368 dark. h1 and h2 were then mixed at 6 nM in amplification buffer and incubated with 1369 samples for 12-16 hrs at RT in the dark. Before imaging, samples were washed three 1370 times in SSCT for 15 min, and stained with DAPI. All HCR images were acquired and 1371 processed with a Nikon Ti2E inverted microscope. All HCR probes with associated 1372 hairpins are listed in the Table S1. 1373 Single-cell RNA-seq 1374 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 76 hTEMs (hTEM.v1; day3 = 35, day 4, n = 31, day 5, n = 22, day7, n = 11. hTEM.v2; day 1375 5, n = 22, day 7, n = 16; hTEM.v3; day 3, n = 35, day 4, n = 23, day 5, n = 12, day 6, n = 1376 9, day 7, n =14. NOTO-LOF; hTEM.v3 day 6, n = 11. SANT-1 (250 nM at days 4-6) 1377 hTEM.v3 day 6, n = 8. VANGL2-LOF; hTEM.v3 day 6, n =18.) were washed with N2B27 1378 basal medium twice and then dissociated using 1:1 TrypLE (Thermo) and Accutase at 1379 37 °C for 10-15 min. Single cells were counted and adjusted to 106/ml in 1 ml N2B27 1380 basal medium, filtered through a 40 μm cell strainer and loaded onto Chromium Single 1381 Cell 3’ Library and Gel Bead Kit v3.1 or v4 (10× Genomics). Following the cDNA-1382 amplification reaction, quality control and quantification was performed on the Agilent 1383 4200 Tapestation using the High Sensitivity D5000 kit (Agilent Technologies). Illumina 1384 sequencing libraries were constructed by fragmentation, end repair, A-tailing and 1385 double-sided size selection, adaptor ligation and sample-index PCR. Quality control and 1386 quantification of final libraries were performed on the Agilent 4200 Tapestation using the 1387 D1000 kit (Agilent Technologies) and Qubit 4 (Thermo Fisher Scientific). Libraries were 1388 then sequenced on NextSeq 2000 (Illumina) with a customized sequencing run format 1389 until sufficient saturation was reached. 1390 1391 Pre-processing of scRNA-seq reads 1392 Single-cell RNA-seq data generated in this study: 1393 All sequencing reads in this study were mapped to the reference genome GRCh38 1394 using cellranger (v9.0.1) with default parameters. Quality control and downstream 1395 analysis were performed with in R (v4.3.3) with Seurat (v4.3.0.1). For each dataset, 1396 cells with fewer than 200 genes expressed or > 5% expressed mitochondrial genes 1397 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 77 were removed. Doublets wre identified and removed using DoubletFinder (v2.0.4) 1398 implemented in scutilsR (v0.1.1). Next, ambient RNA contamination was estimated, 1399 counts were corrected using celda (v1.18.2) implemented in scutilsR (v0.1.1). Cells with 1400 less than 200 genes expressed after count correction were filtered out. Cell cycle phase 1401 scores were estimated using the corrected counts and these features were regressed 1402 out prior to dimension reduction and UMAP construction using Seurat. 1403 Public Single-cell RNA-seq data: 1404 Public and pre-processed datasets from E-MTAB-6967 (human CS7), HRA005567 1405 (human CS8), GSE155121 (human CS10), GSE193007 (Cynomolgus monkey 1406 CS8/9/11), E-MTAB-6967 (E7.0 to E8.5 with all known lineages within this period) were 1407 downloaded and included for cross-species analysis. Data from human and monkey 1408 embryos were integrated based on ortholog using biomaRt (v2.62.0) and Seurat’s 1409 default integration pipeline. Cell-cycle genes were regressed out using the matrix 1410 generated by ScaleData function, followed by dimension reduction, UMAP visualization, 1411 and clustering. Cells belonging to extra-embryonic tissues were excluded from 1412 downstream analysis to reduce the complexity of annotation and data visualization 1413 1414 Dataset integration and batch effect correction: 1415 To analyze scRNA-seq datasets created in this study and public H-M embryo data 1416 generated on different platforms, Seurat, BBKNN or Harmony were used for integration 1417 and batch effect correction, based on best performance. Each dataset from different 1418 experiments (this study or public) is considered a batch and contains at least one 1419 shared cell type. Briefly, log-normalized and scaled matrices from each dataset were 1420 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 78 integrated using the best-performed approaches. Harmony (v1.0.1) was used for 1421 hTEM.v1 (days 4/5/7) and hTEM.v2 (days 5/7) data integration. BBKNN (v1.1.1) was 1422 used for hTEM.v3 (days 3-7) data integration. The reciprocal PCA from Seurat was 1423 used to integrate ortholog human (CS7/8/10) and monkey (CS8/9/11) embryo datasets. 1424 The integration performance is justified given that the gene-expression profiles are from 1425 well-studied cell populations that closely match with the identified cell type datasets 1426 reported here. 1427 1428 RNA velocity analysis 1429 For RNA velocity analysis, the raw fastq sequence data using scvelo (v0.2.5) were 1430 firstly reanalyzed to obtain the count matrices containing the spliced and unspliced 1431 reads, followed by filtering out cells not subjected to UMAP projection and clustering 1432 analysis. Then, RNA velocity was analyzed with velocyto (v0.17.17) in the Python 3.7 1433 environment. Parameters were min_shared_counts= 50 and n_top_genes= 2000 for 1434 scv.pp. filter_and normalized function, n_pcs= 30 and n_neighbors= 30 for scv.pp. 1435 moments function. Mode was set to be stochastic when computing velocities. The 1436 velocity was projected to the UMAP generated previously. 