Different Neuronal Activity Patterns Induce Different Gene Expression Programs

Abstract

A vast number of different neuronal activity patterns could each induce a different set of activity-regulated genes. Mapping this coupling between activity pattern and gene induction would allow inference of a neuron's activity-pattern history from its gene expression and improve our understanding of activity-pattern-dependent synaptic plasticity. In genome-scale experiments comparing brief and sustained activity patterns, we reveal that activity-duration history can be inferred from gene expression profiles. Brief activity selectively induces a small subset of the activity-regulated gene program that corresponds to the first of three temporal waves of genes induced by sustained activity. Induction of these first-wave genes is mechanistically distinct from that of the later waves because it requires MAPK/ERK signaling but does not require de novo translation. Thus, the same mechanisms that establish the multi-wave temporal structure of gene induction also enable different gene sets to be induced by different activity durations.

Keywords: MAPK; RNA-seq; activity-regulated enhancers; activity-regulated transcription; coupling map; eRNA; immediate early genes; mitogen-activated protein kinase; neuronal activity duration; neuronal activity patterns; primary response genes.

Figure 1. Brief neuronal activation selectively induces the first of three waves of gene induction

(A) Experimental system for comparing sustained and brief neuronal activation in vitro. Except where indicated otherwise, neuronal activation is accomplished with brief (1-min) or sustained KCl-depolarization of cortical neurons silenced 14-16h before stimulation with APV and NBQX. (B) Comparison of gene induction upon sustained or brief neuronal activation using activity-regulated-gene-capture-based RNA-sequencing (ARG-seq) (means, n=3-6 biological replicates). Only induced genes are shown. Gene categories are defined based on kinetics of gene induction, as well as induction in the presence of the translation-inhibitor cycloheximide (Figure S1B). PRG = primary response gene. SRG = secondary response gene. Genes induced by brief neuronal activation are enriched for rapid PRGs (rPRGs) (p<10^-13, Fisher's exact test). (C) Three kinetically distinct temporal waves of gene induction as detected by high-throughput microfluidic qPCR. Points represent the mean expression of the median gene for each class. Shading covers the middle quartiles of mean expressions (25%-75%) (n=6 biological replicates). Each wave is kinetically distinct from the other waves (rPRG vs. delayed PRG (dPRG)/SRG induction at 1h, dPRG vs. SRG induction at 2h, p<0.003, rank-sum test). Plotted are 15, 37, and 9 genes from waves 1-3, respectively. (D) Experimental system for comparing the duration of neuronal activation in the visual cortex in vivo. Mice were dark-housed for three days prior to visual stimulation consisting of lights flashing in a repeated pattern: 60s on, 20s off. (E) Gene induction in the visual cortex following visual stimulation, as measured by qPCR. Colored points are means of n=3 biological replicates. Grey points are values from individual biological replicates. Gene categories defined as in (B). *significant induction compared to 0h time point, p<0.05 unpaired, two-sided t-test, fold induction>1.5. Related to Figures S1 and S2, Tables S2 and S3

Figure 2. Neuronal activity patterns can be inferred from ARG expression

(A) A classifier trained on in vitro gene expression data to infer activity histories of 12 in vitro samples (6 brief, 6 sustained). The classifier identified test samples as having undergone either brief or sustained activity based on based on Euclidean distance to training samples. *p = 0.007, exact binomial test. (B) A similar (in vitro-trained) classifier used to infer the activity histories of 12 in vivo visual cortex samples (3 brief, 3 sustained, and 6 unstimulated). *p<0.04, exact binomial test. (C) Method for scRNA-seq-based inference of BRIEF and SUSTAINED activity histories of individual visual cortex excitatory neurons from mice exposed to 1h of sustained visual stimulation. scRNA-seq data from Hrvatin et al., 2017. (D) 1h of visual stimulation significantly increased the fraction of excitatory neurons with BRIEF and SUSTAINED inferred activity states (p<10^-15, Fisher's exact test). (E) Expression of four layer markers in BRIEF and SUSTAINED neurons in scRNA-seq data. Data plotted are imputed mRNA reads after using DECENT (Ye et al., 2017) to account for the presence of technical zeroes. *FDR<0.1, rank-sum test. (F) Differential expression (DE) of all genes (excluding ARGs) in BRIEF compared to SUSTAINED neurons. P-value determined using the rank-sum test. Color of the points represent the log of the ratio of gene expression in deep layers (Layers 5 and 6) to that in upper layers (Layers 2/3 and 4). (G) Fraction of stimulated neurons in each layer that are BRIEF. *More BRIEF neurons in deep vs. upper layers, p<10^-15, Fisher's exact test. ⁺Significant population of brief neurons, p<0.001 based on a Fisher's exact test comparing the number of rPRG-ON neurons among dPRG-OFF neurons in the stimulated cortex to the number of rPRG-ON neurons in unstimulated cortex. Related to Figure S3

