A Transcription Factor Pulse Can Prime Chromatin for Heritable Transcriptional Memory

ABSTRACT

Short-term and long-term transcriptional memory is the phenomenon whereby the kinetics or magnitude of gene induction is enhanced following a prior induction period. Short-term memory persists within one cell generation or in postmitotic cells, while long-term memory can survive multiple rounds of cell division. We have developed a tissue culture model to study the epigenetic basis for long-term transcriptional memory (LTTM) and subsequently used this model to better understand the epigenetic mechanisms that enable heritable memory of temporary stimuli. We find that a pulse of transcription factor CCAAT/enhancer-binding protein alpha (C/EBPα) induces LTTM on a subset of target genes that survives nine cell divisions. The chromatin landscape at genes that acquire LTTM is more repressed than at those genes that do not exhibit memory, akin to a latent state. We show through chromatin immunoprecipitation (ChIP) and chemical inhibitor studies that RNA polymerase II (Pol II) elongation is important for establishing memory in this model but that Pol II itself is not retained as part of the memory mechanism. More generally, our work reveals that a transcription factor involved in lineage specification can induce LTTM and that failure to rerepress chromatin is one epigenetic mechanism underlying transcriptional memory.

INTRODUCTION

Transcriptional memory in yeast and mammalian cells can be induced by signaling events culminating in transcription factor (TF) recruitment to target gene promoters and enhancers (1–4). Mechanistic studies of long-term transcriptional memory (LTTM) in mammalian cells are currently limited (5–11); however, short-term transcriptional memory (STTM) has been studied in greater detail. For example, primary macrophages display transcriptional memory of cytokine stimulation and pathogen exposure (1, 9, 10), yet the underlying molecular events are not established. Memory of this type in macrophages is thought to contribute to the “trained immunity” phenotype, whereby macrophages display a heightened and adapted response to repeat pathogen exposures (10). One epigenetic event thought to contribute to STTM in macrophages is the conversion of gene enhancers from a latent/off state to a poised state by transcription factor recruitment and decoration with active enhancer-associated histone modifications (1). Based on these studies, we reasoned that a short pulse of transcription factor activity could mimic a temporary signal transduction event and that such an approach could be used to study the mechanisms underlying long-term transcriptional memory in mammalian cells. Furthermore, we hypothesized that the use of a system that utilizes transcription factors with the ability to invade closed chromatin sites (i.e., pioneering factors) could result in enhanced memory effects. Indeed, binding of these factors can make previously inaccessible chromatin sites transcriptionally competent and aid in the conversion of gene regulatory elements from a latent/off to a primed or active state (12). CCAAT/enhancer-binding protein alpha (C/EBPα) and AP-1 TF families have been reported to display pioneering factor activity in the myeloid lineage (13, 14), while C/EBPα overexpression in conjunction with endogenous PU.1 works as a module to mediate complete B cell-to-macrophage transdifferentiation (TD) (15, 16). Therefore, we reasoned that the inducible TD system developed previously (15) could be employed to decipher the epigenetic mechanisms contributing to the transcriptional memory phenotype in immune cells. We asked whether a short C/EBPα pulse, insufficient to commit B cells toward the macrophage lineage, might be sufficient to induce LTTM of target genes by initiating chromatin changes that are maintained through cell division. Here, we show that LTTM is indeed observed for a subset of C/EBPα target genes, many of which are important mediators of the pathogen- and interferon-induced monocyte/macrophage inflammatory response, e.g., Cd14, Ifitm1, and Lrp1. The TD model was then employed to define the molecular mechanisms responsible for memory and mitotic inheritance of chromatin state changes. Transcription factor recruitment and histone acetylation are not sufficient to induce memory; however, memory acquisition is blocked with RNA polymerase II (Pol II) elongation inhibitors and tends to occur on genes with a more repressed chromatin state upon the initial stimulus. We present data supporting a critical role for demethylation of histone H3 at K27 (H3K27) in establishing LTTM of Ifitm1 through a mechanism of loss and failure to regain a repressive chromatin landscape. Taking our findings together, we demonstrate that epigenetic priming can indeed confer heritable cellular memory of temporary TF activity in mammalian cells.

RESULTS

A C/EBPα pulse induces long-term transcriptional memory at a subset of target genes.

In the B cell-to-macrophage TD model (15), exogenous C/EBPα is induced and collaborates with endogenous PU.1 at enhancer elements to activate the macrophage gene expression program (14). Interestingly, a commitment point during the TD process is reached between 18 and 24 h whereby the induced, exogenous C/EBPα expression can be removed, but the cells continue to convert toward macrophage specification (15). Using this knowledge, we designed a pulse-chase-restimulation protocol to determine if any target genes display long-term memory after temporary activation by C/EBPα (Fig. 1A). The protocol includes an initial pulse period of 6 or 12 h, sufficient time for the induction of chromatin changes (14) but not for commitment, and a chase period of 6 days (doubling time of 15.4 h, resulting in 9.3 cell divisions in 144 h) (see Fig. S1A in the supplemental material). To validate the protocol design, we determined if the C/EBPα transgene or other known TFs induced in the pulse (17) were maintained at higher levels during the chase. To test this, we performed Western blot analyses on nuclear and cytoplasmic extracts to gauge C/EBPα levels (to track the shuttling of the transgene) and on nuclear extracts to gauge PU.1 and Runx1 levels. By day 3 of the chase period, total levels of tested TFs were comparable to levels in the control cells, suggesting that our memory protocol was designed with adequate time in the chase period for induced TFs to return to baseline and for C/EBPα-estrogen receptor (ER) to return to the cytoplasm (Fig. 1B). To generate a comprehensive list of possible memory events, transcriptome sequencing (RNA-seq) was performed to measure transcript levels before and after 6 h of C/EBPα stimulation from (i) naive cells (vehicle-treated control [CT] cells), (ii) cells previously pulsed for 6 h (6hP), and (iii) cells previously pulsed for 12 h (12hP) (average from 3 replicates). RNA-seq data were used to interrogate the following: (i) to determine if any transcripts induced by the initial pulse remain elevated after the chase period, as these would be genes displaying persistent memory of induction, and (ii) to determine if any of the genes that return to baseline levels display more robust induction upon restimulation. The latter group of genes would therefore display characteristics of LTTM and are the primary interest in this study. Working under the hypothesis that chromatin changes occurring during the pulse are responsible for establishing LTTM, we reasoned that a longer pulse might enhance the memory effect by allowing more time for chromatin changes to occur and accumulate. Therefore, we also used the RNA-seq data to determine if a longer pulse time (12 h versus 6 h) increases memory.

