A single-nucleus transcriptomic atlas of the developing mouse whisker cortex

To begin, we used droplet-based snRNA-seq to establish a reference atlas of wS1 neurons at P22 in mice with normal whisker experience. By P22, all established CPs are complete (Fig 1B, 1C). Next, to evaluate the influence of natural development and whisker experience on neuronal transcriptomic profiles, we used the same approach to profile wS1 in three additional experiments to compare with the reference atlas (Fig 1B). First, to identify temporally regulated genes and study cell type maturation from the onset of active whisking, we profiled wS1 from whisker-intact mice at P12, which coincides with peak synaptogenesis and the onset of CP1. Second, to study the role of whisker experience in guiding cell type maturation, we performed all-whisker deprivation (AWD) by plucking all whiskers bilaterally from P12 to P22, and tested how this alters the composition of cell types and/or their gene expression programs. Third, we tested whether brief (1-day) deprivation of the B and D rows of whiskers at P21 (row-based deprivation, RWD), a manipulation known to drive plasticity of excitatory and inhibitory circuits within deprived whisker columns [18,20], caused cell type-specific transcriptomic changes that could explain this physiological plasticity.

Data from each of these four experiments consisted of 2–4 snRNA-seq biological replicates, with each replicate consisting of pooled cells collected from three mice (Materials and methods). The resulting gene expression matrices were filtered to remove low-quality cells and cell-doublets [21], cells from non-neuronal populations, cells with a high proportion of mitochondrial transcripts (>1%), and cells that mapped poorly to other cortical datasets (Materials and methods). In total, we obtained 81,462 high-quality nuclear transcriptomes corresponding to neurons across the four conditions profiled (Figs 1C and S1A–S1C). We used standard dimensionality reduction and clustering approaches to hierarchically taxonomize wS1 neurons into 2 classes, 12 subclasses, and 53 molecularly distinct neuronal types. Annotations were performed by leveraging known markers, natural clustering, and supervised mapping to established cortical datasets (Fig 1D, 1E, Materials and methods). The relative frequencies of the neuronal subclasses, which spanned two orders of magnitude, were highly consistent across biological replicates within an experimental condition and across experimental conditions (Fig 1F).

Developmental gene expression changes are subclass-specific

To characterize gene expression patterns during normal development, we compared data from the P12 and P22 (normal experience) experiments within each cortical subclass. We computed two scores for each gene: a subclass variability (SV) score, based on its maximum expression fold change among the subclasses, and a temporal differential expression (tDE) score, based on its maximum fold change between P12 and P22 among the subclasses. A total of ~4000 genes, each expressed in >40% of cells in at least one time point and at least one subclass, were included in this analysis (S1 Table). Based on their scores (fold-change (FC) >2 along each axis), genes were stratified into four quadrants Q1–Q4 (Fig 2A, Materials and methods). Glutamatergic and GABAergic subclasses showed a similar proportion of genes (~70%–80%) with low SV and tDE scores (Q3 in Fig 2B). However, glutamatergic subclasses were > 3× more enriched than GABAergic subclasses (5.3% versus 1.5%) in genes with high subclass-variability and high tDE (Q1) and 1.5× more enriched (4.5% versus 3%) in genes with low subclass-variability and high tDE (Q2). These results indicate that glutamatergic neurons undergo greater transcriptional changes during this period of postnatal maturation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Gene expression changes between P12 and P22 are enriched in neurodevelopmental processes.

(A) Overview of analyses to classify genes based on subclass variability and temporal differential expression. P12 contains two biological replicates, and P22 contains four biological replicates. Each replicate contains cells from three mice. (B) Scatter plot of subclass variability (SV) scores and temporal differential expression (tDE) scores of genes in glutamatergic (left) and GABAergic neurons (right). Scores along each axis are capped at the value of 2 (S1 Data). (C) Same as B for glutamatergic neurons with four gene categories highlighted. Subpanels (clockwise, starting from top left) correspond to cell adhesion molecules (CAMs), transcription factors (TFs), housekeeping genes (HKs), and ion channels (ICs). Boxes on the right of each panel list the odds ratio (OR) and adjusted p-value (P_adj) for the enrichment of the corresponding gene category in each quadrant (Fisher’s exact test). Gray values indicate neither enrichment nor depletion, while red and green indicate depletion and enrichment, respectively (see Materials and methods, S1 Data for details). (D) Q1 is enriched in gene ontology (GO) programs associated with neurodevelopment, while Q3 is enriched in general housekeeping processes (S1 Data). (E) Stacked violin plot of example genes (rows) from Q1 with the highest tDE score in L2/3. Columns correspond to subclasses at P12 and P22, violins represent the expression distribution, and color represents median expression. Genes are colored according to the functional categories as in panel C (S1 Data). (F) FISH for tDE genes. Representative images (top) from an ‘across-row’ (see Materials and methods) section in wS1. Cortical layers 2/3 and 4 are indicated on the sections. Quantification (bottom) of mean intensity in nuclei revealed strong temporal regulation of gene expression in L2/3 consistent with what was measured with snRNA-seq (see Fig 2E). Mann–Whitney test, ****p < 0.0001. L2/3_Col19a1: n = 1,390, 1,042 nuclei, L4_Col19a1: n = 1,028, 813 nuclei, L2/3_Sorcs3: n = 1,455, 1,102, L4_Sorcs3: n = 1,021, 813 L2/3_Gabrg3: n = 1,433, 1,311, L4_Gabrg3: n = 1,026, 812. Three mice per condition, two slices per mouse (S1 Data).

