Molecular cues from distinct neuron classes drive differential myelination in the neocortex

SUMMARY

Oligodendrocytes deposit different amounts of myelin in each neocortical layer, but the regulating process remains unclear. We present a single-cell map of oligodendrocyte lineage cells purified from different layers of the neocortex across developmental stages. We find that each layer contains a similar compendium of oligodendrocyte classes, and differ primarily in the proportions of maturation states, suggesting that oligodendrocyte heterogeneity cannot explain layer-specific myelination. We show that signals from different classes of pyramidal neurons can control oligodendrocyte maturation and differential distribution of myelin across cortical layers. We generated a ligand-receptor interactome to predict interactions between projection neuron types and oligodendrocyte states, across cortical layers and time, and validated candidates in vivo. We find that neuronal expression of Fgf18, Ncam1, and Rspo3 promotes cortical myelination. This work provides a comprehensive molecular description of oligodendrocyte development in the mouse cortex, pointing at mechanisms whereby neuron-class-linked signals modulate myelin distribution in the neocortex.

eTOC Blurb

Domínguez-Iturza et al. characterize oligodendrocyte lineage diversity across cortical layers and developmental stages using single-cell RNA sequencing and generate a molecular interactome to uncover neuronal signals affecting myelination. In vivo experiments identify two signals that regulate cortical myelination, providing insight into the mechanisms that govern myelin patterning in the neocortex.

Graphical Abstract

INTRODUCTION

Over the course of vertebrate evolution, the development of the myelin sheath has contributed to the expansion of the central nervous system (CNS) and the emergence of complex brain function. In the CNS, the lipid-rich myelin membrane is produced by oligodendrocytes (OL) that wrap the axons of neurons, forming myelinated segments (internodes), required to increase the velocity of action potentials and thereby enhance the efficiency of long-distance neuronal transmission^1,2. Disrupted myelination can lead to many debilitating neurological disorders, including multiple sclerosis (MS) and schizophrenia^3,4. Yet, our understanding of myelin formation and regulation at the molecular level is still poor. The observation that oligodendrocytes can wrap myelin around paraformaldehyde-fixed axons and inorganic substrates^5–7 suggests that oligodendrocytes may be non-selective, and use physical cues such as axon caliber size as main determinants of myelination. However, the in vivo ensheathment of axons within neuronal networks requires precise patterning and spatial organization of myelin. In vivo, only axons become myelinated, while glial cells, vascular structures, neuronal dendrites, or cell bodies remain unmyelinated, and different neuronal subtypes show class-specific myelination patterns^8–11. This supports a model of regulated interaction between oligodendrocytes and neurons to finely control myelin production and distribution.

The cortex is organized into six radially-oriented layers which are marked by the presence of different projection neuron (PN) populations. Myelination in the mouse cortex initiates in the deep layers around P10 and progresses into the upper layers over postnatal development^12–14; however, the deep layers remain more highly myelinated at all ages^15,16. This differential myelination is at least in part related to differences in myelin density on the axons of the neuronal classes that occupy the different layers ^11,17. Excitatory projection neurons (PNs) in different layers differ greatly in their longitudinal distribution of myelin¹¹, with deep-layer neurons being generally more extensively and uniformly myelinated than upper-layer neurons. Similarly, parvalbumin interneurons of the cortex are more extensively myelinated than somatostatin or vasoactive intestinal polypeptide interneurons^8,9, and display unique patterns of longitudinal distribution of myelin, contributing to layer variation in myelin content.

In principle, layer-specific differences in myelination could be due to differential expression of cues by the different neuronal subclasses that populate each layer, but also to the presence of different types of OLs^18,19, or their distinct layer-specific maturation^5,20. Recent studies have identified cues that modulate the cell-intrinsic capacity of OLs to myelinate their targets²¹; for example, inhibitory signals that prevent myelination, such as Jam2, are expressed on neuronal somas and dendrites⁷. Moreover, mis-positioning of PN subtypes from the deep layers to the upper layers of the cortex results in increased ectopic myelination¹¹, suggesting the presence of neuron-derived cues. The degree to which differential myelination is driven by OL heterogeneity vs layer-specific neuronal signals is unknown.

Here, we present a comprehensive resource of layer-specific oligodendrocyte development in the mouse cortex over time. We profiled OLs and their progenitors across cortical layers from early postnatal stages of myelination to adulthood, using single-cell RNA sequencing and spatial transcriptomics. We find that cortical oligodendrocytes differ primarily in maturation state across layers, suggesting that layer-specific myelination is not due to layer-specific oligodendrocyte subtypes. To test whether neuronal subtypes may differentially communicate with OLs to regulate their maturation and myelination, we built an atlas of predicted ligand-receptor interactions between OL states and projection neuron subtypes, and identified candidate neuron-driven modulators of OPC maturation and/or myelination. In vivo testing of selected candidates showed that Ncam1, Fgf18, and Rspo3 can induce myelination when ectopically expressed in upper-layer callosal neurons. Furthermore, silencing of Ncam1 and Fgf18 in deep-layer neurons leads to a reduction in myelin. The work builds a detailed resource of layer-specific oligodendrocyte development in the cortex, and identifies molecular signals differentially expressed by projection neurons in the deep cortical layers that mediate myelination. Our results support the hypothesis that neuronal diversity directly impacts myelin distribution across the neocortical layers.

RESULTS

Oligodendrocyte heterogeneity across cortical layers is predominantly driven by maturation state

To understand the role of OLs in layer-specific myelination, we profiled OLs and their progenitors across cortical layers over a time-course spanning from the early postnatal to the adult cortex, in mice, using single-cell RNA sequencing (Figure 1A). We used a compound transgenic mouse line incorporating an inducible Cre driver and reporter (Fezf2^CreERT x CAG^{floxStop-tdTomato}), which upon tamoxifen injection at P3 results in the preferential, strong labeling with tdTomato of Layer 5 (L5) corticofugal projection neurons (CFuPN). We crossed this line to the Plp1-EGFP reporter line, where oligodendrocytes are labeled at all stages of development by EGFP. This allowed us to cleanly micro-dissect Layers 1–4 (L1–4) (above the tdTomato⁺ cells), Layer 5 (L5) (tdTomato⁺), and Layer 6 (L6) (below the tdTomato⁺ cells) of the somatosensory cortex. Mice were sacrificed at multiple stages: i) early postnatally (postnatal day 7, P7), as OLs mature but before myelination begins, ii) juvenile (P14), when OLs begin to wrap axonal targets on L5 and L6 neurons, and iii) post-weaning and adult timepoints (P30 and P90), corresponding to myelination of deep-layer (DL, L5 and L6) and upper-layer (UL, L2–4) neurons, respectively (Figure 1A). We dissociated and FACS-isolated Plp1-EGFP⁺ OL populations from each layer and age for scRNA-seq (Figure 1A and Mendeley Data Figure 1A; see STAR Methods). In addition, to enrich for the most mature populations of oligodendrocytes, we also isolated a separate sample at P90 containing only the cells with the highest EGFP intensity (“EGFP⁺-high”, Mendeley Data Figure 1A; see STAR Methods). This approach allowed us to enrich for putatively mature oligodendrocytes, allowing a more detailed characterization of this population. Following the removal of neuronal and astrocytic cell populations and low-quality cells (Mendeley Data Figure 1B-I; see STAR Methods), the datasets together comprised a total of 38,496 cells (~5000–14000 cells per time point; Table S1). Cells were clustered by a reference-based approach²² using a previously published dataset (Marques et al., 2016)¹⁸. We assigned cell identities to the clusters based on the expression of known marker genes (Figure 1B-D and Mendeley Data Figure 2A). We were able to identify populations corresponding to all developmental stages of OL maturation, namely progenitor states (OPC: OL precursor cells expressing Pdgfrα and Cspg5, COP: committed OL progenitors expressing Bmp4 and Fyn,) premyelinating states (NFOL: newly-formed OL expressing Tcf7l2 and Tmem2), myelinating states (MFOL: myelin-forming OL expressing Opalin and Ctps, and MOL: mature OL expressing Ptgds and Apod) (Figure 1B-D; see Table S1 for a list of DE genes across oligodendrocyte states, ages and layers). We employed Monocle3 to order the OLs along a pseudotime trajectory and identify variation in gene expression along pseudotime (see STAR Methods). The pseudotime-variable genes were grouped into several modules (cNMF) (Table S2), some of which are enriched in specific stages of OL maturation (Mendeley Data Figure 3A-D). From those modules, we identified transcription factors, co-factors, and signaling molecules, as potential regulators of oligodendroglia lineage progression (Table S2). Among them, we recognized molecules with established roles in early OL differentiation; for example, Hmgn1²³, which was highly expressed in OPC. Additionally, we observed key regulators of OL maturation, including Bmp4, a negative regulator of maturation enriched in COP, and Tcf7l2, which is enriched in NFOL and promotes oligodendroglia lineage progression by repressing Bmp4^24–26. Furthermore, Nkx6.2, predominantly expressed in MOL, was also identified as a potential regulator in later stages of OL development. Notably, we also identified molecules enriched in each module with poorly understood roles in OL maturation and myelination, which may offer an opportunity to uncover additional factors influencing OL development and maturation. These include Traf4 and Tsc22d1 in cNMF_1 (OPC-related module), Mycl in cNMF_7 (COP-related module), Bag1 and Lcorl in cNMF_9 (NFOL-related module), Hist1h1c, Gas7 and Zdhhc9 in cNMF_3 (MFOL-related module), and Csrp1 and Litaf in cNFM_2 (MOL-related module) (Mendeley Data Figure 3E).

Figure 1. Single-cell profiling of cortical oligodendrocytes demonstrates positional heterogeneity of oligodendrocyte states.

(A) Schematic of S1 cortex layer micro-dissection, EGFP⁺ OL isolation by FACS, and scRNA-seq from P7, P14, P30, and P90 Plp1-EGFP mice (pooled male and female for each age).

(B) Heatmap of top 10 differentially-expressed genes for each OL population.

(C) Violin plot of expression of key marker genes and Plp1 across different OL populations.

(D) UMAP of combined OL scRNA-seq datasets from micro-dissected layers, across all ages and experiments, color-coded by state.

(E) Top: UMAPs of P7, P14, P30, P90 and P90_EGFP⁺-high scRNA-seq data, color-coded by OL state. Bottom: bar plot of P7, P14, P30, P90 and P90_EGFP⁺-high OL cluster proportions for each micro-dissected layer and experiment. Color-coding indicates the OL state. Data are normalized to the total cell count in each micro-dissected layer.