FIG 1.

A C/EBPα pulse induces long-term transcriptional memory at a subset of target genes. (A) Experimental timeline showing the memory protocol, with time points analyzed in this figure indicated by arrows. For the initial pulse, C/EBPα activity was induced by the addition of β-estradiol (β-E2). Control cells (CT) were treated with vehicle (ethanol [EtOH]). After the pulse period, cells were washed three times in medium to remove the β-estradiol and relieve C/EBPα activity. (B) Cellular fractionation was performed, and lysates were analyzed by Western blotting to determine the relative levels of transgene C/EBPα-ΕR in the cytoplasm and nucleus after the pulse and throughout the chase period. Levels in the nucleoplasm of PU.1 and Runx1, two transcription factors induced in the pulse, were also analyzed. The numbers 1 and 2 indicate biological replicates. D1, day 1 of the chase, etc. (C) RNA-seq was performed in vehicle-treated (CT) and β-estradiol-treated pulsed cells (6 h and 12 h pulsed) after the 6-day chase period and 6 h into restimulation (RS). At left is a heat map of the 35 genes displaying highly significant LTTM at 6 h into restimulation with a 12-h initial pulse (FDR < 0.01). After chase and restimulation, transcript levels are shown for control, 6-h-pulsed (6hP), and 12-h-pulsed (12hP) samples. At right is a heat map displaying the fold difference, or memory effect, of 12-h-pulsed samples over control samples at 6 h into restimulation. Genes are listed in order from most to least significant, starting at the top. Locations in the heat map for the Cd14, Lrp1, and Ifitm1 genes used as examples are highlighted with asterisks. Three biological replicates were sequenced for this analysis. (D) RT-qPCR confirmation of the RNA-seq results for LTTM genes Cd14 and Ifitm1, as well as control genes Ccl2 and Klf6 that do not display significant memory. Data are displayed as fold induction during the 6-h restimulation period, with the first pulse being the initial induction of the CT cell sample during the restimulation period. Data are the averages from three biological replicates; a two-sided t test was performed to determine statistical significance. (E) Expression after the first pulse. Box plot diagrams show the average transcript levels in vehicle-treated CT cell samples and pulsed cells after the initial pulse (AP) for significant memory genes (LTTM, n = 35) and control genes that are induced by the pulse but do not acquire memory (no LTTM, n = 1,562).

In total, only one gene (Ccl9) induced by the 12-h pulse remained significantly elevated after the chase period, while no gene reached this significance threshold with a 6-h pulse (log fold change, 2.2; false discovery rate [FDR], ≤0.01) (Table S1). Six hours into restimulation, 35 genes displayed significant LTTM after a prior 12-h pulse (termed memory genes; FDR, ≤0.01) (Table S1), representing approximately 3% of all genes induced by C/EBPα (Fig. 1C). Highlighted with asterisks in Fig. 1C are memory genes chosen here for further mechanistic study, namely, Cd14, Lrp1, and Ifitm1. Ccl2 and Klf6 were selected as control genes as both are induced by C/EBPα but do not display transcriptional memory as defined by the stringent criteria used to generate the list of 35 memory genes outlined in Fig. 1C.

To verify that transcripts induced at the pulse do indeed return to baseline throughout the chase, we performed reverse transcription-quantitative PCR (RT-qPCR) analysis of representative LTTM gene transcripts (Cd14, Lrp1, and Ifitm1) in a detailed time course through the chase period (Fig. S1B), which confirmed that transcripts returned to prestimulation levels by day 3 of the chase. RT-qPCR performed after an initial pulse in CT cells and during the secondary pulse in previously stimulated cells confirmed the existence of memory at representative genes and confirmed the increase in memory between a 6-h and 12-h initial pulse time (Fig. 1D and S1C). Interestingly, memory genes (LTTM) exhibit on average a greater fold induction at the initial pulse than control genes (no LTTM), as depicted by box plots of expression levels (Fig. 1E).

C/EBPα-pulsed cells display a more robust LPS response.

The most significant memory gene in the TD model is Cd14. To confirm that Cd14 displays memory at the protein level in addition to the transcript level, we performed flow cytometry analysis after the chase period and 24 h into restimulation with an antibody that recognizes membrane-bound Cd14. Following the 6-day chase, the signal intensities from CT and pulsed cells were indistinguishable. However, memory could be observed 24 h into restimulation as the percentage of cells in the population expressing high levels of Cd14 (Cd14^high) was significantly greater in the pulsed population (Fig. 2A). In the TD system, as cells convert from B cells to macrophages, they become increasingly responsive to lipopolysaccharide (LPS) stimulation, as measured by levels of cytokine production (15). Present on the host cell plasma membrane, Cd14 plays a major role in Toll-like receptor 4 (TLR4) sensing of bacterial wall molecule lipopolysaccharides and is essential for downstream signaling events and cytokine production (18). We therefore tested the ability of pulsed and control cells to respond to LPS during the early stages of TD (Fig. 2B shows the experimental timeline), expecting that the enhanced levels of Cd14 in the pulsed population could mediate more robust cytokine production. Confirming this hypothesis, pulsed cells transcribed significantly more of the LPS-responsive cytokine genes Ccl2, Ccl3, and Il1b than control cells in response to LPS treatment (Fig. 2C).