1437 1438 Pseudotime analysis 1439 Before pseudotime analysis, the scaled matrix in Seurat was converted to the h5ad 1440 format using coverFormat function in sceasy (v0.0.7) package. Palantir (v1.3.3) was 1441 then used with default parameters in python (v3.9). Note that markers (NKX1-2 for 1442 neural, TBX6 for somitic and NODAL for notochord lineage) for differentiation initiation 1443 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 79 of each lineage were predetermined for each lineage. The resulting pseudotime scores 1444 were displayed based on UMAP embeddings from sub-clustering of each lineage to 1445 elucidate developmental dynamics. The A-P axis inference is based on transcriptomic 1446 profile of each lineage through time and cell fate specifications in a pseduotime ranked 1447 manner. 1448 1449 SCENIC analysis 1450 SCENIC (v1.3.1) analysis was performed on the integrated hTEM.v3 (days 3-7) scRNA-1451 seq data using R (v4.3.3) and Python (v3.13.2) with arboreto (v0.1.6). Firstly, SCENIC 1452 was performed based on motif annotations (hgnc v9) and cisTarget resources of human 1453 ‘hg38_refseq-r80_500bp_up_and_100bp_down_tss.mc9nr.feather’ and ‘hg38_refseq-1454 r80_10kb_up_and_down_tss.mc9nr.feather’ following default pipeline in 1455 https://htmlpreview.github.io/?https://github.com/aertslab/SCENIC/blob/master/inst/doc/1456 SCENIC_Running.html. Next, the predicated regulon activity scores (RAS) represented 1457 by major transcription factors were associated to cell types of hTEM.v3 and grouped by 1458 hierarchical clustering to generate the regulon module enrichment heatmap. To this 1459 heatmap, correlation matrix was calculated using scaled cell type-associated RAS and 1460 clustered by using ‘method = pearson’ from clusterProfiler (v4.14.4) in R. Genes from 1461 each regulon module were subject to Gene Ontology analysis and visualized using 1462 igraph (v2.0.3) in R. 1463 Cell-cell communication analysis 1464 CellChat (v2) was used to analyze intercellular communication following cluster 1465 annotation. The normalized gene expression matrix from the Seurat object was supplied 1466 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 80 as input and processed with default parameters. The CellChatDB. human database was 1467 used to infer the probability of cell-cell communication with “type = "truncatedMean", 1468 trim = 0.1”. Significant ligand–receptor pairs (p < 0.05) were identified and assigned to 1469 signaling pathways. For visualization of overall communication patterns, the summed 1470 expression profiles of ligands and receptors were integrated with the predicted 1471 communication probability. 1472 1473 Cross-species cell type projection and comparative analysis 1474 Single-cell transcriptomic datasets were obtained from CS8-CS11 cynomolgus monkey 1475 embryos and E7.0-E8.5 mouse embryos, together with the corresponding cell 1476 annotations. Gene symbols from monkey and mouse were converted to their human 1477 orthologues using the biomaRt package. To map hTEM.v3 cells onto UMAPs of H-M 1478 and mouse embryos, cells of hTEM.v3 were down-sampled to 20,000 and then 1479 projected using reference mapping approach described in Seurat pipeline 1480 https://satijalab.org/seurat/articles/integration_mapping. Briefly, the conserved label 1481 anchors between selected objects were identified using FindTransferAnchors function, 1482 followed by MapQuery function. The overlapped cell types between objects were 1483 determined by using TransferData function. Cross-species cell type projection was 1484 subsequently visualized in UMAP and Sankey plot to assess cell type conservation 1485 among datasets. ClusterSimilarity was used to determine the pairwise correlation co-1486 efficient across cell types from hTEM.v3 and embryos. 1487 1488 Preparation of Visium HD spatial gene expression assay 1489 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 81 PFA fixed and OCT frozen sectioning (hTEM.v2 and hTEM.v3) slides were prepared as 1490 described in the previous section. Sectioning follows the Visium HD Fixed Frozen 1491 Tissue Preparation Handbook (10x Genomics, CG000764) for the workflows. 1492 Sequencing was conducted by the Single Cell & Spatial Omics Core, School of 1493 Biomedical Sciences at The Chinese University of Hong Kong 1494 (https://www3.sbs.cuhk.edu.hk/en/core_laboratories/single-cell-omics-core/.). 1495 1496 Processing of Visium HD data and visualization 1497 Raw hTEM.v2 and hTEM.v3 Visium HD sequencing reads were mapped to the 1498

genome using spaceranger (v3.1.3) with default settings then, processed and 1499 analyzed with Seurat (5.2.1) running on R (4.4.2). For quality control, spots containing 1500 less than 700 total UMI counts were filtered out, as these spots represent areas outside 1501 of the actual tissue. Similarly, low-quality spots containing less than 200 total UMI 1502 counts and spots containing > 5% mitochondrial counts were filtered out. Following 1503 quality control, 3091 spots of transversely sectioned and 39212 spots of longitudinally 1504 sectioned day-7 hTEM.v2 were retained with a median transcript read out of 4180 and 1505 2122 unique molecular identifiers (UMI), respectively. For the spatial transcriptome of 1506 hTEM.v3, a total of 34,884 (median UMI 2,360) and 58,646 (median UMI 1,260) spots 1507 were retained from day 4 and day 6.5 sections, respectively. 