(F) Heatmap of top 10 differentially-expressed genes for each MOL state.

(G) UMAP of MOL subtypes, color-coded by the state identities identified in this study (MOL-a, MOL-b, and MOL-c).

(H) UMAP of OLs from this study (left), and the projection of Marques et al. 2016 cortical OLs using Azimuth (right).

(I) Left, violin plots of strongly-myelinating marker genes and NFOL marker genes in our MOL substates (color-coded). Right, violin plots of module score for myelinating and NFOL related genes in our MOL substates.

See also Mendeley Data Figure 1–3, Table S1 and S3.

Figure 2. Spatial transcriptomics reveals positional heterogeneity of oligodendrocytes and MOL states.

(A) Schematic summarizing the process of mapping the cell types found in our scRNA-seq study onto a matching spatial transcriptomic tissue previously published (Slide-seqV2)²⁷ using TACCO⁷⁵.

(B) Spatial features plot of Slide-seqV2 of P56 somatosensory cortex, showing the module score of curated marker genes for each neuronal subtype (see Table S4), used to generate Figure 2E.

(C) Gene module score of OL states (see Table S4) plotted onto a UMAP representation of the scRNA-seq data.

(D) Gene module score of OL states plotted onto the spatial distribution of beads in the Slide-seq data, used to generate Figure 2F.

(E) Spatial distribution of projection neuron types in a coronal section of P56 mouse cortex by Slide-seq. The graph shows the proportion of each cell type (y-axis) along various distances from the pia (x-axis). Dashed lines indicate cortical layers.

(F) As in E, for OL populations (OPC, COP, NFOL, MFOL, and MOL).

(G) Gene module score for the MOL states identified in this study (MOL a-c), plotted onto the spatial distribution of beads in the P56 somatosensory cortex Slide-seq data, used to generate Figure 2H.

(H) As in E and F, spatial distribution of mature MOLs, classified by the state identities identified in this study (MOL-a, MOL-b, and MOL-c).

(I) Gene module score for MOL states plotted onto the UMAP.

See also Table S4.

Figure 3. Interactome of pyramidal neuron-oligodendrocyte communication in the neocortex.

(A) Schematic of hypothesized Ligand-Receptor (L-R) interaction. Neurons producing inhibitory (blue) or attractive (red) cues and oligodendrocytes (green) are represented.

(B) Schematic summarizing the generation of the predicted interactome using transcriptomic data from projection neurons (PNs) and oligodendrocytes (OLs) as input to CellPhone DB to identify predicted ligand (L) and receptor (R) pairs.

(C) Dot plot of L-R expression between PN subtypes and OL states at P7 (left), P21 (middle), and P56 (right). L-R pairs that pass significance threshold are shown for each selected grouping, based on cell-type specificity and age. On the x-axis, PN subtypes are represented in colored rectangles, and OL states are shown in colored circles. The size of each data point represents −log₁₀ adj. P values, and the color indicates the average log₂ mean expression level of the interacting molecules from the two participating cell populations. L-R pairs that are not significant are represented with the smallest circle, while those that are not expressed have no circle. The label for each row gives the molecule expressed in the oligodendrocyte population followed by the molecule expressed in the neuronal population.

See also Figure S2, Mendeley Data Figure S4 and S5 and Table S5.

In the scRNA-seq dataset, we noted that, at all ages, myelinating OL populations were disproportionately present in all cortical layers, and especially highly represented in the deep layers (L5 and L6) (Figure 1E). It is possible that differential layer myelination could be due in part to class diversity of mature myelinating OL populations. Prior studies have identified transcriptionally-distinct populations of MOLs¹⁸, but whether these represent different cell types or merely different cell states is unknown. To assess transcriptional diversity in mature OLs that may correlate to differential myelination across layers, we examined specifically the MOL population. We applied an unsupervised hierarchical clustering method to identify transcriptionally-distinct populations of MOLs in our data. This identified 3 subclusters, which we termed MOL-a, MOL-b, and MOL-c (Figure 1F and G). These populations were very similar to each other transcriptionally (Table S3); there were fewer than 100 DEGs (see STAR Methods), which differed mainly in the level of expression rather than binary absence or presence of expression. We then compared our dataset to previously-described cortical OL populations from Marques et al., 2016 ¹⁸; as that study sampled OLs from various brain regions, we used only the cortically-derived cells from their dataset, which were derived from the same region used in the current study, the S1 somatosensory cortex (Figure 1H, Table S1 and S3). We used a random forest classifier to assign MOL identities from the Marques dataset¹⁸ to the MOLs in our dataset (see STAR Methods). We found that our MOL-a, MOL-b, and MOL-c showed only weak correlation to any of the Marques cortical mature MOL populations (Mendeley Data Figure 2B-E), likely due to the different ages analyzed by the two studies and the comparatively low representation of cortically-derived MOLs in the Marques dataset (Table S3).

Assessing the expression of the genes related to OL maturation and myelination suggests that MOL-a and MOL-c represent more highly myelinating states, while MOL-b may represent a less mature myelinating MOL (Figure 1I). We examined the distribution of these three MOL clusters across ages and layers in our scRNA-seq dataset and found that the MOL-b population was more prevalent in L6 than in L1–5 (Mendeley Data Figure 2F and G).

In order to more closely relate cell identities to their topographical organization in the cortex, we analyzed a previously-published in situ transcriptomics (Slide-seqv2) dataset of P56 somatosensory cortex²⁷. For each of the projection neuron and oligodendrocyte cell types, we assayed the combined expression of the top differentially-expressed genes (DEGs) for that cell type (Figure 2A-D, Table S4). Expression of these gene sets in the Slide-seq dataset was found to be consistent with our scRNA-seq dataset: while OPCs and COPs are distributed across all cortical layers, maturing and mature OLs are predominantly present in the deep layers (Figure 1E, Figure 2E and F). Given that deep-layer neurons are more extensively myelinated at all ages, the data suggests that the spatial distribution of OL maturation states may be related to the local neuronal environment. We then examined the spatial distribution of each MOL cluster in the adult Slide-seq dataset, using a gene module score calculated from the top 10 DEGs per cluster, and found that all MOL states were more prevalent in the deeper layers: MOL-a, which is also the most abundant MOL substate, was disproportionately present in L5, and MOL-b and MOL-c in L6 (Figure 2G-I). This suggests that the overall greater myelination observed in the deep layers may be associated with a higher density of mature myelinating OLs. Future studies will be required to clarify whether each MOL state plays a distinct role in cortical myelination or if they merely represent stages along a maturation trajectory.

Our lab has previously shown that myelin density and the laminar location of mature oligodendrocytes correlates with the positioning of deep-layer PN in young adult mice¹¹. To understand whether myelin distribution across cortical layers was altered from the first stages of myelination, we examined whether mis-positioning of deep-layer PN in upper cortical layers at P14–when upper-layer PN are not yet myelinated–would influence myelination distribution. We used the Reeler mouse model, which has a nearly inverted cortical lamination due to abnormal migration of neuronal subtypes²⁸ and which has been reported to have defects in OPC proliferation, distribution, and myelination^29,30. We performed immunohistochemistry in the somatosensory cortex of P14 mice and quantify myelin distribution by dividing the cortex (pia to white matter) into horizontal bins and measuring fluorescence intensity of myelin basic protein (MBP), CTIP2 (L5/6 CFuPN) and CUX1 (L2/3 CPN) in each bin. In the control cortex, MBP intensity strongly correlated with CTIP2 signal (Pearson correlation, r = 0.963, P<0.0001), and negatively with CUX1 signal (r = −0.67, P<0.01) (Figure S1, upper panels). In the reln^−/− cortex, despite disrupted laminar positioning of the neurons, we continued to observe a strong correlation of MBP with CTIP2 neurons (r = 0.77, P<0.05) and a negative correlation with CUX1 neurons (r = −0.74, P<0.01) (Figure S1, lower panels). The correlation of myelination density (MBP signal) with the local density of deep-layer PNs (CTIP2⁺) suggests that the microenvironment created by these neuronal populations may regulate oligodendrocyte maturation and myelination.

Definition of a molecular code controlling interaction among projection neuron subtypes and oligodendrocytes

Given that the location of deep layer neurons can influence the distribution of myelin, we hypothesized that differential myelination is governed, at least in part, by a neuronal subtype-associated signaling code of surface-bound or secreted molecules that affect oligodendrocyte maturation, myelin formation, and ultimately the distribution of myelin in the cortex. Neurons could in theory employ OL-facing signals such as inhibitory and/or attractive molecules to regulate differential myelination (Figure 3A). Thus, an understanding of molecules differentially expressed in weakly- vs. highly-myelinated PN subtypes could enable the identification of signals that play a role in neuron-OL communication.

Given that developmental myelination is a protracted process and continues well into adulthood, we examined neuron-OL interaction over an extended time course. We generated single-cell sequencing datasets from WT P7 and P21 whole somatosensory cortex, and identified all cell types present (Mendeley Data Figure 4). In addition, we also analyzed a publicly-available scRNA-seq dataset from the primary somatosensory (S1) cortex at P56 from the Allen Brain database³¹. Within these datasets, we extracted all PN profiles and oligodendrocyte states, and classified them based on gene expression and laminar location (Mendeley Data Figure 5).

Figure 4. Ligand and receptor pairs show specificity across layers and ages.

(A) Violin plots of normalized gene expression level (in natural log scale) among PN subtypes from scRNA-seq data at P7 (left), P21 (middle), and P56 (right). PN subtypes are color-coded.

(B) Average expression levels over time of selected candidate genes from L-R interaction analysis in neurons. Neuron types are color-coded. Up, genes with higher expression in upper-layer PN. Down, genes with higher expression in deep-layer PN and/or more mature substates of oligodendrocytes.

(C) Expression levels in oligodendrocytes of the cognate ligand or receptor for the molecules in B.

(D) Box plots showing expression levels of Ncam1 and Fgf18 in upper- and deep-layer neurons. Boxes represent the first and the third quartile, with the median indicated by a thick horizontal line (p-values from linear model, Ncam1: P7 p=0.086, P21 p=0.0001, P56 p=0.027; Fgf18: P7 p=0.446, P21 p=0.018, P21 p=0.086) (* p<0.05; ** p<0.01; *** p<0.001).

See also Figures S2 and Table S5.

Figure 5. In vivo screen of candidate mediators of myelination.