FIG 2.

C/EBPα-pulsed cells display a more robust LPS response. (A) Flow cytometry analysis was performed after the chase period (AC) and 24 h into restimulation (RS) on control (CT) and pulsed (P) cells to measure Cd14 levels. The percentage of Cd14^high cells is graphed on the right. Allophycocyanin (APC)-conjugated Cd14 antibody was used. Data are the averages from three biological replicates; a two-sided t test was performed to determine statistical significance. (B) Experimental timeline for LPS stimulation test. After the chase, cells were simultaneously treated with β-estradiol and vehicle or LPS (1 μg/ml) to start the restimulation period. (C) Transcript levels at 6 h of restimulation. Pulsed and control cells were taken for transcript analysis after stimulation with vehicle or LPS for 6 h. Transcript levels of the cytokine genes Cd14, Ccl2, Ccl3, Tnf, and Il1b and the control gene Cd11b were measured. Data are the averages from three biological replicates; a two-sided t test was performed to determine statistical significance.

Transient transcription factor binding primes chromatin for enhancer activation.

While LTTM can be induced with a transcription factor pulse in the TD model, the downstream events resulting in the acquisition and retention of memory are unknown and are of primary interest for the epigenetics community. The prevailing, albeit simplified, model of gene activation initiated by TF binding to enhancer sites consists of the following steps: (i) TF recruitment and possible nucleosome remodeling, (ii) histone acetyltransferase and mediator recruitment, (iii) BRD2/4 recruitment, and (iv) Pol II C-terminal domain (CTD) phosphorylation resulting in transcriptional elongation of target genes. Pol II elongation can be coupled to histone demethylation at H3K27 (19, 20) and methylation at H3K4 (enhancers and promoters) and H3K36 (gene body) (20–22) (Fig. 3A). Based on this scheme, we performed a stepwise analysis to determine which events are critical for memory formation.

FIG 3.

Transient transcription factor binding primes chromatin for establishment of active enhancer sites. (A) Stepwise model of chromatin events occurringdownstream of TF recruitment used to dissect the molecular events responsible for memory acquisition. (B) ChIP-seq for C/EBPα at prestimulation levels (CT cells), after a 12-h pulse, and after the chase (pulse-chase) visualized as average peak trace files for LTTM and control gene enhancers and promoters. Three biological replicates were combined for a single sequencing run. (C) ChIP-seq for PU.1 at prestimulation levels (CT cells), after a 6-h pulse, and after the chase (pulse-chase) visualized as average peak trace files for LTTM and control gene enhancers and promoters. Three biological replicates were combined and sequenced as one sample for analysis. (D) ChIP-qPCR for PU.1 after chase. ChIP-qPCR was performed at LTTM and no-LTTM gene enhancer sites in control (CT) cells and 6-h-pulsed (P) cells after the chase. Data are the averages from three biological replicates, and a two-sided t test was performed comparing the CT-chase to pulsed-chase values at each site. (E) H3K27ac ChIP-qPCR after the chase (AC) and after 6 h of restimulation (RS) in CT and 12-h-pulsed samples. Both memory and control gene enhancer sites were surveyed. Data are the averages from three biological replicates; a two-sided t test was performed to determine statistical significance.

Past studies using experimental models with Saccharomyces cerevisiae have shown that short- and long-term transcriptional memory is observed at select galactose-responsive genes when yeast previously switched from glucose to galactose medium are reexposed to galactose. The mechanism attributed to LTTM in this system is debated, with some studies reporting histone variant H2AZ deposition and gene tethering to the nuclear pore as important events in memory formation (2, 23), while other studies demonstrate that cytoplasmic inheritance of transcription factors is the responsible mechanism (3, 4). We found that the global levels of predominant TFs C/EBPα and PU.1 returned to prestimulation levels in the chase (Fig. 1B); however, it is possible that some degree of TF induced by the pulse remained stably bound to chromatin throughout the chase period to mediate transcriptional memory. To test this, we performed chromatin immunoprecipitation with high-throughput sequencing (ChIP-seq) analysis for C/EBPα and PU.1 from control prestimulation, pulsed, and pulsed-chase samples. For both C/EBPα and PU.1, a small amount of retention was observed at target enhancer sites (see Materials and Methods for how enhancers were defined). However, this phenomenon was not unique to the memory genes (Fig. 3B to D). Interestingly, the gain of C/EBPα after the pulse at LTTM enhancers is greater than that at control gene enhancers, consistent with the observation that LTTM genes are induced to a higher degree (Fig. 1E). Previous work showed that during B cell-to-macrophage TD, the levels of C/EBPα on target enhancers become saturated 3 h after induction (14); however, we observed a time-dependent increase in memory between 6 and 12 h of the initial pulse. Together, these observations argue against TF retention as the key mechanism responsible for memory in this system.

