Cortical RORβ is required for layer 4 transcriptional identity and barrel integrity

Retinoic acid-related orphan receptor beta (RORβ) is a transcription factor (TF) and marker of layer 4 (L4) neurons, which are distinctive both in transcriptional identity and the ability to form aggregates such as barrels in rodent somatosensory cortex. However, the relationship between transcriptional identity and L4 cytoarchitecture is largely unknown. We find RORβ is required in the cortex for L4 aggregation into barrels and thalamocortical afferent (TCA) segregation. Interestingly, barrel organization also degrades with age in wildtype mice. Loss of RORβ delays excitatory input and disrupts gene expression and chromatin accessibility, with down-regulation of L4 and up-regulation of L5 genes, suggesting a disruption in cellular specification. Expression and binding site accessibility change for many other TFs, including closure of neurodevelopmental TF binding sites and increased expression and binding capacity of activity-regulated TFs. Lastly, a putative target of RORβ, Thsd7a, is down-regulated without RORβ, and Thsd7a knock-out alone disrupts TCA organization in adult barrels.


Introduction
Localization of function is a fundamental principle organizing mammalian brain circuitry. Structure to function mapping is particularly striking in sensory input to L4 of the neocortex (Woolsey and Van der Loos, 1970;Catania and Kaas, 1995). L4 neurons are distinctive in their propensity to form cellular aggregates, or modules, that receive segregated thalamic inputs and represent features of the sensory periphery. Whisker barrels in the rodent somatosensory cortex are a prototypical example, but other somatosensory modules within L4 are also present in the cortices of insectivores, carnivores and primates (Krubitzer and Seelke, 2012), and columns receiving segregated input are present in the visual cortices of carnivores and primates, and in other cortical regions (Mountcastle, 1997). At the same time, gene expression studies in mouse and human show that L4 neurons also have a distinctive transcriptional identity that includes expression of RORb (Zeng et al., 2012). Despite these two striking features, little is known about the relationships between transcriptional identity, the mechanisms that establish and regulate that identity, and features of L4 cytoarchitecture.
Researchers have long used the rodent whisker pathway to study cytoarchitecture development (Hand and Strick, 1982;Fox, 1992;Yang et al., 2018). The whisker map is organized into microcolumnar units called barrels located in primary somatosensory cortex (S1). In mice, L4 cortical neurons assemble into columns that form barrel walls and input is relayed via thalamocortical afferents (TCAs), which cluster in the center of barrel hollows. Each whisker is projected through corollary maps in the brainstem and ventrobasal thalamus (Van Der Loos, 1976) before reaching S1.
Barrel formation is also regulated transcriptionally. Transcription factors (TFs) such as Bhlhe22/ Bhlhb5 and Eomes are involved in the early stages of cortical arealization and barrel development (Arnold et al., 2008;Joshi et al., 2008;Elsen et al., 2013). Downstream of these early developmental processes activity-dependent TFs, including Lmo4, NeuroD2, and Btbd3 regulate aspects of barrel organization in response to TCA inputs (Ince-Dunn et al., 2006;Kashani et al., 2006;Huang et al., 2009;Matsui et al., 2013). In addition, the TFs retinoic acid-related orphan receptor alpha (RORa) and beta (RORb), are also implicated in barrel formation. RORa and RORb are expressed in regions of the somatosensory barrel map, with RORa expressed in brainstem, thalamus and cortex, and RORb in thalamus and cortex (Nakagawa and O'Leary, 2003). Recently, RORa was shown to be required in the thalamus and cortex for proper TCA segregation and barrel wall formation (Vitalis et al., 2018). Mis-expression of RORb in neocortex is sufficient to drive cortical neuron clustering and TCA recruitment to ectopic barrel-like structures (Jabaudon et al., 2012). Together these studies have identified multiple TFs with major roles in early barrel development that likely set the stage for more downstream terminal differentiation TFs and activity-regulated TFs to hone the network. Early cortical development, TCA pathfinding, and activity dependent gene regulation are prolific areas of research. However, the later stages of neuronal specification and the molecular mechanisms of TFs involved in barrel development are currently underexplored. TFs such as Bhlh5 and Eomes have broad roles and are widely expressed in the cortex while the more narrowly expressed TFs such as Btbd3 are downstream of activity input leaving a gap in our understanding of the intermediate steps that connect cortical development to activity driven processes. Given the restricted layer-specific expression of RORb and its up-regulation concomitant with the final stages of barrel formation and the onset of input activity, we hypothesized it would be a good candidate to study transcriptional mechanisms connecting cellular specification in L4 with cytoarchitecture and network development.
We show that in addition to being sufficient, RORb is also required for both pre-and postsynaptic barrel organization. Without RORb in the cortex, L4 neurons fail to migrate tangentially and do not organize into barrel wall structures. This also reduced TCA segregation shortly after barrel formation would have normally occurred. Interestingly, TCA segregation also declined as animals aged. Without RORb, L4 gene expression and chromatin accessibility were disrupted, with L4-specific genes down-regulated and L5-specific genes up-regulated suggesting a disruption in terminal cellular identity. This involved complex changes in the expression and/or chromatin accessibility at binding motifs for many TFs in addition to RORb, including developmental regulators and activityregulated TFs. L4 neurons also received delayed excitatory input, a key step in barrel development. Lastly, we identify a putative direct gene target of RORb, Thsd7a, that is down-regulated without RORb and is required for maintained TCA organization in adulthood. Together these data characterize the role of RORb across multiple levels to connect molecular and transcriptional mechanisms to cortical organization and place RORb as a key regulator of a complex developmental transition orchestrating terminal L4 specification and initiating activity responsiveness.

Results
Cortical barrels in mice are complex structures. Cell-sparse barrel hollows are where thalamic projections are concentrated. Barrel walls are formed by cortical cell aggregates that surround the TCAs. Barrel septa consist of the intermediate spaces between barrel walls (Woolsey and Van der Loos, 1970). To assess the impact of RORb loss on barrel organization we used two staining methods. Barrel walls were visualized by Nissl staining (Van der Loos and Woolsey, 1973) and barrel hollows were visualized by vesicular glutamate transporter 2 (VGLUT2), which is strongly expressed in TCAs (Fremeau et al., 2001;Liguz-Lecznar and Skangiel-Kramska, 2007), or as clusters of reporter expressing afferents from VPM neurons. This strategy allowed clear identification of changes in either structure independently. Cytochrome oxidase (CO) staining was also used in some conditions, but the presence of CO signal in both barrel walls and TCAs made it less useful.