Figure 3. Requirement for MAPK/ERK signaling and an open chromatin state distinguish first and second waves of gene induction

(A) Chromatin state in unstimulated neurons shown in metaplots of the geometric mean signal for all genes in each category. All measures of chromatin state are significantly different between rPRGs and dPRGs or SRGs (p<0.009, rank sum test on the area under the curves shown). ChIP-seq data are from cultured cortical neurons, Telese et al., 2015. DNaseI hypersensitivity data are from the 8w cerebrum (Consortium et al., 2012). (B) Transcription factor binding in unstimulated neurons from ChIP-seq, shown in metaplots as in (A). SRF and MEF2: significantly different between rPRGs and dPRGs or SRGs; CREB: not significantly different between rPRGs and dPRGs (p=0.2), but is different between rPRGs and SRGs (p<0.009, rank sum test). Data from cultured cortical neurons, Kim et al., 2010; Telese et al., 2015. (C) ERK activation kinetics with KCl-mediated depolarization. Representative (1 of n=3) western blot for phosphorylated ERK (pERK). Phosphorylation of ERK paralogs, p44 and p42 (upper and lower bands), is kinetically similar (r² = 0.97, Pearson correlation). (D) Similar to (C), but rat cortical neurons treated with sustained or brief bicuculline/4AP. One of n=3-4 representative biological replicates is shown. (E) Same as (D), but from isolated nuclei. (F) Quantification of (C), n=3 biological replicates. The inset is a magnified version of the first ten minutes. pERK induction at its peak (five minutes) is not different between brief and sustained stimulus (p=0.3, paired, two-sided t-test). Error bars represent +/- SEM. (G) rPRG but not dPRG induction in response to sustained activity is dependent on MAPK/ERK. ARG-seq-based gene expression of three representative rPRGs and three representative dPRGs following sustained KCl depolarization of mouse neurons with and without 10μM of the MEK inhibitor, U0126. n=3-7 biological replicates. Error bars are +/- S.E.M. *p<0.01, rank-sum test. (H) Data from the same experiment as (G), showing all ARGs. *significantly different from 1, p<0.01, rank-sum test; +p = 0.02, rank-sum test. Expression of rPRGs is more affected by MEK inhibition than expression of dPRGs (p = 0.002; rank-sum test on 17 rPRGs versus 110 dPRGs using the mean for each gene across n=3-7 biological replicates at its most induced time point). (I) Data the same as in (H), but showing the geometric mean of gene expression. Error bars are +/- SEM from each of n=3-7 biological replicates of all genes in the category. *p<0.03, rank-sum test. (J) rPRG but not dPRG induction in response to brief activity is dependent on MAPK/ERK. Same as (G), top row, but with 1-min KCl depolarization. (K) Same as (H), top row, but with 1-min KCl depolarization. (L) Same as (I), top row, but with 1-min KCl depolarization. Related to Figures S4-S6, Table S2

Figure 4. MAPK/ERK is required for the first wave but not subsequent waves of gene induction in vivo

(A) Visual-stimulus-mediated gene induction of representative genes in the visual cortex upon sustained stimulation in mice injected intraperitoneally with corn oil vehicle or the MEK inhibitor SL327 (100mg/kg), based on ARG-seq. D: dark, no visual stimulation. L: light, with visual stimulation. Error bars are 95% confidence intervals across n=2-3 mice. (B) Same experiment as (A), but showing all rPRGs or dPRGs detected by ARG-seq from n=2-3 biological replicates. *p<0.01 from rank-sum test, significant difference from 1. Induction of rPRGs is more affected by MEK inhibition than induction of dPRGs (p=0.02; rank-sum test, 16 rPRGs vs. 14 dPRGs using the mean for each gene at its most induced time point across n=2-3 biological replicates). (C) Same as (A) but with brief visual stimulation. (D) Same as (B) but with brief visual stimulation. Related to Figure S6, Table S2

Figure 5. MAPK/ERK mediates fast recruitment of Pol2 to rapid PRG promoters

(A) RNA Polymerase 2 (Pol2) binding (ChIP-seq) at the promoters of rPRGs. Lines represent the mean and shading the S.E.M. across loci. Data shown are from n=1 of 2 biological replicates. Pol2 binding to rPRG promoters is blunted by MEK inhibition (see (B)). The KCl-dependent fold-increase in mean Pol2 density (-300bp to +300bp) is significant under both vehicle and MEK inhibitor treatments (FDR<0.001 in each of two biological replicates, paired rank sum test). MEK inhibition does not affect Pol2 occupancy in unstimulated neurons (FDR>0.05 in each of two biological replicates, paired rank sum test). (B) ChIP-seq-based time course of fold-change in Pol2 occupancy at rPRG promoters (-300bp to +300bp). Shown are mean fold-change values across genes, with +/- S.E.M error bars. *FDR <0.01 in each of two replicates, paired rank-sum test on fold-change values. (C) Pol2 binding at the promoter of the representative rPRG Fos upon sustained neuronal activation. Data normalized prior to visualization. (D) Plotting and statistics same as (A) but showing dPRG promoters. (E) Plotting and statistics as in (B) but showing dPRG promoters. (F) Plotting as in (C) but showing representative dPRG, Sertad1. Related to Figure S7, Table S7