Following our model (Fig. 3A), we next asked if histone acetylation induced by the pulse remained at elevated levels through the chase. We performed ChIP for histone H3 acetylated at K27 (H3K27ac) to gauge the level of enhancer activation after the chase period and during restimulation. We observed no difference in the levels of H3K27ac between pulsed and CT cells after the chase, suggesting that the H3K27ac deposited during the pulse (14) is lost. However, upon restimulation, we observed that the enhancers of the LTTM genes Lrp1 and Ifitm1 displayed higher levels of H3K27ac in pulsed cells (second stimulation) than in CT cells (initial stimulation), with a similar trend for the LTTM gene Cd14 enhancer (Fig. 3E). Together, these data support the conclusion that memory genes are epigenetically primed for activation but not by augmented and retained levels of H3K27ac at enhancer sites. Therefore, in subsequent studies we focused our investigations on events occurring downstream of TF binding and H3K27 acetylation, including transcriptional elongation and histone methylation, as chromatin events potentially inherited to maintain LTTM.

Pol II elongation during the pulse is required for memory formation.

Following histone acetylation, BRD4 is recruited to transition RNA polymerase II (Pol II) into the elongation phase by employing pTEFb to mediate CTD phosphorylation (24). Therefore, we speculated that LTTM establishment might require Pol II elongation during the pulse. To test this, during C/EBPα induction we treated cells with flavopiridol (Flavo), which is known to inhibit the release of poised Pol II into the elongating form capable of transcribing through the gene body. After the pulse, cells were washed and released into the memory protocol. Importantly, multiple groups have demonstrated that Flavo treatment during gene induction does not hinder recruitment of TFs, Pol II, and cofactors or histone acetylation (22, 25). Thus, while Flavo treatment impairs successful transcription of memory genes at the pulse (Fig. S2A), it does not prevent the engagement of chromatin at memory gene loci in response to TF induction. Strikingly, Flavo treatment at the pulse essentially blocked memory formation (Fig. 4A), suggesting that memory formation requires successful Pol II elongation.

FIG 4.

Pol II elongation but not retention is required for memory formation. (A) Flavopiridol (300 nM) was added to the cell culture simultaneously with C/EBPα induction and washed out after the pulse (P), and the cells were carried through the memory protocol. RNA was taken at 6 h of restimulation for qPCR analysis to determine the memory of selected LTTM genes. Data are represented as the fold induction of P/CT transcripts, where a value of 1 (dashed line) signifies no memory; data are the averages from three biological replicates. A two-sided t test was performed comparing the P/CT values for the dimethyl sulfoxide control to those for Flavo. (B) Flavo was added at 500 nM for 4 h during the chase, after which the cells were washed three times in medium and replated to continue the memory protocol. RT-qPCR analysis was performed in pulsed and control cells at 6 h into restimulation to measure transcript levels of the memory genes Cd14, Lrp1, and Ifitm1 as well as of the control genes Ccl2 and Klf6. Data are represented as fold induction of P/CT transcripts, where a value of 1 (dashed line) signifies no memory; data are the averages from three biological replicates. (C) ChIP-seq for total Pol II (unphosphorylated form using antibody 8WG16) at prestimulation levels (CT cells), after a 12-h pulse, and after the chase (12-h pulse-chase) visualized as average peak trace files for LTTM and control gene enhancers and promoters. Two biological replicates were combined and sequenced as one sample for analysis. (D) ChIP-seq for poised and elongating Pol II (phosphorylated at serine 5 [s5p] of the CTD using antibody 4H8) at prestimulation levels (CT cells), after a 12-h pulse, and after the chase (12-h pulse-chase) visualized as average peak trace files for LTTM and control gene enhancers and promoters. Two biological replicates were combined and sequenced as one sample for analysis.

Retention of low levels of elongating Pol II or retention of docked or poised polymerase at promoters may drive transcriptional memory (11, 26). Supporting this hypothesis, it has been demonstrated that the presence of poised Pol II at promoters enables rapid immediate early gene activation in neurons (27). It has been documented that within 1 h of Flavo treatment, total Pol II is lost from the gene body of actively transcribed genes (28). To first determine if low levels of elongating Pol II are retained and could thus bolster reactivation of LTTM genes, pulsed and control cells were treated for 4 h on day 4 of the chase to abrogate any elongating Pol II still present on memory genes. Posttreatment, cells were washed and released into fresh medium to recover and then on day 6 restimulated for 6 h to assess memory. While the treatment period was sufficient to temporarily impair Pol II transcription (Fig. S2B), Flavo treatment during the chase did not dampen the memory of tested LTTM genes (Fig. 4B). This led us to conclude that, for the genes tested, retained low levels of elongating Pol II are not responsible for LTTM. To determine if Pol II was retained in any form at LTTM gene enhancers or promoters, we performed ChIP assays for Pol II using antibodies that recognize (i) the unphosphorylated CTD to survey total Pol II (clone 8WG16) and (ii) the CTD phosphorylated at serine 5 (clone 4H8) to survey Pol II in the poised and elongating forms (29). We found that at LTTM gene promoters and enhancers, Pol II levels increased with the pulse; however, after the chase the levels were indistinguishable from those of the control (prestimulation) (Fig. 4C and D). ChIP-qPCR analysis confirmed that after the chase period, the levels of poised Pol II at surveyed gene enhancers and promoters were comparable between pulsed and control cells (Fig. S2C), leading us to conclude that retained Pol II is not a contributing mechanism to LTTM in this model of transcriptional memory.

Memory gene regulatory regions are not marked by sustained H3K4 methylation.