RORb is required for postnatal barrel wall formation and influences segregation of thalamocortical afferents (TCAs)
To begin exploring RORb function in barrel organization, we used a global, constitutive knock-out (KO), which contains a GFP expression cassette knocked-in to the Rorb locus. Rorb GFP/+ mice express GFP in RORb expressing cells allowing identification of barrel cortex without significant disruption to barrel structures or neuronal function (Liu et al., 2013). Rorb GFP/+ mice were used as controls (Ctl), while Rorb GFP/GFP mice disrupt both copies of Rorb to generate a KO. Controls showed no detectable disruption to barrel organization compared to WT animals ( Figures 1A and 2A).
Barrels form around postnatal day 5 (Rice and Van der Loos, 1977). Nissl staining of barrel walls at P7, P30, and P60 showed that RORb is required for barrel wall formation. Representative images of Nissl and GFP are shown in Figure 1A where the lack of barrel wall organization is clearly visible at P7 and remains disrupted at P30. Figure 1B quantifies this effect as the contrast between barrel hollows and barrel wall/septa fluorescence intensity. Contrast was calculated as (barrel -septa) / (barrel + septa) where septa includes barrel walls (see methods for details). Quantification demonstrated a near complete lack of contrast in KO barrel cortex supporting a lack of cortical organization.
While TCAs have been shown to instruct cortical cell organization we hypothesized the lack of barrel walls might reciprocally affect TCA organization. TCAs visualized by VGLUT2 staining showed an intact pattern of barrel hollows at P7 in KO animals, Figure 2A. However, careful quantification of the VGLUT2 contrast between hollows and septa showed a significant decrease in the KO suggesting loss of RORb and/or the lack of barrel walls had a mild but measurable effect on TCA segregation. Interestingly, as animals aged into adulthood TCA segregation also declined in control as well as Rorb KO animals. Disorganization in the Rorb KO was characterized by both loss of quantifiable VGLUT2 contrast as well as the qualitative barrel patterning most obvious at P60 between Ctl and KO in Figure 2A. Both genotype and age significantly affected VGLUT2 contrast (genotype p=4.5e-07 and age p=2.6e-06 by two-way ANOVA) but did not interact significantly. Comparing pairwise across ages we find a significant decline in TCA organization between P7 and P20 controls, with no significant change from P20 to P60. This suggests that while both age and loss of RORb significantly reduced contrast, loss of RORb did not significantly change the time course of TCA desegregation.  Ctl) and Rorb knock-out (KO) animals were age matched at P7, P30, and P60. (B) Quantification of barrel hollow to barrel walls/septa contrast (Barrel-Septa Contrast) from Nissl staining. N = 4 age-matched animals for each genotype (Ctl or KO). Two tissue sections containing the largest portions of whisker barrel field identified by GFP signal were averaged per animal. Whisker plots show the median per animal ± standard deviation. Gray points show mean contrast for each animal. P-value by independent sample t-test, between Ctl and KO at each timepoint.
To examine whether loss of VGLUT2 contrast could be due to late arrival of VGLUT2 + inputs from outside the VPM we injected AAV expressing mCherry under the hSyn promoter specifically into the VPM (Figure 2-figure supplement 1A). The VGLUT2 barrel-septa contrast was comparable to the barrel-septa contrast in the VPM-specific mCherry filled afferents at P30 strongly suggesting loss of VGLUT2 contrast with age is due to loss of TCA organization (  Rorb KO reduces thalamocortical afferent (TCA) segregation. (A) VGLUT2 staining of excitatory thalamic axon terminals in cortical whisker barrels shows normal initial TCA patterning at P7 but with reduced barrel-septa contrast in Rorb KO, and further reductions in contrast with age in both KO and Ctl. Ctl and Rorb KO animals were age matched. (B) Quantification of barrel hollow to barrel walls/septa contrast (Barrel-Septa Contrast) in VGLUT2. N = 4-6 age-matched animals for each genotype (Ctl or KO; each section shown is from a separate animal). Two tissue sections containing the largest portions of whisker barrel field identified by GFP signal were averaged per animal. Whisker plots show median contrast per animal ± standard deviation. Gray points show mean contrast for each animal. P-value by independent sample t-test, between Ctl and KO at each timepoint. The online version of this article includes the following figure supplement(s) for figure 2: Together these data show that RORb is critical for normal whisker barrel formation and, loss of TCA segregation into adulthood suggests that time/age continues to affect cytoarchitecture.

RORb is required in the cortex but not the thalamus for barrel organization
In addition to L4 excitatory neurons, RORb is expressed in the thalamic neurons that project to barrel hollows. To assess whether the disruption of barrels is dependent on RORb expression in thalamus and/or locally in cortex we used a floxed allele of Rorb (Rorb f/f ) crossed to Cre driver lines generating tissue-specific disruption of RORb as diagrammed in Figure 3A. A knock-in line expressing Cre from the serotonin transporter gene, Sert (Slc6a4 or 5-HTT) locus was used to knock-out Rorb in the thalamus. The Sert Cre line alone showed a mild disruption to TCA organization without disrupting barrel walls, suggesting the Cre knock-in might be hypomorphic ( Figure 3B-C). However, thalamic KO of Rorb (Sert Cre Rorb f/f ) showed no additional disruption to TCAs or barrel walls. This is consistent with the observation that Rorb KO also did not disrupt barreloid organization ( Figure 3-figure supplement 1A). Thus, loss of RORb in thalamic neurons was not responsible for the loss of cortical wall organization or the majority of TCA disorganization observed in the global Rorb GFP/GFP KO. A knock-in line expressing Cre from the Emx1 locus removed RORb specifically in forebrain structures. Emx1 Cre alone showed no significant disruption to barrel organization ( Figure 3D-E). However, barrel organization was significantly disrupted by cortical KO of Rorb (Emx1 Cre Rorb f/f ). In addition, a CamK2a Cre diver line that removes RORb in the cortex after barrel formation, showed no effect. CamK2a Cre activated expression of a tdTomato reporter from the Rosa26 locus in only a subset of GFP + L4 neurons (Figure 3-figure supplement 1B), therefore it is not clear whether late expression of RORb is expendable or whether expression in a subset of L4 neurons is sufficient for barrel organization. Together these data demonstrate that RORb is required in the cortex prior to barrel formation. Loss of RORb in the thalamus does not disrupt barrel architecture, suggesting RORb drives barrel wall organization through cell-intrinsic mechanisms within layer 4.

