Capturing Environmental Plant Memories in DNA, with a Little Help from Chromatin

Abstract

Plants are eukaryotes living mostly immotile in harsh environments. On occasion, it is beneficial for their survival to maintain a transcriptional response to an environmental stress longer than the stress lasts (transcriptional memory) and even to reiterate such a response more quickly or more strongly when the same stress is re-encountered (priming memory). In eukaryotes, transcription takes place in the context of chromatin, the packaging material of DNA. Chromatin regulation is often invoked when it comes to environmental transcriptional and priming memory in plants, but rarely chromatin-based regulation can be accurately assigned to a given aspect of transcription in vivo. The conserved eukaryotic chromatin-modifying system Polycomb/Trithorax can support both long-term stability and flexibility of gene expression in Drosophila. The main principles of Polycomb/Trithorax regulation will be outlined and illustrated with the best-studied case of environmental memory from Arabidopsis. Despite being complex, the Polycomb/Trithorax system relies on experimentally tractable elements in the form of DNA, termed Polycomb/Trithorax Responsive Elements. PREs/TREs are essentially memory DNA elements. Here, relevant information to identify PRE/TRE-like elements in plants is highlighted. Examples of priming memory in plants are discussed in relation to the first two reported putative memory DNA elements. Arguably, similar cases from plants can be conducive in dissecting the contribution of DNA-based from chromatin-based regulation of transcription, when two types of DNA elements are defined: those representing binding sites for the transcription factors determining the environmental response and those controlling memory by regulating chromatin modification dynamics, ultimately maintaining the corresponding transcriptional state.

Introduction

Natural environments are dynamic and exert a high degree of stimulation especially on immotile inhabitants. Despite encountering many unfavorable conditions (stresses), some of which often re-occur during their life, plants are outstanding survivors. They possess stress memory mechanisms that can be established at the transcriptional level in the form of a response outlasting its initial trigger (transcriptional memory) or a stronger/faster response upon a second stress (priming memory). They also possess mechanisms to adjust the length of memory, i.e. to erase it at the appropriate time, as maintaining the memory indefinitely can be detrimental. A frequent hypothesis is that memory responses are mediated at the level of chromatin.

In eukaryotes, including plants, transcription takes place in the context of chromatin, the packaging material of DNA. Consequently, the DNA-based and the chromatin-based regulation of transcription are highly intertwined. Coding and regulatory DNA sequences are major determinants of transcriptional outcomes. General transcription factors (TFs) associate with DNA usually near the transcription start site (TSS) at basal promoter elements to assemble RNA polymerase II complexes for transcription initiation (Conaway and Conaway 1997, Horikoshi et al. 1989). Specific TFs recognize diverse DNA elements located in promoters/enhancers and function to confer differential gene expression as well as to stabilize transcription pre-initiation complexes. Another determinant of transcriptional outcome is the chromatin state (Cosgrove et al. 2004). The basic unit of chromatin is the nucleosome composed of globular proteins called histones. Each nucleosome wraps about 147 bp of DNA around one histone octamer containing two copies each of H2A H2B, H3 and H4, having protruding flexible N-termini or tails. Post-transcriptional modifications of histones, mostly on tails (therein ‘histone modifications’), are known to alter the chromatin state. Currently there are few experimental systems where chromatin-based regulation can be accurately assigned to a given aspect of a transcriptional program in vivo, despite the recent burst of interest in epigenetics, an intensive research area where histone modifications are often in the spotlight.

The highly conserved Polycomb group (PcG)/Trithorax group (TrxG) system is well characterized in different eukaryotes (Steffen and Ringrose 2014). The PcG and TrxG proteins form several large multiprotein complexes that covalently modify histone tails and cause structural changes in chromatin. Two PcG complexes are termed Polycomb repressive complex 1 (PRC1) and Polycomb repressive complex 2 (PRC2). PRC1 monoubiquitinates Lys119 on histone H2A (H2Aub) and PRC2 trimethylates Lys27 on histone 3 (H3K27me3). In contrast, TrxG complexes positively affect transcription and trimethylate Lys4 and Lys36 on histone 3 (H3K4me3 and H3K36me3). PcG and TrxG were initially characterized genetically in fruit flies as antagonistic regulators of Hox genes (Lewis 1978) encoding TFs that specify the identity of each segment along the body axis of the animal. The transcription status of Hox genes persists throughout the life of the animal in appropriate patterns, even though the signals initiating these patterns are only present for hours in the embryos. The best-studied case of PcG/TrxG regulation in plants comes from the process of vernalization, whereby the prolonged cold of winter provides competence to flower in spring, after the cold stimulus disappears (Amasino 2004). Venalization is based on a mitotically stable PcG/TrxG program operating at the potent inhibitor of the flowering gene Flowering Locus C (FLC). Besides these intensively studied examples, PcG/TrxG have hundreds of other targets genes in plant and animal genomes, many of which change dynamically during development, illustrating that PcG/TrxG regulation of gene expression can also be flexible (Ringrose 2007). As I will explain later, these two features result from multiple mechanisms to couple tightly and antagonize the silencing and activating activities of PcG and TrxG.