https://doi.org/10.1371/journal.pbio.3003176.g002

To understand the functional significance of Q1–Q4 for glutamatergic neurons, we computed the enrichment of curated lists of transcription factors (TFs), cell adhesion molecules (CAMs), ion channels (ICs), and housekeeping genes (HKs) within glutamatergic subclasses using Fisher’s exact test (Fig 2C, Materials and methods). HKs were primarily enriched in Q3, consistent with their constitutive, non-specific expression. TFs were not significantly enriched in any quadrant, reflecting their broad role in transcriptional regulation across all subclasses during development [22,23]. CAMs, many of which regulate circuit formation [24], were strongly enriched in Q1 and, to a lesser degree in Q4, consistent with their cell type-specific and dynamic expression patterns during circuit formation reported previously in other contexts [23,25–27]. ICs were also enriched in Q1 and Q4, which aligns with physiological changes in neurons during this developmental period. Gene ontology (GO) analysis further supports these trends: Q1 genes were enriched for “cell adhesion” and “axon guidance”, while Q3 genes were associated with general cellular processes such as “protein/mRNA transport” and “chromatin reorganization” (Figs 2D and S1D). Q2 was enriched in “synaptic translation” terms, with ~60% of its enriched GO terms (8/13) overlapping with Q3 (S1D and S2A Figs). Q4 shared terms with Q1 and Q3, including “ion transport”, “cell–cell adhesion”, and “protein phosphorylation” (S2A and S1D Figs). Together, these results suggest that Q1 is a subclass-specific program related to circuit development, Q2 is a global developmental program, Q3 is a static “housekeeping” program, and Q4 is a subclass-specific program associated with neuronal identity. Similar trends were observed in GABAergic subclasses, albeit with fewer Q1 and Q2 genes (Figs 2B and S2B).

Next, we examined subclass-specific temporal gene regulation by identifying genes significantly up- or down-regulated between P12 and P22 (FC > 2 between P12 and P22, FDR < 0.05, Wilcoxon rank-sum test) (S2 Table). Of ~421 genes identified, > 60% exhibited subclass-specific expression changes (S2C, S2D Fig), with most subclass-specific genes enriched in Q1, and shared genes enriched in Q2 (Figs 2B and S2E, S2F). We detected ~ 1.5× more downregulated than upregulated genes (S2D Fig), a pattern observed in several studies of fly brain development, where gene repression plays a prominent developmental role [28–30].

Fig 2E shows examples of temporally regulated genes enriched in L2/3 glutamatergic neurons, including CAMs, TFs, ICs, and other genes. To validate the snRNA-seq findings, we performed multiplexed FISH using RNAscope (Fig 2F). We verified assay reliability using control probes. As expected, positive control probes showed widespread expression in the brain, while negative control probes produced no detectable signal (S2G, S2H Fig). We then probed three temporally regulated genes enriched in L2/3 (Col19a1, Sorcs3, and Gabrg3) at P12 and P22 in wS1 (Fig 2E, 2F). The pattern of temporal regulation of gene expression was consistent between snRNA-seq and FISH (Fig 2E, 2F), confirming subclass-specific temporal gene regulation in wS1 during postnatal development.

Developmental changes in L2/3 cell type composition

We next analyzed the maturation of cell types between P12 and P22. Using a decision tree-based classifier [31] trained on the cell type transcriptional profiles at P22, we assigned P22 type labels to each P12 transcriptome (Figs 3A and S3A, S3B; Materials and methods, S3 Table). Based on transcriptomic similarity, each neuron at P12 could be unequivocally classified into one of the 45 cell types at P22. This mapping allowed us to compare each cell type between the two ages.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. L2/3 pyramidal cell type composition and gene expression are selectively regulated during development.