RORb is required for expression of a layer four gene profile and repression of layer five genes
Because RORb is a transcription factor we hypothesized loss of function would change gene expression in L4 neurons. To test this, RNA-seq was performed on sorted GFP + cells from micro-dissected L4 S1. We were careful in this dissection to exclude a small population of GFP + L5 neurons. Differential expression analysis between Rorb GFP/+ and Rorb GFP/GFP cells identified many dysregulated genes (fold change !2, adjusted p-value<0.01). At postnatal day 2 (P2) and prior to barrel formation, 246 genes were significantly disrupted with 51% down-regulated in the KO. At P7, just after barrel formation, 433 genes were disrupted with 36% down-regulated. At P30, 286 genes were disrupted with 37% down-regulated. Examining the overlap between ages we find very few genes significantly disrupted in the same direction across time points, suggesting highly dynamic and complex regulation, Figure 4A, B.
RORb expression is a key feature distinguishing L4 neurons (Lein et al., 2007). To examine the effect of RORb loss on layer-specific transcriptional identity we assessed the layer specificity of genes differentially expressed between control and Rorb KO (DEGs). The Allen Brain Atlas was used to manually screen all DEGs for layer-specific expression in the neocortex. Genes were considered layer-specific if the in-situ hybridization (ISH) signal appeared at least three-fold higher in one layer (considering layers 2 and 3 together). Many genes had complex specificities showing enrichment in two or more layers. These were not included for simplicity. Grouping DEGs based on the layer they are normally expressed within, we see that DEGs which should be expressed in upper layers were generally down-regulated and DEGs that should be expressed in deep layers were generally up-regulated in the Rorb KO, Figure 4A-D. The strongest effects were loss of many L4 genes and increased expression of many L5 genes. While many L4 and L5 genes were affected, this was not a global identity switch. Many L4 and L5 genes identified from the Allen Brain Atlas were not differentially expressed. In order to assess the statistical significance of the down-regulation of L4 genes and up-regulation of L5 genes we used the Allen Atlas differential search function to contrast L4 to L5 of primary somatosensory cortex (SSp) and included all genes with >1.5 fold change and expression threshold >1.6 ( Figure 4E-F). Of the 102 L4-specific genes 26% were down-regulated in the KO, a Figure 3. RORb is required in the cortex but not the thalamus for barrel organization. (A) Diagram and timeline of Cre driver line tissue-specific expression in cortex versus thalamus and timing relative to barrel formation and consolidation. Color indicates expression in cortex (red) or thalamus (purple). (B) VGLUT2 and Nissl staining of whisker barrel cortex at P30 from floxed Rorb control without Cre (Rorb f/f Ctl), Sert Cre control (Sert Ctl) without floxed Rorb and the cross (Rorb f/f Sert Cre ), which knocks out Rorb specifically in thalamus during embryonic development. Whisker plots as described for Figure 1B. (C) Quantification of VGLUT2 Barrel-Septa Contrast in genetic lines from B. N = 3-5 P30 animals. Quantification and plotting as described in Figure 2B. P-value by ANOVA. (D) VGLUT2 and Nissl staining of whisker barrel cortex from Emx1 Cre control (Emx1 Ctl) without floxed Rorb, and the cross (Rorb f/f Emx1 Cre ) from P7 and P30 animals, and a P60 animal from floxed Rorb crossed to a CamK2a Cre driver line. Emx1 Cre knocks out Rorb specifically in forebrain during embryonic development, and CamK2a Cre knocks out Rorb in forebrain neurons at postnatal weeks 2-3. (E) Quantification of VGLUT2 Barrel-Septa Contrast in genetic lines from D. N = 3-5 animals per age group. Quantification and plotting as described in Figure 2B. P-values by independent sample t-test, between Ctl and KO at each time point. CamK2a Cre showed no difference from Rorb f/f Ctl. Whisker plots as described for Figure 1B.   Lines plot the mean ± SE. P by moderated t-test adjusted for multiple comparisons (Benjamini-Hochberg). (E) Additional L4 and L5 genes were identified using the Allen Brain Atlas differential search contrasting L4 SSp structures to L5 SSp. Genes with >1.5 fold change and expression threshold >1.6 were selected. Genes already shown in A-C were removed. Hence each gene shown does not meet statistical criteria for differential expression in Ctl/KO by RNA-seq. Line plots show RNA-seq LFC for each layer-specific gene. The solid black line is the mean across genes and the solid gray line indicates no change. Negative LFC indicates decreased expression in Rorb KO, and LFC >0 indicates increased expression in Rorb KO. (F) Overall (first bar), 1% of genes were downregulated (blue) and 2% were upregulated (red). Downregulated genes were overrepresented (26%) among the 102 L4-specific genes (middle bar), while upregulated genes were overrepresented (19%) among the 240 L5-specific genes. Both overrepresentations were significant (p<2.2e-16) by fisher exact test.
single gene was up-regulated, and the remainder were unchanged. Conversely, up-regulated genes were overrepresented (19%) among the 240 L5-specific genes, and a fisher exact test revealed that these overrepresentations were highly significant (p<2.2e-16). Thus, although only a portion of the L4 gene expression profile is altered by loss of Rorb, it is disproportionately weighted towards down-regulation of L4 genes and upregulation of L5 genes.
Several L5 genes are worth noting. Bcl11B/Ctip2, is a marker of thick-tufted L5B-type neurons and significantly up-regulated at P2 in the KO, but silenced at P7 and P30 similar to control ( Figure 4D). Fezf2, another L5B marker and regulator of Bcl11B (Chen et al., 2005), was similarly silenced over barrel development, but was overexpressed at P30 in the KO. Foxo1 is mainly expressed in L5 at younger ages (Allen Developing Mouse Brain Atlas) declining over barrel development, but in the KO was significantly overexpressed at P7. Etv1, also a L5A marker (Doyle et al., 2008), was upregulated in the KO at both P2 and P30. Lastly, Egr4 was up-regulated at P30 in the KO, and has been associated with Etv1 expressing neurons (Doyle et al., 2008). RNAscope (Wang et al., 2012) in situ analysis against two L5 genes confirmed up-regulation in L4 ( Figure 5, Figure 5-figure supplement 1A). Together these data support a disorganized partial shift in layer identity with many different factors implicated at distinct time points.