(A) Schematic of experimental setup. In-utero electroporation (IUE) was performed at embryonic stage 14.5 (E14.5) with an expression vector containing the target gene of interest and the pCAG-EGFP vector. Progeny was analyzed by immunohistochemistry at postnatal day 21 (P21). Genes tested are listed, with significant hits in bold.

(B) Quantification of myelin basic protein (MBP) intensity (left) and CC1⁺ (mature OL marker) cell density (right) of electroporated (IUE, red dots for MBP and blue dots for CC1) and contralateral (contra, black) cortical regions at P21 after in utero electroporation (IUE) of candidate mediators of myelination at E14.5. Each data point represents one brain section (MBP: EGFP n=5 mice, N=23 sections; Gas6 n=3 mice, N=22 sections; Fgf18 n=4 mice, N=39 sections; Ncam1 n=3 mice, N=12 sections; Sema5a n=5 mice, N=15 sections, Rspo3 n=5 mice, N=34 sections. CC1: EGFP n=5 mice, N=23 sections; Gas6 n=3 mice, N=13 sections; Fgf18 n=4 mice, N=20 sections; Ncam1 n=3 mice, N=12 sections; Sema5a n=3 mice, N=10 sections, Rspo3 n=5 mice, N=34 sections) (p-values from linear models, MBP: EGFP p=0.0589, Gas6 p=0.006, Fgf18 p=0.000009, Ncam1 p=0.048, Sema5a p=0.053, Rspo3 p=0.0023; CC1: EGFP p=0.986, Fgf18 p=0.043, Ncam1 p=0.025, Sema5a p=0.213, Rspo3 p=0.218) (* p<0.05; ** p<0.01; *** p<0.001). Lines connecting two data points represent corresponding IUE and contralateral regions of each section.

(C) Top, representative section of IUE cortex for control (EGFP) and overexpressed candidates (Gas6, Fgf18, Ncam1, Sema5a and Rspo3). Sections were immunolabelled for MBP in red and CC1 (mature oligodendrocyte marker) in blue. Electroporated cells express EGFP. Images show a stitched composite spanning the entire brain section. Analyzed electroporated and contralateral regions are highlighted with a yellow box (scale bar: 500μm). Bottom, representative images of IUE and contralateral P21 S1 cortical regions analyzed for MBP intensity (red) and CC1 density (gray). Over-expressed candidate genes (Fgf18, Ncam1, Sema3a, and Rspo3) are indicated for each panel. Inset: magnification of upper cortical layers, showing MBP intensity and CC1 count (scale bar: 100μm).

See also Figure S3, S4 and Mendeley Data Figure 6.

We explored Ligand-Receptor (L-R) interactions between PN subtypes and oligodendrocyte states by employing a curated molecular database, CellPhoneDB³², with stringent cutoffs (see STAR Methods). To identify genes expressed across different PN subtypes that have cognate ligands or receptors on distinct states of OL maturation, PN subtypes were tested against the different states of oligodendrocytes in the same scRNA-seq datasets for each of the three ages. Analysis of these datasets resulted in >120 L-R pairs with significant interaction in at least one OL-PN pairwise test (Table S5, Figure 3B and C).

The results included well-established signaling molecules that control oligodendrocyte differentiation, such as Gas6-Tyro3 and Cxcl12-Ackr3 (where Gas6 and Cxcl12 are expressed in PN, and Tyro3 and Ackr3 are expressed in OLs). Gas6 stimulates oligodendrogenesis and myelination, and loss of Gas6 is associated with loss of oligodendrocytes³³. Cxcl12 promotes OPC differentiation in vitro³⁴ and remyelination in vivo in the adult CNS³⁵. The presence of these expected interacting candidate molecules adds validity to the other predicted interactions, many which have not been previously implicated in myelination.

We identified candidate L-R interactions predicted to involve specific cell populations and ages: i) nonspecific interactions shared between most or all PN subtypes with all oligodendrocytes (termed “general”), ii) interactions with upper-layer PNs (UL PN), iii) interactions with deep-layer PNs (DL PN), interactions involving either iv) young or, v) mature oligodendrocytes, and vi) interactions specific to a particular age. To narrow down possible targets, we focused on those interactions where the candidate molecules were differentially expressed between upper- and deep-layer neurons, as these are the most likely to represent candidates that could mediate differential distribution of myelination across layers. We found several genes that were differentially expressed between the upper-layer and deep-layer neurons, either at a specific age or independently of age (Figure 3C, Figure 4A-C, Figure S2A). Interestingly, Gas6 was differentially expressed across PNs, being abundantly present in deep-layer PNs at all ages, while Cxcl12 was predominantly expressed in deep-layer PNs only at younger ages (P7 and P21) (Figure 4A and B). Other genes, previously unclassified as myelinogenic, such as Fgf18, Ncam1, Sema5a or Glra2, had a higher expression in deep-layer PNs at all ages, and therefore may represent possible candidates to promote layer-specific myelination. Conversely, genes such as Ptn or Epha4 had greater expression in upper-layer PNs, and thus may represent candidate molecules that prevent myelination in upper cortical layers. By contrast, the expression of their corresponding oligodendrocyte partner molecules remained relatively uniform across cortical layers (Figure S2B), suggesting that projection neurons may drive layer-associated effect in myelination. Interestingly, genes such as Nectin1/3 or Xpr1 had layer-specific expression only at certain postnatal stages. In contrast, Bdnf showed a dynamic expression pattern over time, being enriched in deep layers at P7, coinciding with the onset of myelination. At later stages, as the upper layers also undergo some myelination, Bdnf shows high expression in both upper and deep layers (Figure 4). Altogether, these genes may represent candidate effectors of a neuron subtype-specific molecular code governing differential OL maturation and / or myelination in distinct layers.

In vivo functional screen validates candidate molecular regulators of layer-specific myelination in the neocortex

To test the functional effects of our candidate L-R genes on differential myelination, we applied an in vivo overexpression screen. We selected candidates that were more highly expressed in deep-layer neurons (L5 and L6), and which may thus represent “inductive” signals, and overexpressed them in upper-layer neurons (L2/3 CPN) via in utero electroporation at E14.5 (Figure 5A, Figure S3A). Mice were sacrificed at P21, and we used immunohistochemistry to examine OL maturation (using CC1 as a marker of mature OLs) and myelination (using myelin basic protein, MBP, as a marker) (Figure 5A). Given that myelination in the cortex varies at distinct rostro-caudal and medio-lateral positions, we used the matching area of the un-electroporated, contralateral hemisphere as control. We note that a limitation of this approach is that contralateral axonal projections of the electroporated L2/3 CPN, could influence myelination in the contralateral region, thereby reducing any difference in myelin levels.

We first tested a known modulator of myelination, Gas6, as a positive control³³. As expected, Gas6 overexpression led to increased myelination in the electroporated region. The empty EGFP expression vector (negative control) resulted in no difference in myelination or oligodendrocyte maturation between electroporated and contralateral cortical regions (Figure 5B and C, Figure S3B and C). We tested four predicted “inductive” candidates from our interactome; three of the candidates (Fgf18, Ncam1, and Sema5a) are enriched in deep-layer neurons, and while the expression of the fourth candidate, Rspo3, is similar in all PN sub-types, its ligand (Lgr4) is specifically enriched in mature oligodendrocytes (Figure 3C, Figure 4, Figure S2C and D). Fgf18, Ncam1, and Rspo3, significantly increased myelination in the electroporated region compared to the contralateral hemisphere; the fourth molecule, Sema5a, approached but did not reach significance (p=0.053). Two candidates, Fgf18 and Ncam1, were also able to promote OPC maturation (Figure 5B and C, Figure S3C). While FGF18 is a mitogen that promotes proliferation of OPCs³⁶ and, in vivo, activation of one of its receptors, FGFR2, regulates myelin growth^37,38, Ncam1 and Rspo3 have not been previously implicated in myelinogenesis, and may represent candidate mediators of neuron-driven myelination in the neocortex. We thus proceeded to investigate whether the identified myelin-promoting molecules also influenced OPC proliferation. Immunohistochemistry for PDGFRα, a marker of OPCs, upon overexpression of either Fgf18, Ncam1 or Rspo3 revealed that Fgf18 overexpression led to an increased in OPC density, while Ncam1 or Rspo3 overexpression had no significant effect (Figure S4). We next calculated a gene module score for the receptor genes expressed by oligodendrocytes that could mediate a response to Fgf18, Ncam1 or Rspo3 ligands (Lgr4, Fgfr1, and Fgfr2)). Supporting a role for these ligands in mediating myelination, we found that the module score increased with age, and was most highly expressed in oligodendrocytes of the deep layers, which have the most myelin (Figure S3D).

Given the small effect size of our candidate genes in stimulating OPC maturation and myelination when ectopically expressed as single genes, and the fact that myelination is likely regulated by a combination of genes, we next tested whether co-expression of candidate pro-myelinating factors could induce a stronger phenotype. We thus co-electroporated Fgf18 and Ncam1 at E14.5 and examined mice at P14 and P21 (Figure 6, Figure S3E). We found a strong combined effect of the two genes for both increasing OPC maturation (increased CC1⁺ cell density) and myelination (increased MBP intensity). Interestingly, StringDB, a curated database of all protein-protein interactions³⁹, indicates that both the FGFR1 and FGFR2 receptors can respond to both Fgf18 and Ncam1. This further supports a model in which FGF cascade activation stimulates oligodendrocyte maturation and myelination near deep-layer neurons, which express these ligands more abundantly (Figure 3 and 4).

Figure 6. Co-expression of Fgf18 and Ncam1 cooperatively enhance myelination.

(A) Left, representative section of IUE cortex at P14 overexpressing Fgf18 and Ncam1 together. Sections were immunolabelled for MBP (red) and CC1 (blue). Electroporated cells express EGFP. The image show a stitched composite spanning the entire brain section. Analyzed electroporated and contralateral regions are highlighted with a yellow box (scale bar: 500μm). Center, representative electroporated and contralateral S1 cortical region at P14 after IUE at E14.5 of combined Fgf18 and Ncam1 (scale bar: 100μm). Inset: magnification of upper cortical layers, showing MBP intensity and CC1 count (scale bar: 100μm). Right, quantification at P14 of myelin basic protein (MBP) intensity and CC1⁺ cell density in upper layers of electroporated (IUE, red for MBP and blue for CC1) and contralateral (contra, black) cortical regions after IUE. Each data point represents one brain section (n=5 mice, N=30 sections total). (p-values from linear models: Total: MBP p=0.0006; CC1 p=0.0000008) (* p<0.05; ** p<0.01; *** p<0.001).