A distinguishing feature of PcG/TrxG is that they act via specialized DNA elements called Polycomb/Trithorax Response Elements (PREs/TREs). The main property of PREs/TREs is to recruit PcG/TrxG to specific genomic locations, based on the DNA binding ability of proteins interacting with the PcG/TrxG complexes at the time of binding. PREs/TREs are essentially memory DNA elements, largely responsible for maintaining chromatin states that instruct outcomes on transcription in the absence of initiating signals. Initially they were discovered and characterized in flies (Simon et al. 1993, Busturia et al. 1997), where they have properties that have not been tested in other systems. Nevertheless, PRE-like elements have also been postulated to exist in vertebrates and plants, and the recruitment property was verified for a number of them (Schubert et al. 2006, Sing et al. 2009, Woo et al. 2010, Berger et al. 2011, Lodha et al. 2013, Sun et al. 2014).

Arguably, priming memory responses in plants at PcG/TrxG targets have multiple potential experimental advantages for dissecting eukaryotic mechanisms of chromatin-based memory. The initiating signals are external, and therefore can be manipulated. While cellular processes with transcriptional memory are generally difficult to identify, in priming assays, the need for an information storage system is obvious and, moreover, three intricate memory features become separable. The establishment, length and manifestation of memory each take place in different experimental stages: pre-treatment, recovery and subsequent treatment(s). Notably, the initial stress responses and the priming memory are genetically separable, as documented for example for heat stress priming memory (Charng et al. 2007, Baurle 2016). Finally, in plants, priming is widespread, i.e. it can be trigged by a variety of stresses affecting multiple sets of genes, many of which are PcG/TrxG targets (Jaskiewicz et al. 2011, Ding et al. 2012, Sani et al. 2013, Virlouvet et al. 2014, Feng et al. 2016, Lamke et al. 2016). The experimental systems under which these target genes are currently investigated are also diverse. For example, the length of the recovery period, i.e. length of memory, can vary from 2 h (Ding et al 2012) to 10 d (Sani et al 2013). As we learn more from these systems, it is important to decide if all of them can satisfy a unifying definition of memory, as discussed in a series of thought-provoking publications (Bird 2007, Muller-Xing et al. 2014).

After briefly informing on the complex principles of PcG/TrxG regulation learned from animal systems, I will outline PRE properties and their assay in Drosophila, and recent advances on PRE-like activity in Arabidopsis.

In What Way Can Histone Modifications Instruct Gene Expression?

It is important first to consider this question because the correct answers may be different from expected when we apply the same logic used with generic regulators of transcription (TFs). Perhaps this is one reason why the role of histone modifiers such as PcG/TrxG is often but erroneously reduced to mere silencers/activators of transcription, and correlation between H3K27me3/H3K4me3 and gene silencing/activation is interpreted as a causality.

One way to simplify the question and reveal the distinct contribution of histone modifications to gene expression is in the absence of other specific regulatory features natively present in eukaryotic promoters/enhancers. Using this strategy, a recent mammalian study provided a surprisingly simple answer that can also unify the mode of action of four chromatin regulators. The authors fused each of these chromatin regulators to a protein that binds to the DNA of an active promoter in an inducible manner, and followed the timing of silencing in individual cells using time-lapse microscopy (Bintu et al. 2016). They observed that for all regulators, upon recruitment to the promoter, cells switched from active to silent states with no intermediated states, with some delay. The delay was different among cells, thus revealing a stochastic process. To understand how permanent the gene expression changes are, the chromatin regulators were released from the promoter. Gene reactivation, like silencing, also occurred in stochastic, all-or-none events. What distinguished different regulators was the timing and the rate of silencing and re-activation. For example, embryonic ectoderm development (EED), a core PcG protein, was qualified as a ‘slow silencer’ because silencing events were initiated several cell division cycles after its recruitment to the artificial promoter in a fraction of cells that increased with time. Once EED was released from the promoter, a fraction of cells soon reactivated whereas the remaining fraction remained silenced for a longer time, making EED a ‘partial committer’. Overall, the study demonstrates that chromatin regulators/histone modifications instruct time-dependent outcomes on transcription. It also highlights a contrast in gene regulation directed by regulators modifying histones and binding of TFs to promoters. TF-dependent control takes place in a graded non-random manner, while chromatin regulators stochastically control ‘duration-dependent fractional control’ (Bintu et al. 2016).

Is the time-dependent control of gene expression by histone modifications employed in actual biological systems? Temporal control is best evidenced by striking heterochronic phenotypes of PcG/TrxG mutants in different eukaryotes, i.e. homeotic transformation of one organ type into another in flies and incorrect timing of developmental phase transitions in plants (Lewis 1978, Kinoshita et al. 2001), and it is supported by histone modification-based mechanisms in many cases in flies. To name a recent example from plants, the work by Sun et al. (2014) delineated a timing mechanism involving histone modifications at a PcG target gene for termination of floral stem cells. When considering the cases where gene expression programs are flexible during development at genes targeted by PcG/TrxG, we have to explain how the same system can confer seemingly opposite outcomes.