Rorb KO disrupts transcription factor binding sites near DEGs
RORb, Bcl11b, Foxo1, Etv1, and Egr4 are TFs that often regulate gene expression by binding to distal regulatory sites such as enhancers. There are many chromatin features of enhancers, one of which is that they are open and accessible to enzymatic fragmentation in assays such as the Assay for Transposase Accessible Chromatin (ATAC) (Buenrostro et al., 2015). To begin examining mechanisms involved in changing gene expression, we performed ATAC-seq on sorted GFP + L4 neurons from control and Rorb KO animals at P30 ( Figure 6A). High confidence ATAC-seq peaks were assessed for differential accessibility between control and KO samples. We identified 5210 peaks with !2 fold change in accessibility (FDR < 0.02). Nearly 4-times as many regions lost accessibility (N = 4123 closed) than increased (N = 1087 opened), (Figure 6-figure supplement 1A). Differential ATAC peaks were primarily located in introns and intergenic regions Figure 6-figure supplement 1B suggesting loss of RORb function resulted in closure of many more regulatory regions than opening.
We hypothesized that many of the closed regions might contain a RORb binding motif while regions that opened may have binding potential for other TFs. To assess this possibility, two software algorithms (MEME and HOMER) were used to identify de novo enriched motifs from the DNA sequences of differential ATAC peaks separating closed and opened regions. This unbiased analysis also identifies which enriched sequences match known TF binding motifs. RORb was the top motif from closed regions, Figure 6-figure supplement 1C. Considering only expressed TFs, the potent neurogenic factors NeuroD1 and Ascl1 were also among the top motifs in closed regions. In regions that opened, the top motifs from expressed TFs were Nfil3, Hlf, Jun, Fos, Trps1, Mef2a/c/d and Irf2. Similar analysis was performed on ATAC peaks near up or down-regulated DEGs as well as L4 and L5 DEGs. To confirm enrichment and identify motif locations we used MEME FIMO and HOMER to scan for instances of a given set of motifs. This was done for all expressed TFs either enriched in the de novo motif analysis or differentially expressed, for which high quality motif models existed. Motif instances were cross-validated by retaining only those found by both MEME and HOMER. Figure 6B plots the odds ratio of motifs significantly enriched compared to control regions. Many of the motifs found by de novo analysis were confirmed, including RORb in regions that closed.
To assess which TFs might play a significant role in up or down-regulation of DEGs we varied a distance window around the transcription start site (TSS) to identify nearby ATAC or control regions containing a DNA motif. We tested for enrichment of motifs in ATAC regions near DEGs compared to motifs in control regions. We also tested whether DEGs with a nearby motif were significantly enriched compared to a control group of genes that did not change expression in the Rorb KO. In essence, we tested whether motifs were enriched around certain DEGs and whether a significant portion of those DEGs had a nearby motif. To reduce false positives, only motifs with significant enrichment in both tests are shown in Figure 6C-D.
Genes down-regulated at P30 showed significant enrichment of nearby RORb motifs suggesting RORb is important for gene activation ( Figure 6C). Motifs for Nr4a1 and Nfil3 were enriched near up-regulated DEGs at P2 and P7 respectively consistent with an early role for these TFs in activating . Two regions containing S1 were averaged per animal. Whisker plots show the median per animal ± standard deviation. Gray points show mean number of puncta per cell for each layer in each animal. P-value by Figure 5 continued on next page expression. Foxo1 motifs were enriched near genes down-regulated at P2 and P7. Consistent with a role in early gene regulation, Foxo1 was highly expressed at P2 and declined with age in control neurons ( Figure 4D). However, in the KO, Foxo1 remained significantly elevated at P7 eventually decreasing to levels comparable to control at P30. The close proximity of Foxo1 binding sites to down-regulated genes and its elevated expression at younger ages suggests it may act as a repressor that is normally silenced just after barrel formation to allow proper gene induction in L4 neurons. Without RORb, silencing of Foxo1 is delayed allowing it to aberrantly repress targets at younger ages.
Interestingly, we did not find RORb motifs enriched near L4 genes suggesting the shift in layerspecific gene expression is a downstream effect of RORb loss. While RORb does not appear to directly regulate layer-specific genes, Zfp281 motifs were enriched near L4 genes in the de novo motif search and confirmed by specific mapping (Figure 6-figure supplement 1C and Figure 6D). Zfp281 was highly expressed in both samples, at all ages, and unchanged by Rorb KO (Figure 6figure supplement 1C). Zfp281 motifs were also enriched in regions that closed in the Rorb KO suggesting it might be a novel activator of L4-specific genes and dependent on some other factor to maintain accessible chromatin at its binding sites.
Nfe2l and NeuroD1 motifs were enriched near L5 genes. NeuroD1 motifs were also enriched in regions that closed suggesting it might act as an inhibitor of L5-specific genes as these genes increased expression when NeuroD1 sites closed. Nfe2l consists of a family of TFs that share a binding motif. Nfe2l1 was expressed at younger ages and increased in the adult while Nfe2l3 was highly expressed at P2 and silenced by P7 ( Figure 6-figure supplement 1D). Rorb KO did not significantly disrupt expression of either, but the motif was enriched in regions that opened suggesting Nfe2l1 and/or three may be novel activators of L5-specific genes.
The TF motifs enriched near up-regulated DEGs were noteworthy for possible relationships with neuronal activity. Nr4a1 is an activity induced TF that regulates the density and distribution of excitatory synapses . Nfil3 and Hlf bind and compete for similar DNA motifs (Mitsui et al., 2001), and may also be involved in activity-regulated transcription. Nfil3 is up-regulated in human brain tissue following seizures (Beaumont et al., 2012), and mutations in Hlf are linked to spontaneous seizures (Gachon et al., 2004;Hawkins and Kearney, 2016). In addition, motifs for the classic immediate early genes, Jun and Fos, were enriched in regions that opened. These observations led us to examine the expression of other activity-regulated TFs. Many were significantly up-regulated at P30 while Lmo4 and its binding partner Lbd2 were up-regulated at P7 (Figure 6-figure supplement 1E; Matsui et al., 2013). Lmo4 expression is induced by calcium signaling and is required for TCA patterning in barrel cortex (Kashani et al., 2006;Huang et al., 2009). Another activity-regulated TF, Btbd3, which drives L4 neurons to orient their dendrites into barrel hollows, was significantly down-regulated ( Figure 6-figure supplement 1E). Lmo4 and Btbd3 are the only genes previously shown to disrupt barrels that were also dysregulated in the Rorb KO ( Figure 6-figure supplement 1F). In the Rorb KO Lmo4 was up-regulated, but Lmo4 KO disrupts barrels, suggesting that Rorb KO disrupts barrels through a divergent mechanism from what has been previously described.
Interestingly, the protein product of S100A10, p11, is involved in serotonin signaling via binding to the serotonin receptors Htr1b, Htr1d, and Htr4 (Warner-Schmidt et al., 2009). S100A10 was down-regulated at P7 and P30 (Figure 6-figure supplement 1G). Htr1b was the only of the three serotonin receptor subunits known to interact with p11 expressed in our samples and was also significantly down-regulated at P7 and P30. These data suggest that in addition to altered layer identity, Rorb KO may also disrupt serotonergic signaling, an important pathway in TCA communication with cortex (Kawasaki, 2015). Together with up-regulation of activity-regulated TFs, L4 neurons in the Rorb KO likely have significantly altered responses to activity. These analyses paint a complex picture where gene expression in L4 Rorb KO neurons is disrupted by multiple mechanisms. Loss of RORb results in closure of many RORb binding sites which are also enriched near genes with reduced expression in adults consistent with an activator role for RORb. Other regulatory changes involve complex combinations of altered TF expression and/or altered binding potential at sites that opened or closed in the KO likely due to downstream effects Figure 6. Rorb KO disrupts transcription factor binding sites near DEGs. (A) ATAC-seq normalized reads per million (RPM) for biological replicates, y-axis scaled 0-2. Samples collected from GFP + S1 L4 Rorb GFP/+ neurons (Ctl, blue) and GFP + S1 L4 Rorb GFP/GFP neurons (KO, red). Arrows indicate differential peaks (fold change !2, FDR < 0.02). Open arrows indicate differential peaks with transcription factor motif sequences as in (B). (B) Crossvalidated motifs with significant enrichment in ATAC peaks with differential accessibility. Closed; regions with significantly reduced access, Opened; regions with significantly increased access in the Rorb KO. Motif instances were cross-validated between MEME and HOMER algorithms. Odds ratio and p-value calculated comparing to motif frequency in control regions. (C-D) Cross-validated motif enrichment in ATAC peaks near the TSSs of (C) upregulated or down-regulated DEGs and (D) L4-or 5-specific genes. Bars plot odds ratio over control regions. Asterisk indicates significant motif enrichment (p<0.03 by Fisher exact test) in nearby ATAC peaks compared to control regions and separately significant enrichment (p<0.03 by Fisher exact test) of DEGs with a nearby motif compared to an independent group of control genes. The online version of this article includes the following figure supplement(s) for figure 6: of RORb loss. These changes impact both known neurodevelopmental regulators as well as activityregulated TFs.