(B) As in A, but at P21. Each data point represents one brain section (n=5 mice, N=30 sections total) (p-values from linear models, Total: MBP p=0.00000001; CC1 p=0.00004) (* p<0.05; ** p<0.01; *** p<0.001).

See also Figure S3, S7 and Mendeley Data Figure 6.

Finally, to determine whether these molecules are necessary for normal myelination in deep layers, we silenced their expression in vivo using shRNA. As Ncam1 and particularly Fgf18 already have substantially lower expression in upper-layer projection neurons compared to deep-layer projection neurons, we targeted deep-layer neurons in this experiment. Plasmids carrying shRNAs targeting Ncam1 or Fgf18 were generated and validated for knockdown efficiency in vitro (Figure S5A). These constructs were then delivered by in utero electroporation at E12.5 to specifically target projection neurons in the deep layers (Figure 7, Figure S3A). Silencing of each gene individually did not alter myelin levels (Figure 7A). Despite the very high starting levels of myelin in the electroporated layers, which makes detection of any change in myelin very difficult, when both targets were silenced at the same time, we observed a reduction in myelination as well as in the number of mature oligodendrocytes within the electroporated region at P14 (Figure 7B, Figure S5 B-D). Knockdown of Ncam1 alone also led to a significant reduction in the number of mature oligodendrocytes (Figure 7A, Figure S5B and C). To confirm these results using an alternate method, we performed CRISPR-mediated disruption of Ncam1 and Fgf18 in deep-layer neurons, which similarly led to a reduction in myelination in the deep layers (Figure S6). These results support a model whereby myelination is regulated by the combined action of multiple signaling factors. While these factors may act through shared or distinct downstream molecular pathways, their combined activity exerts a strong effect on myelination, as demonstrated by both the knockdown and overexpression experiments. Together, these findings demonstrate that neuronal expression of Ncam1 and Fgf18 has an important role in facilitating normal myelination in the deep cortical layers.

Figure 7. NCAM1 and FGF18 are essential for effective myelination in deep cortical layers.

(A) Top left, schematic of experimental setup. In-utero electroporation (IUE) was performed at embryonic stage 12.5 (E12.5) with vectors containing the shRNAs for Ncam1 and Fgf18 under the control of the U6 promoter. Progeny was analyzed by immunohistochemistry at postnatal day 14 (P14). Middle and bottom left, representative sections of IUE cortex at P14 silencing Ncam1 or Fgf18. Sections were immunolabelled for MBP (red) and CC1 (blue). Electroporated cells express sfGFP. Images show a stitched composite spanning the entire brain section. Analyzed electroporated and contralateral regions are highlighted with a yellow box (scale bar: 500μm). Center, representative electroporated and contralateral S1 cortical region at P14 after IUE at E12.5 (scale bar: 100μm). Inset: magnification of deep cortical layers, showing MBP intensity and CC1 count (scale bar: 100μm). Right, quantification of myelin basic protein (MBP) intensity and CC1+ cell density in deep layers of electroporated (IUE, red for MBP and blue for CC1) and contralateral (contra, black) cortical regions after IUE. A scramble (scr) sequence was electroporated and quantified as control. Each data point represents one brain section (shFgf18: n=4 mice, N=14 sections total; shNcam1: n=4 mice, N=15 sections total; scr: n=3 mice, N=14 sections total) (p-values from linear models: MBP: shNcam1 p=0.578, shFgf18 p=0.964, scr p=0.530; CC1: shNcam1 p=0.003, shFgf18 p=0.580, scr p=0.633) (* p<0.05; ** p<0.01; *** p<0.001).

(B) As in A but silencing Ncam1 and Fgf18 together. Each data point represents one brain section (shNcam1+shFgf18: n=5 mice, N=22 sections total; scr: n=3 mice, N=14 sections total). (p-values from linear models: MBP: shNcam1+shFgf18 p=0.000009, scr p=0.530; CC1 shNcam1+shFgf18 p=0.000237, scr p=0.633) (* p<0.05; ** p<0.01; *** p<0.001).

See also Figures S5-S7 and Mendeley Data Figure 6.

To validate the changes observed with MBP using an independent marker and assess effects across myelination stages–namely, the earlier, less compact stage identified by myelin-associated glycoprotein expression, MAG, versus the later, compact stage, labeled with MBP–we performed immunostaining for MAG. Co-overexpression of Ncam1 and Fgf18 led to a significant increase in MAG levels in upper, electroporated cortical regions, with no effect in deeper, un-electroporated layers, while their silencing in deep layers resulted in a reduction of MAG (Figure S7A and B). Similar effects were observed for MBP (Figure S7C). Together, these findings indicate that neuronal expression of NCAM1 and FGF18 regulates multiple stages of myelination, from early to mature stages.

Altering neuronal levels of NCAM1 and FGF18 could affect neuronal properties that influence myelination, such as axon caliber. To test this, we measured axonal diameter in high-resolution images of electroporated neurons. Neither overexpression nor silencing of Ncam1 or Fgf18–alone or in combination–altered axonal caliber compared to controls (Mendeley Data Figure 6). Although we cannot exclude the possibility that other neuronal or axonal properties may be affected, these findings collectively suggest that neuronal NCAM1 and FGF18 function primarily as molecular cues regulating oligodendrogenesis and myelination.

Taken together, our results indicate that different cortical projection neuron classes differentially express signaling molecules that communicate with oligodendrocytes and influence myelination, resulting in layer-specific differences in myelination. Our screen allowed us to identify Fgf18 and Ncam1 as myelin-stimulating candidate genes produced preferentially by PNs of the deep layers, pointing to pathways underlying differential OL maturation and myelination and the role played by PN diversity in affecting the development of myelin maps in the neocortex, and pointing at the value of this resource in capturing molecular mechanisms controlling myelination.

DISCUSSION

Different regions of the CNS are known to be differentially myelinated, but the mechanisms that control the establishment of these myelin maps are poorly defined. Here, we present a resource comprising a comprehensive single-cell molecular map of cells of the oligodendrocyte lineage within different layers of the neocortex across different developmental stages of myelination. As a first application of this resource, we tested the hypothesis that distinct neuronal classes may use differential signaling to control OL differentiation and myelin deposition to drive layer-specific myelination patterns. Our data show that OL diversity cannot per se account for differential distribution of myelin between the deeper and the upper layers of the neocortex. On the contrary, the composition of PN subtypes in the different cortical layers is a determinant of myelin density and OL state distribution. This adds to the existing evidence that projection neuron diversity in the neocortex forms the framework for instructing the localization, states, and function of other cell types, including interneurons and microglia^40–42. These findings put forward the principle that neuronal diversity in the CNS is not only important per se, but drives interactions with glia to develop higher-order features of the brain, including layer-specific maps of myelin distribution.

Functional testing of candidate genes allowed us to identify three pro-myelinating proteins, NCAM1, FGF18, and R-spondin-3, which could guide differential myelination in the cortex (Figure 5 and 6). Our findings reveal that deep-layer neurons exhibit higher levels of Ncam1 and Fgf18 compared to upper-layer neurons (Figure 4D, Figure S2C and D), and that silencing them in deep-layer neurons is sufficient to reduce myelination in that region (Figure 7B). Although enriched in deep-layer neurons, these proteins are not restricted to a specific cell type; in future work, it would be interesting to test if their expression in other cell types may also influence myelination.

Our findings implicating these genes in myelination are supported by related evidence. For example, R-spondin 3 has not been directly associated with myelination but it binds the leucine-rich repeat-containing G-protein coupled receptor 4 (LGR4), activating Wnt-signaling pathways that, when activated at specific developmental stages, regulate oligodendrocyte differentiation and myelination^43–47.

Co-expression of Ncam1 and Fgf18 results in an enhanced effect on OLs maturation and myelination (Figure 6), supporting the hypothesis that myelination is a very nuanced process likely controlled by a combinatorial code of signals. The precise downstream mechanisms engaged by NCAM1 and FGF18 remain to be defined, whether they converge on shared pathways or act through distinct signaling cascades to enhance myelination requires further study. Nevertheless, existing literature offers potential mechanistic clues. Both proteins bind fibroblast growth factor receptors (FGFR1 and FGFR2), suggesting that activation of the FGF signaling pathway in oligodendrocytes promotes maturation and myelination. Consistently, activation of FGFR2 in oligodendrocytes regulates myelin thickness and the expression of myelin-related genes in the spinal cord⁴⁸. NCMA-NCAM interactions between axons and OPCs during the initiation of myelination may trigger a cis interaction between NCAM and FGFR at the OPC membrane, activating MAPK signaling, and promoting OPC survival⁴⁹. Consistent with our knock-down results, knockout of Ncam1 leads to delayed developmental myelination in the corpus callosum, a phenotype that is normalized by adulthood⁵⁰.

Signaling from projections neurons is likely multifactorial, influencing not only the regulation of myelination itself but also the proliferation and maturation capacity of OPCs and oligodendrocytes. Although additional experiments (e.g., OPC transplantation between cortical layers) would be necessary to fully disentangle these effects, our current findings indicate that the tested molecules exert distinct influences. R-spondin 3 appears to primarily regulate myelination, whereas NCAM1 and FGF18 affect oligodendrocyte maturation and myelination, and FGF18 additionally impacts OPC proliferation.

Altogether, although both molecules can engage FGF signaling, the selective effect of FGF18 on OPC proliferation supports the notion that their effects are mediated, at least in part, by distinct downstream molecular mechanisms.

This study provides an initial interactome and evidence of specific molecules that different projection neurons could utilize to modulate myelination, from affecting OPCs dynamics to promoting myelin deposition. The oligodendrocyte developmental resource and OL-PN interactome presented here are intended to enable future studies to expand our understanding of the complexity of PN-OPC/OL interaction in the cerebral cortex and their contributions to the establishment of myelin pattens.

A clinical goal in demyelinating disorders has been to promote efficient myelin repair by directly promoting OLs differentiation and remyelination⁵¹. Some demyelination phenotypes appear to be associated with a lack of myelination activity rather than a lack of OLs^52,53: for example, multiple sclerosis lesions contain mature OLs but fail to remyelinate axons. Notably, the unsialylated form of NCAM1 is enriched in the cerebrospinal fluid of multiple sclerosis patients that respond well to steroid therapy compared to those that do not^54,55, suggesting that increased NCAM1 levels could be promoting remyelination in those patients. Over normal development, NCAM1 in the brain transitions from a highly-polysialylated embryonic form to a less-polysialylated adult form⁵⁶ altering its binding properties⁵⁷. Polysialylated NCAM1 (PSA-NCAM) reduces OPC maturation and prevents myelination⁵⁸, and its downregulation coincides with the onset of myelination in the human fetal forebrain⁵⁹. Together, these findings suggests that regulation of NCAM1 sialylation may be an important factor in the control of myelination.