(A) Schematic for transferring P22 cell type labels (reference data) to P12 cells to facilitate cell type-level comparisons. (B) Within glutamatergic neurons (~80% of all neurons), all cell types except for L2/3_A and L2/3_B have approximately the same relative frequency between P12 and P22. Pearson correlation coefficient between the relative frequencies is indicated on top (S1 Data). (C) Scatter plot of PC1 vs. PC2 for L2/3 neurons at P12 (top row) and P22 (bottom row). Within each row, the leftmost panel highlights cells colored by their type-identity (L2/3_A, L2/3_B, and L2/3_C). In the remaining three panels within each row, cells are colored based on their aggregate expression levels of markers for each type (see Materials and methods). Between P12 and P22, L2/3_A markers decrease in expression and increase in specificity, while L2/3_B markers increase in expression during development, driving cell type identity maturation (S1 Data). (D) Representative FISH images showing labeling of cell type markers Cdh13, Trpc6, and Chrm2 at P12 and P22 within wS1 L2/3. Arrowheads indicate putative Cdh13-expressing interneurons (See S5A Fig). (E) Summary plots based on overlay of all images of L2/3 visualizing expression of Cdh13, Trpc6, and Chrm2 at P12 and P22. Circles represent individual excitatory cells within L2/3, colored based on their expression of one or more marker genes (color code as in panel E). To the right of each summary plot is the kernel density estimate for each type along the pial-ventricular axis. Cdh13+  cells dominate in upper L2/3 at P12, whereas Trpc6+ cells are more abundant at P22. N = 10–12 slices from 3 mice per time point. (F) Quantification of the fraction of excitatory (Slc17a7+) L2/3 cells expressing one or more markers Cdh13, Trpc6, and Chrm2 at P12 and P22. N = 10–12 slices from 3 mice per time point (S1 Data).

https://doi.org/10.1371/journal.pbio.3003176.g003

Despite spanning two orders of magnitude, cell type frequencies correlated tightly between P12 and P22 within all four GABAergic subclasses and 7/8 glutamatergic subclasses (Figs 3B and S3C). This suggests that most wS1 cell types acquire a distinct and stable transcriptional signature prior to P12. The only exception was the L2/3 glutamatergic neuronal subclass (Fig 3B). Similar to what was observed in V1 [10], L2/3 glutamatergic neurons in wS1 can be clustered into three nominal types that we label L2/3_A, L2/3_B, and L2/3_C. The relative frequency of L2/3_C was similar between the two ages, but L2/3_A halved from P12 to P22, while L2/3_B doubled from P12 to P22 (Fig 3B). Despite this prominent change in cell type composition, a principal component analysis (PCA) performed using cell type markers in L2/3 glutamatergic neurons revealed that the underlying transcriptomic programs remained highly similar between P12 and P22. The top two principal eigenvectors, which accounted for ~35%–40% of the overall variance showed a 1:1 correspondence between P12 and P22 (S3D, S3E Fig), but with a slight reduction in the correlation coefficient, likely due to the compositional shift of L2/3 neurons from P12 to P22. This temporal variation in cell type frequency coincided with (i) a marked increase in the specificity of type A and C markers and (ii) increased expression strength of B markers from P12 to P22 (Figs 3C and S4A). Consistent with the shift in cell type composition, > 20% of L2/3 type-specific markers were temporally regulated, the highest among all subclasses (S4B Fig).

Genes distinguishing L2/3 cell types and their temporal regulation

To validate the cell type compositional shift, we performed FISH in wS1, targeting the marker Cdh13 for L2/3_A, Trpc6 for L2/3_B, and Chrm2 for L2/3_C (S4C, S4D Fig). These marker genes were also previously identified in V1 [10]. FISH confirmed that between P12 and P22, Cdh13+  L2/3_A cells decreased in frequency, Trpc6+ L2/3_B cells increased in frequency, and Chrm2+ L2/3_C cells remained stable (Fig 3D–3F). At P12, L2/3_A cells were concentrated at the L1/2 border, with a small population of L2/3_B cells and a large population of L2/3_C cells. At P22, L2/3_A cells were largely absent at the L1/2 border and instead concentrated in the middle of L2/3 (Fig 3E). As in V1, Cdh13+  cells in the center of L2/3 likely represent inhibitory neurons, as they do not co-express Slc17a7 (vGlut1) (S5A Fig). Between P12 to P22, the proportion of Chrm2+ L2/3_C cells remained stable, while Trpc6+ L2/3_B cells became more prevalent compared to Cdh13+ L2/3_A cells (Fig 3F). In contrast to V1, where L2/3_A cells remained localized to L1/2 at ~P21 despite decreasing in frequency, in wS1, L2/3_A cells decrease in frequency at the L1/2 border while Trpc6+ L2/3_B cells become more prevalent (Figs 3F and S5B, S5C). This mirrored the compositional trends observed in the snRNA-seq data (Fig 3B). As another validation of L2/3 compositional changes, we targeted Adamts2 for L2/3_A, Bdnf for L2/3_B, and Chrm2 for L2/3C [10,32]. The expression of Adamts2 and Bdnf showed the expected trend between P12 and P22 (S5D, S5E Fig), with Adamts2+ cells decreasing in abundance while Bdnf+ cells increased, agreeing with the snRNA-seq results (S4C Fig).

