1. Physics of Living Systems
Download icon

Hybrid protein assembly-histone modification mechanism for PRC2-based epigenetic switching and memory

  1. Cecilia Lövkvist
  2. Pawel Mikulski
  3. Svenja Reeck
  4. Matthew Hartley
  5. Caroline Dean
  6. Martin Howard  Is a corresponding author
  1. Computational and Systems Biology, John Innes Centre, United Kingdom
  2. Cell and Developmental Biology, John Innes Centre, United Kingdom
Research Article
Cite this article as: eLife 2021;10:e66454 doi: 10.7554/eLife.66454
5 figures, 13 tables and 1 additional file

Figures

Figure 1 with 4 supplements
Existing model cannot explain persistence of metastable memory at FLC.

(A) Schematic of FLC and VIN3 expression (left) and chromatin states (right, H3K36me3 modifications in green and H3K27me3 in red) over time, in the warm (red background), in the cold (blue background) and subsequent warm (red background), with wild-type at 10 and 20 days post-cold shown first and lhp1 mutant at 10 and 20 days post-cold below. (B) MN-ChIP data for H3K27me3 profile (normalised to H3 levels) at FLC nucleation region (H3 and H3K27me3 levels with error bars are shown separately in Figure 1—figure supplement 1) after a 6-week cold treatment at 5°C. Error bars are sem, from n=3 biological replicates, with separate populations of seedlings harvested from MS plates. (C) Fraction of FLC-ON cells in simulated lhp1 mutant from model in Berry et al., 2017a in warm after cold (full curves, each averaged over 4000 realisations). Number of nucleation-region nucleosomes varied, with assumption that 1/6 of the nucleation region covered in H3K27me2/me3 is sufficient for full silencing, with cells being FLC-ON if this is not satisfied at one or both of the FLC copies. Simulations compared to experimental data (black diamonds) from Yang et al., 2017 and newly acquired data (pink circles) (see also Figure 1—figure supplements 2, 3 and 4 and Materials and methods).

Figure 1—figure supplement 1
MN-ChIP experimental results across the FLC locus after a 6-week cold treatment at 5°C.

Positions of H3 shown in (A) and H3K27me3 in (B) (error bars: sem, n=3 biological replicates, with separate populations of seedlings harvested from MS plates).

Figure 1—figure supplement 2
Simulated fraction of FLC-ON cells in lhp1 as a function of time in the warm after cold.

(A,B) Simulated data from histone-feedback only model: full colored lines, with varying number of nucleation region nucleosomes, as indicated on right of each panel. Experimental data: same as Figure 1C. (A) As in Figure 1C, but here at least 1/3 of the H3 histones must carry K27me2/me3 for full silencing, with cells being FLC-ON if this is not satisfied at one or both of the FLC copies. (B) As in Figure 1C, but here the nucleosomes are regularly distributed after DNA replication (one to each daughter strand in turn), and where at least 1/6 of the H3 histones must carry K27me2/me3 for full silencing, with cells being FLC-ON if this is not satisfied at one or both of the FLC copies. (C) Fraction of FLC-ON cells in simulated lhp1 mutant using hybrid assembly/histone modification model, where nucleosomes are also regularly distributed after DNA replication (one to each daughter strand in turn). Full coloured curves: varying number of non-histone modification memory elements, as indicated on right of panel, in addition to modifications on three nucleosomes (each curve averaged over 4000 realisations). Experimental data: same as Figure 1C. At least 1/3 of the H3 histones must carry K27me2/me3 for full silencing, with cells being FLC-ON if this is not satisfied at one or both of the FLC copies.

Figure 1—figure supplement 3
Analysis of proportion of FLC-ON cells in lhp1 after a 10-week cold treatment.

Cold at 5°C was followed by different number of days (where Tx = x days) in long-day, warm conditions (at 20°C with 16 hr of light). Pink and gray bars are from same imaging data but where a variable threshold is used to determine the fraction of ON-cells (40th percentile, 85th percentile, 98th percentile, see ‘Experimental details’). Also shown is prediction from histone-feedback only model (blue), with three nucleosomes and at least 1/6 of the H3 histones must carry K27me2/me3 for full silencing, with cells being FLC-ON if this is not satisfied at one or both of the FLC copies, as in Figure 1C.

Figure 1—figure supplement 4
Images of lhp1 mutant roots.