Many neuropsychiatric conditions are characterized by white matter abnormalities^4,60. Notably, several myelinogenic molecules identified in this study have been implicated in neuropsychiatric conditions. SNPs in NCAM1 contribute to differential risk of developing bipolar disorder and schizophrenia⁶¹, and autoantibodies against NCAM1 were found in patients with schizophrenia⁶². In addition, disruption in Wnt and FGF signaling, which are activated by Rspo3 and NCAM1 / Fgf18, have been observed in schizophrenia, bipolar disorder, and major depressive disorder^63,64.

Our results support a model in which distinct classes of cortical projection neurons use different signaling molecules to modulate OL development and myelination, ultimately guiding differential distribution of myelin in the cortical layers. A more comprehensive investigation of the contribution of neuronal diversity, across brain regions, to OL biology and myelin is bound to expand our understanding of developmental myelination and the process of myelin dysfunction in disease states.

LIMITATIONS OF THE STUDY

Our study produced an extensive ligand-receptor interactome atlas built from scRNA-seq data derived from cortical projection neuron subtypes and distinct states of oligodendrocyte lineage cells. However, in vivo validation of cell type-specific enrichment of these ligands and receptors at the protein level remains technically challenging due to low abundance and limited antibody reliability, and current single-cell proteomic approaches lack the sensitivity and specificity required to robustly identify and quantify such proteins (e.g., FGF18). To validate two candidates, we performed immunofluorescence-based quantification and found higher expression in deep-versus upper-layer projection neurons. In addition, manipulation of gene expression by in utero electroporation was validated at the mRNA level using qPCR in NE-4C cells, which provides a more sensitive and reliable assessment of knockdown efficiency than in vivo protein-level quantification in electroporated neurons.

We demonstrate that modulation of Fgf18 and Ncam1 levels exerts a strong effect on cortical myelination. However, the downstream molecular pathways involved, and whether these factors act through shared or distinct signaling mechanisms remain to be determined. Although we detected no changes in neuronal migration, axonal projections, or diameter, we cannot fully exclude the possibility that manipulation of Fgf18 and Ncam1 alters intrinsic neuronal properties that secondarily influence myelination. Nevertheless, given the absence of detectable structural neuronal changes, such indirect effects appear less likely, and our data more strongly support a model in which FGF18 and NCAM1 influence myelination through interactions with OPCs and/or oligodendrocytes.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Paola Arlotta (paola_arlotta@harvard.edu).

Materials availability

This study did not generate any new unique reagents.

Data and code availability

• Raw FASTQ files from single-cell RNA-Seq data of oligodendrocytes from micro-dissected layers across all time points, and from all cells in the mouse cortex at P7 and P21 are available at GEO accession number GSE233255.

• Additional data figures were deposited in Mendeley Data: 10.17632/khnrd29pyy.1

• All original code is available at github: https://github.com/kimkh415/oligo_myelination and deposited at Zenodo: http://doi.org/10.5281/zenodo.19829308

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR METHODS

Experimental model and study participant details

Mice and husbandry

For FACS-isolation of oligodendrocyte-lineage cells from different cortical layers, we used the following mouse lines: Plp1-EGFP⁶⁵, Fezf2^{CreERt2 66}, conditional reporter line CAG^{floxStop-tdTomato} Ai14-Gt(ROSA)26Sor^{tm9(CAG-tdTomato)Hze} (IMSR Cat# JAX:007909, RRID:IMSR_JAX:007909)⁶⁷. Mice from these different lines were crossed to generate Fezf2^CreERT2/+; td-Tomato^+/−; Plp1-EGFP^+/− mice to simultaneously label oligodendrocytes and corticofugal projection neurons (CFuPN). Mice were injected at postnatal day 3 with 4-hydroxytamoxifen (Sigma H6278) in corn oil (single dose of 100 mg tamoxifen per kg body weight). For PN mis-localization experiments, Reeler KO (Reln^−/− RRID:IMSR_JAX:000235) mice were used. Transgenic mice were maintained on a C57BL/6J background, and both male and female mice were used for experiments. In utero electroporation procedures were performed on CD-1 timed pregnant females purchased from Charles River Laboratories. All procedures were designed to minimize animal suffering and approved by the Harvard University Institutional Animal Care and Use Committee (IACUC) and performed in accordance with institutional and federal guidelines. Mice were housed in individually ventilated cages using a 12-hour light/dark cycle and had access to water and food ad libitum.

Method details

Cortical microdissection, single cell suspension and FACS

Microdissection of layers from S1 cortex (for scRNA-seq for OL)

For each sample, one male and one female Fezf2^CreERT2/+; td-Tomato^+/−; Plp1-EGFP^+/− mouse at each time point were used for S1 cortex layer microdissections. Mice at different time points (P7, P14, P30, and P90) were anesthetized with isoflurane, followed by quick decapitation, and brains were dissected and transferred into ice-cold low-fluorescence Hibernate-A (Brain Bits, HALF 500). Brains were coronally sectioned at 200–400 μm in Hibernate-A on a Lecia vibratome, and sections were selected exclusively from the S1 cortex: specifically, from the point where the corpus callosum crosses the midline (bregma +1.1mm) to the point where the hippocampus dips below the dentate gyrus (bregma −1.94mm). Using a fluorescence dissecting microscope, the S1 region was further micro-dissected in cold Hibernate-A into 3 layer compartments (L1–4, L5 and L6) using Fezf2-tdTomato⁺ cells to demarcate L5. Care was taken to exclude the corpus callosum since it is known to have differential OLs from the cortex⁶⁸.

Cortical dissection (for scRNA-seq of all cell types)

Somatosensory and motor cortex from wild-type animals was dissected at P7 and P21. Mice were anesthetized with isoflurane, followed by quick decapitation, and brains were dissected and transferred into ice-cold media. The somatosensory and motor cortex were excised and transferred to fresh ice-cold media.

Preparing single-cell suspensions

Live single-cell suspensions were prepared for fluorescence activated cell sorting (FACS) using an adapted protocol from Worthington Papain Dissociation System kit (Worthington Biochemical, LK003150)⁶⁹. Briefly, micro-dissected brain tissues were finely chopped using a blade and incubated in EBSS with Papain and DNAse for 30 mins (P7) to 1.5 hours (P90) at 37°C. This was followed by gentle trituration to break up the tissue into a single-cell suspension, and papain activity was blocked using the ovomucoid inhibitor. The dissociated cells were resuspended in ice-cold Hibernate-A containing 10% fetal bovine serum (FBS, SH30070.03HI).

Fluorescence-activated Cell Sorting (FACS)

Oligodendrocyte purification

Plp1-EGFP⁺ oligodendrocytes were sorted on a MoFlo Astrios EQ Cell Sorter (Beckman Coulter) using a 100 μm nozzle. For most samples, all EGFP⁺ populations of different EGFP intensity were collected, since they likely represent different stages of OL maturation. For P30, we collected two independent biological replicates, P30_1 and P30_2. For the second sample, we collected cells with slightly higher EGFP intensity, with the aim of enriching in more mature oligodendrocyte populations. For P90, we also collected two samples, one with all-EGFP⁺-cells, and a second sample (from a different biological replicate) gated on the population with the highest GFP signal (labelled “P90_EGFP⁺-high”), with the aim of enriching for mature oligodendrocytes to allow detailed characterization. On average, 5,000 to 8,000 EGFP⁺ cells were sorted per layer and age. The sorted cells were spun down at 200g G for 5 min and resuspended in PBS containing 0.04% BSA (NEB B9000S).

All cortical populations

Live cells were isolated by sorting on a MoFlo Astrios using a 100μm nozzle, as DAPI-negative and Vybrant DyeCycle Ruby (Thermo Fisher V10273)-positive events.

Single-cell sequencing

Oligodendrocytes at P7, P14, P30 and P90

Approximately 6–8,000 sorted EGFP⁺ oligodendrocytes were loaded onto a Chromium^™ Single Cell 3’ Chip (10x Genomics, PN-120236) and processed through the Chromium Controller to generate single-cell gel beads in emulsion. Single-cell RNA-Seq libraries were prepared with the Chromium^™ Single Cell 3’ Library & Gel Bead Kit v2 (10x Genomics, PN-120237) or v3 for the second experiment at P30 and P90 (10x Genomics, PN-1000121). Libraries from different layers were pooled based on molar concentrations and sequenced on a NextSeq 500 instrument (Illumina) to an average read depth of 30,000 reads per cell.

All cell populations at P7 and P21

Approximately 10,000 cells from dissociated cortexes were loaded onto a Chromium^™ Single Cell 3’ Chip (10x Genomics, PN-120236) and processed through the Chromium Controller to generate single-cell gel beads in emulsion. Single-cell RNA-Seq libraries were prepared with the Chromium^™ Single Cell 3’ Library & Gel Bead Kit v2 (10x Genomics, PN-120237). Libraries from different replicates were pooled based on molar concentrations and sequenced on a NextSeq 500 instrument (Illumina) to an average read depth of 30,000 reads per cell.

The read distribution was as follows: 26 bases for read 1, 57 bases for read 2 and 8 bases for Index 1. A sequencing saturation of 70–80% was targeted for all samples.

scRNA-seq Data Analysis Pipeline

Single-cell RNA-seq binary cell call (BCL) files for experiments from P7, P14, P30 and P90 oligodendrocytes and all cell types from P7 and P21 mice were converted to FASTQs using the mkfastq function from Cell Ranger software version 3.0.1 for oligodendrocytes and 3.0.2 for all cell types, with the default parameters. FASTQs were aligned to the mm10 (Ensembl 93) reference transcriptome, and gene expression matrices were obtained using the Cell Ranger count function of with the default parameters. CellBender⁷⁰ was then applied to remove ambient RNA (remove-background method with epochs=150) and Scublet⁷¹ to compute doublet score for each cell (expected_doublet_rate=0.05, min_counts=2). R version 4.0.3 and Seurat package version 4.0.0⁷² were used to perform all downstream analyses.