Rorb KO delays excitatory input to barrel cortex
To examine whether RORb loss impacts network activity, we examined inhibitory and excitatory synaptic properties of L4 neurons. We found no change in inhibitory innervation at P14 or P24 as measured by miniature inhibitory postsynaptic currents (mIPSCs), Figure 7-figure supplement 1A-B. However, synaptic function as measured by miniature excitatory postsynaptic currents (mEPSCs) revealed a significant delay in excitatory input, Figure 7A-C. At P5, the frequency of mEPSCs was low and comparable in control and KO, Figure 7B-C. At P7, around the time when recurrent cortical synapses begin to sharply increase (Ashby and Isaac, 2011) and LTP has just ended (Crair and Malenka, 1995), controls showed increased mEPSC frequency. However, Rorb KO animals had a significantly lower mEPSC frequency at P7 ( Figure 7A-C), suggesting decreased functional synaptic input. At P10, Rorb KO neurons increased mEPSC frequency to levels comparable with controls. This suggests synaptic connections were delayed by Rorb KO mostly likely affecting recurrent excitatory connections. At P10, this defect in frequency is mostly corrected, but Rorb KO also showed significantly increased mEPSC amplitude at P10, possibly compensating for the delay at P7. By P19, both frequency and amplitude of mEPSCs were similar between control and KO ( Figure 7B). These data support a subtle functional disruption to the barrel circuit in Rorb KO animals that is consistent with the transcriptional changes.
The putative RORb target, Thsd7a, is required for adult TCA, but not barrel wall organization To begin exploring the relationship between disrupted gene expression in the Rorb KO and barrel organization, we examined known functions of genes differentially expressed at multiple developmental time points. Two candidates were identified with potential roles in cell migration and synaptogenesis. PlexinD1 (Plxnd1) is a cell signaling molecule known to play a role in pathfinding and synaptogenesis (Chauvet et al., 2007;Wang et al., 2015). Thrombospondin 7a (Thsd7a) regulates endothelial cell migration (Wang et al., 2010), but its role in the brain is unknown. In controls, expression of both genes followed a similar developmental trajectory as Rorb, peaking around P7 ( Figure 8A). In the Rorb KO, Plxnd1 was significantly lower at P2 and P7 while Thsd7a was significantly lower at all three time points. In addition, we identified several differential ATAC peaks near Thsd7a with reduced accessibility ( Figure 8B). This included a peak containing a strong RORb motif just downstream of the transcription start site, suggesting Thsd7a might be a direct target of RORb regulation.
There was no detectable disruption to barrel organization in Plxnd1 conditional KO mice (Plex-inD1 flox crossed to Emx1 cre , Figure 8C-D). A Thsd7a constitutive KO also showed no disruption to barrel wall organization at P7 or P30. Interestingly, Thsd7a KO did show decreased VGLUT2 contrast between barrels and septa at P30 but not P7, suggesting Thsd7a is important for maintenance of TCA organization in adulthood ( Figure 8C-D). The barrel phenotype of Thsd7a KO was qualitatively different from Rorb KO barrels. Specifically, the overall barrel pattern remained more intact in the Thsd7a KO despite the quantitative decrease in VGLUT2 contrast. As before, desegregation of VPM afferents was confirmed by VPM injection of AAV-hSyn-mCherry (Figure 8-figure supplement 1). Thsd7a KO may maintain sharper barrel borders than Rorb KO due to intact barrel walls. Reduction in VGLUT2 contrast in the Thsd7a KO could be due to increased TCA localization in the septa and/ or decreased TCA localization in the barrels. To distinguish these two possibilities, three regions of low VGLUT2 staining adjacent to the barrel field were quantified and used for within tissue slice normalization of barrel and septa intensities. Thsd7a KO resulted in a 24% decrease in barrel hollow VGLUT2 signal and a 56% increase in the septa ( Figure 8E-F). High resolution imaging showed a clear increase in VGLUT2 puncta located in the septa ( Figure 8G). Thus, loss of Thsd7a after Rorb KO likely contributes to the decrease in TCA segregation in adulthood.