Impairing transcriptional elongation at gene bodies and enhancer RNAs (eRNAs) during cytokine stimulation in macrophages is known to also impair deposition of H3K4 methylation on de novo activation sites (22). As blocking transcriptional elongation abrogated memory in this study (Fig. 4A), we reasoned that deposition of H3K4 methylation at promoters and/or enhancer sites could be important for LTTM establishment and retention. Consistent with this hypothesis, high levels of histone H3 di- or trimethylated at K4 (H3K4me2/3) at gene promoters in yeast are known to persist after cessation of transcription (30), and short-term transcriptional memory in primary mouse macrophages is correlated with the deposition and retention of H3K4me1 at de novo-activated enhancers (1). Furthermore, both yeast and human studies of long-term transcriptional memory have demonstrated inheritance of elevated H3K4me2 deposited during the initial pulse (7, 26). To determine if H3K4 methylation acquired during the pulse persisted through the chase period, we performed ChIP-seq analysis to map H3K4me1, H3K4me2, and H3K4me3 at various points along the memory protocol. To identify chromatin events unique to memory genes, we performed analyses on the set of LTTM genes (n = 35) as well as on a set of control genes that are induced by C/EBPα but do not display LTTM (n = 1,562). At putative LTTM gene enhancers after the chase, H3K4me1 levels were slightly elevated in pulsed cells compared to levels in control cells. However, this same level of retention was seen at control gene enhancers as well (Fig. 5A, compare purple line to black line). ChIP-qPCR at model gene enhancer sites at the pulse and through the chase period revealed that while H3K4me1 was indeed deposited during the pulse, the levels fell gradually through the chase to prestimulation levels (Fig. S3A). No discernible difference was observed in the levels of H3K4me3 between control and pulsed cells at LTTM gene promoters (Fig. 5A, right, compare purple line to black line). Furthermore, ChIP-seq and ChIP-qPCR at selected LTTM gene regulatory regions confirmed that while H3K4me2 levels increase at the pulse, the increased levels return to prestimulation conditions after the chase (Fig. 5B and C and S3B). Hence, we conclude that LTTM in our model system does not involve significant retention of H3K4 methylation at promoters or enhancers.

FIG 5.

LTTM genes exist in a more repressed chromatin state prior to stimulation. (A) Average profiles for H3K4me1 levels at probable enhancers (centered on p300 binding sites) of LTTM and control genes are shown on the left. Two biological replicates were combined for a single sequencing run. Average profiles for H3K4me3 levels centered on LTTM and control gene promoters are shown on the right. Two biological replicates were combined for a single sequencing run. (B) Average profiles for H3K4me2 levels centered on LTTM and control gene enhancers or promoters. In this experiment, the signal in prestimulation control cells (CT), 6-h-pulsed cells, and pulsed cells after the chase (pulse-chase) are depicted. Two biological replicates were combined for a single sequencing run. (C) H3K4me2 ChIP-seq. Genome browser snapshots of H3K4me2 ChIP-seq in prestimulation control (CT), 6-h-pulsed (AP), and after-chase (AC) samples for representative memory (Cd14, Lrp1, and Ifitm1) and control (Ccl2 and Klf6) genes. (D) Violin plots display the difference in H3K4 methylation levels before the pulse at LTTM and control gene enhancers and promoters. LTTM, n = 35; CT, n = 1,562. A two-sided Wilcoxon test was used to estimate P values. (E) H3K27me3 prestimulation. Violin plots display the difference in H3K27me3 levels before the pulse at LTTM and control genes at enhancers and promoters and in gene bodies. LTTM, n = 35; CT, n = 1,562. A two-sided Wilcoxon test was used to estimate P values.

LTTM genes display a more silent chromatin signature prior to stimulation.

In our analysis of H3K4 methylation states on LTTM genes and control genes, we observed that LTTM genes generally displayed lower levels of H3K4 methylation at enhancers and especially at promoters (Fig. 5D). To ask if the lack of active histone marks indicated a higher level of repression, we analyzed the levels of the repressive mark H3K27me3 on LTTM versus control genes. This revealed that before the pulse, LTTM genes displayed higher levels of H3K27me3 across enhancers, promoters, and gene bodies (Fig. 5E). In addition, the prestimulation levels of total Pol II at LTTM gene promoters were lower than those of control no-LTTM genes (Fig. 4C and D). Taken together, these observations indicate that LTTM genes are in a more repressed chromatin state (low H3K4me, high H3K27me3, and low Pol II) than control genes that do not acquire memory with a pulse of C/EBPα activity.

Demethylation of H3K27me3 contributes to the mechanism of LTTM for Ifitm1.

Our previous work with the TD model showed that upon C/EBPα induction, there is a gradual loss of H3K27me3 at C/EBPα binding sites, particularly at de novo enhancers that harbor high levels of H3K27me3 and low levels of H3K4me prior to induction (14). Furthermore, we along with others have shown that effective H3K27 demethylation is linked to active Pol II elongation (20, 31, 32), an event critical for the establishment of LTTM (Fig. 4A). Therefore, we asked if LTTM gene enhancers and/or promoters were depleted of H3K27me3 in the pulse and if this depletion was sustained through the chase and could enable a quicker activation upon restimulation. When we looked across LTTM and control genes for retained loss of H3K27me3, we did not observe loss at gene promoters but did observe moderate loss of H3K27me3 at both control and LTTM gene enhancers (Fig. S4A). Looking specifically at the top 10 LTTM genes, here again we see that the enhancer regions show greater sustained loss of H3K27me3 than promoter regions (Fig. S4B). Although the difference is subtle, we reasoned that the removal of H3K27me3 at LTTM enhancers could be biologically significant as LTTM genes are decorated with higher baseline levels of H3K27me3 than those of the control gene group (Fig. 5E). Interestingly, when we studied the density of H3K27me3 signal at LTTM enhancers in control and pulsed cells after the chase, most LTTM genes displayed sustained loss of H3K27me3 at enhancer sites, with a particular enrichment for Ifitm1 enhancer sites (Fig. 6A, asterisks mark enhancers for the genes Cd14, Lrp1, and Ifitm1). Sustained loss of H3K27me3 was confirmed by ChIP-qPCR at Cd14 and Ifitm1 enhancers, where H3K27me3 levels remained lower throughout the chase in pulsed cells (Fig. 6B and S4C).