1508 Filtered read counts were then processed following the standard Seurat protocol 1509 (https://satijalab.org/seurat/articles/visiumhd_analysis_vignette). Briefly, samples were 1510 normalized and scaled, followed by dimension reduction using PCA and UMAP. The 1511 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 82 FindNeighbors function was followed by RunUMAP function with 50 dimensions. The 1512 primary clusterings of Visium HD data were identified via the FindClusters function for 1513 crude cell type annotations. Unsupervised clustering was conducted using a bin size of 1514 8 μm and Seurat default settings. All other parameters used for each function were kept 1515 at default values. 1516 To deconvolute the spot-level data for accurate cell type annotations, the Robust Cell 1517 Type Decomposition (RCTD) approach from Seurat was applied using the reference cell 1518 types from integrated hTEM.v3 (days 3-7) scRNA-seq data. RCTD clusters were 1519 annotated by leveraging the cell type reference in Figure S3L. To visualize spatial 1520 UMAP of marker gene expression, customized scripts were devised using 1521 SpatialDimPlot function in Seurat for displaying weight normalized read counts listed in 1522 Figure 4L-4N, 6G, S5E-S5H. 1523 1524 Quantification and statistical analysis 1525 All statistical results and graphs were generated by Graphpad Prism. The numbers of 1526 samples and types of statistical analyses are given in the figure captions 1527 and results sections. 1528 1529 1530 1531 1532 1533 1534 1535 1536 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 83

1537 1538 1. Miao, Y., and Pourquié, O. (2024). Cellular and molecular control of vertebrate 1539 somitogenesis. Nature Reviews Molecular Cell Biology 25, 517-533. 1540 https://doi.org/10.1038/s41580-024-00709-z. 1541 2. Martinez Arias, A., Rivron, N., Moris, N., Tam, P ., Alev, C., Fu, J., Hadjantonakis, 1542 A.-K., Hanna, J.H., Minchiotti, G., and Pourquie, O. (2024). Criteria for the 1543 standardization of stem-cell-based embryo models. Nature cell biology 26, 1625-1544 1628. https://doi.org/10.1038/s41556-024-01492-x. 1545 3. Shao, Y., and Fu, J. (2022). Engineering multiscale structural orders for high-1546 fidelity embryoids and organoids. Cell Stem Cell 29, 722-743. 1547 https://doi.org/10.1016/j.stem.2022.04.003. 1548 4. Beccari, L., Moris, N., Girgin, M., Turner, D.A., Baillie-Johnson, P., Cossy, A.-C., 1549 Lutolf, M.P., Duboule, D., and Arias, A.M. (2018). Multi-axial self-organization 1550 properties of mouse embryonic stem cells into gastruloids. Nature 562, 272-276. 1551 https://doi.org/10.1038/s41586-018-0578-0. 1552 5. Xu, P.-F., Borges, R.M., Fillatre, J., de Oliveira-Melo, M., Cheng, T., Thisse, B., 1553 and Thisse, C. (2021). Construction of a mammalian embryo model from stem 1554 cells organized by a morphogen signalling centre. Nature communications 12, 1555 3277. https://doi.org/10.1038/s41467-021-23653-4. 1556 6. Li, H., Guan, W., Huang, J., Shen, P., Wu, J., Luo, H., Yang, Y., Ning, S., Chang, 1557 L., and Zhao, H. (2025). A complete model of mouse embryogenesis through 1558 organogenesis enabled by chemically induced embryo founder cells. Cell. 1559 https://doi.org/10.1016/j.cell.2025.07.018. 1560 7. Miao, Y., Djeffal, Y., De Simone, A., Zhu, K., Lee, J.G., Lu, Z., Silberfeld, A., Rao, 1561 J., Tarazona, O.A., and Mongera, A. (2023). Reconstruction and deconstruction 1562 of human somitogenesis in vitro. Nature 614, 500-508. 1563 https://doi.org/10.1038/s41586-022-05655-4. 1564 8. Yamanaka, Y., Hamidi, S., Yoshioka-Kobayashi, K., Munira, S., Sunadome, K., 1565 Zhang, Y., Kurokawa, Y., Ericsson, R., Mieda, A., and Thompson, J.L. (2023). 1566 Reconstituting human somitogenesis in vitro. Nature 614, 509-520. 1567 https://doi.org/10.1038/s41586-022-05649-2. 1568 9. Sanaki-Matsumiya, M., Matsuda, M., Gritti, N., Nakaki, F., Sharpe, J., Trivedi, V., 1569 and Ebisuya, M. (2022). Periodic formation of epithelial somites from human 1570 pluripotent stem cells. Nature communications 13, 2325. 1571 https://doi.org/10.1038/s41467-022-29967-1. 1572 10. Anand, G.M., Megale, H.C., Murphy, S.H., Weis, T., Lin, Z., He, Y., Wang, X., Liu, 1573 J., and Ramanathan, S. (2023). Controlling organoid symmetry breaking 1574 uncovers an excitable system underlying human axial elongation. Cell 186, 497-1575 512. e423. https://doi.org/10.1016/j.cell.2022.12.043. 1576 11. Gribaudo, S., Robert, R., van Sambeek, B., Mirdass, C., Lyubimova, A., Bouhali, 1577 K., Ferent, J., Morin, X., van Oudenaarden, A., and Nedelec, S. (2023). Self-1578 organizing models of human trunk organogenesis recapitulate spinal cord and 1579 spine co-morphogenesis. Nature biotechnology, 1-11. 1580 https://doi.org/10.1038/s41587-023-01956-9. 1581 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 84 12. Yaman, Y.I., and Ramanathan, S. (2023). Controlling human organoid symmetry 1582 breaking reveals signaling gradients drive segmentation clock waves. Cell 186, 1583 513-527. e519. https://doi.org/10.1016/j.cell.2022.12.042. 1584 13. Hamazaki, N., Yang, W., Kubo, C.A., Qiu, C., Martin, B.K., Garge, R.K., 1585 Regalado, S.G., Nichols, E.K., Pendyala, S., and Bradley, N. (2024). Retinoic 1586 acid induces human gastruloids with posterior embryo-like structures. Nature cell 1587 biology 26, 1790-1803. https://doi.org/10.1038/s41556-024-01487-8. 1588 14. Rito, T., Libby, A.R., Demuth, M., Domart, M.