Self-Reinforcing and Antagonizing Activities of PcG and TrxG Support Stable but Flexible Gene Expression States

The molecular mechanisms of PcG/TrxG-mediated regulation derived from Drosophila and mammalian studies were recently covered in detail (Steffen and Ringrose 2014, Bauer et al. 2016) and remain best characterized from the genetic to the biochemical level in flies. Here, three types of activities were reported: PcG self-reinforcing, TrxG self-reinforcing and PcG–TrxG antagonizing, each supported by multiple mechanisms. It is proposed that at PcG/TrxG gene targets, activation and silencing are in dynamic balance. Once either of the states is established, the positive feedback mechanisms ensure its stability. However, when one type of counteracting activities is favored, the state can switch. This is related to the bistable behavior described by Bintu et al. (2016) where cells can only adopt two extreme states while the intermediate states are not stable due to momentum from reinforcing activities.

The existence of multiple variants of most of the individual components assembling PcG/TrxG complexes in mammals and plants undoubtedly adds a degree of complexity to these systems. While a lot of the biochemical activities underlying PcG/TrxG regulation are not detailed in plants, the overall remarkable PcG/TrxG conservation allows us to predict that stable and switchable states characterize PcG/TrxG genes in plants too. This can be exemplified with accumulated evidence for FLC that can overall point to the existence of both stable and switchable states (Fig. 1).

Fig. 1

Stable and switchable PcG/TrxG states at FLC during vernalization. (A) Quantitative repression during cold, and stable maintenance upon return to warm of FLC. Intervals (∼10 d) of warm and cold exposures are represented by yellow and blue rectangles. FLC mRNA is represented with black lines, interrupted by dotted lines for shorter exposures to cold (data from Sheldon et al. 2002). (B) Cell population encoding memory of winter based on two possible states of each cell: ON and OFF (model from Angel et al. 2011, Satake and Iwasa 2012). (C–F) Conversion of H3K4/36me3 into H3K27me3 during the transition between stable states via nucleation (D, E) and spreading (F) of H3K27me3. Distribution relative to the FLC locus and amount of different histone modifications is in orange, red and green (data from Finnegan and Dennis 2007).

FLC mRNA is highly abundant in the warm season before winter in ecotypes with a winter habit, based on the activity of a suite of TrxG-like proteins that load H3K4me3 at the TSS and H3K36me3 in the gene body. In this active state, one likely mechanism by which TrxG counteracts PcG is by allosteric inhibition of PRC2 action by H3K4/36me3, as described in Schmitges et al. (2011). This high transcription state and all intermediate states down to a saturation level reached during sufficient cold exposure remain stable upon return to warm conditions (Fig. 1A); these states represent what is classically known as FLC winter memory (Amasino 2004). They are established as a result of two PcG processes. Nucleation is initiated by a cascade of events in the cold and corresponds to three types of quantitative changes: accumulation of H3K27me3, reduction of active marks and FLC expression. Spreading takes place upon return to warm in the gene body (Finnegan and Dennis 2007), and it is a type of PcG-reinforcing activity most probably involving binding of PCR2 to H3K27me3 (Margueron et al. 2009) during DNA replication via co-operation with DNA polymerase α subunits (Hyun et al. 2013). Together, nucleation and spreading lead to gradual conversion of the active histone modifications before cold exposure into H3K27me3 after cold (Fig. 1C–F). An algorithmic evaluation of histone modifications at FLC (Angel et al. 2011, Satake and Iwasa 2012) based on a model of positive feedback of histone modifications (Dodd et al. 2007) resulted in the prediction that quantitative states during cold may be determined at the cell population level when each cell behaves in a bistable manner and their timing of switching is heterogeneous and proportional to the duration of cold (Fig. 1B). This would moreover render the response stable against noisy signals and allow inheritance of winter memory during cell divisions.

How Do PcG/TrxG Find Their Target Genes?

PcGs target hundreds to thousands of genes in different eukaryotes; their activity follows cell-specific genomic patterns despite the fact that the core PcG proteins do not have DNA binding activity and are ubiquitous. In Drosophila, chromatin domains of H3K27me3 extend over several hundreds of kilobases to include clusters of co-regulated genes and sharp binding peaks of two or more PcG proteins at individual genes (Schwartz et al. 2006). In contrast, mammalian and plant PRC2 are more tightly co-localized with H3K27me3. PcG-binding sites and H3K27me3 are mostly located close to the TSS and sometimes ‘blanket’ the transcription unit in mammals (Ringrose 2007). In plants, the association of FERTILIZATION INDEPENDENT ENDOSPERM (FIE) and H3K27me3 is strong in gene bodies (Deng et al. 2013). There are currently two, non-mutually exclusive, models addressing mainly the recruitment of PcGs. Comparably less is known about how TrxGs select their targets, except for Drosophila.

The ‘chromatin sampling’ or responsive model postulates that, at the majority of targeted loci, H3K27me3 derived from PCR2 is influenced mainly by chromatin features, including transcription (Riising et al. 2014, Hosogane et al. 2016), DNA methylation and other histone modifications (reviewed by Klose et al. 2013, Wiles and Selker 2016). This model has gained support especially from data in vertebrates where it is linked with a specific genomic feature: most promoters contain CG dinucleotide islands. Both PcG and TrxG associate with different zing finger CxxC domain proteins capable of reading the unmethylated status of CpG islands (Blackledge et al. 2010, Thomson et al. 2010). In this model, both PcG and TrxG complexes continuously sample unmethylated CpG islands irrespective of the transcriptional status of the gene. Based on the positive feedback of each complex and the antagonizing function, chromatin sampling will result in accumulation of PcG at inactive regions and TrxG at transcribed genes without the need for direct interactions with specific DNA sequences. Some data from Arabidopsis can corroborate elements of this model. For example, in Arabidopsis, DNA methylation was suggested to prevent targeting by PcG in vegetative tissues at transposable elements and protein-coding genes specifically targeted in the endosperm (Weinhofer et al. 2010), and H3K27me3 redistribution was observed in the absence of maintenance DNA methylation (Zhang et al. 2007). Also, redistribution of H3K27me in response to transcription was documented in Arabidopsis at individual genes (Kwon et al. 2009, Buzas et al. 2011).