FIG 6.

Loss of H3K27me3 is a heritable event that contributes to LTTM of Ifitm1. (A) Heat map showing the density of the H3K27me3 signal at LTTM geneenhancers in control and 12-h-pulsed cells after the chase derived from H3K27me3 ChIP-seq analysis. Three biological replicates were combined for a single sequencing run and used to calculate the intensity of signal over each enhancer region. Enhancer sites are ranked by the most to least amount of H3K27me3 lost with a 12-h pulse. Asterisks indicate enhancers associated with the memory genes chosen for study here (Cd14, Lrp1, and Ifitm1). (B) ChIP-qPCR of H3K27me3 at LTTM genes and associated regulatory regions from control cells (CT) and 6-h- and 12-h-pulsed cells after the chase. Data are the averages from three biological replicates; a two-sided t test was performed to determine statistical significance. (C) UNC1999 experiment depicted in the schematic. The Ezh1/2 inhibitor UNC1999 (or dimethyl sulfoxide [DMSO]) was added to cells at 3 μM for 48 h, and then the compound was washed out, and the cells were left to recover for 6 days (D6). Samples were collected after the treatment (AT), at day 2 of the chase period (D2), following the chase period (D6), and 6 h after C/EBPα was induced (D6 stim). (D) Profile plots show H3K27me3 ChIP-seq signal after treatment (AT) and after the chase (day 6, D6) at all genes induced by C/EBPα, in both CT (dimethyl sulfoxide)- and UNC1999-treated cells. Three biological replicates were combined for a single sequencing run. (E) RT-qPCR was performed to determine transcript levels for LTTM genes Cd14 and Ifitm1 at time points throughout the UNC1999 experiment. The fold changes in transcript levels between UNC1999- and vehicle (DMSO)-treated samples are depicted over the graphs at each time point. Data are the averages from three biological replicates; a two-sided t test was performed to determine statistical significance. (F) ChIP-qPCR of H3K27me3 at LTTM genes and associated regulatory regions from control cells (DMSO), UNC1999-treated cells after the 48-h treatment period (AP), and UNC1999-treated cells after the 6-day chase (AC). Data are the averages from three biological replicates; a two-sided t test was performed to determine statistical significance.

The level of H3K27me3 at enhancers is generally low, making it difficult to draw conclusions based exclusively on ChIP studies. Therefore, to substantiate the importance of H3K27 demethylation on memory establishment and retention, we used small-molecule inhibitors that impair the enzymatic activity of the H3K27me3 demethylases and methylases, namely, Jmjd3/Utx (inhibited by glycogen synthase kinase [GSK]-J1/J4) (33) and Ezh1/2 (inhibited by UNC1999) (34). If H3K27 demethylation and retention contribute to memory, then blocking demethylation during the pulse should reduce the memory effect. Indeed, treatment of cells with GSK-J1/J4 during the pulse (added to cells at the same time as C/EBPα induction) significantly reduced memory observed for the LTTM genes Cd14, Lrp1, and Ifitm1 but did not alter expression of Ccl2 or Klf6 (Fig. S4D). However, when we looked at the transcript levels during the pulse under GSK-J1/J4 treatment, the induction of LTTM genes was reduced by compound treatment (Fig. S4E). Therefore, it is not possible to conclude that the diminution of memory by compound is a result of changing H3K27me3 levels as reduced transcription could also be at play. We then asked if globally reducing the levels of H3K27me3 with the Ezh2 inhibitor UNC1999 (34) in the absence of a C/EBPα pulse could impart the same transcriptional memory phenomenon. Following the experimental design depicted in Fig. 6C, cells were treated for 48 h to allow for ∼3 rounds of cell division under Ezh2 inhibition; the compound was washed out, and the cells were allowed to recover for a 6-day chase period. ChIP-seq analysis confirmed that H3K27me3 levels were reduced with compound treatment at C/EBPα target gene enhancers and, to a lesser extent, promoters (Fig. 6D, AT). After the 6-day chase, we observed a modest sustained loss of H3K27me3 (Fig. 6D, D6). Treatment with UNC1999 did indeed result in significant derepression of Cd14 and Ccl2 (Fig. 6E and S4F), but we observed a gradual return to pretreatment levels for all genes tested by day 6 of the chase (Fig. 6E and S4F). After the chase, cells were induced with C/EBPα to determine if prior Ezh2 inhibition resulted in a transcriptional memory phenotype. Upon C/EBPα induction, Cd14 and Ifitm1 retained significant memory of UNC1999 treatment (Fig. 6E). ChIP-qPCR studies at candidate gene regulatory regions confirmed that the Ifitm1 enhancer site lost H3K27me3 after UNC1999 treatment and that this loss was retained over the 6-day chase period (Fig. 6F). These results indicate that prior Ezh1/2 inhibition can substitute for a C/EBPα pulse to impart memory at Ifitm1 and that the resulting chromatin state is maintained for at least 6 days to mediate greater induction upon C/EBPα stimulation. Together, these data support the conclusion that for some genes under investigation in this study, H3K27 demethylation is critical for LTTM, either through maintained loss of H3K27me3 and/or through an unknown event linked to demethylation.