Additionally, S6 Fig highlights temporally regulated TFs, ICs, and CAMs differentially expressed among L2/3_A-C types. Although gene enrichment for each type is not strictly discrete, reflecting continuous transcriptional variation [10,19,33], some genes (e.g., Meis2, Foxp1, Kcnh7, Dscaml1) retain their expression patterns from P12 to P22, suggesting roles in the initial establishment and/or maintenance of cell type identity. Other genes (e.g., Rfx3, Zbtb16, Scn9a, Ncam2) are significantly temporally regulated and may be involved in the refinement of these cell type identities, including the maturation of their circuitry and physiology.

Taken together, our results suggest that for most wS1 neuronal types, transcriptomic distinctions and relative composition are established before P12 and persist despite developmental gene expression changes during this period of early sensory experience (Fig 2). The exceptions to this rule were L2/3 types whose composition and type-specific gene signatures are significantly altered between P12 and P22, mirroring trends reported earlier in V1 [10]. This raises the question of whether V1 and wS1 share global and subclass-specific gene expression programs related to maturation, which we now address.

Differences and similarities between wS1 and V1 in cell type development and gene expression changes

The results thus far have highlighted multiple similarities between wS1 and V1 maturation. For example, both regions share the same transcriptomic subclasses and markers, L2/3 is selectively regulated by development in both, and both contain equivalent L2/3 cell types and markers [10]. These similarities motivated a systematic comparison of developmental gene expression changes between wS1 and V1.

We identified tDE genes in V1 between P14 and P21—the time points in the published data [10] closest to those used in this study (S4 Table). As in wS1, temporal gene regulation in V1 was predominantly subclass-specific (S7A, S7B Fig). Hypergeometric tests revealed that subclass-specific tDE genes were shared between wS1 and V1 for most subclasses (Fig 4A). For a few subclasses (e.g., inhibitory Sst, Vip and Lamp5, and excitatory L4 and L6b), there were too few tDE genes to conduct the analysis (S2C and S7A Figs). The shared subset of temporally regulated genes was enriched in GO terms related to neuronal development, including “GABAergic synapse”, “glutamatergic synapse”, “axon”, “nervous system development”, “axon guidance”, and other related programs (Figs 4B and S7C). Within L2/3, the overlapping genes were drawn from various functional categories, including CAMs, TFs, ICs, and NTRs (Fig 4C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Comparative transcriptomic analysis of wS1 and V1.

(A) Heatmap highlighting the overlap of tDE genes between all pairs of wS1 subclasses (P12 vs. P22; rows) and V1 subclasses (P14 vs. P21; columns). The left and right panels correspond to downregulated and upregulated genes. Values correspond to Bonferroni-corrected -log₁₀(P_adj) values from a hypergeometric test of overlap of tDE genes. The background set for these tests was the set of all the genes regulated in any subclass. Except for L4, subclasses with fewer than 10 tDE genes (S2C and S7A Figs) showed little to no overlap. The value P_adj = 0.05 is highlighted on the scalebar (right). (B) Top 5 GO terms from shared downregulated and upregulated genes in V1 and wS1 in corresponding subclasses from panel A. See S7C Fig for a full list (S1 Data). (C) Examples of shared genes that are temporally downregulated (left) and upregulated (right) in L2/3 neurons between V1 and wS1 (S1 Data). (D) Scatter plots showing highly similar relative frequencies between matched cell types (see S7D Fig) across V1 and wS1 (S1 Data).

https://doi.org/10.1371/journal.pbio.3003176.g004

Moreover, wS1 and V1 share a striking correspondence in cell type composition. We trained classifiers on the V1 cell types at P14 and P21 and used them to transfer labels onto wS1 at P12 and P22, respectively. All 45 wS1 cell types mapped to the correct V1 subclass, and most cell types mapped 1:1 (S7D Fig). We also found that the relative cell type frequencies were highly concordant between the matched ages, indicating a high degree of overlap in cell type identity between the two cortical regions (Fig 4D). Together, this suggests that the two cortical regions share developmental programs at cell type resolution.

