Main 主要的

Neuronal circuits in the neocortex underlie our ability to perceive our surroundings, integrate various forms of sensory information and support cognitive functions. Cortical computation relies on assemblies of excitatory and inhibitory neuron types that are joined into canonical microcircuit motifs. The synaptic innervation and intrinsic properties of fast-spiking parvalbumin-expressing inhibitory interneurons (PV interneurons) have emerged as key parameters for controlling cortical circuit stability and plasticity1,6. During development, sensory experience shapes the synaptic innervation of PV interneurons in an afferent-specific manner, and synaptic input to PV interneuron dendrites is a crucial node for cortical dysfunction in disorders7,8,9,10,11. In the adult brain, neuronal-activity-dependent regulation of the recruitment and excitability of PV interneurons is fundamental for maintaining the balance between excitation and inhibition, and has been implicated in gating cortical circuit plasticity during learning processes1,2,12,13,14,15. However, the molecular mechanisms that underlie these features—in particular, the transcellular signalling events that relay alterations in neuronal network activity and adjust PV interneuron function—are poorly understood.
新皮质中的神经元回路是我们感知周围环境、整合各种形式的感觉信息和支持认知功能的能力的基础。皮层计算依赖于兴奋性和抑制性神经元类型的组装,这些神经元类型被连接到规范的微电路图案中。表达小白蛋白的快速尖峰抑制性中间神经元(PV 中间神经元)的突触神经支配和内在特性已成为控制皮质回路稳定性和可塑性的关键参数 1,6 。在发育过程中,感觉体验以传入特异性方式塑造 PV 中间神经元的突触神经支配,而 PV 中间神经元树突的突触输入是疾病中皮质功能障碍的关键节点 7,8,9,10,11 。在成人大脑中,PV中间神经元的招募和兴奋性的神经元活动依赖性调节是维持兴奋和抑制之间平衡的基础,并且与学习过程中的门控皮质回路可塑性有关 1,2,12,13,14,15 。然而,这些特征背后的分子机制,特别是传递神经元网络活动变化和调整 PV 中间神经元功能的跨细胞信号传导事件,人们知之甚少。