DISCUSSION

Epigenetic changes, by definition, survive cell division and are therefore able to transmit to progeny information gained in parental cells. This mechanism can provide progeny with an adapted response to environmental cues. Here, we show that a short pulse of transcription factor C/EBPα activity can induce transcriptional memory on a subset of genes, and this memory survives multiple rounds of cell division. Many of the identified memory genes play important roles in the antimicrobial response of the innate immune system (e.g., Cd14, Ifitm1, Abca1, Lrp1, Cd33, and Serpinb2). We show that transcriptional memory of Cd14 expression translates to the protein level, where pulsed cells display higher levels of Cd14 on the cell membrane. As Cd14 is a critical mediator of Toll-like receptor signaling in response to bacterial infection, pulsed cells in our model also transcribe larger amounts of Ccl2, Ccl3, and Il1b in response to LPS stimulation. These data suggest that memory of Cd14 expression could enable a more robust response to bacterial infection.

The transcriptional memory we observed in this model increases in a manner dependent on the time of the initial pulse. This simple observation lends some critical insight into the mechanisms at play as it suggests that the epigenetic features that maintain memory are cumulative. For example, we know that peak binding of C/EBPα is achieved within the first 3 h of induction. If simply the binding of target sites by C/EBPα was sufficient to establish memory, then we would not observe an increase in memory with a 12-h pulse compared to the level with a 6-h pulse. Supporting these results, when we looked at C/EBPα binding by ChIP, we did not observe retention through the chase period.

Memory formation imparted by C/EBPα is blocked in the presence of the Pol II elongation inhibitor flavopiridol. Therefore, we focused our mechanistic studies on chromatin events dependent on Pol II elongation, including histone methylation and demethylation events. From previous work, we had insight into which events occur in the first 12 h of TD that correlate well with the kinetics of our model, and these include H3K4 methylation (increasing steadily in the first 12 h) and H3K27 demethylation (decreasing steadily in the first 12 h of TD). Therefore, we analyzed in depth the addition or removal of these marks and their maintenance through the memory protocol.

Memory formation has been associated with long-lived retention of H3K4me1/2 at enhancers or of H3K4me3 at gene promoters in previous models of transcriptional memory (14, 30). However, in our system, retained H3K4me1/2/3 histone modifications were not observed but were steadily lost throughout the chase. In regard to H3K4 methylation, however, we did observe a significant difference in the baseline levels of this mark between LTTM and control gene regions where LTTM gene enhancers and promoters had lower levels of H3K4 methylation before the pulse. This led us to interrogate the levels of repressive H3K27me3 to understand if LTTM gene regions are in a more repressed state in naive cells. Indeed, we observed higher levels of H3K27me3 at LTTM genes. Interestingly, prior to the first stimulus, we also observed less total Pol II at LTTM gene promoters than at control gene promoters. Together these findings indicate that the characteristic of LTTM genes compared to control genes (those induced by C/EBPα but show no memory) is germane to that of de novo versus primed enhancers. This observation supports the hypothesis that chromatin-priming events result in transcriptional memory, where LTTM genes are epigenetically primed during the pulse for greater induction upon restimulation.

In our model, we postulate that the priming event conferring memory occurs in conjunction with Pol II elongation and at least in part involves H3K27 demethylation at certain gene regulatory regions (Ifitm1). We draw this conclusion from the results of different experimental approaches. First, small but detectible levels of H3K27me3 are lost with the pulse at enhancers of the most significant LTTM genes (including genes Cd14 and Ifitm1), and this loss is retained through the chase. Second, treating cells with an Ezh2 inhibitor for 2 days in the absence of a C/EBPα pulse can induce a phenotype similar to transcriptional memory. Therefore, we conclude that LTTM seen in this model system is maintained at least in part by a passive mechanism, where loss of H3K27 methylation occurring in conjunction with gene activation is not reestablished. This could allow faster reactivation by removing one step in the transcriptional activation process and, for example, allow faster acetylation at H3K27 during active enhancer establishment. Indeed, we show that memory gene enhancer sites gain more H3K27ac upon restimulation. We propose that, in our model, removal of repressive histone marks is more critical for long-term memory than the addition of active histone marks. In line with our observation, a recent paper studying long-term memory in primary macrophages concluded that removal of repressive H3K9 methylation enabled transcriptional memory induced by LPS exposure (35). In future studies, it will be interesting to survey the nucleosome landscape to query whether nucleosome depletion occurring during Pol II elongation is another event that is inherited, as suggested in recent studies of transcriptional memory in mammalian cells (6, 7).

Polycomb-mediated repression is known to play a pivotal role in defining and securing cell fate by silencing lineage-specific genes in progenitor cells and then subsequently silencing developmental genes during terminal differentiation (36). Questions stemming from the data presented here concern which stimuli and what cell types in vivo have the ability to acquire transcriptional memory that can be transmitted to daughter cells. It is tempting to speculate that a biological event within innate immune cells with the ability to establish LTTM in vivo could be signaling in myeloid progenitor cells that induce transcription coupled to H3K27 demethylation at macrophage-specific genes prior to terminal macrophage differentiation. Interestingly, in response to type I interferon exposure, hematopoietic stem cells (HSCs) exit quiescence and produce progenitor cells to populate different blood lineages (37), and common myeloid progenitors are activated for emergency myelopoiesis to generate large numbers of inflammatory monocytes and neutrophils (38). In this scenario, myeloid progenitor cells could acquire memory of pathogen exposure through interferon-induced gene activation, pass transcriptional memory to daughter cells for many generations, and thereby display the phenotype of trained immunity.