All-whisker deprivation from P12 to P22 does not alter cell type development

In mouse V1, 1–2 weeks of dark rearing during development selectively alters the transcriptomes of L2/3 cell types, revealing a requirement of visual experience for cell type development [10,19]. To test whether WD has a similar effect on cell type development in wS1, we plucked all whiskers on the face from P12 to P22 and analyzed the resulting gene expression changes. We refer to this manipulation as 10-day AWD (10d AWD) (Fig 5A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Full-face 10-day all-whisker deprivation (10d AWD) does not influence cell type maturation.

(A) All whisker deprivation is conducted for 10 days, from P12 to P22 (10-d AWD). All whiskers on both sides of the face are plucked, then checked every other day and plucked if there is regrowth. Controls are sham mice, which are anesthetized the same amount of time as 10d AWD but not plucked. (B) Glutamatergic neuronal types have approximately the same relative frequency in P22 10d AWD (y-axis) and normal P22 (x-axis) (S1 Data). (C) PC1 vs. PC2 scatter plot for L2/3 neurons at P22 10d AWD (top) and normal P22 (bottom), highlighting the location of types and the type-specific marker scores. Representation as in Fig 3C. Scores are similar between the two datasets (see also S8D Fig) (S1 Data). (D) L2/3 markers genes, as in S4C Fig, are shown as a function of cells’ position along PC1 comparing patterns between P22 with normal experience and P22 10d AWD (S1 Data).

https://doi.org/10.1371/journal.pbio.3003176.g005

We repeated the analysis outlined in Fig 3A, 3B, training a classifier on the P22 cell types and using it to transfer P22 labels to 10d AWD transcriptomes. L2/3 composition and cell type frequency at P22 were minimally affected by 10d AWD (Figs 5B and S8A). Similar to the developmental analysis in Fig 3, we used PCA to study the influence on cell type composition in 10d AWD. For normal versus 10d AWD P22, the top two principal eigenvectors, which captured >40% of the variance, exhibited a more robust 1:1 mapping compared to the case of the normal P22 versus normal P12 (S8B, S8C, and S3E Figs), consistent with 10d AWD having a minimal effect on the overall expression levels and the correlation structure of gene expression. Furthermore, the expression patterns of L2/3 type markers along the first two principal components (PCs) were highly similar between normal and 10d AWD mice at P22 (Figs 5C, 5D, and S8D).

Analysis of differentially expressed genes (DEGs) further confirmed that 10d AWD had a minimal influence on gene expression in all subclasses (S5 Table). We found 51 DEGs, which is ~ 8x smaller than the number of temporally regulated DEGs between P12 and P22 (S8E–S8G Fig). For most of DEGs, regulation was subclass-specific, with Lamp5 + interneurons exhibiting the largest number of changes. Altogether, we conclude that full-face WD between P12 and P22 minimally impacts the maturation of cell types and subclass-specific transcriptional programs in wS1.

Brief row-based whisker deprivation (P21–P22) upregulates column-specific activity-dependent gene expression programs in wS1

Previous work has shown that depriving a single row of whiskers for one day induces rapid plasticity in excitatory and inhibitory circuits in the columns corresponding to the deprived whiskers. This includes Hebbian and homeostatic synaptic plasticity and changes in the intrinsic excitability of neurons, with different mechanisms at play depending on the cell type and cortical layer [18,20,34–36]. The molecular factors mediating these layer-specific changes in response to WD are unknown.

We performed 1-day deprivation of the B and D whisker rows on P21 (1d RWD), a manipulation that induces competitive whisker map plasticity [18], and assessed transcriptomic cell types and gene expression changes in S1 using snRNA-seq and FISH at P22 (Fig 6A). Cell type distributions were highly correlated between normal and 1d RWD datasets at P22 (S9A, S9B Fig). As in the normal and 10d AWD P22 datasets, variation within L2/3 neurons was captured well by the top two PCs (S9C Fig), which correlated 1:1 with the top two PCs of normal P22 data (S9D Fig). Furthermore, the expression patterns of L2/3 type markers along the first two PCs were highly similar between normal and 1d RWD mice at P22 (S9E–S9G Fig). Altogether, this suggests that 1d RWD has a minimal influence on cell type identity and composition in wS1, as expected for such a brief sensory manipulation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Brief row whisker deprivation (1d RWD) upregulates activity-dependent gene expression programs in deprived columns.