Top: Images of representative roots from an early timepoint, 10WT1, 1 day after 10 weeks of cold at 5°C. The first root from the left has no FLC-ON cells. The other roots show Venus signal that is not located in the nucleus, with the white arrow pointing to a particular example. Bottom: Images of representative roots from a later timepoint, 10WT10, 10 days after 10 weeks of cold at 5°C. The first root from the left has no FLC-ON cells. The other roots show nuclear-localised FLC-Venus signal in some cells/cell files. The red arrow marks a file of ON-cells.

Figure 2 with 2 supplements
Model with extra assembly memory elements can explain nucleation dynamics and metastable memory at FLC.

(A) Schematic of the model with protein assembly as extra memory elements (see Figure 2—figure supplement 1 for a detailed schematic of the model). (B) VIN3 mRNA levels (black circles) in 8°C cold treatment for ColFRI, replotted data from Hepworth et al., 2018, blue line: linear fit. (C) Simulated H3K27me3 nucleation peak dynamics in cold (fraction of maximum possible occupancy; red line, averaged over 4000 realisations), with extra protein assembly memory elements, compared to data from Yang et al., 2014 (black circles, error bars: sem) for wild-type ColFRI. (D) Fraction of FLC-ON cells (full curves, each averaged over 4000 realisations) in simulated lhp1 mutant using model with varying number of non-histone modification memory elements, in addition to modifications on three nucleosomes, and with the assumption that 1/3 of the nucleation region covered in H3K27me2/me3 is sufficient for full silencing, with cells being FLC-ON if this is not satisfied at one or both of the FLC copies. Comparison shown to experimental data (black diamonds) from Yang et al., 2017, and newly acquired data (pink circles) (see also Figure 1—figure supplement 3 and Materials and methods). Details of the model, simulations and parameters are found in Materials and methods and Figure 2—figure supplement 2.

Figure 2—figure supplement 1
Detailed schematic of feedbacks in mathematical model including assembly dynamics.

States me0 to me3 refer to methylation state of H3K27. Neutral marks are me0/me1, repressive marks are me2/me3. States Pbound and Punbound represent whether a site in the nucleation region is bound by an assembly subunit or not. Black arrows represent state transitions; coloured arrows represent feedback interactions. Feedbacks from me2 state not included for clarity.

Figure 2—figure supplement 2
Varying pdem and pex in the histone-feedback only model to find a bistable region.

The fmin and fmax values justified in Table 5 are used with kme=kmehigh (Table 6). See 'Model parameters and initial conditions' section for definition and explanation of Bi, where strong bistability corresponds to Bi1. Simulations averaged over 4000 realisations.

Simulation of cold treatment followed by warm conditions in model with assembly positive feedback, including both nucleation region and gene body.

Simulated dynamics (lines) for assembly, as well as nucleation and body region H3K27me3 levels, for wild-type ColFRI. (A) Cold treatment (blue background) is followed by warm conditions (red background). Simulated H3K27me3 nucleation levels in cold same as Figure 2C. Experimental H3K27me3 ChIP data (black circles, error bars: sem) (Yang et al., 2014). (B) Simulated dynamics in the warm, after 2, 4, 6, or 8 weeks of cold treatment. The 8 week data in the warm is the same as in (A). Simulated assembly and H3K27me3 levels are fraction of maximum possible occupancy in relevant region, each averaged over 4000 realisations.

Figure 4 with 1 supplement
PRE excision simulation and simulated reactivation in Lov-1 accession.

(A) H3K27me3 levels over the entire FLC locus in a simulated PRE excision experiment. A fully spread state is simulated; the nucleation region (with any assembly elements) is removed (at day 11 in the simulations), thereby interrupting the looping reactions and compromising the spread state. Varying levels of PRC2 feedbacks (kmeex, right) are simulated in the body region when the nucleation region is removed, with each level averaged over 4000 realisations. (B,C) Simulated dynamics (lines) of assembly, as well as nucleation and body region H3K27me3 levels, for Lov-1 accession. Lov-1 is simulated with the parameter for transcription activation (α) in the warm higher than in ColFRI (specified in Materials and methods). (B) Cold treatment (blue background) is followed by warm conditions (red background). Experimental H3K27me3 ChIP data (black circles, error bars: sem) (Qüesta et al., 2020). (C) Simulated dynamics in the warm after 2, 4, 8, or 12 weeks of cold treatment. The 12-week data in the warm is the same as in (B). Simulated assembly and H3K27me3 levels are fraction of maximum possible occupancy in relevant region, each averaged over 4000 realisations. In Figure 4—figure supplement 1, non-replicating Lov-1 cells are simulated.