Neuronal network activity mobilizes BMP signalling
神经元网络活动调动 BMP 信号传导

To identify candidate transcellular signals that are regulated by neuronal network activity in mature neocortical neurons, we examined secreted growth factors of the bone morphogenetic protein (BMP) family, which have been implicated in cell-fate specification and neuronal growth during development16,17,18,19,20,21,22,23,24. We examined four bone morphogenetic proteins (BMP2, BMP4, BMP6 and BMP7) in mice, and found that Bmp2 mRNA was significantly upregulated in glutamatergic neurons after stimulation (3.5 ± 0.5-fold; Extended Data Fig. 1a–d). A similar activity-dependent increase in BMP2 was observed at the protein level in neurons derived from a Bmp2 HA-tag knock-in mouse (Bmp2HA/HA; Extended Data Fig. 1e–g and Supplementary Information). As developmental morphogens, BMPs direct gene regulation in recipient cells through SMAD transcription factors25,26,27,28,29 (Fig. 1a). Notably, the canonical BMP target genes Id1 and Id3 were significantly upregulated in stimulated neocortical cultures, and this process was blocked by the addition of the extracellular BMP antagonist Noggin (Extended Data Fig. 1h,i). In the neocortex of adult mice, key BMP signalling components continue to be expressed, with the ligand BMP2 exhibiting the highest mRNA levels in glutamatergic neurons (Extended Data Fig. 2a–c). To test whether the transcription of BMP target genes is activated in response to increased neuronal network activity in adult mice, we chemogenetically silenced upper-layer PV interneurons in the barrel cortex (Fig. 1b). This local reduction of PV-neuron-mediated inhibition results in increased neuronal network activity30,31 accompanied by a four- to eightfold transcript increase in the activity-induced primary response genes Fos and Bdnf (Fig. 1c). Of note, this chemogenetic stimulation also resulted in the upregulation of BMP target genes (Id1 and Smad6, and an increase in Id3 when compared with mCherry + clozapine N-oxide (CNO) negative controls) (Fig. 1c). We then mapped neuronal cell populations in which BMP target genes were activated in response to neuronal network activity, using a novel temporally controlled BMP signalling reporter (BMP-responsive Xon; BRX) (Fig. 1d). We combined BMP-response element sequences (4×BRE) from the Id1 promoter32 with the small molecule (LMI070)-gated miniXon cassette33 to drive a nucleus-targeted eGFP (Extended Data Fig. 3). Thus, the level of nuclear eGFP reports the activation of BMP signalling during a time window specified by LMI070 application (Extended Data Fig. 4a–f). Chemogenetic stimulation resulted in a selective increase in the activity of the BRX reporter in PV interneurons, whereas the mean reporter output in glutamatergic cells and non-PV interneurons was unchanged (Fig. 1f,g, but note that a sparse subpopulation of NeuN+Gad67 glutamatergic neurons did show a significant reporter signal). Genetic restriction of the BRX reporter to PV interneurons revealed a threefold increase in the BRX signal in response to chemogenetic stimulation (Extended Data Fig. 4g–i). Together, these results show that increased cortical network activity mobilizes BMP2 and selectively activates the BMP signalling pathway in PV interneurons in the barrel cortex of adult mice.
为了识别成熟新皮质神经元中受神经元网络活动调节的候选跨细胞信号,我们检查了骨形态发生蛋白 (BMP) 家族的分泌生长因子,该家族与发育过程中的细胞命运规范和神经元生长有关 16,17,18,19,20,21,22,23,24 ;扩展数据图 1e-g 和补充信息)。作为发育形态发生素,BMP 通过 SMAD 转录因子 25,26,27,28,29 指导受体细胞中的基因调控(图 1a)。值得注意的是,经典 BMP 靶基因 Id1 和 Id3 在受刺激的新皮质培养物中显着上调,并且通过添加细胞外 BMP 拮抗剂 Noggin 来阻断这一过程(扩展数据图 1h,i)。在成年小鼠的新皮质中,关键的 BMP 信号传导成分继续表达,配体 BMP2 在谷氨酸能神经元中表现出最高的 mRNA 水平(扩展数据图 2a-c)。为了测试成年小鼠中 BMP 靶基因的转录是否因神经元网络活动增加而被激活,我们用化学遗传学方法沉默了桶状皮质中的上层 PV 中间神经元(图 1b)。 PV神经元介导的抑制的局部减少导致神经元网络活动增加 30,31 ,同时活动诱导的初级反应基因Fos和Bdnf的转录本增加四到八倍(图1c)。 值得注意的是,这种化学遗传学刺激还导致 BMP 靶基因上调(Id1 和 Smad6,以及与 mCherry + 氯氮平 N-氧化物 (CNO) 阴性对照相比,Id3 增加)(图 1c)。然后,我们使用新型时间控制的 BMP 信号报告基因(BMP 响应 X on ; BRX)绘制了神经元细胞群,其中 BMP 靶基因响应神经元网络活动而被激活(图 1d)。我们将来自 Id1 启动子 32 的 BMP 响应元件序列 (4×BRE) 与小分子 (LMI070) 门控 miniX on33 组合起来以驱动细胞核靶向 eGFP(扩展数据图 3)。因此,核 eGFP 的水平报告了 LMI070 应用程序指定的时间窗口内 BMP 信号传导的激活(扩展数据图 4a-f)。化学遗传学刺激导致 PV 中间神经元中 BRX 报告基因的活性选择性增加,而谷氨酸能细胞和非 PV 中间神经元中的平均报告基因输出没有变化(图 1f,g,但请注意 NeuN + Gad67 谷氨酸能神经元确实显示出显着的报告信号)。 BRX 报告基因对 PV 中间神经元的遗传限制揭示了响应化学遗传学刺激的 BRX 信号增加了三倍(扩展数据图 4g-i)。总之,这些结果表明,皮层网络活动的增加会动员 BMP2,并选择性地激活成年小鼠桶状皮层 PV 中间神经元中的 BMP 信号通路。