Discussion
While somatotopic maps were one of the earliest and most obvious forms of cytoarchitecture, our understanding of the role neuronal identity plays in module formation is largely unknown. Studies have long approached the question of what drives cortical organization from the perspective of network activity and, in the case of barrel cortex, from the perspective of key structures and pathways needed to relay sensory input. More recent studies characterizing transcription factors required for barrel organization point to the importance of molecular mechanisms regulating transcriptional programs. However, many of these TFs are part of the pathways that carry sensory input or are fundamental regulators of broad developmental programs. It was unclear whether a TF such as RORb, a highly restricted marker of L4 identity, could influence macro-scale processes such as module formation. Indeed, we show that while RORb is clearly regulating only a fraction of the phenotypic and transcriptional properties of L4 neurons, it is necessary for terminal specification of L4 identity and proper organization of L4 cytoarchitecture.
Specifically, RORb is required in the cortex for barrel wall formation and full TCA segregation. This differs from earlier work focusing on the role of TCA patterning and activity as instructive for barrel wall formation. Instead, we find that loss of RORb specifically in the cortex affects TCA segregation shortly after barrel walls should have formed, suggesting that bidirectional signaling between L4 neurons and TCAs is involved in establishing proper organization. That such signaling occurs was first suggested by cortex-specific knockout of NMDA receptor subunits (Iwasato et al., 2000;Lee et al., 2005). A second study highlighting this role of cortical influence on TCA organization knocked out the metabotropic glutamate receptor Grm5 (Ballester-Rosado et al., 2016) in cortical neurons. In contrast, cortex-specific knockout of another member of the ROR family of transcription factors, Rora (Vitalis et al., 2018) disrupts the cellular organization of cortical barrels, but appears to leave TCA segregation intact.
While loss of RORb function affected TCA segregation from the time of formation we note that loss of the putative RORb gene target, Thsd7a, primarily affected TCA segregation in adults despite maximal expression at P7. While additional studies are needed, we speculate one possible explanation could be that Thsd7a functions around the time of barrel formation to establish long lasting TCA structures that only manifest aberrant phenotypes later in life. Alternatively, the moderate expression level of Thsd7a at P30 may be sufficient for a role in adult maintenance. In either case, a role for Thsd7a in the nervous system has not been described previously. In endothelial cells, Thsd7a localizes to the membrane of the leading edge of migrating cells where it functions to slow or inhibit migration (Wang et al., 2010). Perhaps in somatosensory cortex it inhibits movement of nearby projections such as dendrites or axons allowing cortical neurons to 'corral' TCAs in barrel hollows. Thsd7a is not the only potential downstream target of RORb worthy of further investigation. Pcdh20 has a role in L4 identity in regulating appropriate laminar positioning of L4 cells. Without Pcdh20, cells migrate to L2/3 instead (Oishi et al., 2016). In RORb KO cells, Pcdh20 is down-regulated but cells still migrate to L4 suggesting a possible novel role for Pcdh20 downstream of RORb function.
Our observation that barrel organization declined with age in wildtype animals is very interesting and possibly the first description of this phenomenon in mice (Rice, 1985). It suggests continued plasticity or degradation of maintenance mechanisms over time. Few studies have examined plasticity within this structure in adulthood. This is in part because studies have shown a decline in the capacity to rewire sensory input to the cerebral cortex with age in certain systems. In the visual system, loss of sensory input has been shown to alter TCAs during a critical postnatal period (Antonini and Stryker, 1993;Erzurumlu and Gaspar, 2012). It is thought that once this critical period closes, TCA organization is fixed. Thus, developmental processes in the visual and somatosensory systems are assumed to stabilize TCAs and restrict learning and memory related changes to plasticity among cortical connections (Fox, 2002;Feldman and Brecht, 2005;De Paola et al., 2006;Karmarkar and Dan, 2006). However, there is some evidence to support a shift in this model of adult plasticity in both the visual and somatosensory cortex (Khibnik et al., 2010;Wimmer et al., 2010). In particular, Oberlaender et al. showed that a mild form of sensory deprivation induced by whisker trimming in 3 month old rats substantially altered TCAs in barrel cortex (Oberlaender et al., 2012). However, because adult TCA plasticity has garnered limited attention, we currently lack genetic studies examining the molecular mechanisms behind these processes. The natural decline in barrel organization and the mechanism of Thsd7a influence on TCA segregation merit further investigation as exciting new contexts to study both the functional roles of cortical organization and the impact of age.
Recent studies are revealing that neuronal identity in certain structures remains plastic during early postnatal periods. For example, mistargeted L4 neurons that migrate to layer 2/3 take on characteristics of their surroundings (Oishi et al., 2016) and misexpression of some TF can alter the identity of postnatal neurons (Rouaux and Arlotta, 2010;Rouaux and Arlotta, 2013). We find that loss of RORb disrupts the transcriptional identity of L4 neurons, which lose expression of many L4 genes and aberrantly express many L5 genes. While this shift to a more L5-like transcriptional profile is not a global identity switch, it suggests L4 identity continues to be refined relative to deeper layer profiles late into postnatal development.
The complex expression changes observed likely occur through a multi-tiered reorchestration of gene regulation. Up-regulation of known L5 TFs such as Bcl11b/Ctip and Etv1 at P2 may help drive an early diversion down an L5-like trajectory. Regulatory signatures detected in adult neurons such as closure of binding sites for Zfp281 enriched near L4 genes and opening of Nfe2l1/3 motifs enriched near L5 genes may represent the tip of the developmental iceberg. In addition, our stringent motif analysis aimed to keep false positives low may also miss relevant regulators with more minor roles. While we detect changes in binding capacity for many TFs, including RORb, the complexity of dysregulation spread out across early postnatal development means there are certainly additional mechanisms driving this shift in cellular identity to be discovered. Here we combine the power of genetic knock-out strategies with multiple molecular profiling assays to interrogate the transcriptional network influenced by RORb. We found RNA-seq paired with ATAC-seq provided a rich picture of the transcriptional changes occurring in Rorb KO neurons and insight into both developmental and adult functioning. Changes to the transcriptional network involved both differentially expressed TFs and TFs whose only perturbation was increased or decreased access to binding sites. Without these complementary perspectives, proteins such as Zfp281 and Nfe2l1/3 TFs might have been overlooked.
We identify several other TFs worthy of further investigation for their role in cortical development. Ascl1 and NeuroD1 are potent TFs that can induce transdifferentiation of mouse embryonic perinatal fibroblasts or microglia into neurons (Vierbuchen et al., 2010;Matsuda et al., 2019). NeuroD1 binds a different motif than NeuroD2, which is known to regulate barrel formation (Ince-Dunn et al., 2006), suggesting a distinct role. In addition, Trps1 was strongly up-regulated by RORb loss at P7 and P30, and it was enriched in regions that opened. Its role in neurons is not clear, but it has been characterized as a transcriptional repressor that inhibits cell migration making it a tempting target to explore the lack of L4 neuron migration necessary to form barrel walls (Wang et al., 2018).
In addition to disrupted layer identity, we also detect a significant disruption in the potential for Rorb KO cells to transcriptionally respond to activity connecting cellular identity, module formation and molecular responsiveness to input. In the adult Rorb KO, many activity-regulated TFs were upregulated, with the exception of Btbd3, and their DNA motifs showed increased accessibility. Around P7, when activity is critical for instructing cortical reorganization, we see reduced mEPSC frequency in L4 Rorb KO neurons, which is rectified by P10. Some of the transcriptional changes in the Rorb KO may be a form of compensation for the lack of input at P7. Failed up-regulation of Htr1b and down-regulation of S100a10/p11 may also be an attempt to increase activity in KO neurons. More is known about the role of Htr1b in TCAs where it is transiently expressed and, when stimulated, inhibits thalamic neuronal firing (Bennett-Clarke et al., 1993;Rhoades et al., 1994) and disrupts barrel formation (Young-Davies et al., 2000). TCA inhibition is thought to be the mechanism by which excess 5-HT disrupts barrels. While it is difficult to infer the role of Htr1b and p11 without characterizing cellular localization in S1 L4 neurons, down-regulation of p11 resulting in less Htr1b localizing to the membrane coupled with reduced Htr1b expression could relieve inhibition in L4 Rorb KO neurons. Barrel formation and the ability to respond to activity inputs corresponds with increased RORb expression and this increase is attenuated when TCA inputs are eliminated (Pouchelon et al., 2014). Together this suggests terminal differentiation and migration of neurons within L4 to form barrel walls are closely synchronized to excitatory input and require RORb for proper establishment.
Although few other studies have examined the transcriptional targets and molecular mechanisms of TFs that regulate barrel formation, our study suggests RORb is likely involved in the later stages of cellular specification and implicates several new TFs. RORb also appears to function by distinct mechanisms from TFs previously characterized to regulate barrel formation. Loss of Bhlhe22 disrupts both barrel wall formation and TCA segregation but results in down-regulation of Lmo4 (Joshi et al., 2008) unlike Rorb KO, which increased Lmo4. Interestingly, Eomes is required for barrel wall organization but does not appear to affect TCA segregation (Elsen et al., 2013). Lhx2 and RORa are more broadly expressed than RORb. Lhx2 KO results in moderate down regulation of RORb suggesting it is also likely upstream of RORb in barrel development (Wang et al., 2017). Loss of Lhx2 greatly reduced TCA branching producing smaller barrels and barrel field. This phenotype is very similar to Rora KO barrels (Vitalis et al., 2018) suggesting RORa's mechanism may be more similar to earlier developmental TFs than to RORb. Disruption of barreloid development in Rora KO thalamus is also consistent with a role in earlier stages of development (Vitalis et al., 2018). However, Rora was down-regulated in our Rorb KO dataset suggesting it may also have a role downstream of RORb. Several additional TFs appear to be downstream of RORb. For example, Btbd3 is important for dendritic orientation and is down-regulated in the Rorb KO. It may be that dendritic orientation occurs after L4 cells have migrated to form barrel walls and provide an organized reference point for orientation. Thus, we have characterized in depth the molecular and transcriptional mechanism of RORb as it orchestrates a critical juncture in barrel development where terminal differentiation and activity inputs are integrated to drive cellular organization in the cortex.  (Liu et al., 2013;Koch et al., 2017;Byun et al., 2019). Rorb GFP mutation deletes the RORb1 isoform, the predominant isoform in brain, and not the RORb2 isoform (Liu et al., 2013). The Rorb f/f allele deletes both isoforms. The following mice were obtained from Jackson Laboratories:

Imaging and fluorescence quantification
Tissue was imaged on a Leica DMI 6000B Inverted Widefield Imaging Fluorescence Microscope or a Zeiss LSM 880 confocal microscope. All genotypes and age groups contained roughly even numbers of males and females. A minimum of two slices containing at least five intact barrels between rows B-D were quantified per animal. Experimenters were blinded to age and genotype during imaging and quantification. Regions of interest (ROIs) were drawn manually by a blinded researcher around 5-6 intact barrels from rows B, C, or D using Fiji (Schindelin et al., 2012). An ROI including the total space around selected barrels up to the edges of adjacent barrels was drawn to be used for calculating septa intensity ( Figure 9). For Thsd7a KO and controls, three additional ROIs were drawn in the region adjacent to barrel cortex with low VGLUT2 signal to be used as background to normalize barrel and septa intensity. Custom MATLAB code was used to quantify the average fluorescence in ROIs. Septa intensity was calculated as septa total ROI intensity -sum(barrel ROIs). Contrast = (barrel -septa) / (barrel + septa). For absolute barrel or septa intensity, measurements were normalized to background regions ( Figure 9) within each tissue section. This was not necessary for contrast calculations because contrast is a ratio. Contrast and normalized barrel and septa intensity were averaged for two slices per animal. Two-way ANOVA was used to test for a significant effect of genotype and/or age as well as for an interaction between the two variables. Independent sample t-test was used to test for significant differences between genotypes at each age. No power analysis was performed and numbers of replicates performed were the minimum needed to demonstrate reproducibility, consistent with practices in similar published studies.