Transcription-responsive chromatin features can be illustrated for FLC (Fig. 2). When the FLC transcript was interrupted by a T-DNA insertion in the gene body, maximum H3K27me3 levels accumulated in the area where there was no transcript and the same was true when transcription was obstructed by deletion of the TSS (Fig. 2A). These situations show that lack of transcription per se leads to an increase in H3K27me3. The kinetics of removal and addition of H3K27me were measured upon transcriptional activation and shut down at an FLC gene body fragment (Fig. 2B). The slow addition rate after cessation of transcription is reminiscent of the spreading of H3K27me3 after cold exposure, which requires cell divisions to ‘lock in’ the repressed state (Finnegan and Dennis 2007). Similarly, a mechanism of transcription-responsive H3K27me3 may operate at many genes that diminish expression during development, in cells that transition though a lineage. The presence of H3K27me3 can maintain a repressed state outside cellular types where the gene product is required, while buffering against low level activating activities that may be present in the newly differentiated cells.

Fig. 2

Re-distribution of histone modifications at FLC in response to changes in transcription. (A) Interruption of transcription with the T-DNA insertion in the flc20 mutant (left, the T-DNA is represented in black) or deletion of the TSS in the flc2 mutant (right, the 1.8 kb of DNA including the ATG deletion is represented by an interrupted line) leads to high levels of H3K27me3 (green bar) in the FLC gene body area where the transcript is absent in both mutants. In the same area, H3K36me3 becomes undetectable in flc20 (not investigated in flc2). The FLC gene body and promoter regions are represented in gray. (B) Kinetics of H3K27me3 following changes in transcription in an inducible transgenic system. The 5′ end portion of a 10 kb transgenic FLC (tFLC) and the inducible promoter are represented in gray. Vertical and horizontal axes are H3K27me3 abundance and time in days, respectively. Black and green indicate timing of OFF/ON and ON/OFF transitions of transcription, respectively, in the experimental system used in Buzas et al. (2011): FLC transcription level reached a maximum 6 h after addition of dexamethasone (Dex) and a minimum 2 d after removal of Dex. Scale bar = 1 kb. Experiments were performed at warm temperature. Data are from Buzas et al. (2011).

Based on Hox gene regulation in Drosophila, an instructive model was proposed in which sequence-specific binding factors and/or non-coding RNAs directly interact with PcG proteins to define their site of action. The DNA elements at these sites, termed PREs/TREs, not only fulfill the recruitment function but also have other special properties in Drosophila (outlined below). To understand if the instructive model can be generalized to other organisms, appropriate assays have to be available in these other systems. I outline PRE properties and their assays and discuss these in relation to relevant data from Arabidopsis in the remaining part of this review.

The decision on which of these models is in operation and when has important consequences on our understanding of gene regulation. When histone modifications are linked with PREs, they can instruct transcriptional outcomes, and when chromatin-modifying complexes sample the chromatin, they respond to changes in the transcription.

Polycomb Response Elements (PREs) in Drosophila

Discovery

Long-term stability of gene expression is required for proper development of the body segments along the anterior–posterior axis of the fly embryo. The factors controlling segmentation, termed gap proteins, provide positional information to establish a repressed state of homeotic genes in specific regions of the embryo. However, gap proteins are only present within the initial hours of embryo development. PcG proteins are ubiquitous and they are the ones known to ‘lock in’ the repressed states stably throughout cell divisions. It was already known decades ago that PcG proteins maintain silencing by somehow modifying chromatin, and the question was whether they establish a state of chromatin compaction that can then become self-propagating or whether they play an active role and are required continuously to maintain repressed states. An example that provided the answer involved experiments with a 725 bp DNA fragment called MCP725 (Busturia et al. 1997). PcG proteins are required at MCP725 to maintain position-specific silencing of the homeotic gene Abdominal-B during late development. When this DNA element was excised at any larval stage, using a conditional recombinase system, silencing of an associated reporter was lost, i.e. the expression reappeared. Therefore, this DNA element is continuously required to maintain the repressed state. Other similar DNA elements were characterized in flies at about the same time (Simon et al. 1993).

Characteristics

PREs can be several hundred base pairs long without a clear consensus. Instead they have a modular structure with multiple DNA sequence motifs of more than one DNA-binding protein. These proteins include, for example, Pleiohomeotic (Pho) and related Pho-like, GAGA factor (GAFl, Trithorax-like), Pipsqueak, Dorsal switch protein, Zeste, etc. Importantly, while most of these can tether PcG to their targets, they can also be found in other locations. Pho, for example, plays crucial roles in PcG silencing via interaction with both PRC1 and PRC2, but a large proportion of Pho-binding sites are associated with promoter regions of genes marked with active histone modifications (Schuettengruber et al. 2009). The combinatorial action of several DNA-binding proteins, the lack of factors that favor transcription, transcription itself, non-coding RNAs and as yet unidentified factors are expected to participate in PRE function.