Fig. 1: Increased neuronal activity elicits BMP signalling in PV interneurons of the adult barrel cortex.
figure 1

a, Illustration of BMP pathway components. BMPRs, BMP receptors. b, Schematic representation of the protocol for chemogenetic manipulation of neuronal activity in the adult barrel cortex. P42, postnatal day 42; P56, postnatal day 56. c, Expression of the immediate early genes Fos and Bdnf and the SMAD1/5 target genes Id1, Id3, Smad6 and Smad7 in the barrel cortex of chemogenetically stimulated and control mice (n = 3–6 mice per group, mean ± s.e.m., two-way ANOVA with Tukey’s post-hoc test). d, Schematic representation of the viral BRX reporter. Nucleus-targeted eGFP (NLS-eGFP) is expressed under the control of regulatory elements from the Id1 gene (4×BRE), a minimal SV40 promoter and the miniXon splicing cassette. ITR, inverted terminal repeat. e, Experimental paradigm. f, Representative images of the BRX reporter signal in barrel cortex layer 2/3 of PVCre mice. Cre-dependent mCherry identifies PV cells, NeuN identifies neurons and the somatic–perinuclear GAD67 signal identifies GABA neurons. Scale bar, 20 μm. g, BRX reporter-driven nuclear eGFP intensity per mouse (n = 3 mice per group, cell numbers indicated in columns, mean ± s.e.m., one-way ANOVA with Tukey’s multiple comparisons) and cumulative distribution of eGFP reporter intensity per cell for glutamatergic and PV-positive neurons (Kolmogorov–Smirnov test).

Source Data

BMP–SMAD1 signalling regulates synaptic proteins

During development, the combinatorial action of various BMP ligands and receptors directs the cell-type-specific regulation of target genes through SMAD transcription factors, but SMAD-independent functions have also been described16,20,22,34,35,36. In neocortical neurons, stimulation with BMP2 (20 ng ml−1 for 45 min) resulted in the activation of SMAD1 and SMAD5 (hereafter, SMAD1/5) in both glutamatergic and GABAergic (γ-aminobutyric-acid-producing) neurons (Extended Data Fig. 5a–c). To uncover neuronal SMAD1 target genes, we performed chromatin immunoprecipitation followed by sequencing (ChIP–seq) for SMAD1/5 in adult mouse neocortex and neocortical cultures (Fig. 2a). We identified 239 and 543 sites that were bound in the mouse neocortex and in cultured neocortical neurons, respectively (Fig. 2b and Supplementary Table 1). Notably, 77% of the binding sites in vivo were reproduced in the cultured neuron preparations. To specifically map sites that are acutely regulated by BMP–SMAD1/5 signalling, we stimulated cortical cultures by adding recombinant BMP2. After stimulation, we identified another 353 BMP2-responsive SMAD1/5-binding sites. Most of the BMP2-responsive peaks were associated with promoter elements. To investigate whether SMAD1/5 trigger the de novo activation of target genes or, rather, modifies the transcriptional output of active genes, we mapped histone H3 acetylated at lysine 27 (H3K27ac) marks, a chromatin modification at active promoters and enhancers. By intersecting H3K27ac ChIP–seq signals with SMAD1/5 peaks (Fig. 2b,c), we found that most BMP2-responsive elements contain significant H3K27ac marks, which are slightly increased after stimulation. This suggests that many of these sites are already active without BMP2 stimulation. By comparison, constitutively bound regions exhibited a lower H3K27ac signal (Fig. 2b,c). Sequence analysis identified an enrichment of different motifs for SMAD1/5 DNA binding in the constitutive (tissue and neuronal culture) and in the BMP2-responsive gene-regulatory elements, suggesting that DNA binding involves different co-factors (Fig. 2d). The effect of the BMP2-induced recruitment of SMAD1/5 on transcriptional output was examined by RNA sequencing (RNA-seq). Differential gene expression analysis identified 30 and 147 upregulated transcripts 1 h and 6 h after BMP2 stimulation, respectively (Extended Data Fig. 5d and Supplementary Table 2). Fifty per cent of the regulated genes 1 h after BMP2 stimulation had direct SMAD1/5 binding at their promoters. These genes included known negative-feedback-loop genes of the BMP signalling pathway (Id1, Id3 and Smad7). Twenty-five per cent of differentially regulated genes 6 h after BMP2 stimulation had direct SMAD1/5 binding (Extended Data Fig. 5d). Conditional knockout of Smad1 in postmitotic neurons was sufficient to abolish the upregulation of these genes in response to BMP2 signalling and reduce their expression in naive (unstimulated) neurons (Fig. 2f, Extended Data Fig. 5e,f and Supplementary Table 3). Direct transcriptional targets of BMP–SMAD1 signalling in neocortical neurons included an array of activity-regulated genes such as Junb, Trib1 and Pim3, as well as genes that encode key components of the extracellular matrix (Bcan and Gpc6) and glutamatergic synapses (Lrrc4 and Grin3a) (Fig. 2e and Extended Data Fig. 5g,h). Moreover, neuronal ablation of Smad1 was accompanied by broad changes in gene expression beyond the deregulation of direct SMAD1 target genes (Extended Data Fig. 5i). Top gene ontology (GO) terms enriched amongst the upregulated genes were ‘glutamatergic synapse’ and transcription factors under the term ‘nucleus’ (Extended Data Fig. 5j). Furthermore, deregulated genes included the majority of neuronal-activity-regulated rapid primary response genes (rPRGs) and secondary response genes (SRGs) (Extended Data Fig. 5k). Thus, SMAD1 is a key downstream mediator of BMP signalling in mature neurons and its neuronal loss of function results in a substantial upregulation of neuronal activity response genes in vitro.