-C., Cornwall-Scoones, J., and 1589 Briscoe, J. (2024). Timely TGFβ signalling inhibition induces notochord. Nature 1590 637, 673-682. https://doi.org/10.1038/s41586-024-08332-w. 1591 15. Robles-Garcia, M., Thimonier, C., Angoura, K., Ozga, E., MacPherson, H., and 1592 Blin, G. (2024). In vitro modelling of anterior primitive streak patterning with 1593 human pluripotent stem cells identifies the path to notochord progenitors. 1594 Development 151, dev202983. https://doi.org/10.1242/dev.202983. 1595 16. del Corral, R.D., Olivera-Martinez, I., Goriely, A., Gale, E., Maden, M., and 1596 Storey, K. (2003). Opposing FGF and retinoid pathways control ventral neural 1597 pattern, neuronal differentiation, and segmentation during body axis extension. 1598 Neuron 40, 65-79. https://doi.org/10.1016/s0896-6273(03)00565-8. 1599 17. Dias, A., Pascual-Mas, P., Robertson, G., Torregrosa-Cortés, G., Stelloo, S., 1600 Casaní-Galdón, P., Babin, S., Romaniuk, Y., Mayran, A., and Wehmeyer, A.E. 1601 (2025). Opposing Nodal and Wnt signalling activities govern the emergence of 1602 the mammalian body plan. bioRxiv. https://doi.org/10.1101/2025.01.11.632562. 1603 18. Miura, S., Davis, S., Klingensmith, J., and Mishina, Y. (2006). BMP signaling in 1604 the epiblast is required for proper recruitment of the prospective paraxial 1605 mesoderm and development of the somites. Development 133, 1767-3775. 1606 https://doi.org/10.1242/dev.02552. 1607 19. Zhai, J., Guo, J., Wan, H., Qi, L., Liu, L., Xiao, Z., Yan, L., Schmitz, D.A., Xu, Y., 1608 and Yu, D. (2022). Primate gastrulation and early organogenesis at single-cell 1609 resolution. Nature 612, 732-738. https://doi.org/10.1038/s41586-022-05526-y. 1610 20. Xiao, Z., Cui, L., Yuan, Y., He, N., Xie, X., Lin, S., Yang, X., Zhang, X., Shi, P., 1611 and Wei, Z. (2024). 3D reconstruction of a gastrulating human embryo. Cell 187, 1612 2855-2874. e2819. https://doi.org/10.1016/j.cell.2024.03.041. 1613 21. Kondoh, H., and Takemoto, T. (2012). Axial stem cells deriving both posterior 1614 neural and mesodermal tissues during gastrulation. Current opinion in genetics & 1615 development 22, 374-380. https://doi.org/10.1016/j.gde.2012.03.006. 1616 22. Kimelman, D. (2016). Tales of tails (and trunks): forming the posterior body in 1617 vertebrate embryos. Current topics in developmental biology 116, 517-536. 1618 https://doi.org/10.1016/bs.ctdb.2015.12.008. 1619 23. Cambray, N., and Wilson, V. (2007). Two distinct sources for a population of 1620 maturing axial progenitors. Development 134, 2829-2840. 1621 https://doi.org/10.1242/dev.02877. 1622 24. Koch, F., Scholze, M., Wittler, L., Schifferl, D., Sudheer, S., Grote, P., 1623 Timmermann, B., Macura, K., and Herrmann, B.G. (2017). Antagonistic activities 1624 of Sox2 and brachyury control the fate choice of neuro-mesodermal progenitors. 1625 Developmental cell 42, 514-526. e517. 1626 https://doi.org/10.1016/j.devcel.2017.07.021. 1627 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 85 25. Morabito, R.D., Tatarakis, D., Swick, R., Stettnisch, S., Schilling, T.F., Horsfield, 1628 J.A., and Martin, B.L. (2025). The ratio of Wnt signaling activity to Sox2 1629 transcription factor levels predicts neuromesodermal fate potential. bioRxiv. 1630 https://doi.org/10.1101/2025.01.16.633481. 1631 26. Wood, T.R., Kyrsting, A., Stegmaier, J., Kucinski, I., Kaminski, C.F., Mikut, R., and 1632 Voiculescu, O. (2019). Neuromesodermal progenitors separate the axial stem 1633 zones while producing few single-and dual-fated descendants. BioRxiv, 622571. 1634 doi: https://doi.org/10.1101/622571. 1635 27. Yuan, Y., Wang, X., Yan, X., He, N., Lu, X., Yang, J., Xie, X., Yuan, H., Chen, N., 1636 and Liu, Y. (2025). 3D reconstruction of a human Carnegie stage 9 embryo 1637 provides a snapshot of early body plan formation. Cell Stem Cell 32, 1006-1024. 1638 e1005. https://doi.org/10.1016/j.stem.2025.04.007. 1639 28. Gunhaga, L., Jessell, T.M., and Edlund, T. (2000). Sonic hedgehog signaling at 1640 gastrula stages specifies ventral telencephalic cells in the chick embryo. 1641 Development 127, 3283-3293. https://doi.org/10.1242/dev.127.15.3283. 1642 29. Park, C.H.T., Pruitt, J.H., and Bennett, D. (1989). A mouse model for neural tube 1643 defects: the curtailed (Tc) mutation produces spina bifida occulta in Tc/+ animals 1644 and spina bifida with meningomyelocele in Tc/t. Teratology 39, 303-312. 1645 https://doi.org/10.1002/tera.1420390312. 1646 30. Kemmler, C.L., Smolikova, J., Moran, H.R., Mannion, B.J., Knapp, D., Lim, F., 1647 Czarkwiani, A., Hermosilla Aguayo, V., Rapp, V., and Fitch, O.E. (2023). 1648 Conserved enhancers control notochord expression of vertebrate Brachyury. 1649 Nature communications 14, 6594. https://doi.org/10.1038/s41467-023-42151-3. 1650 31. Liem, K.F., Tremml, G., Roelink, H., and Jessell, T.M. (1995). Dorsal 1651 differentiation of neural plate cells induced by BMP-mediated signals from 1652 epidermal ectoderm. Cell 82, 969-979. https://doi.org/10.1016/0092-1653 8674(95)90276-7. 1654 32. Tozer, S., Le Dréau, G., Marti, E., and Briscoe, J. (2013). Temporal control of 1655 BMP signalling determines neuronal subtype identity in the dorsal neural tube. 1656 Development 140, 1467-1474. https://doi.org/10.1242/dev.090118. 1657 33. Kahane, N., and Kalcheim, C. (2020). Neural tube development depends on 1658 notochord-derived sonic hedgehog released into the sclerotome. Development 1659 147, dev183996. https://doi.org/10.1242/dev.183996. 1660 34. Nguyen, P.D., Hollway, G.E., Sonntag, C., Miles, L.B., Hall, T.E., Berger, S., 1661 Fernandez, K.J., Gurevich, D.B., Cole, N.J., and Alaei, S. (2014). Haematopoietic 1662 stem cell induction by somite-derived endothelial cells controlled by meox1. 