Figure 4—figure supplement 1
H3K27me3 levels are stable after cold in Lov-1 in simulated non-replicating cells.

Simulated dynamics of hybrid assembly/histone modification model for nucleation region and body H3K27me3 in Lov-1 in the warm after 2, 4, 8, or 12 weeks of cold treatment. As in Figure 4B,C, but here replication is suspended in the warm. Simulated H3K27me3 levels are fraction of maximum possible occupancy in relevant region, each averaged over 4000 realisations.

Schematic of nucleation, spreading, and maintenance at FLC.

First, stochastic nucleation of an assembly occurs in the cold and permits the de-novo addition of nucleation region H3K27me3. On return to warm, a long-range looping interaction mediates spreading of H3K27me3 across the locus. Eventually, the assembly memory storage elements at the locus are lost, but H3K27me3 is in most cases maintained at the locus, but can eventually be lost if the transcriptional push is high enough (e.g. in Lov-1 variety).

Tables

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional information
Gene (Arabidopsis thaliana)FLCTAIRAT5G10140
Strain, strain background (Arabidopsis thaliana, Columbia-0)‘Col-FRI; Col-FRIsf2Yang et al., 2017 DOI:10.1126/science.aan1121Columbia-0 background with active FRI
Strain, strain background (Arabidopsis thaliana, Columbia-0)lhp1-3 FRIsf2
FLC-Venus
Yang et al., 2017 DOI: 10.1126/science.aan1121lhp1 mutant in Columbia-0 background with active FRI, a mutated FLC and complementing FLC-Venus transgene
Antibodyanti-H3 (rabbit polyclonal)AbcamCat# ab1791,
RRID:AB_302613
‘3 µg/IP’
Antibodyanti-H3K27me3 (rabbit, polyclonal)MilliporeCat# 07–449,
RRID:AB_310624
‘4 µg/IP’
Sequence-based reagent−86This studyMN-ChIP primersF: CACTCTCGTTTACCCCCAAA
R: TCCTTTTCTCGCTTTATTTCTTTC
Sequence-based reagent−49This study and Yang et al., 2017
DOI: 10.1126/science.aan1121
FLC_−49_F
MN-ChIP primersF: GCCCGACGAAGAAAAAGTAG
R: TTCAAGTCGCCGGAGATACT
Sequence-based reagent67This studyMN-ChIP primersF:
AGGATCAAATTAGGGCACAAA
R:
TCAATTCGCTTGATTTCTAGTTTTTT
Sequence-based reagent98This studyMN-ChIP primersF:
GAGAGAAGCCATGGGAAGAA
R:
AGCTGACGAGCTTTCTCTCGAT
Sequence-based reagent117This studyMN-ChIP primersF: AAAAAACTAGAAATCAAGCGAATTGA
R:
CTTTCTCGATGAGACCGTT
Sequence-based reagent348This studyMN-ChIP primersF: GTGCTCTTTTACTTTTCTGAG
R: AGAGATCCGCCGGAAAAA
Sequence-based reagent415This study and Yang et al., 2017
DOI:10.1126/science.aan1121
FLC_416_F
MN-ChIP primersF: GGCGGATCTCTTGTTGTTTC
R:
TTCTTCACGACATTGTTCTTCC
Sequence-based reagent581This studyMN-ChIP primersF: TGCATGGATTTCATTATTTCCT
R: TCACTCAACAACATCGAGCA
Sequence-based reagent651Yang et al., 2017 DOI: 10.1126/science.aan1121
FLC_652_F FLC_809_R
MN-ChIP primersF:
CGTGCTCGATGTTGTTGAGT
R: TCCCGTAAGTGCATTGCATA
Sequence-based reagent785This studyMN-ChIP primersF: TCATTGGATCTCTCGGATTTG
R:
AGGTCCACAGCAAAGATAGGAA
Sequence-based reagent923This studyMN-ChIP primersF:
TTCCTATCTTTGCTGTGGACCT
R:
GAATCGCAATCGATAACCAGA
Chemical compound, drugProtease inhibitor cocktail, cOmpleteSigma- AldrichCat#5056489001
Chemical compound, drugMicrococcal nucleaseTakara BioCat#2910A200 U/ml
Software, algorithmImageJhttps://imagej.nih.gov/ij/Fiji, RRID:SCR_002285
Software, algorithmImage analysis pipelineThis study
Yang et al., 2017
(with modifications)https://github.com/JIC-Image-Analysis/lhp1-analysis
OtherPropidium iodide stain (PI)Sigma-AldrichCat# P48642 µg/mL
Table 1
Definitions of the simulated system, the histones, and the protein assembly at FLC.
Definitions of the simulated system
TypeStatus
Protein assembly occupancy k[1,Np]Pk{unbound,bound}
Histones i[1,N]Si{me0,me1,me2,me3}
Regions of histonesRm={Histones[1,...,NNR]whenm=NR(nucleationregion)Histones[NNR+1,...,N]whenm=B(bodyregion),NB=NNNRHistones[1,...,N]whenm=L(entireregion),NL=N
Table 2
Definitions and functions used in the propensity functions in Table 3.
Definitions
The Kronecker deltaδx,y=1,x=y0,xy
Neighbour effectEi=jMi(ρme2δSj,me2+δSj,me3)
Neighbouring setMi={{i3,i2,i1,i+1,i+2},ieven,{i2,i1,i+1,i+2,i+3},iodd}
Fraction of histones in me3 or me2P(Rm)me2/me3=1NmjRm(δSj,me2+δSj,me3)
Sum of bound proteinsQ=j=1NpδPj,bound
Protein feedbackG={1,Qv,0,Q<v}
Spontaneous protein bindingγbind(w)=γVIN3w(1+ww0)w0 is timescale for VIN3 binding saturation (weeks)w is weeks in cold
Table 3
Propensity functions used in the simulations.
Propensities
 Non-assembly methylation propensitybime=β(δSi,me0(γme0-1+kme01Ei)+δSi,me1(γme1-2+kme12Ei)+δSi,me2(γme2-3+kme23Ei))
 Demethylation propensityridem=γdem(δSi,me1+δSi,me2+δSi,me3)
 Transcriptionf={α(fmaxP(Rm)me2/me3PT(fmaxfmin))αfmin,P(Rm)me2/me3PT,P(Rm)me2/me3<PT
m={NR,P(RNR)me2/me3>P(RB)me2/me3B,P(RNR)me2/me3P(RB)me2/me3
iff>flimthen f=flim
 Protein binding propensityIf Q<NP then rbind=γbind(w)+kbindGIf Q=NP then rbind=0
 Protein unbinding propensityrkunbind=δPk,boundγunbind
 Total methylation propensityrime={bime+kpmeG(δSi,me0+δSi,me1+δSi,me2),iRNRbime,iRB
Table 4
Specification of parameters for different conditions.