Fig. 2: Neuronal BMP2–SMAD1 signalling regulates synaptic components.
figure 2

a, Schematic representation of ChIP–seq and RNA-seq experiments from mouse neocortex and neocortical cultures. qRT–PCR, quantitative PCR with reverse transcription. b, ChIP–seq analysis of neocortical tissue and naive (0 h) or growth-factor-stimulated (1 h 20 ng ml−1 BMP2) neocortical neuron cultures at DIV14 (14 days in vitro). Heat maps in purple show the peak strength of SMAD1/5 binding; heat maps in green show H3K27ac binding at SMAD1/5 peak regions. The right column (in black) shows the position of promoter elements. Each binding site is represented as a single horizontal line centred at the SMAD1/5 peak summit; the colour intensity correlates with the sequencing signal for the indicated factor. Peaks are ordered by decreasing SMAD1/5 peak intensity. c, Mean normalized ChIP–seq signal for SMAD1/5 and H3K27ac plotted for BMP2-responsive and constitutive SMAD1/5-binding sites. Grey lines indicate signal obtained from vehicle-treated cultures and purple lines indicate signal from BMP2-stimulated cultures. d, Top enriched motifs detected for BMP2-responsive (left) and constitutive (right) SMAD1/5 peaks. e, Examples of IGV genome browser ChIP–seq tracks showing the H3K27ac (green), SMAD1/5 (purple) and RNA-seq (grey) signals for the SMAD1/5 targets Id3 and Bcan in naive (−) and BMP2-stimulated cultures. f, qPCR analysis of the mRNA expression of Id3 and Bcan in AAV-Syn-eGFP infected versus AAV-Syn-Cre infected Smad1fl/fl neocortical cultured neurons. Fold change (FC) relative to unstimulated cells is shown for 1 h and 6 h stimulation with 20 ng ml−1 BMP2. Bar graphs show mean ± s.e.m. (n = 5 independent cultures per condition, one-way ANOVA with Tukey’s multiple comparisons).