1663 Nature 512, 314-318. https://doi.org/10.1038/nature13678. 1664 35. Imaz-Rosshandler, I., Rode, C., Guibentif, C., Harland, L.T., Ton, M.-L.N., 1665 Dhapola, P., Keitley, D., Argelaguet, R., Calero-Nieto, F.J., and Nichols, J. (2024). 1666 Tracking early mammalian organogenesis–prediction and validation of 1667 differentiation trajectories at whole organism scale. Development 151, 1668 dev201867. https://doi.org/10.1242/dev.201867. 1669 36. Zeng, B., Liu, Z., Lu, Y., Zhong, S., Qin, S., Huang, L., Zeng, Y., Li, Z., Dong, H., 1670 and Shi, Y. (2023). The single-cell and spatial transcriptional landscape of human 1671 gastrulation and early brain development. Cell Stem Cell 30, 851-866. e857. 1672 https://doi.org/10.1016/j.stem.2023.04.016. 1673 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 86 37. Morimoto, M., Sasaki, N., Oginuma, M., Kiso, M., Igarashi, K., Aizaki, K.-i., 1674 Kanno, J., and Saga, Y. (2007). The negative regulation of Mesp2 by mouse 1675 Ripply2 is required to establish the rostro-caudal patterning within a somite. 1676 Development 134, 1561–1569. https://doi.org/10.1242/dev.000836. 1677 38. Kraus, F., Haenig, B., and Kispert, A. (2001). Cloning and expression analysis of 1678 the mouse T-box gene Tbx18. Mechanisms of development 100, 83-86. 1679 https://doi.org/10.1016/S0925-4773(00)00494-9. 1680 39. Neidhardt, L.M., Kispert, A., and Herrmann, B. (1997). A mouse gene of the 1681 paired-related homeobox class expressed in the caudal somite compartment and 1682 in the developing vertebral column, kidney and nervous system. Development 1683 genes and evolution 207, 330-339. https://doi.org/10.1007/s004270050120. 1684 40. Ensini, M., Tsuchida, T.N., Belting, H.-G., and Jessell, T.M. (1998). The control of 1685 rostrocaudal pattern in the developing spinal cord: specification of motor neuron 1686 subtype identity is initiated by signals from paraxial mesoderm. Development 1687 125, 969-982. https://doi.org/10.1242/dev.125.6.969. 1688 41. Swindell, E.C., Thaller, C., Sockanathan, S., Petkovich, M., Jessell, T.M., and 1689 Eichele, G. (1999). Complementary domains of retinoic acid production and 1690 degradation in the early chick embryo. Developmental biology 216, 282-296. 1691 https://doi.org/10.1006/dbio.1999.9487. 1692 42. Gibb, S., Maroto, M., and Dale, J.K. (2010). The segmentation clock mechanism 1693 moves up a notch. Trends in cell biology 20, 593-600. 1694 https://doi.org/10.1016/j.tcb.2010.07.001. 1695 43. Deschamps, J., and Duboule, D. (2017). Embryonic timing, axial stem cells, 1696 chromatin dynamics, and the Hox clock. Genes & development 31, 1406-1416. 1697 https://doi.org/10.1101/gad.303123.117. 1698 44. Sandell, L.L., Sanderson, B.W., Moiseyev, G., Johnson, T., Mushegian, A., 1699 Young, K., Rey, J.-P., Ma, J.-x., Staehling-Hampton, K., and Trainor, P.A. (2007). 1700 RDH10 is essential for synthesis of embryonic retinoic acid and is required for 1701 limb, craniofacial, and organ development. Genes & development 21, 1113-1124. 1702 https://doi.org/10.1101/gad.1533407. 1703 45. Pringle, N.P., and Richardson, W.D. (1993). A singularity of PDGF alpha-receptor 1704 expression in the dorsoventral axis of the neural tube may define the origin of the 1705 oligodendrocyte lineage. Development 117, 525-533. 1706 https://doi.org/10.1242/dev.117.2.525. 1707 46. Ikeya, M., and Takada, S. (1998). Wnt signaling from the dorsal neural tube is 1708 required for the formation of the medial dermomyotome. Development 125, 4969-1709 4976. https://doi.org/10.1242/dev.125.24.4969. 1710 47. Le Dréau, G., Garcia-Campmany, L., Rabadán, M.A., Ferronha, T., Tozer, S., 1711 Briscoe, J., and Martí, E. (2012). Canonical BMP7 activity is required for the 1712 generation of discrete neuronal populations in the dorsal spinal cord. 1713 Development 139, 259-268. https://doi.org/10.1242/dev.074948. 1714 48. Fontenas, L., and Kucenas, S. (2021). Spinal cord precursors utilize neural crest 1715 cell mechanisms to generate hybrid peripheral myelinating glia. Elife 10, e64267. 1716 https://doi.org/10.7554/elife.64267. 1717 49. Xi, H., Fujiwara, W., Gonzalez, K., Jan, M., Liebscher, S., Van Handel, B., 1718 Schenke-Layland, K., and Pyle, A.D. (2017). In vivo human somitogenesis guides 1719 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 87 somite development from hPSCs. Cell reports 18, 1573-1585. 1720 https://doi.org/10.1016/j.celrep.2017.01.040. 1721 50. Borycki, A.-G., Brunk, B., Tajbakhsh, S., Buckingham, M., Chiang, C., and 1722 Emerson Jr, C.P. (1999). Sonic hedgehog controls epaxial muscle determination 1723 through Myf5 activation. Development 126, 4053-4063. 1724 https://doi.org/10.1242/dev.126.18.4053. 1725 51. Yamada, T., Placzek, M., Tanaka, H., Dodd, J., and Jessell, T. (1991). Control of 1726 cell pattern in the developing nervous system: polarizing activity of the floor plate 1727 and notochord. Cell 64, 635-647. https://doi.org/10.1016/0092-8674(91)90247-V. 1728 52. Calhoun, J.D., Rao, R.R., Warrenfeltz, S., Rekaya, R., Dalton, S., McDonald, J., 1729 and Stice, S.L. (2004). Transcriptional profiling of initial differentiation events in 1730 human embryonic stem cells. Biochemical and biophysical research 1731 communications 323, 453-464. https://doi.org/10.1016/j.bbrc.2004.08.117. 1732 53. Shen, M.M. (2007). Nodal signaling: developmental roles and regulation. 1733 Development 134, 1023-1034. https://doi.org/10.1242/dev.000166. 