AAV injection into VPM
50 nl of AAV-hSyn-mCherry (Addgene 114472-AAV8, RRID:Addgene_114472) was delivered by stereotactic injection to the dorsal VPM of P18-20 animals. Mice were euthanized by cardiac perfusion of 4% paraformaldehyde solution at P30. Cortex was removed and flattened for tangential sectioning of barrel field into 50 mm slices on a freezing Microtome (Leica SM 2010R). Subcortical structures were embedded in agarose and sectioned into 50 mm coronal slices on a vibratome (Leica VT1000S), counterstained with DAPI, and imaged (Keyence BZ-X700). Barrel cortex was stained, imaged and contrast calculated as described above. We required two slices with a minimum of two mCherrry saturated barrels and no mCherry outside of the barrel field. Saturated barrels were defined as adjacent barrels surrounded by barrels with mCherry signal. Only saturated barrels were quantified and the same ROIs were used to quantify mCherry and VGLUT2.

Multiplex fluorescent RNA in situ hybridization (RNAscope) with immunohistochemistry
Mice were euthanized by cardiac perfusion of 4% paraformaldehyde solution at P30. Brain tissue was pretreated according to the RNAscope Sample Preparation and Pretreatment Guide for Fresh Frozen Tissue (Manual RNAscope assay; Advanced Cell Diagnostics). Tissue was sectioned at 12 mm and subsequent staining Figure 9. Example of quantification method. Regions of interest (ROIs) were drawn in Fiji by a researcher blinded to genotype and age. performed according to the manufacturer's instruction for the RNAscope Fluorescent Multiplex kit (Advanced Cell Diagnostics, 320850) with two protocol modifications. Antigen retrieval was carried out in an autoclave set to a 5 min 'fast' cycle, 121˚C, 15 psi. After protease III digestion, probe solutions containing 313301-C2 (Fezf2) or 484781 (Tox) also contained 10% NGS and 3% BSA to allow the probe binding step to also serve as the IHC blocking step. After developing the fluorescent in situ signal, slides were protected from light and stained overnight at room temperature with 1:250 chicken anti-GFP (Aves labs GFP-1020, RRID:AB_10000240) diluted in 1X Tris-borate-EDTA (TBE) buffer containing 10% NGS and 3% BSA. Slides were washed four times in 1X TBE for 2-5 min and incubated for two hours at room temperature with 1:500 Goat Anti-Chicken Alexa Fluor 488 (Invitrogen A-11039, RRID:AB_2534096). Slides were washed four times in 1X TBE for 2-5 min, counterstained with DAPI and coverslips mounted according to the instructions for the RNAscope Fluorescent Multiplex kit. Batches of staining were balanced to contain equal numbers of control and Rorb KO samples per batch.
Stained tissue was imaged on a Zeiss LSM 880 confocal microscope. Two regions of neocortex containing S1 were imaged for each animal with automated image stitching so that layers 2 through six were contained in a single image. Images of RNA signal were background subtracted in ImageJ (Fiji) using a rolling ball radius of 5 pixels. GFP signal was used to draw ROIs within L4 and L5. A custom CellProfiler (Lamprecht et al., 2007) pipeline identified cells by identifying nuclei from DAPI images and expanding ROIs, and identified RNA puncta. RNA puncta were associated with the nearest cell in R using X,Y coordinates output from CellProfiler. RNA puncta were tallied per cell and the mean calculated per image then per animal and plotted. P-values were calculated by independent sample t-test between Ctl and KO L4.
Spiny stellate neurons were recognized based on their compact, GFP + cell bodies within the GFP + cell-dense layer 4. Input resistance was measured every 10-20 s with a small hyperpolarizing pulse and data were discarded if input or series resistance changed by >20%. P-values were calculated by 2-way ANOVA and adjusted for multiple comparisons by Tukey post hoc correction.

RNA-seq
RNA-seq was performed as described previously (Sugino et al., 2019). Briefly, 1000-1500 GFP + cells were isolated by FACS (BD FACSAria Flow Cytometer) from micro dissected L4 S1 live tissue (N = 4 biological replicates per age and genotype). Figure 10 shows examples of the region micro dissected out to exclude L5. The four independent biological samples were collected from a pool generated by combining tissue from one male and one female mouse for a total of 8 animals used per time point. Cells Figure 10. Example of micro dissected region for L4 S1 from coronal slices. Yellow dashed line indicates the tissue retained for FACS. Layer five is labeled for reference.
were sorted directly into extraction buffer and RNA stored at À80C for <three weeks. All libraries were prepared and sequenced in a single batch to prevent batch effects. Total RNA was purified (Arcturus PicoPure RNA Isolation kit, KIT0204) according to manufacturer's specifications. Libraries were prepared using Ovation Trio RNA-Seq library preparation kit with mouse rRNA depletion (0507-32) according to manufacturer's specifications and sequenced on a NextSeq Illumina platform (NextSeq 500/550 High Output (1 Â 75 cycles)) obtaining 27 ± 2 million reads (mean ± SE). Reads were mapped by STAR with 90 ± 0.3% unique mapping (mean ± SE) and quantified with feature-Counts (Liao et al., 2014). Differentially expressed genes were identified by Limma (Ritchie et al., 2015) using a fold change cutoff of 2 and padj <0.01 from a moderated t-test adjusted for multiple comparisons using FDR (Benjamini-Hochberg).

Data access
Raw and processed RNA-seq and ATAC-seq files are available at GEO accession GSE138001.

Motif analysis
Motifs identified de novo from the sequences underlying ATAC peaks was carried out using MEME AME with shuffled input sequences as control and default settings (Fraction of maximum logodds = 0.25, E-value threshold 10) (McLeay and Bailey, 2010), and HOMER findMotifsGenome.pl function masking repeats and -size given (Heinz et al., 2010). Scanning for specific motif matches in the sequences underlying ATAC peaks was carried out using MEME FIMO used the default threshold of p-value<1e-4 (Grant et al., 2011) and HOMER findMotifsGenome.pl -find function. When possible 2-3 PWMs were obtained from Jaspar (Khan et al., 2018) and Cis-BP (Weirauch et al., 2014) prioritizing PWMs from direct data sources such as ChIP-seq. The R package GenomicRanges (Lawrence et al., 2013) was used to identify overlapping motifs between the two algorithms for cross validation. The overlap criteria allowed a 1 bp difference in the start or end position of the motif to accommodate ambiguity among motif models. Fisher Exact tests were calculated in R to test for enrichment of motifs in ATAC regions compared to control regions and to test for enrichment of genes with a nearby motif from a DEG group compared to a control group of genes. The set of control regions was generated by shuffling ATAC peaks throughout the genome excluding sequence gaps using BedTools (Quinlan and Hall, 2010) and the control group of genes were defined as expressed above 5 TPM but unchanged by age or Rorb KO. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.