In Drosophila, PREs can fulfill a number of tasks, based on which they can be identified (Fig. 3). (i) They recruit PcG and can create an ectopic H3K27me3 domain at new sites of integration in the genome. (ii) They can establish a repressed state. In transgenic assays, this property can be monitored using a cassette where the PRE is intercalated between a minimal promoter and a reporter gene. Other PRE tasks from Drosophila have not been verified in other organisms: (iii) PREs maintain a memory of the transcription state over cell divisions. To test this property, a suitable experimental system where the transcriptional response can be monitored over cell divisions is required (Fig. 3D i) as well and another verified PRE (Fig. 3D ii). Notably, unlike enhancers, PREs do not determine patterns of gene expression. (iv) Most PRE sites coincide with TRE sites (Schwartz et al. 2010)

Fig. 3

PRE properties and assays in Drosophila. (A) A PRE contains sites for multiple DNA-binding proteins. White and gray bars in (A–C) represent DNA motifs of two PcG-recruiting proteins. (B) When PcG complexes associate with PRE, they create a H3K27me3 domain that can extend to the neighboring sequences [property (i)]. (C) PREs silence their targets in association with H3K27me3 [property (ii)]. Properties (i) and (ii) can be tested in transgenic assays (B and C). A PRE alone or a cassette where PRE is inserted between a minimal promoter (P) and a reporter are fused are inserted at ectopic locations. The minimal promoter drives transcription of the reporter (black arrow) while the cassette including the PRE silences the reporter (black bars). (D) A PRE can maintain a pattern established in response to a signal by a TF, after the signal disappears. (I) Illustration of maintenance of a pattern in an anatomical structure over cell divisions; (II) distinct contribution of an enhancer and its associated PRE. A specific enhancer establishes the pattern in response to a signal. Its associated PRE (blue), or a PRE from another locus (pink), can maintain the pattern over cell divisions.

Yet other PRE properties may be less general or are just emerging. These include long-range interactions (Cheutin and Cavalli 2014) and stabilizers/amplifiers of transcriptional output (Ringrose 2007, Steffen and Ringrose 2014).

PRE-Like Activity in Arabidopsis

In Arabidopsis, PRE-like activity was studied mainly in a developmental context at only a handful of individual loci (Schubert et al. 2006, Berger et al. 2011, Lodha et al. 2013, Sun et al. 2014) as discussed in several reviews (Schatlowski et al. 2008, Grossniklaus and Paro 2014). A summary of these studies in relation to the main Drosophila PRE properties (Table 1) indicates that in plants, as in mammals, PREs have been mostly recognized based on their ability to confer silencing of reporter genes and create H3K27me3 domains [properties (i) and (ii)]. Therefore, it is important to identify systems capable of probing other PRE properties, especially maintenance [property (iii)]. Some supporting data exist at FLC, and potentially at two other genes with priming memory.

Table 1

Partial PRE activity at target genes in plants

Reference	Target gene	PRE property
Development
Schubert et al. (2006)	AGAMOUS	(i)
Berger et al. (2011)	LEAFY COTYLEDON2	(i), (ii)
Lodha et al. (2013)	BREVIPEDICELLUS, KNAT2	(i)
Sun et al. (2014)	KNUCLES	(i)
Parent of origin
Grossniklaus and Paro (2014)	MEDEA	(i), (ii)
Environmental response
Sung et al. (2006)  Sheldon et al. (2002)  Buzas et al. (2011)	FLOWERING LOCUS C	(i), (ii), (iii)
Virlouvet et al. (2014)	RD29B	Potentially (iii)
Feng et al. (2016)	P5CS1	Potentially (iii)

Some early work on FLC indicates that each of the two halves of the first FLC intron contains sequences required for maintaining FLC repression after cold (Sheldon et al. 2002), i.e. they can fulfill property (iii). Other studies (also reviewed in Buzas et al. 2012) have confirmed that cis-elements within the promoter-proximal part of the first FLC intron may fulfill this role via non-coding RNAs (Heo and Sung 2011) and via the recently identified RY elements (Yuan et al. 2016), and that the distal half of the FLC intron recruits PcG activity via an unknown mechanism. Therefore, FLC most probably represents a case where chromatin-responsive and instructive features, including PRE properties (i–iii), co-exist and their mechanisms are still insufficiently defined.

Demonstrating the maintenance property (iii) is especially cumbersome for developmental contexts where it is difficult to distinguish persisting vs. vanishing initiating signals. Conversely, priming memory assays offer unique opportunities to elucidate whether a DNA element confers memory or not. Two recent pioneering studies on priming advanced further beyond reporting correlation between chromatin modifications and transcriptional response by defining DNA elements with a role in memory (Virlouvet et al. 2014, Feng et al. 2016). In both cases, the target gene involved is under PcG/TrxG regulation, although they were not specifically investigated in this context.