1734 54. Pijuan-Sala, B., Griffiths, J.A., Guibentif, C., Hiscock, T.W., Jawaid, W., Calero-1735 Nieto, F.J., Mulas, C., Ibarra-Soria, X., Tyser, R.C., and Ho, D.L.L. (2019). A 1736 single-cell molecular map of mouse gastrulation and early organogenesis. Nature 1737 566, 490-495. https://doi.org/10.1038/s41586-019-0933-9. 1738 55. Luo, T., Liu, C., Cheng, T., Zhao, G.-Q., Huang, Y., Luan, J.-Y., Guo, J., Liu, X., 1739 Wang, Y.-F., and Dong, Y. (2025). Establishing dorsal-ventral patterning in human 1740 neural tube organoids with synthetic organizers. Cell Stem Cell 32, 1071-1741 1086.E1078. https://doi.org/10.1016/j.stem.2025.04.011. 1742 56. Ybot-Gonzalez, P., Gaston-Massuet, C., Girdler, G., Klingensmith, J., Arkell, R., 1743 Greene, N.D., and Copp, A.J. (2007). Neural plate morphogenesis during mouse 1744 neurulation is regulated by antagonism of Bmp signalling. Development 134, 1745 3203–3211. https://doi.org/10.1242/dev.008177. 1746 57. Sharma, R., Shafer, M.E., Bareke, E., Tremblay, M., Majewski, J., and Bouchard, 1747 M. (2017). Bmp signaling maintains a mesoderm progenitor cell state in the 1748 mouse tailbud. Development 144, 2982-2993. 1749 https://doi.org/10.1242/dev.149955. 1750 58. Krammer, T., Stuart, H.T., Gromberg, E., Ishihara, K., Cislo, D., Melchionda, M., 1751 Perez, F.B., Wang, J., Costantini, E., and Lehr, S. (2024). Mouse neural tube 1752 organoids self-organize floorplate through BMP-mediated cluster competition. 1753 Developmental Cell 59, 1940-1953. e1910. 1754 https://doi.org/10.1016/j.devcel.2024.04.021. 1755 59. Diez del Corral, R., and Morales, A.V. (2017). The multiple roles of FGF signaling 1756 in the developing spinal cord. Frontiers in cell and developmental biology 5, 58. 1757 https://doi.org/10.3389/fcell.2017.00058. 1758 60. Ghyselinck, N.B., and Duester, G. (2019). Retinoic acid signaling pathways. 1759 Development 146, dev167502. https://doi.org/10.1242/dev.167502. 1760 61. Cao, Y., Zhao, J., Sun, Z., Zhao, Z., Postlethwait, J., and Meng, A. (2004). fgf17b, 1761 a novel member of Fgf family, helps patterning zebrafish embryos. 1762 Developmental biology 271, 130-143. 1763 https://doi.org/10.1016/j.ydbio.2004.03.032. 1764 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 88 62. Naiche, L., Holder, N., and Lewandoski, M. (2011). FGF4 and FGF8 comprise the 1765 wavefront activity that controls somitogenesis. Proceedings of the National 1766 Academy of Sciences 108, 4018-4023. https://doi.org/10.1073/pnas.1007417108. 1767 63. McCann, M.R., Tamplin, O.J., Rossant, J., and Séguin, C.A. (2012). Tracing 1768 notochord-derived cells using a Noto-cre mouse: implications for intervertebral 1769 disc development. Disease Models & Mechanisms 5, 73-82. 1770 https://doi.org/10.1242/dmm.008128. 1771 64. Lee, J.D., and Anderson, K.V. (2008). Morphogenesis of the node and notochord: 1772 the cellular basis for the establishment and maintenance of left–right asymmetry 1773 in the mouse. Developmental Dynamics 237, 3464-3476. 1774 https://doi.org/10.1002/dvdy.21598. 1775 65. Adams, D.S., Keller, R., and Koehl, M. (1990). The mechanics of notochord 1776 elongation, straightening and stiffening in the embryo of Xenopus laevis. 1777 Development 110, 115-130. https://doi.org/10.1242/dev.110.1.115. 1778 66. Bancroft, M., and Bellairs, R. (1976). The development of the notochord in the 1779 chick embryo, studied by scanning and transmission electron microscopy. 1780 Development 35, 383-401. https://doi.org/10.1242/dev.35.2.383. 1781 67. Leeson, T., and Leeson, C. (1958). Observations on the histochemistry and fine 1782 structure of the notochord in rabbit embryos. Journal of anatomy 92, 278. 1783 68. Cui, L., Lin, S., Yang, X., Xie, X., Wang, X., He, N., Yang, J., Zhang, X., Lu, X., 1784 and Yan, X. (2025). Spatial transcriptomic characterization of a Carnegie stage 7 1785 human embryo. Nature Cell Biology 27, 360-369. https://doi.org/10.1038/s41556-1786 024-01597-3. 1787 69. Sweetman, D., Wagstaff, L., Cooper, O., Weijer, C., and Münsterberg, A. (2008). 1788 The migration of paraxial and lateral plate mesoderm cells emerging from the late 1789 primitive streak is controlled by different Wnt signals. BMC developmental biology 1790 8, 63. https://doi.org/10.1186/1471-213X-8-63. 1791 70. Aulehla, A., and Pourquié, O. (2010). Signaling gradients during paraxial 1792 mesoderm development. Cold Spring Harbor perspectives in biology 2, a000869. 1793 https://doi.org/10.1101/cshperspect.a000869. 1794 71. Gouti, M., Delile, J., Stamataki, D., Wymeersch, F.J., Huang, Y., Kleinjung, J., 1795 Wilson, V., and Briscoe, J. (2017). A gene regulatory network balances neural 1796 and mesoderm specification during vertebrate trunk development. 1797 Developmental cell 41, 243-261. e247. 1798 https://doi.org/10.1016/j.devcel.2017.04.002. 1799 72. Diez del Corral, R., and Morales, A.V. (2014). Retinoic acid signaling during early 1800 spinal cord development. Journal of Developmental Biology 2, 174-197. 1801 https://doi.org/10.3390/jdb2030174. 1802 73. Peyrot, S.M., Wallingford, J.B., and Harland, R.M. (2011). A revised model of 1803 Xenopus dorsal midline development: differential and separable requirements for 1804 Notch and Shh signaling. Developmental biology 352, 254-266. 1805 https://doi.org/10.1016/j.ydbio.2011.01.021. 1806 74. Lukaszewicz, A.I., and Anderson, D.J. (2011). Cyclin D1 promotes neurogenesis 1807 in the developing spinal cord in a cell cycle-independent manner. Proceedings of 1808 the National Academy of Sciences 108, 11632-11637. 