(A) Schematic representation of the 1d row-whisker deprivation (RWD) manipulation and ‘across-row’ S1 section in which all barrel columns are identifiable. A representative image of DAPI labeling in ‘across-row’ S1 section. (B) Dotplots of genes upregulated by 1d RWD in glutamatergic subclasses (panels). Within each panel, rows indicate condition and columns indicate genes. The size of each circle corresponds to the % of cells with nonzero expression, and the color indicates average expression level (S1 Data). (C) Representative wide-field images of wS1. Barrels and barrel fields are indicated with light gray rectangles and labeled. Boxes indicate the locations of ROIs used for analysis in G. (D) Npas4 intensity inside Slc17a7+  L2/3 excitatory cells is increased after 1d RWD. Violin plots show median (dashed line) and quartiles (dotted lines) across individual cells. Mann–Whitney test, p < 0.0001 ****, n = 1,852, 1,699 cells, respectively, from 5 mice per group (S1 Data). (E) Junb intensity inside Slc17a7+ L2/3 cells is increased after 1d RWD. Mann–Whitney test, p < 0.0001 ****, n = 1,864, 1,703 cells, respectively, from 5 mice per group (S1 Data). (F) Btg2 intensity inside Slc17a7+ L2/3 cells is increased after 1d RWD. Mann–Whitney test, p < 0.0001 ****, n = 1,836, 1,706 cells respectively, from 5 mice per group (S1 Data). (G) L2/3 C and D columns were compared to examine whether gene expression changes are specific to the deprived (D) column. Pseudo-colored outlines of Npas4, Junb, and Btg2 expressing cells. Each plot is an overlay of 5 images of L2/3 S1 from 3 mice. Only Slc17a7+ cells are shown. (H) Quantification of column-specific gene expression changes in 1d RWD. Plotted values are the difference between the fraction of Npas4/Junb/Btg2 expressing cells among all excitatory (Slc17a7+) cells in the L2/3 D column and the corresponding fraction in the L2/3 C column. Symbols represent individual sham or 1d RWD mice. For each condition, mean and SEM (error bars) are shown. Two-way RM-ANOVA, Whisker Experience: F(1,8) = 11.41, **p = 0.0097 with Šidák correction for multiple comparisons, Npas4: **p = 0.002. JunB: n.s., p = 0.20. Btg2: *p = 0.014 (S1 Data).

https://doi.org/10.1371/journal.pbio.3003176.g006

It is well-established that sensory manipulation induces activity-dependent gene expression programs that result in long-lasting cellular adaptations essential for learning and maintaining circuit homeostasis [37,38]. We hypothesized that 1d RWD may elicit an activity-dependent transcriptional response (Fig 6A). Within each neuronal subclass, we sought genes expressed in at least 20% of cells that were upregulated in 1d RWD mice (FC > 2, FDR < 0.05, Wilcoxon rank-sum test) (S6 Table). We detected 59 such genes across all 12 subclasses. Fig 6B shows that the upregulated genes contain well-known “activity-dependent” genes such as Npas4, Btg2, Junb, and Nr4a1 [37]. Importantly, these genes were different from the few regulated by 10d AWD (S8 Fig), indicating their specificity to 1d RWD plasticity.

We performed FISH experiments to confirm these gene expression changes and assess their laminar and columnar organization in wS1 (Fig 6C). We used a plane of section that contains one whisker column from each of the five rows (A–E), enabling whisker row identity to be recognized since each whisker row maps to an array of cortical columns in wS1 [39] (Fig 6A). We probed for three IEG candidates shown in Fig 6C: Npas4, Btg2, and Junb. These mRNAs were significantly increased in L2/3 of wS1 as measured by intensity per L2/3 excitatory cell (Fig 6D–6F). To determine if the mRNAs were upregulated in a whisker column-specific manner, we separately analyzed the number of cells expressing each mRNA in excitatory L2/3 neurons in the deprived D versus non-deprived C column of the same tissue section from sham and deprived mice (Fig 6G, 6H). Npas4+ and Btg2+ cells were significantly upregulated in L2/3 of the D column compared to L2/3 of the C column in whisker-deprived mice but not in sham controls. This indicates whisker column-specific gene regulation during competitive map plasticity (Fig 6H). Junb+ cells showed a non-significant trend toward increase in D compared to C in 1d RWD mice. This suggests that the elevation in overall Junb intensity in L2/3 (Fig 6E) may be less column-specific than for Npas4 and Btg2. Together, these results suggest that 1d RWD triggers rapid and selective induction of genes functionally associated with activity-dependent plasticity in a spatially localized pattern wherein mRNAs are more strongly upregulated in the deprived columns than in neighboring spared columns. Such gene expression patterns may constitute an early molecular mechanism of competitive map plasticity.