The term epigenetic memory most commonly refers to chromatin changes in response to external cues that result in permanent cell fate commitment, usually in regard to cellular differentiation or reprogramming. For example, many models of differentiation, reprogramming, or transdifferentiation that involve the addition or removal of key lineage-specifying TFs harbor an intrinsic commitment point where the initiating event (TF induction) can be relieved, but the cells continue toward the newly specified fate (39). Often, a broad assumption is made that any chromatin changes occurring before the commitment point are lost as the overall cellular phenotype reverts to the original state. However, the data presented here support the idea that some chromatin changes occurring before the commitment point are indeed inherited and can be preserved as latent memory. Phenotypic consequences of this latent memory may not be revealed until cell descendants are presented with a similar stimulus or differentiation cue, upon which an adapted response is observed. Clinically, this finding could be relevant to understand the developmental origins of health and disease (DOHaD) (40), drug sensitization and addiction (41), and genomic priming for developmental competency (42). For example, fetal and neonatal exposures to abnormally high levels of hormone or endocrine disruptors could epigenetically prime target genes for altered responsiveness later in life, when similar hormonal signals increase during puberty. Long-lived epigenetic priming could also play a critical role in establishing immunological memory within innate immune cell progenitors that have the capacity to proliferate in response to pathogen challenge. Together, our data suggest that altering the repressive chromatin landscape has the capacity to impart long-lived priming effects and has implications in trained immunity and in DOHaD.

MATERIALS AND METHODS

Transdifferentiation and memory assay.

C10 (HAFTL cells with a stably integrated C/EBPα transgene) cells were seeded at a concentration of 2e6 cells/ml in B cell medium (RPMI medium lacking phenol red, 10% charcoal-stripped fetal bovine serum [FBS], l-glutamine, 50 μM β-mercaptoethanol, penicillin-streptomycin). To initiate transdifferentiation (TD), a 2× TD mix was made by adding the following components to B cell medium: β-estradiol at 200 nM, interleukin-3 (IL-3; Peprotech) at 20 nM, and macrophage colony-stimulating factor (M-CSF; Peprotech) at 20 nM. The 2× mixture was added 1:1 to cell-containing medium to start the TD process. After the pulse period, cells were washed three times in warm B cell medium and replated in fresh B cell medium. Medium was refreshed every 2 days. TD was initiated as above at day 6 to start the restimulation period.

ChIP protocol.

Cells at 10e6 cells/ml were fixed in 1% formaldehyde for 10 min at room temperature. See the experimental procedures in the supplemental material for the detailed protocol. Approximately 10 million cells were used to make chromatin for each immunoprecipitation. The antibodies and amounts used were as follows: 6 μl of PU.1 (sc-352x), 10 μl of C/EBPα p300 (sc-585x), 3 μl of H3K27me3 (07-449; Millipore), 5 μg of H3K4me1 (C15410037; Diagenode) 2 μl of H3K4me2 (04-790; Millipore), 5 μl of H3K4me3 (05-1339; Millipore), 2 μl of H3K27ac (39133; Active Motif), 5 μl of RNA polymerase II CTD repeat YSPTSPS antibody (8WG16) (ab817; Abcam), and 5 μl of RNA polymerase II phospho-S5 4H8 (39097; Active Motif).

For ChIP-seq library preparation, 10 ng of ChIP DNA was used for each library. An NEBNext DNA library prep reagent set for Illumina (E6000L) was used to generate the libraries, along with the standard Illumina barcoding primers according to the manufacturer's instructions.

RNA-seq.

RNA was extracted using TRIzol (15596-026) according to the manufacturer's instructions. Five micrograms of RNA was used as starting material for rRNA depletion using an Ambion RiboMinus kit (A1083708). rRNA-depleted samples were analyzed on an Agilent Bioanalyzer to ensure that the majority of rRNA was depleted before the library was prepared. A total of 100 ng of rRNA-depleted RNA was then used in an NEBNext Ultra Directional RNA-seq kit (E7420L) to generate sequencing libraries according to the manufacturer's instructions.

Bioinformatic analysis.

All parameters, software versions, and references can be found in Table S2 in the supplemental material. For RNA-seq, reads were mapped using STAR with the Ensembl gene annotation gtf (GRCm38.75). bigwig tracks were made using DeepTools BamCoverage. For quantitative analysis, reads were counted on gene exons using HTSeq-count (union mode). Sample scaling and statistical analysis were performed using the Rpackage DESeq2 (version 1.6.3) with default parameters. LTTM genes were selected for showing a significant difference between pulse-chase-restimulation and control-chase-restimulation values (P value of <0.01, negative binomial Wald test), no significant difference between pulse-chase and control-chase values (P value of >0.01), and strongly significant induction upon restimulation (pulse-chase-restimulation/pulse-chase of >1 and FDR of <0.01).

For ChIP-seq, reads were mapped using STAR. Duplicate reads were removed using picard, while bigwig tracks were made using DeepTools BamCoverage. Peak calling was performed using macs2.

For all quantitative analyses, reads were counted on merged regions for each time point using the R package csaw (function regionCounts). In order to estimate sequencing depth, we counted reads on 10-kb windows covering the whole genome (using the csaw package, function WindowCounts) and then used DEseq2 on these counts to compute the scaling factors. This ensures robustness of the scaling with respect to differences in peak numbers and to highly enriched outlier regions. Scaled counts were then transformed to a log₂ scale (after adding a pseudocount of 1) and used in subsequent analyses.

For enhancer analysis, we merged peaks of p300 found at 0 h and 12 h after C/EBPα induction and selected those showing at least a 2-fold increase at 12 h. All p300 increasing peaks in ±100 kb of gene transcription start sites (TSS) were selected as potential associated enhancers. We then measured the levels of the other ChIP-seq signals in these regions to make box plots. For promoter analysis, we used the region of ±2 kb around the TSS and proceeded as for enhancer regions.

Average plots and heat maps were made using DeepTools (parameter, -binSize 10).

Accession number(s).

Data are available under Gene Expression Omnibus (GEO) accession number GSE72488.