1809 https://doi.org/10.1073/pnas.1106230108. 1810 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 89 75. Marinopoulou, E., Biga, V., Sabherwal, N., Miller, A., Desai, J., Adamson, A.D., 1811 and Papalopulu, N. (2021). HES1 protein oscillations are necessary for neural 1812 stem cells to exit from quiescence. Iscience 24. 1813 https://doi.org/10.1016/j.isci.2021.103198. 1814 76. Bi, W., Deng, J.M., Zhang, Z., Behringer, R.R., and De Crombrugghe, B. (1999). 1815 Sox9 is required for cartilage formation. Nature genetics 22, 85-89. 1816 https://doi.org/10.1038/8792. 1817 77. Stelloo, S., Alejo-Vinogradova, M.T., van Gelder, C.A., Zijlmans, D.W., van 1818 Oostrom, M.J., Valverde, J.M., Lamers, L.A., Rus, T., Alcaraz, P.S., and Schäfers, 1819 T. (2024). Deciphering lineage specification during early embryogenesis in 1820 mouse gastruloids using multilayered proteomics. Cell Stem Cell 31, 1072-1090. 1821 e1078. https://doi.org/10.1016/j.stem.2024.04.017. 1822 78. Aibar, S., González-Blas, C.B., Moerman, T., Huynh-Thu, V.A., Imrichova, H., 1823 Hulselmans, G., Rambow, F., Marine, J.-C., Geurts, P., and Aerts, J. (2017). 1824 SCENIC: single-cell regulatory network inference and clustering. Nature methods 1825 14, 1083-1086. https://doi.org/10.1038/nmeth.4463. 1826 79. Alten, L., Schuster-Gossler, K., Beckers, A., Groos, S., Ulmer, B., Hegermann, J., 1827 Ochs, M., and Gossler, A. (2012). Differential regulation of node formation, nodal 1828 ciliogenesis and cilia positioning by Noto and Foxj1. Development 139, 1276-1829 1284. https://doi.org/10.1242/dev.072728. 1830 80. Manning, C.S., Biga, V., Boyd, J., Kursawe, J., Ymisson, B., Spiller, D.G., 1831 Sanderson, C.M., Galla, T., Rattray, M., and Papalopulu, N. (2019). Quantitative 1832 single-cell live imaging links HES5 dynamics with cell-state and fate in murine 1833 neurogenesis. Nature communications 10, 2835. https://doi.org/10.1038/s41467-1834 019-10734-8. 1835 81. Zhou, Q., and Anderson, D.J. (2002). The bHLH transcription factors OLIG2 and 1836 OLIG1 couple neuronal and glial subtype specification. Cell 109, 61-73. 1837 https://doi.org/10.1016/S0092-8674(02)00677-3. 1838 82. Cho, G., Lim, Y., Cho, I.-T., Simonet, J.C., and Golden, J.A. (2014). Arx together 1839 with FoxA2, regulates Shh floor plate expression. Developmental biology 393, 1840 137-148. https://doi.org/10.1016/j.ydbio.2014.06.012. 1841 83. Navis, A., Marjoram, L., and Bagnat, M. (2013). Cftr controls lumen expansion 1842 and function of Kupffer’s vesicle in zebrafish. Development 140, 1703-1712. 1843 https://doi.org/10.1242/dev.091819. 1844 84. Tyser, R.C., Mahammadov, E., Nakanoh, S., Vallier, L., Scialdone, A., and 1845 Srinivas, S. (2021). Single-cell transcriptomic characterization of a gastrulating 1846 human embryo. Nature 600, 285-289. https://doi.org/10.1038/s41586-021-04158-1847 y. 1848 85. Morimoto, Y., Yoshida, S., Kinoshita, A., Satoh, C., Mishima, H., Yamaguchi, N., 1849 Matsuda, K., Sakaguchi, M., Tanaka, T., and Komohara, Y. (2019). Nonsense 1850 mutation in CFAP43 causes normal-pressure hydrocephalus with ciliary 1851 abnormalities. Neurology 92, e2364-e2374. 1852 https://doi.org/10.1212/wnl.0000000000007505. 1853 86. Vallstedt, A., Muhr, J., Pattyn, A., Pierani, A., Mendelsohn, M., Sander, M., 1854 Jessell, T.M., and Ericson, J. (2001). Different levels of repressor activity assign 1855 redundant and specific roles to Nkx6 genes in motor neuron and interneuron 1856 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 90 specification. Neuron 31, 743-755. https://doi.org/10.1016/S0896-1857 6273(01)00412-3. 1858 87. Kong, J.H., Yang, L., Dessaud, E., Chuang, K., Moore, D.M., Rohatgi, R., 1859 Briscoe, J., and Novitch, B.G. (2015). Notch activity modulates the 1860 responsiveness of neural progenitors to sonic hedgehog signaling. 1861 Developmental cell 33, 373-387. https://doi.org/10.1016/j.devcel.2015.03.005. 1862 88. Cable, D.M., Murray, E., Zou, L.S., Goeva, A., Macosko, E.Z., Chen, F., and 1863 Irizarry, R.A. (2022). Robust decomposition of cell type mixtures in spatial 1864 transcriptomics. Nature biotechnology 40, 517-526. 1865 https://doi.org/10.1038/s41587-021-00830-w. 1866 89. White, P.H., Farkas, D.R., McFadden, E.E., and Chapman, D.L. (2003). Defective 1867 somite patterning in mouse embryos with reduced levels of Tbx6. Development 1868 130, 1681–1690. https://doi.org/10.1242/dev.00367. 1869 90. Jin, S., Guerrero-Juarez, C.F., Zhang, L., Chang, I., Ramos, R., Kuan, C.-H., 1870 Myung, P., Plikus, M.V., and Nie, Q. (2021). Inference and analysis of cell-cell 1871 communication using CellChat. Nature communications 12, 1088. 1872 https://doi.org/10.1038/s41467-021-21246-9. 1873 91. Wang, Y., and Serra, R. (2012). PDGF mediates TGFβ-induced migration during 1874 development of the spinous process. Developmental biology 365, 110-117. 1875 https://doi.org/10.1016/j.ydbio.2012.02.013. 1876 92. Forlani, S., Lawson, K.A., and Deschamps, J. (2003). Acquisition of Hox codes 1877 during gastrulation and axial elongation in the mouse embryo. Development 130, 1878 3807–3819. https://doi.org/10.1242/dev.00573. 1879 93. Deschamps, J., and van Nes, J. (2005). Developmental regulation of the Hox 1880 genes during axial morphogenesis in the mouse. Development 132, 2931-2942. 1881 https://doi.org/10.1242/dev.01897. 1882 94. Lawrence, J.E., Roberts, K., Tuck, E., Li, T., Mamanova, L., Balogh, P., Usher, I., 1883 Piapi, A., Mazin, P., and Anderson, N.D. (2024). HOX gene expression in the 1884 developing human spine. Nature communications 15, 10023. 