For initial quality filtering, genes expressed in more than three cells were kept, and cells expressing more than 500 genes were kept. Cells with greater than 10% mitochondrial gene expression (percent.mt) were removed. The SCTransform function from Seurat (sctransform package version 0.2.1) with vars.to.regress parameter (to account for batch effects) set to percent.mt and sample (two biological replicates for each micro-dissected cortical layer) was used to normalize, find variable genes, and scale the dataset. The identified list of variable genes was used to perform principal component analysis (RunPCA function with the default parameters). Cell clusters were identified using the top 30 principal components and Louvain clustering algorithm (FindNeighbors function with dims=1:30 and FindClusters function with resolution=0.5 unless otherwise indicated). For visualization, the top 30 PC dimensions were further reduced to two uniform manifold approximation and projection (UMAP) dimensions (RunUMAP function with dims=1:30). The procedure, from SCTransform to dimensionality reduction for visualization, was repeated each time unwanted cell populations were removed, allowing more-accurate capture of the variation among the remaining cells, until we were left with only the cells of interest. Visualization of scRNA-seq was performed with ggplot2 (version 3.2.1, https://ggplot2.tidyverse.org)⁷³.

Differentially expressed genes (DEGs)

Cluster-specific and cortical layer-specific DEGs were identified by fitting a negative binomial generalized linear model (glm.nb function from MASS R package v7.3–53.1) to each gene while considering the differences in total number of UMIs per cell. The p-value for each gene was obtained by Wald test. The resulting list of DEGs were filtered to have adjusted p-value less than 0.05.

Oligodendrocytes Across Ages

The two experiments went through separate QC, because they were processed at different times using different 10X chemistry kit versions. For replicate1, the initial cell clustering identified 17 different clusters. Clusters with high expression of neuron-, microglia- or astrocyte-specific genes were removed (211 excitatory neurons with Neurod2, Satb2, and Cux1; 30 inhibitory neurons with Gad1 and Gad2; 31 microglia with Aif1; 1313 Astrocytes with Aqp4, Aldh1l1). Clusters with low-quality signals (low UMI/cell, low genes/cell and high percent.mt; 144 cells) were also removed. For the second experiment, we only found one cluster of Astrocytes (109 cells), which was removed.

After QC, the remaining OLs (22189 from experiment 1 and 16307 from experiment 2) were merged into the same Seurat object and processed together. We performed reference based approach²² (code available on Github), using a previously-published dataset¹⁸ as a reference to co-embed oligo substates (OPC, COP, NFOL, MFOL and MOL) from different ages. In this reduced dimension, new clusters were identified and annotated with OL type by the DEGs.

The MOL population was isolated and processed to further identify subtypes. A new set of variable genes was found, and sample-specific covariates (animal and cortical layer) were regressed out when scaling the data. For batch correction, Harmony (R package v1.0) was applied to age. To evaluate the significance of MOL sub-clustering, we applied the testClusters function from scSHC (R package v0.1.0)⁷⁴, reducing the initial seven clusters (FindCluster function from Seurat with resolution=0.3) to three significant ones. We noted that some of the genes with highest effect size among sub-clusters were immediate early genes, which could be contributing to the clustering.

All cell types across ages

Initial cell clustering identified 32 and 34 different clusters in P7 and P21 datasets, respectively. Clusters with high doublet scores and those low in quality (low UMI/cell, low gene/cell, and high percent.mt) were removed (2803 cells in P7, and 1272 cells in P21). After processing the remaining cells for each age (36,898 cells for P7, and 53,732 cells for P21), the identified cell clusters were assigned cell type labels based on their cluster-specific marker genes (Table S6). Clusters that expressed markers of glutamatergic neurons (ExN; Slc17a7 and Neurod2) and oligodendrocytes (OL) were further processed to annotate the subtypes. ExN and OL were individually processed to yield cell clustering. Clusters were then assigned a subtype label (ExN subtypes: UL_CPN, L4_Stellate, DL_CPN, NP, SCPN, CThPN, and Subplate; OL subtypes: OPC, COP, NFOL, MFOL, MOL) based on the DEGs (Table S6). MFOL and MOL subtypes are only present in the P21 dataset because these more mature OL populations emerge after P7. Annotated ExN and OL subtypes were subsequently used to predict cell type-specific cell-cell interactions.

Predicting MOL subtype labels using random forest classifier

To assign MOL subtype labels from a previously published dataset¹⁸ to MOLs in our layer-dissected scRNA-seq data, we used the Marques et al. MOL categories to train a random forest (RF) classifier. As the Marques et al. study spans multiple brain regions, we first extracted all MOLs of cortical origin to create a reference dataset. One of the mature oligodendrocyte populations in the reference dataset (MOL2, a white-matter associated MOL) was removed from the comparison because of its very low (9 cells) representation in the cortical samples in the Marques et al. dataset. As the number of cells in each Marques et al. MOL population varied widely (Table S3), the reference data was then downsampled to have an equal number of cells per MOL subtype; this retained 165 out of an initial 429 cells. Variable genes from the two datasets were combined and used to train the model. Training was performed with the tuneRF function in the randomForest R package (Liaw and Wiener 2002) v4.6–14, with doBest = T. For validation, 20% of the cells were reserved as a cross-validation set, and compared to a null model prepared by shuffling the cell type labels on the training data set. The trained classifier was then applied to predict MOL subtypes.

To compare between RF prediction and manually-labeled MOL subtypes, we applied Spearman correlation using the top 200 DE genes of manually identified MOL subclusters. Only the genes with adjusted p-value of < 0.05 and effect size of > 0.5 were used. Correlation is visualized using corrplot R package v0.84 (Table S3).

Co-embedding OLs from different datasets using Azimuth

To visualize our scRNA-seq dataset of cortical layer dissociated OLs together with cortical OLs from the Marques et al. dataset, we applied the Azimuth algorithm via Seurat (R package version 4.0.0). UMAP was first computed on our data using the first 20 Harmony reduced dimensions with return.model=T. Between the two datasets, a set of anchors (mutual nearest neighbor cells) was identified using FindTransferAnchors function with reduction=‘pcaproject’ and reference.reduction=‘spca’, dims=1:50. IntegrateEmbeddings function followed by ProjectUMAP function were used to integrate the embeddings using the identified anchor set and to obtain projection onto the reference UMAP for each query cell.

Developmental trajectory analysis

We used Monocle3 (R package version 0.2.2 with default parameters) to infer developmental trajectory (learn_graph function) on a pre-computed UMAP and assign pseudotime (order_cells function) to cells. A starting point for the trajectory was manually selected to be an endpoint of the trajectory graph closest to the earliest cell population (OPC). To study patterns of coordinated gene regulation across OL developmental states, we applied consensus NMF (cNMF; k=15 components; code available on github). Genes that change in expression along pseudotime were identified (Monocle3 graph_test function) and filtered by significance (q-value < 0.05) and spatial autocorrelation (Moran’s I > 0) before used in the cNMF. Mitochondrial genes were excluded from this module analysis. Some modules are enriched in specific stages of OL maturation. From the genes in these modules, we identified transcription factors (TF), co-factors, and signaling molecules (see Table S2 for the top 30 TF in each module). Representative molecules with enriched expression in a particular OL state are shown in expression feature plots.

Slide-seq data analysis

We used a previously-published in situ transcriptomics (Slide-seqv2) dataset of P56 mouse somatosensory cortex in coronal section²⁷. The region corresponding to the cortex was isolated from the whole puck.

Leveraging the distinct gene expression profiles and spatial connectedness of different cortical layers, beads were grouped into spatial regions (L1/Pia, L2–3, L4, L5, L6, L6b and WM) using “find_regions” function from TACCO⁷⁵ (v0.2.2 using python v3.9.13 with annotation_key=None, position_weight=1.5, and resolution=0.7). To deconvolve cell types present in each bead, we applied “annotate” function from TACCO (with max_annotation=5 and min_log2foldchange=1). There were two rounds of cell type deconvolution. First, we used the reference dataset from Allen Brain³¹ and subset it to include somatosensory only, to identify OL-containing beads along with other cell types. Then, to label distinct OL types and MOL subtypes, we used gene sets identified from our scRNA-seq data of OLs. This second round of deconvolution only affected beads that had some contribution of OL from the first round, by replacing OL contribution with the further-deconvoluted OL and MOL subtypes. Lastly, the density of each cell type as a function of distance from the pia was computed using “annotation_coordinate” function (with max_distance=2000 and delta_distsance=100) and plotted using a modified version of “plot_annotation_coordinate” function (available on Github).

Cells in the Slide-seq dataset were annotated using the curated marker genes for each cell type of interest (marker genes used are listed in Table S4). Modules are plotted in their spatial location using the SpatialFeaturePlot function of Seurat v4.0.0 in R v4.0.3 with min.cutoff='q10'.

CellPhoneDB

Differential ligand-receptor (L-R) expression and interaction predictions were performed between PN subtypes and OLs. We filtered P7 and P21 scRNA-seq datasets generated in the laboratory and the P56 single cell dataset from the Allen Institute³¹ to systematically identify L-R pairs. We applied CellphoneDB version 2.0³² to obtain the interactomes at the three time points. CellphoneDB was executed by invoking the statistical_analysis method with the following parameters: --subsampling --subsampling-log=true –threads=8. Since CellphoneDB generates raw P-values as output, the Benjamini and Hochberg method was applied to correct for multiple testing. Output interactions were filtered using an adjusted P threshold of 0.05. Predicted interactions are inferred based on the mean expression of the cognate ligands and receptors from RNA-seq data of each PN sub-type and OL state. This approach captures high-confidence interaction predictions; however, weakly-expressed transcripts or drop-out expression values may result in missed predictions. Because CellPhoneDB is built using human references, we converted mouse gene symbols to human Ensembl IDs using the getLDS function from biomaRt package version 2.40.5. Human gene names in the returned predictions were then converted back to mouse gene names.

In utero screen

We selected candidates predicted to promote myelination, namely genes that were more highly expressed in deep layer neurons. We chose to focus on candidates that may promote myelination, because our ligand and receptor interaction analysis is more likely to identify attractive interactions, and pro-myelination factors are also clinically more relevant for future studies. In the method chosen for delivery, in utero electroporation (IUE), plasmids remain episomal and become diluted through successive cell divisions, enabling “birthdate” labeling of specific cell types. By targeting the timing of electroporation to periods when only neuronal progeny are being produced, the rapid dilution of the plasmid in replicating (non-neuronal) cell types, combined with the low proportion of cells that initially receive the plasmid, act to restrict sustained expression to the targeted neuronal population and limit any off-target effects.