Source Data

SMAD1 controls the innervation of PV interneurons

In neocortical circuits, the excitation–inhibition balance is regulated by glutamatergic input synapses onto PV interneurons, and perineuronal nets (PNNs) surrounding these cells are modified in response to changes in neuronal network activity37,38. To test whether pyramidal-cell-derived BMP2 modifies the innervation of PV interneurons, we generated Bmp2 conditional knockout mice in which Bmp2 is selectively ablated in upper-layer glutamatergic neurons (Cux2creERT2::Bmp2fl/fl; referred to as Bmp2ΔCux2 mice). We then adopted genetically encoded intrabodies (fibronectin intrabodies generated by mRNA display; FingRs) to quantitatively map the synaptic inputs to PV interneurons39,40 (Extended Data Fig. 6a–c and Supplementary Video 1). A FingR-PSD95 probe was selectively expressed in PV interneurons in layer 2/3 of the barrel cortex under the control of a PV-cell-specific enhancer41 (Fig. 3a–d). Notably, the density of synapses onto PV interneurons was reduced after genetic ablation of Bmp2 in upper-layer pyramidal cells of Bmp2ΔCux2 mice (Fig. 3e,f). We then generated PV-interneuron-specific Smad1 conditional knockout mice to examine whether BMP2 acts through SMAD1. Postnatal ablation of Smad1 (PVcre/+::Smad1fl/fl; referred to as Smad1ΔPV mice) did not alter the density or distribution of PV cells in the somatosensory cortex (Extended Data Fig. 7a–c). Using a Cre-recombinase-dependent form of the FingR-PSD95 probes (Fig. 4a), we observed a 40% reduction in the density of glutamatergic synapses as observed by morphology onto Smad1ΔPV interneurons (Fig. 4b,c). This was accompanied by a comparable reduction in the frequency of miniature excitatory postsynaptic currents (mEPSCs), but there was no change in mEPSC amplitude in acute slice recordings (Fig. 4d–f). The density of perisomatic PV–PV synapses (identified by synaptotagmin-2 and a FingR-gephyrin probe39) was also reduced (Fig. 4g and Extended Data Fig. 7d,e). However, there was no significant change in the frequency or amplitude of miniature inhibitory postsynaptic currents (mIPSCs) in PV cells of Smad1ΔPV mice, owing probably to compensatory inhibition derived from other interneuron classes (Fig. 4h–j). Thus, SMAD1 is required for normal functional glutamatergic innervation of layer 2/3 PV interneurons, and the loss of SMAD1 results in reduced glutamatergic input to these cells in Smad1ΔPV mice.
在新皮质回路中,兴奋-抑制平衡由 PV 中间神经元上的谷氨酸输入突触调节,并且围绕这些细胞的神经周围网络 (PNN) 会根据神经元网络活动的变化进行修改 37,38 。为了测试锥体细胞衍生的 BMP2 是否会改变 PV 中间神经元的神经支配,我们生成了 Bmp2 条件敲除小鼠,其中 Bmp2 在上层谷氨酸能神经元中被选择性消除 (Cux2 creERT2 ::Bmp2 fl/fl 小鼠)。然后,我们采用基因编码的胞内抗体(由 mRNA 显示生成的纤连蛋白胞内抗体;FingRs)来定量地将突触输入映射到 PV 中间神经元 39,40 (扩展数据图 6a-c 和补充视频 1)。在PV细胞特异性增强子 41 的控制下,FingR-PSD95探针在桶状皮层2/3层的PV中间神经元中选择性表达(图3a-d)。值得注意的是,在 Bmp2 ΔCux2 小鼠的上层锥体细胞中进行 Bmp2 基因消融后,PV 中间神经元上的突触密度降低(图 3e,f)。然后,我们生成了 PV 中间神经元特异性 Smad1 条件敲除小鼠,以检查 BMP2 是否通过 SMAD1 发挥作用。出生后去除 Smad1(PV cre/+ ::Smad1 fl/fl ;称为 Smad1 ΔPV 小鼠)不会改变体感中 PV 细胞的密度或分布皮质(扩展数据图 7a-c)。使用 Cre 重组酶依赖性形式的 FingR-PSD95 探针(图 4a),我们观察到 Smad1 ΔPV 中间神经元的形态学观察到谷氨酸突触密度减少了 40%(图 4b) ,C)。 这伴随着微型兴奋性突触后电流(mEPSC)频率的相当减少,但急性切片记录中 mEPSC 振幅没有变化(图 4d-f)。体周PV-PV突触的密度(由synaptotagmin-2和FingR-gephyrin探针 39 识别)也降低了(图4g和扩展数据图7d,e)。然而,Smad1 ΔPV 小鼠PV细胞中微型抑制性突触后电