The ‘memory element’ ME at RD29B

Priming memory is well documented during reoccuring dehydrations stress (Ding et al. 2012, Avramova 2015). Water loss is reduced in plants subjected to more than one cycle of dehydration/hydration, and transcriptional responses of critical genes are different in each cycle. In assays where repetitive treatments of air-drying dehydration for a couple of hours is followed by rehydration for 22 h, approximately 300 dehydration-induced genes show higher transcript levels in the second treatment as compared with the first and return to low levels during recovery. One of these genes that recently became a marker for dehydration priming memory is RD29B (RESPONSIVE TO DESSICATION 29B). While it may be disputable that a short recovery period of as little as 2 h may indeed monitor memory, a major advantage of the system is that the RD29B transcription rate reaches extremely high levels during the 2 d of the standard assays. This wide transcription interval allows clear monitoring of priming memory and of factors influencing it to a different degree.

A key mediator of dehydration stress is ABA. The RD29B promoter contains two adjacent ABA-responsive elements (ABREs; ACGTGG/TC) bound by ABRE-binding factors (ABFs). Based simply on mRNA quantification in the first and third dehydration treatment, multimerized versions of 50 bp sequences containing ABREs, termed MEs (memory elements), were demonstrated to confer priming memory in transgenic assays. The presence of both motifs and the flanking context were found to determine priming memory (Virlouvet et al. 2014).

In a series of elegant genetic and physiological approaches, the authors attempted to identify the trans factors acting on ME to confer memory. Three ABFs that target ABREs were only necessary for the strong dehydration response at RD29B (Virlouvet et al. 2014) as suggested earlier (Ding et al 2012). Instead, ABA signaling components were essential, although not sufficient, for priming memory at RD29B. The additional factors contributing to priming memory were attributed to chromatin-derived mechanisms; specifically, the retention of H3K4me3 and stalled RNA polymerase II at native RD29B during the recovery phase when transcription at RD29B becomes inactive. A role for H3K27me has so far been difficult to establish. There were no substantial changes in H3K27me3 during the 48 h of the two dehydration treatments at the regions investigated (Liu et al. 2014). The role of CURLY LEAF, one of the catalytic subunits of PRC2, in RD29B regulation was difficult to interpret in part due to the reduced amount of ABA in curly leaf (Liu et al. 2014). It is moreover possible that detection of critical fluctuations in H3K27me3 requires cell divisions, for example if the recovery period is longer or during more repetitions. Notably the priming memory at RD29B can be monitored even after four treatments (Ding et al. 2012), but H3K27me3 levels were not measured at the stage the response becomes saturated. Alternatively, H3K27me3 may play a critical role at RD29B in another developmental stage. A study showed that dehydration stress induces RD29B more strongly in seedlings than in germinating seeds, while the response to ABA treatment was more acute in germinating embryos (Nakashima et al. 2006). The transition from embryonic to post-germinative growth is now known as a hotspot where PRC1 components initiate repression of the seed maturation program (Charng et al. 2007, Yang et al. 2013, Wang et al. 2016).

To conclude, transgenic MEs are capable of conferring priming memory. If this indeed can represent a PRE, property (iii) can be determined in further experiments while considering principles of PcG/TrxG regulation and/or design principles from Drosophila PREs (Fig. 3).

The ‘essential for memory fragment’ (EMF) element at P5CS1

Proline accumulation in response to moderate environmental stress, including salt, increases the plants tolerance upon a second encounter. This can be monitored by increased accumulation in both free proline and transcription of a proline biogenesis gene termed P5CS1 (pyrroline-5-carboxylate synthetase 1) in assays with successive salt treatments (Fig. 4). Furthermore, a second environmental factor, namely light, was recently implicated in the P5CS1 induction during recurring salt stress (Feng et al. 2016). It was found that the manifestation of transcriptional memory in the second salt treatment, but not the establishment of the response in the pre-treatment, required light. Importantly, this same distinction could also be associated with the basic region/leucine zipper TF named Long hypocotyl 5 (HY5), and a DNA region of the P5CS1 promoter containing one of the confirmed HY5-binding sites. This region, situated in the area between –2.6 and –2.3 kb from the TSS, was indentified in transgenic assays by a series of promoter deletions and it is termed the EMF (‘essential for memory fragment’).

Fig. 4

Features of priming memory, exemplified at P5CS1. Priming memory features can be partitioned by experimental stages in regular priming assays. The memory is established during the pre-treatment stage. The length of memory can be determined by adjusting the period of recovery between treatments. Finally, priming memory is manifested during the second treatment and it can be monitored easily, together with its conditioning factors, by an increased level in comparision with pre-treatment. In the case of P5CS1, priming memory can be monitored at three levels (black line): proline content, P5CS1 mRNA and mRNA of a GUS reporter when fused with a 2.6 kb sequence of the P5CS1 promoter (p2.6GUS). Loss of memory is demonstrated by deletion of 0.3 kb from the same construct (p2.3GUS, interrupted white line) (Feng et al. 2016; see text for details).