1885 https://doi.org/10.1038/s41467-024-54187-0. 1886 95. Rominger, C.M., Bee, W.-L.T., Copeland, R.A., Davenport, E.A., Gilmartin, A., 1887 Gontarek, R., Hornberger, K.R., Kallal, L.A., Lai, Z., and Lawrie, K. (2009). 1888 Evidence for allosteric interactions of antagonist binding to the smoothened 1889 receptor. The Journal of pharmacology and experimental therapeutics 329, 995-1890 1005. https://doi.org/10.1124/jpet.109.152090. 1891 96. Moran, C.M., Salanga, M.C., and Krieg, P.A. (2011). Hedgehog signaling 1892 regulates size of the dorsal aortae and density of the plexus during avian 1893 vascular development. Developmental Dynamics 240, 1354-1364. 1894 https://doi.org/10.1002/dvdy.22600. 1895 97. Kibar, Z., Torban, E., McDearmid, J.R., Reynolds, A., Berghout, J., Mathieu, M., 1896 Kirillova, I., De Marco, P., Merello, E., and Hayes, J.M. (2007). Mutations in 1897 VANGL1 associated with neural-tube defects. New England Journal of Medicine 1898 356, 1432-1437. https://doi.org/10.1056/nejmoa060651. 1899 98. Lei, Y.-P., Zhang, T., Li, H., Wu, B.-L., Jin, L., and Wang, H.-Y. (2010). VANGL2 1900 mutations in human cranial neural-tube defects. New England Journal of 1901 Medicine 362, 2232-2235. https://doi.org/10.1056/nejmc0910820. 1902 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 91 99. Feng, X., Ye, Y., Zhang, J., Zhang, Y., Zhao, S., Mak, J.C., Otomo, N., Zhao, Z., 1903 Niu, Y., and Yonezawa, Y. (2024). Core planar cell polarity genes VANGL1 and 1904 VANGL2 in predisposition to congenital vertebral malformations. Proceedings of 1905 the National Academy of Sciences 121, e2310283121. 1906 https://doi.org/10.1073/pnas.2310283121. 1907 100. Song, H., Hu, J., Chen, W., Elliott, G., Andre, P., Gao, B., and Yang, Y. (2010). 1908 Planar cell polarity breaks bilateral symmetry by controlling ciliary positioning. 1909 Nature 466, 378-382. https://doi.org/10.1038/nature09129. 1910 101. Jülich, D., Cobb, G., Melo, A.M., McMillen, P., Lawton, A.K., Mochrie, S.G., 1911 Rhoades, E., and Holley, S.A. (2015). Cross-scale integrin regulation organizes 1912 ECM and tissue topology. Developmental cell 34, 33-44. 1913 https://doi.org/10.1016/j.devcel.2015.05.005. 1914 102. Mole, M.A., Galea, G.L., Rolo, A., Weberling, A., Nychyk, O., De Castro, S.C., 1915 Savery, D., Fässler, R., Ybot-Gonzalez, P., and Greene, N.D. (2020). Integrin-1916 mediated focal anchorage drives epithelial zippering during mouse neural tube 1917 closure. Developmental Cell 52, 321-334. e326. 1918 https://doi.org/10.1016/j.devcel.2020.01.012. 1919 103. Jessen, T.N., and Jessen, J.R. (2017). VANGL2 interacts with integrin αv to 1920 regulate matrix metalloproteinase activity and cell adhesion to the extracellular 1921 matrix. Experimental Cell Research 361, 265-276. 1922 https://doi.org/10.1016/j.yexcr.2017.10.026. 1923 104. Nagaoka, T., Furuse, M., Ohtsuka, T., Tsuchida, K., and Kishi, M. (2019). Vangl2 1924 interaction plays a role in the proteasomal degradation of Prickle2. Scientific 1925 reports 9, 2912. https://doi.org/10.1038/s41598-019-39642-z. 1926 105. Wang, L., Xiao, Y., Tian, T., Jin, L., Lei, Y., Finnell, R.H., and Ren, A. (2018). 1927 Digenic variants of planar cell polarity genes in human neural tube defect 1928 patients. Molecular genetics and metabolism 124, 94-100. 1929 https://doi.org/10.1016/j.ymgme.2018.03.005. 1930 106. Ossipova, O., Kim, K., and Sokol, S.Y. (2015). Planar polarization of Vangl2 in 1931 the vertebrate neural plate is controlled by Wnt and Myosin II signaling. Biology 1932 open 4, 722-730. https://doi.org/10.1242/bio.201511676. 1933 107. Catala, M. (2021). Overview of secondary neurulation. Journal of Korean 1934 Neurosurgical Society 64, 346-358. https://doi.org/10.3340/jkns.2020.0362. 1935 108. Si-Tayeb, K., Noto, F.K., Sepac, A., Sedlic, F., Bosnjak, Z.J., Lough, J.W., and 1936 Duncan, S.A. (2010). Generation of human induced pluripotent stem cells by 1937 simple transient transfection of plasmid DNA encoding reprogramming factors. 1938 BMC developmental biology 10, 81. https://doi.org/10.1186/1471-213X-10-81. 1939 109. Singh, A.M., Sun, Y., Li, L., Zhang, W., Wu, T., Zhao, S., Qin, Z., and Dalton, S. 1940 (2015). Cell-cycle control of bivalent epigenetic domains regulates the exit from 1941 pluripotency. Stem cell reports 5, 323-336. 1942 https://doi.org/10.1016/j.stemcr.2015.07.005. 1943 110. Singh, A.M., Zhang, L., Avery, J., Yin, A., Du, Y., Wang, H., Li, Z., Fu, H., Yin, H., 1944 and Dalton, S. (2020). Human beige adipocytes for drug discovery and cell 1945 therapy in metabolic diseases. Nature communications 11, 2758. 1946 https://doi.org/10.1038/s41467-020-16340-3. 1947 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint 92 111. Tristan, C.A., Hong, H., Jethmalani, Y., Chen, Y., Weber, C., Chu, P .-H., Ryu, S., 1948 Jovanovic, V.M., Hur, I., and Voss, T.C. (2023). Efficient and safe single-cell 1949 cloning of human pluripotent stem cells using the CEPT cocktail. Nature 1950 protocols 18, 58-80. https://doi.org/10.1038/s41596-022-00753-z. 1951 112. Kim, K., Na, M., Oh, K., Cho, E., Han, S.S., and Chang, S. (2022). Optimized 1952 single-step optical clearing solution for 3D volume imaging of biological 1953 structures. Communications biology 5, 431. https://doi.org/10.1038/s42003-022-1954 03388-8. 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 22, 2025. ; https://doi.org/10.64898/2025.12.20.695666doi: bioRxiv preprint