Vectors containing the sequence of the target genes selected from the L-R interaction analysis were purchased from Horizon discovery or amplified from mouse cortical cDNA. Target genes were subcloned by PCR-based cloning (NEBuilder HiFi DNA cloning, E5520S, New England BioLabs) into a mammalian expression vector (PCM153) under the control of the EF1a promoter, which allows robust and long-term expression of the transgene. The expression vector containing the target gene of interest was co-electroporated by in-utero electroporation with the pCAG-EGFP vector to obtain a strong GFP fluorescent signal that allows clear identification of the electroporated neurons (Figure 5A). For the negative control, the empty expression vector was co-electroporated with pCAG-EGFP.

For the silencing experiments, shRNA sequences were obtained from IDT and cloned into a vector under the control of the U6 promoter. This vector also included the EF1α promoter to drive sfGFP expression, enabling visualization of electroporated cells. The plasmids containing shNcam1 and shFgf18 were combined at a concentration of 1 μg/μl each for in utero electroporation, while the scramble control (scr) plasmid was electroporated at a concentration of 2 μg/μl.

The shNcam1 sequence was previously published in Zhou et al 2022⁷⁶: 5’-CGTTGGAGAGTCCAAATTCTT-3’

The shFgf18 sequence was previously published in Chen et al 2023⁷⁷: 5’-CCTGCACTTGCCTGTGTTT-3’

Scramble (scr): 5’-ACCAGTTGCGCCAATCATAT-3’

For CRISPR-mediated loss-of-function (indels), we created vectors containing gRNAs targeting Ncam1 and Fgf18 under the control of the U6 promoter. Two independent gRNAs were designed for each gene using the Benchling sgRNA design tool. Guides were selected based on predicted on-target and off-target efficiency and inclusion in either the Asiago library (Doench et al., 2016) or CRISPick. For each gene, a single plasmid encoding both gRNAs was constructed. Plasmids targeting Ncam1 and Fgf18 were co-electroporated at a concentration of 1ug/ul each plasmid. A non-targeting guide sequence was electroporated (2 μg/μl) and quantified as a control. Constructs were delivered by in utero electroporation (IUE) performed at embryonic stage 12.5 (E12.5). Progeny were analyzed by immunohistochemistry at postnatal day 14 (P14). Periventricular heterotopias were observed in 2/6 mice analyzed after gRNA targeting of Ncam1 and Fgf18, no heterotopias were present in NTg controls. Cortical regions selected for analysis were located outside the heterotopic areas.

g Ncam1_1: 5’- GTATGCCGGCATCGTCGATGT - 3’

g Ncam1_2: 5’- GCCTTGAAGTTGATTTCCCCG - 3’

gFgf18_1: 5’- GGGGCTCGAGATGATGTGAGT - 3’

gFgf18_2: 5’- GTGTGGACTTCCGCATCCACG - 3’

NTg_1: 5’- GTCGTGCCTAGCTCGGTTGAG - 3’

NTg_2: 5’- GGGATTCGTGCGGGAGATACG - 3’

In utero electroporation (IUE) was performed as previously described^78–80. In brief, purified DNA (2 μg/μl) mixed with 0.005% fast green in sterile PBS was injected in utero into the lateral ventricle of CD1 embryos. Five pulses of 35V or 40V (for E12.5 and 14.5 respectively) and 50ms were delivered at 1s intervals with a CYU21EDIT square wave electroporator and 1cm diameter platinum electrodes (Napa Gene).

Electroporated progeny were analyzed at different postnatal stages: P14 (which corresponds to the initial phase of myelination in deep layers), and P21 (when the mouse cortex is actively being myelinated).

Fluorescence immunolabelling

Immunohistochemistry was performed using P14 and P21 mice. Animals were anesthetized with tribromoethanol (Avertin) and transcardially perfused with 0.1 M PBS (phosphate buffered saline, pH 7.4) followed by ice-cold 4% paraformaldehyde (Electron Microscopy Sciences, 15710), as previously described⁶⁹. Cortical tissue was then post-fixed overnight in 4% paraformaldehyde, followed by 3X 10-minutes washes in 0.1 M PBS. Serial coronal sections (40–50 μm thick) were cut using a Leica microtome (VT1000 S), collected in PBS and stored at 4°C. For most comparative analysis, sectioned matched, primary somatosensory cortex (S1 cortex) tissues were used for immunohistochemistry. Free-floating sections were blocked for 1–2 hour at room temperature in blocking buffer (PBS with 0.3–0.5% BSA, 0.3% Triton X-100, and 2–8% donkey serum), and then incubated overnight at 4°C in blocking buffer with the following primary antibodies: rat anti-MBP (1:100, MAB386, Millipore), rabbit anti-Cux1 (1:100, CDP M-222, Santa Cruz Biotechnology), rabbit anti-Cux1 (1:100, 11733–1-AP, Proteintech), rat anti-Ctip2 (1:100, ab18465, Abcam), mouse anti-APC (CC1 clone, 1:100, OP80, Millipore), goat anti-PDGFRα (1:100, Novus Biologicals, AF1062), rabbit anti-GFP (1:500, A11122, Invitrogen), mouse anti-MBP (BioLegend, previously Covance #SMI-99P), mouse anti-NeuN (1:100, MAB377, Millipore), rabbit anti-MAG (D4G3) (1:100, #9043, Cell Signaling), mouse anti-Ncam1 (1:100, AB5032, Millipore), mouse anti-Fgf18 (1:50, sc393471, Santa Cruz Biotechnology), goat anti-Brn2 (1:100, sc-6029, Santa Cruz Biotechnology). Secondary antibody labeling was performed in blocking buffer at room temperature for 2 hours as follows: donkey anti-rat IgG (H+L) Alexa Fluor 647 preabsorbed (1:500–800, ab150155, Abcam), donkey anti-mouse IgG (H+L) Alexa Fluor 546 (1:500, A10036, Thermo Fisher Scientific), donkey anti-rabbit IgG (H+L) Alexa Fluor 488 (1:500, a21206, Thermo Fisher Scientific), donkey anti-rabbit IgG (H+L) Alexa Fluor 647 (1:500, A-31573, Thermo Fisher Scientific), donkey anti-goat IgG (H+L) Alexa Fluor 546 (1:500, A11056, Thermo Fisher Scientific). Sections were mounted using Fluoromount-G (00–4958-02, Thermo Fisher Scientific).

For NCAM1 and FGF18 immunohistochemistry, postfixed P21 brains were submerge in 30% sucrose and 20 μm thick coronal section made on a cryostat (Leica CM1850). Antigen retrieval was performed by incubating the sections in boiling Tris-EDTA buffer (10mM Tris base, 1mM EDTA, 0.05% Tween 20, pH9.0) for 10 min before blocking.

Imaging

Wide-field fluorescence images were acquired using a Zeiss Axio Imager 2 upright microscope. Tile scan images spanning the entire brain slice were captured using a 10x air objective and stitched using Zeiss Zen Blue Software.

Sections stained with NeuN, Cux1 and Ctip2 were imaged in a Zeiss LSM980 confocal microscope with a 20x objective and 2048×2048 pixels per image. A 3um z-stack was acquired and one image of the slice used as representative image.

Sections stained with Brn2, Ctip2, and Fgf18 or Ncam1 were imaged in a Zeiss LSM900 confocal microscope with a 20x objective and 3100×3100 pixels per image. A 2um z-stack was acquired and the sum projection used for analysis.

High resolution imaging of axons for caliber analysis was done on a Zeiss LSM900 using Airyscan 2 detectors. Images were acquired with a 63x oil objective, 1834×1834 pixels per image, and a 0.2um z-stack. Airyscan processing was performed in each image.

Image Analysis

Reelin mutant and control mice

For every cortical section, we divided the cortex from the pia to the white matter into 10 horizontal bins, made the plot profile, and calculated the average pixel value of fluorescence for MPB, CTIP2, and CUX1 in each bin using Fiji⁸¹. The values for each bin were averaged for 9 sections (3 sections each from 3 individual animals from the S1 cortex) for each of the WT and Reeler^−/− to obtain an average plot profile.

Pearson’s correlation coefficient was computed for the average plot profile for MBP, CTIP2 and CUX1.

In vivo screen

For every tissue section, electroporated and contralateral regions of the S1 cortex, from pia to corpus callosum, were analyzed. Cortical thickness was divided in two equal-height bins, termed upper layers (UL) and deep layers (DL). To quantify MBP protein expression levels, we calculated the average pixel value per bin using Fiji (v 2.9.0/1.53t). To quantify CC1 density, images were thresholded, and masked CC1⁺ cell bodies automatically counted using the ‘analyze particles’ function of Fiji. To quantify PDGFRα⁺ density, images were thresholded and positive cells automatically counted using CellProfiler⁸². CC1 and PDGFRα density per mm² was calculated per bin.

The significance of the differences in the fluorescence intensities of MBP and MAG, and CC1 and PDGFRα density between IUE and contralateral was computed using linear models. For each candidate gene tested, linear models were fit on the data, modeling biological replicate (animal) and section as fixed effects, and with or without the electroporation as another fixed effect. Two fitted models were compared using the ANOVA function to evaluate the use of an additional predictor.

NCAM1 and FGF18 intensity

Fluorescence intensity was quantified for individual cells by manually delineating nuclear regions based on Brn2 or Ctip2 staining, which identify upper- and deep-layer neurons, respectively. Each nucleus-defined region of interest (ROI) was then expanded by 2 μm to include the surrounding cytoplasm. The fluorescence intensity of NCAM1 or FGF18 was measured within these expanded ROIs.

Axon caliber analysis

For upper-layer neurons overexpressing Ncam1 and/or Fgf18, axonal measurements were performed within cortical layer IV. For deep-layer neurons in which Ncam1 and/or Fgf18 were silenced, measurements were taken in the corpus callosum. Axonal caliber was assessed at 11 evenly spaced points along a 20 μm segment, and the mean caliber was calculated for each axon.

NE4C cell culture, transfection and qPCR analysis

The neural stem cell line NE-4C (CRL-2925, ATCC) was cultured according to the manufacturer’s protocol. Briefly, culture plates were pre-coated with poly-L-lysine for 2h prior to cell seeding. Cells were maintained in ATCC-formulated Eagle’s Minimum Essential Medium (#30–2003, ATCC) supplemented with 2mM L-Glutamine and 10% fetal bovine serum (FBS). Cells were incubated at 37°C and 5% CO₂.