Post-cold, a time lag of 1 day without replication is first simulated with new, post-cold values of the parameters specified, with the exception of α, which is changed to its post-cold value after that time lag.

Pre-coldColdPost-cold
Cell cycle lengthcwcccw
Trans-acting gene activation (α)αwαcαw
PRC2-mediated methylation rate (kme) at i-th histonekme={kmehigh,iRNRkmehigh,iRB{kmehigh,iRNRkmelow,iRB{kmehigh,iRNRkmehigh,iRB
Spontaneous protein binding0γbindw0
Simulated number of cell cycles Ngen10.9NgenxNgeny
Table 5
Fixed parameters with references justifying values used.
ParameterDescriptionValueJustification
NNumber of histones60Berry et al., 2017a
kme0-1PRC2-mediated methylation rate (me0 to me1) (histone−1 s−1)9kme
kme1-2PRC2-mediated methylation rate (me1 to me2) (histone−1 s−1)6kme
kme2-3PRC2-mediated methylation rate (me2 to me3) (histone−1 s−1)kme
γme2-3Noisy methylation rate (me2 to me3)
(histone−1 s−1)
kme2-3/20
γme1-2Noisy methylation rate (me1 to me2)
(histone−1 s−1)
kme1-2/20
γme0-1Noisy methylation rate (me0 to me1)
(histone−1 s−1)
kme0-1/20
βRelative local PRC2-activity1
ρme2Relative activation of PRC2 by H3K27me20.1
γdemNoisy demethylation rate
(histone−1 s−1)
fminpdem
flimLimit on maximum transcription initiation (s−1)1/60
PTThreshold for full repression of transcription1/3Hepworth et al., 2018; Jadhav et al., 2020
fminMinimum transcription initiation rate (s−1)fmax/25Yang et al., 2014
fmaxMaximum transcription initiation rate (s−1)7.510-4Yang et al., 2014; Ietswaart et al., 2017
pdemDemethylation probability (histone−1 transcription−1)0.17Figure 2—figure supplement 2
pexHistone exchange probability (histone−1 transcription−1)8.310-2
cwCell cycle length in warm (22°C) conditions (h)22Rahni and Birnbaum, 2019; Zhao et al., 2020
ccCell cycle length in cold (5°C) conditions (h)154
Table 6
Free parameters and their values used in the simulations.
ParameterDescriptionValue
NNRNumber of histones in nucleation region6
NpMaximum number of proteins in assembly17
Ngenx
Ngeny
Number of cell cycles in cold conditions
Number of cell cycles post cold
8.7
43.6
kme
kmehigh
kmelow
PRC2-mediated methylation rate (me2 to me3)
High rate (histone−1 s−1)
Low rate (histone−1 s−1)
41051.7105
kp-meAssembly-mediated methylation rate (histone−1 s−1)0.01
γVIN3Noisy addition of protein to the assembly (s−1)1.810-4
w0Timescale for VIN3 binding saturation in weeks3
kbindProtein feedback rate (s−1)0.05
γunbindNoisy unbinding of proteins from the assembly (protein−1 s−1)10-3
vNumber of proteins in assembly required for protein feedback4
α = αw
α = αc
Trans-acting gene activation warm conditions
Trans-acting gene activation cold conditions
1
0.75
Table 7
Additional parameters and equations.