Figure 6. MAPK/ERK is required for rapid eRNA induction but not H3K27 acetylation at enhancers

(A) H3K27ac accumulation (ChIP-seq) at the rPRG Arc locus upon sustained KCl depolarization. The gene expression of Arc based on ARG-seq is shown for comparison. Data normalized by read-depth prior to visualization. (B) Same as (A), but for the dPRG, Rasgrp1. (C) H3K27ac accumulation (ChIP-seq) at enhancers upon sustained KCl depolarization. Plotted are means from n=2 biological replicates. Lines represent the median across enhancers, dark shading the two middle deciles, and light shading the upper and lower quartiles. The increase from 0 to 10 min is significant for both enhancers near rPRGs and those near dPRGs (p<0.00001, rank-sum test). (D) H3K27 accumulation at enhancers near rPRGs and dPRGs is not significantly affected by MEK inhibition (p>0.2, rank-sum test). Data as in (C). The y-axis shows the induction at each enhancer's most-induced time point (10, 30, or 60 min) in each condition. (E) eRNA induction (total RNA-seq) upon neuronal activation. Plotted as in (C). (F) MEK inhibition blocks eRNA induction at enhancers near rPRGs but not dPRGs. Plotting as in (D), except showing the maximum eRNA induction at 20 or 60 min. *p=0.01, rank-sum test, using means for each enhancer from n=2 biological replicates; N.S., p>0.05. Related to Figure S8, Tables S2 and S8

Figure 7. eRNA-seq enables eRNA quantification at individual enhancers, revealing rapid and delayed enhancers

(A) eRNA-seq methodology. (B) Reads in target enhancers: eRNA-seq vs. total RNA-seq. (C) eRNA-seq-based eRNA expression at significantly induced (FDR<0.05) rapid and delayed enhancers upon sustained activation. Rapid enhancers are significantly induced by 20 minutes and delayed enhancers only by 60 minutes. Light lines are means for individual enhancers from n=4 biological replicates, and heavy lines are the geometric means for all enhancers shown. (D) rPRGs compared to dPRGs are enriched for the presence of nearby rapid enhancers (p=0.02, Fisher's exact test), but there are also rapid enhancers near dPRGs. (E) eRNA-seq-based eRNA expression at three enhancers near the rPRG Egr1 revealing two rapid and one delayed enhancer. *p<0.05, paired rank-sum test. Error bars are means +/- S.E.M. (F) Indicators of open chromatin prior to stimulation at rapid versus delayed enhancers, with metaplots showing the geometric mean of all enhancers in each class. All are significantly different between rapid and delayed enhancers (p<10^-7, rank sum test using area under the curve). Histone mark ChIP-seq data from cultured cortical neurons, Telese et al., 2015. (G) Binding of transcription factors, the mediator subunit MED23, and NCoR at rapid versus delayed enhancers prior to stimulation, shown as in (F). All are significantly different between rapid and delayed enhancers (p<10^-4, rank sum test on area under the curve). ChIP-seq data from cultured cortical neurons, Kim et al., 2010; Telese et al., 2015. (H) Rapid enhancers show greater induction in response to brief activity than delayed enhancers, based on eRNA-seq (p < 10^-9, rank-sum test). The y-axis shows the mean fold induction from n=4 biological replicates for each enhancer at its most-induced time point (20 or 60 min). (I) Rapid enhancers are more MAPK/ERK-dependent than delayed enhancers, based on eRNA-seq (p = 0.006, rank-sum test, using means for each enhancer from n=4 biological replicates). For each class of enhancers, the earliest time point at which that class exhibits significant eRNA induction is shown (20 min for rapid and 60 min for delayed enhancers). The y-axis shows the KCl-dependent fold induction with MEK inhibition divided by the same fold-induction with vehicle treatment only (i.e., ratio of fold-inductions). (J) Effect of MEK inhibition on the enhancer function of the Fos enhancer e5, using a luciferase reporter assay in which the enhancer drives transcription from a minimal Fos promoter. *p<0.03 from t-test based on n=3 biological replicates. Related to Figure S8, Table S2

Figure 8. Distinguishing features of first wave genes (rapid PRGs) and second wave genes (delayed PRGs)

rPRGs are distinguished by dependence on MAPK/ERK signaling, proximity to rapid enhancers, and an open chromatin state. Light green check marks indicate partial effects.