A postnatal developmental atlas of mouse wS1

Previous studies have profiled transcriptomic cell diversity in the adult and adolescent somatosensory cortex (wS1) [7,40–42], a canonical model for studying cortical development and plasticity, but early postnatal gene expression changes shaping cell type identity remain unexplored. Here, we used snRNA-seq to analyze ~80,000 single-nucleus transcriptomes in mouse wS1 across P12 and P22 and three rearing conditions. The taxonomy and frequency distribution of ~45 neuronal cell types were conserved across time points and conditions, enabling identification of shared and cell type-specific changes during normal development and WD. Comparing P12 and P22, we identified ~250 temporally regulated genes, with over 60% being subclass-specific and linked to neuronal maturation processes like axon guidance and synapse formation. Glutamatergic neurons exhibited greater temporal regulation than GABAergic neurons, consistent with their increased evolutionary divergence [43]. By uncovering the molecular programs underlying postnatal development of sensory cortex, our research can inform understanding of neurodevelopmental disorders such as autism that are associated with circuit dysfunction and impaired sensory processing.

L2/3 cell types undergo significant changes during normal development

Of the 45 molecularly defined neuronal cell types, 42 were present at identical frequencies at P12 and P22, indicating that most wS1 neuron types are defined before P12. The only cell types that undergo significant changes in composition and cell type-specific gene expression between P12 and P22 are PYR cell subtypes in L2/3 (L2/3_A-C). The molecular identity and developmental changes of the three L2/3 cell types in wS1 mirror those observed in V1 [10], where L2/3 cell types have been shown to possess unique binocular tuning properties and correlate with distinct projections to higher visual areas [10,44]. It is therefore possible that the three L2/3 cell types in wS1 also have distinct functional properties and connectivity patterns. For example, L2/3_A cells were enriched in Cdh13, while L2/3_B/C cells were enriched in Cdh12. Because L5 PYR cells that express Cdh12 or Cdh13 are inhibited by CCK + versus PV+ interneurons, respectively [45], L2/3_A cells and L2/3_B/C cells may show a similar inhibitory circuit specialization. Similarly, in adult mouse S1, L2/3 cells enriched in the Baz1a gene (likely corresponding to our L2/3_B type) undergo robust changes in IEG expression in response to whisker input [46]. The delayed maturation that we observed for L2/3_B cells may be related to this pronounced experience-dependent plasticity in adulthood. Finally, the segregation of L2/3_A, B, and C cell types at different laminar positions suggests that the L2-situated cells (L2/3_A and B) may receive less VPM thalamocortical axonal innervation than L3-situated cells (L2/3_C), where VPM axon density is higher [47,48], as well as different local inhibitory inputs [49].

A potential mechanism underlying the developmental flexibility of L2/3 neurons could be the presence of a transcriptomic gradient. In PC space, we observed that L2/3 neurons form a triangle, with each of the three cell types occupying a vertex. Additionally, many markers were expressed in a gradient fashion, like the functional gradients observed for L2/3 neurons in V1 [50]. This suggests that wS1 L2/3 neurons exhibit an overall gradient pattern while also containing some specialist cell types, mirroring what has been observed in V1 and the intestine [10,19,51]. Occupying a flexible continuum could facilitate the drastic gene expression and compositional shifts observed in these neurons from P12 to P22.

Developmental transcriptomic conservation and divergence between V1 and wS1

Recent studies have shown a high degree of correspondence in neural diversity and cell type markers between cortical areas [52,53], and this was indeed the case between wS1 and V1. Our comparison of P12 to P22 in wS1 and P14 to P21 in V1 revealed similar developmental gene expression patterns. In most cases, there was significant overlap in the up- and down-regulated genes between matched neuronal subclasses between the two regions and little to no overlap across distinct subclasses. The notable exception was that L4 neurons did not share downregulated genes between the two regions (Fig 4A), which may be related to the fact that L4 contains spiny stellate cells in wS1 but not V1.

Despite this broad similarity, there was a laminar difference in the organization of L2/3 neuron subclasses between the two regions. Our data suggests that wS1 undergoes a similar remodeling of cell types from P12 to P22 as V1, with A-like cells decreasing and B-like cells increasing in number. However, unlike V1 where the L2/3_A-C cells segregate into sublaminae along the radial (pial-ventricular) axis at P21, we found that, in wS1, L2/3_A and L2/3_B somas are intermixed in superficial L2/3, but are spatially separated from the C somas, which reside in deep L2/3. The factors driving this difference between V1 and wS1 are unknown but may stem from broader differences in laminae across the sensory cortices [54].

The overall similarity in gene expression programs between V1 and wS1 is expected, as both regions share comparable timelines of postnatal development. Thalamocortical axons innervate L4, and callosal axons cross the midline and undergo activity-dependent rearrangement at similar stages in both regions [55,56]. Adult patterns of sensory input emerge in wS1 and V1 around the same time, with onset of active whisking at P12 and eye-opening at P14. However, the earliest sensory-evoked responses appear earlier in wS1, in the first few postnatal days [57], than in V1, which requires eye opening. Critical periods also differ, with classical critical periods for WD-induced plasticity in wS1 closing by ~ P16, while ocular dominance plasticity in V1 persists until P32 [13,58,59].