Pr in the definitions of POFF and PON refer to the proportion of time (see ‘Model parameters’ section), relevant for Figure 2—figure supplement 2.

ParameterDescriptionValue
NsimNumber of simulations for a gene4000
toutTime step for time course averaging (h)1
DefinitionReference
POFFPr(P(RL)me2/me3>3PT4)Berry et al., 2017a
PONPr(P(RL)me2/me3<PT4)
BiBi=4PONPOFFSneppen and Dodd, 2012; Berry et al., 2017a
Table 8
Additional details for simulations of PRE excision in Figure 4A.

Values for kmeex specified in figure.

Default kmeChange in simulationCondition
kme={kmehigh,iRNRkmehigh,iRBkme={0,iRNRkmeex,iRBSNR{me0}Pk{unbound}After 11 days in warm, post-cold.
Table 9
Summary of other figure specific parameters.
FigureParameterValue
1C, 2D, Figure 1—figure supplements 2 and 3FLC-ON cell
γme2-3
γme1-2
γme0-1
P(RNR)me2/me3<PT
kme2-3/100
kme1-2/100
kme0-1/100
1C, Figure 1—figure supplement 3γVIN3
PT
kme,forhistoneiRB
0 (s−1)
1/6
0 (histone−1 s−1)
Figure 1—figure supplement 2AγVIN3
PT
kme,forhistoneiRB
0 (s−1)
1/3
0 (histone−1 s−1)
Figure 1—figure supplement 2BγVIN3
PT
kme,forhistoneiRB
Regular nucleosome distribution to daughter strands
0 (s−1)
1/6
0 (histone−1 s−1)
N/A
Figure 1—figure supplement 2CγVIN3
PT
kme,forhistoneiRB
Regular nucleosome distribution to daughter strands
γunbind
0 (s−1)
1/3
0 (histone−1 s−1)
N/A
810-4(protein−1 s−1)
2CPT
kme,forhistoneiRB
1/3
1.710-5 (histone−1 s−1)
2DγVIN3
PT
kme,forhistoneiRB
0 (s−1)
1/3
0 (histone−1 s−1)
4 B,CγVIN3
Ngenx
αw
kmelow
2.110-4(s−1)
13.1
2
1.710-5(histone−1 s−1)
Figure 4—figure supplement 1As 4B,C but no replication in warm after coldN/A
Table 10
Primers used to define nucleation and body region in ChIP data for Lov-1 (Qüesta et al., 2020) and ColFRI (Yang et al., 2014).
Lov-1Col-FRI
Nucleation region157_F 314_R
416_F 502_R
157_F 314_R
416_F 502_R
Body2465_F 2560_R
3197_F 3333_R
4322_F 4469_R
5139_F 5244_R
1933_F 2171_R
2465_F 2560_R
3197_F 3333_R
3998_F 4178_R
4322_F 4469_R
5139_F 5244_R
Table 11
Summary of experimental details for lhp1 imaging and MN-ChIP.
lhp1 imagingMN-ChIP
Plant materialFRI lhp1-3 FLC-VenusColFRI
Growth mediaMS-GLU (MS without glucose)MS-GLU (MS without glucose)
PlatesVerticalHorizontal
Pre-growth7 days14 days
Table 12
Summary statistics for quantitative image analysis of lhp1.
Days after cold treatmentNumber of roots imagedNumber of cells quantified
T7111696
T10173598
T14102648

Additional files

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)