For gene silencing, cells were transfected with the silencing plasmids or the scramble control using Lipofectamine 3000 (#L3000001, Thermo Fisher) following the manufacturer’s instructions. After 24h, RNA was extracted using the Quick-RNA MicroPrep Kit (#R1051, Zymo Research) in accordance with the provided protocol.

cDNA was synthesized using the SuperScript IV first-strand synthesis system (#18091050, Thermo Fisher) and quantified by quantitative real time-PCR using the Sso Advanced universal SYBR Green Supermix (#1725270, Bio-Rad) and the following primers: Ncam1 (Fw: AGGAGAAATCAGCGTTGGAG, Rv: TGTAGATGGTGAGGGTAGAGG), Fgf18 (Fw: AGTGGAGACAGATACCTTCGG, Rv: GTACTTGGCAGACATCAGGG), H3f3b (Fw: ACTGACTGTTCACAGACC, Rv: CCATCCCTTCTGCGTATTAG). Relative mRNA expression levels of Ncam1 and Fgf18 were normalized to H3f3b and expressed as fold change relative to the scramble control.

Quantification and statistical analysis

Statistical analyses were performed with R (versions 4.0.3, 4.0.0, and 0.2.2) or GraphPad (version 10.6.1). Statistical details and tests used are described in the methods. The number of animals, brain sections, as well as exact p-values are detailed in the corresponding figure legends. Significance was determined at p-value < 0.05.

All tables are provided as a separated excel file.

Supplementary Material

2

Table S1. Distribution of cell types and differentially expressed genes in the scRNA-seq dataset from microdissected cortical layers, related to Figure 1 and Mendeley Data Figure 1. Gene modules used to identify oligodendrocyte and non-oligodendrocyte populations in the Plp1-EGFP scRNA-seq datasets. Distribution of assigned cell types and assigned MOL states in scRNA-seq datasets from microdissected cortical layers at different ages. Differentially expressed genes in oligodendrocyte populations (overall and stratified by age and microdissected layer) in the scRNA-seq dataset.

3

Table S2. Feature loadings (gene scores) for the 15 pseudotime-variable cNMF dimensions, related to Figure 1 and Mendeley Data Figure 3.

4

Table S3. Differentially expressed genes between MOL states in our scRNA-seq dataset and MOL populations identified in Marques et al., 2016, and overlap between MOL state assignments identified here (manual annotation) and identities reported in Marques et al., 2016 (assigned by random forest classifier), related to Figure 1.

5

Table S4. Marker gene sets used for identification of projection neuron populations, oligodendrocyte lineage cells, and MOL states in the Slide-seq dataset, related to Figure 2.

6

Table S5. Ligand-Receptor pairs identified in using CellPhoneDB, related to Figure 3.

7

Table S6. Marker gene sets used to identify different oligodendrocyte and neuronal populations in the cortical scRNA-seq datasets at P7 and P21, related to Figure 3 and Mendeley Data Figure 4.

8

Table S7. qPCR primer sequences, related to Figure 7 and Mendeley Data Figure 5A.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rat anti-MBP	Millipore	Cat# MAB386; RRID:AB_94975
Rabbit anti-Cux1	Santa Cruz Biotechnology	Cat# CDP M-222
Rabbit anti-Cux1	Proteintech	Cat# 11733-1-AP; RRID:AB_2086995
Rat anti-Ctip2	Abcam	Cat# ab18465; RRID:AB_2064130
Mouse anti-APC (CC1 clone)	Millipore	Cat# OP80; RRID:AB_2057371
Goat anti-PDGFRa	Novus Biologicals	Cat# AF1062
rabbit anti-GFP	Invitrogen	Cat# A11122; RRID:AB_221569
mouse anti-MBP (SMI 99 clone)	BioLegend	Cat# 808403; RRID:AB_2734562
mouse anti-NeuN	Millipore	Cat# MAB377; RRID:AB_2298772
rabbit anti-MAG	Cell Signaling	Cat# 9043; RRID: AB_2665480
mouse anti-Ncam1	Millipore	Cat# AB5032 RRID: AB_11213653
mouse anti-Fgf18	Santa Cruz Biotechnology	Cat# sc393471
goat anti-Brn2	Santa Cruz Biotechnology	Cat# sc-6029 RRID: AB_2167385
donkey anti-rat IgG (H+L) Alexa Fluor 647 preabsorbed	Abcam	Cat# ab150155; RRID:AB_2813835
donkey anti-mouse IgG (H+L) Alexa Fluor 546	Thermo Fisher Scientific	Cat# A10036; RRID:AB_11180613
donkey anti-rabbit IgG (H+L) Alexa Fluor 488	Thermo Fisher Scientific	Cat# a21206; RRID:AB_2535792
donkey anti-rabbit IgG (H+L) Alexa Fluor 647	Thermo Fisher Scientific	Cat# A-31573; RRID:AB_2536183
donkey anti-goat IgG (H+L) Alexa Fluor 546	Thermo Fisher Scientific	Cat# A11056; RRID:AB_2534103
Bacterial and virus strains
DH5a competent E.coli	New England Biolabs	Cat# C2987H
Chemicals, peptides, and recombinant proteins
Hibernate A	Brain Bits	Cat# HALF-500
Fetal Bovine Serum	Cytiva	Cat# SH30070.03HI
4-hydroxytamoxifen	Sigma	Cat# H6278
BSA	NEB	Cat# B9000S
Vybrant DyeCycle Ruby	Thermo Fisher	Cat# V10273
DAPI	Thermo Fisher	Cat# D1306
NEBuilder HiFi DNA cloning	New England BioLabs	Cat# E5520S
Fast green	Sigma	Cat# F7258
Paraformaldehyde	Electron Microscopy Sciences	Cat# 15710
Triton X-100	Sigma	Cat# T9284
Donkey Serum	Sigma	Cat# S30
Fluoromount-G	Thermo Fisher	Cat# 00-4958-02
Lipofectamine 3000	Thermo Fisher	Cat# L3000001
SuperScript IV first-strand synthesis system	Thermo Fisher	Cat# 18091050,
Sso Advanced universal SYBR Green Supermix	Bio-Rad	Cat# 1725270
Critical commercial assays
Worthington Papain Dissociation System kit	Worthington Biochemical	Cat# LK003150
Chromium™ Single Cell 3’ Chip	10x Genomics	Cat# PN-120236
Chromium™ Single Cell 3’ Library & Gel Bead Kit v2	10x Genomics	Cat# PN-120237
Chromium™ Single Cell 3’ Library & Gel Bead Kit v3	10x Genomics	Cat# PN-1000121
EndoFree Plasmid Maxi Kit	Qiagen	Cat# 12362
Quick-RNA MicroPrep Kit	Zymo Research	Cat# R1051
Deposited data
Raw FASTQ files from single-cell RNA-Seq data of oligodendrocytes from micro-dissected layers across all time points and from all cells in the cortex at P7 and P21 are available at GEO.	This paper	https://www.ncbi.nlm.nih.gov/geo/ Accession number GSE233255
scRNA-Seq from adult mouse cortex	Yao et al.31	https://doi.org/10.1016/j.cell.2021.04.021
Slide-seqV2	Stickels et al.27	https://doi.org/10.1038/s41587-020-0739-1
Additional Data Figures are deposited in Mendeley Data	This paper	10.17632/khnrd29pyy.1
Experimental models: Cell lines
NE-4C Neural stem cell line	ATCC	ATCC CRL-2925
Experimental models: Organisms/strains
Mouse: Plp1-EGFP	Mallon et al.65	NA
Mouse: Fezf2CreERt2	Matho et al.66	NA
Mouse: CAGfloxStop-tdTomato Ai14-Gt(ROSA)26Sortm9(CAGtdTomato)Hze	The Jackson Laboratory	RRID:IMSR_JAX:007909
Mouse: Reln−/−	The Jackson Laboratory	RRID:IMSR_JAX:000235
Mouse: CD1 pregnant females	Charles River	Strain code 022
Oligonucleotides
shNcam1 (5’-CGTTGGAGAGTCCAAATTCTT-3’)	IDT	Zhou et al 2022
shFgf18 (5’-CCTGCACTTGCCTGTGTTT-3’)	IDT	Chen et al 2023
Scramble (5’-ACCAGTTGCGCCAATCATAT-3’)	IDT	NA
gNcam1_1 (5’-GTATGCCGGCATCGTCGATGT-3’)	IDT	NA
g Ncam1_2 (5’-GCCTTGAAGTTGATTTCCCCG-3’)	IDT	NA
gFgf18_1 (5’-GGGGCTCGAGATGATGTGAGT-3’)	IDT	NA
gFgf18_2 (5’-GTGTGGACTTCCGCATCCACG-3’)	IDT	NA
NTg_1 (5’-GTCGTGCCTAGCTCGGTTGAG-3’)	IDT	NA
NTg_2 (5’-GGGATTCGTGCGGGAGATACG-3’)	IDT	NA
Primers for qPCR, see Table S11	This paper	NA
Recombinant DNA
Plasmid: pCM153	This paper	NA
Plasmid: pCAG-GFP	Matsuda et al. 2004	Addgene #11150
MGC Mouse Ncam1 cDNA	Horizon Discovery	MMM1013-202763372
MGC Mouse Sema5a cDNA	Horizon Discovery	MMM1013-202858545
MGC Mouse Rspo3 cDNA	Horizon Discovery	MMM1013-202740672
Plasmid: p213-shNcam1	This paper	NA
Plasmid: p213-shFgf18	This paper	NA
Plasmid: p213-scr	This paper	NA
Plasmid: p215-gNcam1	This paper	NA
Plasmid: p215-gFgf18	This paper	NA
Plasmid: p215-NTg	This paper	NA
Software and algorithms
Fiji (v2.9.0/1.53t)	Schindelin et al.81	https://fiji.sc/
CellProfiler 3.0	McQuin et al.82	https://cellprofiler.org/
Prism 10 (10.6.1)	GraphPad	https://www.graphpad.com/
R (versions 4.0.0, 4.0.3, 0.0.2)	Open-source	https://cran.r-project.org/
BioRender.com	BioRender	https://www.biorender.com/
Adobe Illustrator	Illustrator	https://www.adobe.com/products/illustrator.html
Code is available at github and deposited at Zenodo.	This paper	https://github.com/kimkh415/oligo_myelination Zenodo DOI: DOI: 10.5281/zenodo.19829308

Highlights.

• Single-cell developmental atlas of OL reveals layer-specific maturation.

• Molecular communication atlas defines PN-OL interactions across layers and time.

• Neuron class-linked signals drive differential layer myelination.

• NCAM1 and FGF18 are critical for cortical myelination.