Despite the importance of sensory experience in wS1 circuit function, our findings reveal that WD from P12 to P22 does not impact cell type development in wS1, unlike dark rearing in V1, which alters gene expression and cell type composition in L2/3 [10,19]. This suggests that intrinsic programs may play a larger role in wS1 cell type maturation [60]. Why might L2/3 cell type development be regulated by visual experience in V1, but not by 10d AWD WD in wS1? One possibility is that the overall impact of sensory experience on cortical cell and circuit development is milder in wS1 than in V1. Consistent with this idea, ocular dominance shifts of V1 neurons driven by monocular deprivation are often stronger than whisker tuning shifts in wS1 driven by plucking a subset of whiskers [61,62]. During the critical period for ocular dominance plasticity in V1, visual experience shapes thalamocortical axons. It is required to maintain their structure, while in wS1, whisker sensory use does not alter thalamocortical or L4 topography after P4, and before that only nerve or follicle damage is sufficient to drive plasticity [56,63]. The critical period for ocular dominance plasticity has a stark ending at P32 in mice [58], while S1 retains many forms of experience-dependent plasticity into adulthood [3], suggesting that wS1 plasticity may not be as tied to developmental mechanisms as in V1.

Differences in sensory input deprivation paradigms may also explain this discrepancy [13]. Dark rearing in V1 completely eliminates visual sensory drive and statistical patterns of sensory input, leaving only retinal spontaneous activity, which has different properties. On the other hand, whisker plucking reduces but does not fully remove tactile input. For example, grooming and cuddling still provide afferent activation. Thus, plucking is likely to eliminate less sensory input than dark rearing and, hence, may have a less pronounced impact on S1 gene expression.

Competitive whisker deprivation drives plasticity-related gene expression programs in L2/3

Despite the lack of requirement for whisker experience on cell type maturation, we found that brief WD of only two rows of whiskers (1d RWD) upregulates activity-dependent gene expression programs in multiple wS1 cell types, including Npas4, Btg2, Junb, and Nr4a1, which are known to be involved in synaptic plasticity and circuit homeostasis [37]. Typically, expression of these IEGs is thought to be driven by increased levels of activity, as occurs during environmental enrichment (P65-75 animals) in S1 [64,65] or dark rearing followed by light exposure in V1 [66]. Interestingly, dark-rearing alone also increased the expression of several genes, particularly in L2/3 excitatory neurons [66]. Our finding that a subset of IEGs are upregulated in response to deprivation indicates that IEG induction may represent a general plasticity program broadly used to engage plasticity mechanisms rather than a specific consequence of elevated neural activity.

We found that deprived columns showed increased expression of IEGs relative to neighboring spared columns. This deprivation paradigm triggers synaptic plasticity, a component of competitive map plasticity, wherein whisker-evoked responses to deprived whiskers weaken and shrink within the whisker map while responses to spared whiskers strengthen and expand [67]. Thus, the selective upregulation of IEGs in deprived columns may represent an early molecular step underlying competitive map plasticity. Together, the 1d RWD results indicate that whisker experience exerts only relatively subtle effects on the development of cell types but impacts gene expression levels within specific cell types.

Summary and limitations

This work contributes a single-nucleus transcriptomic resource for postnatal wS1 development under normal and deprived sensory conditions. We delineated developmental changes in neuronal subclasses and types, and found that these were highly conserved between mouse V1 and wS1. However, whisker experience served a limited role in wS1 cell type development, which contrasts with the greater role of vision in V1 cell type development. Brief WD induced significant whisker column-specific changes in the expression of IEGs. Thus, while wS1 may rely less on experience and more on hardwired genetic programs than V1 to develop cell types, these cell types can undergo experience-dependent gene expression changes.

Throughout the study, we validated multiple scRNA-seq findings by performing FISH experiments targeting representative genes. Future studies could leverage spatial transcriptomics to quantify gene expression in wS1 barrels to further test the hypotheses established in this work. More research is needed to assess the gene regulatory networks, including chromatin organization and DNA methylation, that mediate experience-driven cell-type-specific transcriptome changes. Moreover, low sampling resolution or protein level changes could explain the limited transcriptomic effects observed during 10d AWD or 1d. Finally, integrating cell type-specific IEG expression with neuronal physiology using slice electrophysiology or in vivo two-photon imaging during whisker sensory behaviors could provide deeper insights into the functional consequences of gene expression changes.

Ethics statement

All methods, including the handling of vertebrate animals, followed NIH guidelines and were approved by the UC Berkeley Animal Care and Use Committee under Protocol ID Number AUP-2016-02-8351-3.

Computational methods

Experimental methods