The PcG/Trx chromatin signatures H3K27me3/H3K4me3 are present at the P5SC1 locus and fluctuate during salt treatments. Feng and colleagues investigated closely the dynamics of H3K4me3 in both the native context of P5CS1 and on two types of transgenes containing promoter fragments either with or without EMF, or EMF itself together with a minimal promoter, upstream of a β-glucuronidase (GUS) reporter. The high levels of H3K4me3 established in the first salt treatment were retained during the recovery stage at both native chromatin and the chromatin adjacent to the EMF-containing, but not EMF-lacking transgenes. H4K3me3 retention at the native loci for a short period appears to be a general feature in yeast (Ng et al. 2003) and was reported in plant studies at specific loci primed by dehydration stress during the recovery phase (Ding et al. 2012, Lamke et al. 2016). However, the presence and retention of H3K4me3 during recovery on a transgene are unique and can be reminiscent of a TRE-like feature. The levels of H3K27me3 were analyzed during the priming response only at some regions of the P5CS1 locus. Across the P5CS1 promoter-proximal gene body, H3K27me3 levels were reduced at the end of the first salt treatment seemingly to a higher extent than the second treatment, and the levels returned to the control level during the recovery period. Unfortunately, the significance of these changes and whether the locus contains other sites registering notable H3K27me3 with salt priming remained unclear. In a separate priming study, H3K27me3 dynamics during recovery were implicated in transcriptional responsiveness in the second treatment (Sani et al. 2013).

The transgenic construct where EMF alone was fused with a minimal promoter and a reporter did not recapitulate priming memory. Therefore, the authors reached the conclusion that the EMF is not sufficient for priming memory. It is not clear whether the EMF is a TRE or a truncated PRE that simply lacks the complete activity from the native enhancer conferring the response. The mRNA level of the reporter was higher at this construct compared with the minimal promoter alone. Again this has been interpreted as lack of property (ii), i.e. EMF does not silence the reporter, or by the presence of positive enhancer elements within the EMF.

In the working model proposed by the authors, two types of TF play roles in P5CS1 regulation. The first unknown TF regulates the salt-induced response, while HY5, probably together with its homolog HYH, is required at the EMF for H3K4me3 retention during recovery in the light. In the future, the interactions between HY5/HYH and EMF-related fragments should be explored in relation to PcG/TrxG recruitment.

PcG recruiting activity to DNA motifs in Arabidopsis

Although gene sets with H3K27me3/H3K4me3 signatures were extensively profiled in Arabidopsis whole seedlings (e.g. Zhang et al. 2007, Oh et al. 2008, Roudier et al. 2011), as well as in more specialized tissues (Weinhofer et al. 2010, Lafos et al. 2011, de Lucas et al. 2016), only in recent years have genome-wide analyses of Arabidopsis implicated different classes of DNA-binding proteins in recruitment of PcG, via PRC1 and PRC2, to putative DNA motifs (summarized in Table 2).

Table 2

Selected cis motifs that may correspond to putative PREs in plants

DNA motif (putative TFs)	Finding (reference)
	Genome-wide	Individual target
GAGA-like (BACIS PENTACYSTINE, BPC)	Enriched in FIE binding sites (Deng et al, 2013)	BPC6 directs SWINGER to GAGA motif at ABI4 promoter in roots (Mu et al. 2017)
	In vivo interaction between BP6 and LHP1/VRN2 via LHP1 Enrichment of GAGAGA promoters in common LHP1 and H3K27me3-targeted genes (Hecker et al. 2015)
	Enriched in LHP1-binding sites (Molitor et al. 2016)
Telo-box (TELOMERE REPEAT BINGING, TBR)	Enriched in FIE-binding sites (Deng et al. 2013)	—
	Enriched in LHP1-binding sites (Molitor et al. 2016)
	TRB1 and TRB3 are genetic enhancers of lhp1 (Zhou et al. 2016)
	Concerted reduction of H3K27me3 in tfl2 and clf (Wang et al. 2016)
CArG (MADS box gene)	Concerted reduction of H3K27me3 in tfl2 and clf (Wang et al. 2016)	SHORT VEGETATIVE PHASE interacts with LHP1 to repress SEPALLATA3 via H3K27me3 (Liu et al. 2009)
		AGAMOUS recruits PcG to WUSCHEL promoter (Liu et al. 2011)
RY elements (VIVIPAROUS1/ABSCISIC ACID3-LIKE, VAL)	AtBMI1 and H3K27me3 co-regulated genes (Merini et al, 2017)	VAL1/2 silence FLC via binding to intronic RYs and interaction with LHP1 (Yuan 2016)
Conserved non-coding sequences (unknown)	In silico analysis (Berke and Snel 2014)	—

The only PCR2 protein profiled genome-wide is FIE (Deng et al. 2013), which is also the only protein present in all PRC2 complexes. The great majority of over a thousand high confidence regions bound by FIE overlapped with H3K27me3, and many of these genes were de-repressed in PcG mutants. Both H3K27me3 and FIE are distributed across gene bodies, with FIE having a bias towards the 5′ end of genes. FIE-binding sites also containing H3K27me3 were enriched in three DNA motifs, similar to a zinc finger protein-binding site, to the telobox factor 1-binding site and to a GAGA factor (GAF)/Trx-like-binding site, which is a known component of PREs from fruit flies (Schuettengruber and Cavalli 2009)

Interestingly, the GAGA and telobox motifs were also over-represented in gene sets targeted by PRC1 components (Hecker et al. 2015, Wang et al. 2016). In Drosophila, GAF was named for the capacity to bind GAGAG stretches and has a dual role in activation and repression (Berger and Dubreucq 2012). In plants, there are no recognizable GAF homologs. However, the members of the BASIC PENTACYSTEINE (BPC) protein family have the ability to bind to the GAGA motif and control developmental expression of homeotic genes, reminiscent of animal GAFs. Based on the ability to form stable dimers and multimers, BPCs were suggested to affect DNA condensation and transcriptional activation in plants (Hecker et al. 2015), as is the case in animals. BPC6 appears to interact directly with LIKE HETEROCHROMATIN PROTEIN (LHP1) (Hecker et al. 2015) and also to direct PRC2 catalytic component SWINGER to the promoter of ABA INSENSITIVE 4 in roots (Mu et al. 2017). It is currently unclear if BPCs can recruit both PRC1 and PRC2 at the same targets and it is difficult to establish what is the relevance of the presence of the telobox in both sets of PRC2- and PRC1-binding sites. TELOMERE REPEAT BINDING (TRB) proteins, interacting with these motifs, were identified from a genetic enhancer screen of the lhp1 phenotype. Both have roles independent of PcG and appear to act as second layer repressive backups at PcG targets (Zhou et al. 2016).

In another study that supports targeting of PcGs by TFs, the authors proposed that different combinations of PRC members regulate developmental programs by selecting their targets using specific TFs. The CArG box, the binding motifs for MADs box transcription factors, is a notable example of a motif identified amongst others in the common targets with reduced H3K27me3 in LHP1 and CLF (Wang et al. 2016). Previous studies implicated MADS box factors in H3K27me H3K27me3 deposition at individual loci (Liu et al. 2009).

An interesting group of candidates for PRC1/PRC2 recruitment include proteins with modules for binding both DNA and histone modifications. VIVIPAROUS1/ABSCISIC ACID INSENSITIVE 3-like 1 and 2 proteins, VAL1/2, contain B3 domains that can bind RY elements (CATGCA) and PHD finger domains that bind methylated lysines on histone tails. VAL1/2-binding sites are enriched at PRC1 component AtBM1 target genes marked with H3K27me3 (Merini et al. 2017). At FLC, VAL1/2 bind to intronic RY sites leading to FLC silencing via physical interaction with LHP1 (Yuan et al. 2016).

The issue of PCR2 recruitment and how its target genes arose during evolution was addressed by an in silico study (Berke and Snel 2014). Most PCR2 targets in plants encode genes belonging to TF families, and some of these families contain higher than expected proportions of H3K27me3-marked genes (Lafos et al. 2011). Retention of H3K27me3 during gene duplication and correlation with conserved non-coding sequences (CNSs), mostly representing TF-binding sites, in the upstream region of these genes may be explained by a DNA sequence requirement for PRC2 recruitment (Berke and Snel 2014). Although only a handful of PREs have been investigated in plants, the proportion of cases where CNSs can be part of PRE function is high (Berke and Snel 2014).

In conclusion, genome-wide studies underpin the existence of DNA sequence-based recruitment of PcG in plants, sometimes verified for individual genes, and provide the first sets of TF candidates and DNA motifs that may operate as part of PRE-like activity.

Final Remarks

The PcG/TrxG field provides a good framework to interrogate complex relationships between histone modifications and transcriptional outcomes. Based on the existence of the regulatory DNA elements PREs/TREs, the instructive model of PcG/TRxG recruitment can help select those cases where a cause–effect relationship can be defined, just like in any clean genetic study where DNA has a central role. The set of PRE properties derived from Drosophila seem to represent a suitable toolkit for this purpose in plants. Combining the PRE principles design with the experimental advantages of a priming assay can lead to the development of improved systems to test the hypothesis that cellular memory can be mediated by chromatin. Cases that will not strictly fit the framework of the instructive model may benefit from additional principles, for example those currently under development at the intersection of mathematical modeling and biology (Ringrose and Howard 2017).

Funding

This work was supported by the Ministry of Education, Culture, Sports, Science and Technology, Japan [a Grant-in-Aid for Young Scientists B (grant No. 15K18558) and for Scientific Research on Innovative Areas (grant No. 16H01459)].

Abbreviations

Abbreviations
 
• ABF

  ABRE-binding factor

 
• ABRE

  ABA-responsive element

 
• BPC

  BASIC PENTACYSTEIN

 
• CNS

  conserved non-coding sequence

 
• EED

  EMBRYONIC ECTODERM DEVELOPMENT

 
• EMF

  essential for memory fragment

 
• FIE

  FERTILIZATION INDEPENDENT ENDOSPERM

 
• FLC

  FLOWERING LOCUS C

 
• GAF1

  GAGA-associated factor 1

 
• GUS

  β-glucuronidase

 
• HY5

  Long hypocotyl 5

 
• HYH

  homolog of HY5

 
• LHP1

  LIKE HETEROCHROMATIN PROTEIN 1

 
• ME

  memory element

 
• P5CS1

  pyrroline-5-carboxylate synthetase 1

 
• PcG

  Polycomb group

 
• Pho

  Pleiohomeotic

 
• PRC1/2

  Polycomb repressive complex 1/2

 
• PRE/TRE

  Polycomb/Trithorax Response Element

 
• RD29B

  RESPONSIVE TO DESSICATION 29B

 
• TF

  transcription factor

 
• TRB

  TELOMERE REPEAT BINDING

 
• TrxG

  Trithorax group

 
• TSS

  transcription start site

 
• VAL1/2

  VIVIPAROUS1/ABSCISIC ACID INSENSITIVE 3-like 1/2

Acknowledgments

I apologize for the possible omissions that were made in citing important contributions to the field.

Disclosures

The author has no conflicts of interest to declare.