H3K27me3 demethylases alter HSP22 and HSP17.6C expression in response to recurring heat in Arabidopsis

Acclimation to high temperature increases plants’ tolerance of subsequent lethal high temperatures. Although epigenetic regulation of plant gene expression is well studied, how plants maintain a memory of environmental changes over time remains unclear. Here, we show that JUMONJI (JMJ) proteins, demethylases involved in histone H3 lysine 27 trimethylation (H3K27me3), are necessary for Arabidopsis thaliana heat acclimation. Acclimation induces sustained H3K27me3 demethylation at HEAT SHOCK PROTEIN22 (HSP22) and HSP17.6C loci by JMJs, poising the HSP genes for subsequent activation. Upon sensing heat after a 3-day interval, JMJs directly reactivate these HSP genes. Finally, jmj mutants fail to maintain heat memory under fluctuating field temperature conditions. Our findings of an epigenetic memory mechanism involving histone demethylases may have implications for environmental adaptation of field plants.

in the wild type and jmjq mutants. a, H3 peaks detected by ChIP-seq at the HSP22 and HSP17.6C loci in the wild type and jmjq mutants without acclimation (above) and 3 days (72 h) after acclimation (below). b and c, Histone H3 levels at HSP22 (b), HSP17.6C (c), and TA3 (d) determined by ChIP-qPCR in the wild type and jmjq mutants. Grey jitter dots represent expression level from each sample. No difference in histone H3 signals at those two loci was observed between the wild type and jmjq mutants by a two-tailed Student's t-test. the post-hoc test that followed. NS, nonsignificant. n > 78. Although fresh weight in acclimaized hsp22 hsp17.6c double mutants was significantly lighter than that in acclimaized wild type after heat shock, no difference in survival rate was observed. The lack of difference in survival rate of hsp22 hsp17.6c double mutants, while there is a difference in jmjq mutants, suggests that other differentially expressed genes, such as HSP21 may also contribute to phenotypic consequence of jmjq mutants for heat acclimation. before heat shock (right). Plants were grown under +ACC +HS condition. Those two panels are both negative controls that show that the b-estradiol treatment itself did not trigger phenotypic rescue. c, Quantification of survival rate shown in Fig. 4b and Supplementary Fig. 26a and b. Ten-day-old seedlings grown under +ACC +HS condition with b-estradiol application before acclimation and before heat shock were categorized into three groups based on phenotypic severity: green, normal growth; light green, partially damaged; white, perished.
Significance was determined by c 2 test. n > 199.

Definition of transition rates between three states of histone modification
We developed a mathematical model describing state transitions of histone modification at the cell-population level to predict the level of HSP expression under changing temperature.
We focus on a locus with units of nucleosomes. Each nucleosome is in one of the following three states, actively modified (A), unmodified (U), and repressively modified (R) (Fig.   3g), as assumed in a previous study 25,26 . Similarly, $→& and %→$ are given as increasing functions of the number of R ( ) as follows: where !→# and !→# are defined similarly as in Eq. (1).

Dynamics of histone modification at the cell-population level
Using the transition rate functions defined in Eqs. (1) and (2) It can be mathematically shown that the above system has a unique and globally stable equilibrium under a constant temperature. We define the equilibrium state at = 22 (°C) as where is a positive constant. In the following analyses, we substitute = 1̂-,0 ⁄ so that (0) = 1. Although we do not observe cell death, a majority of cells are damaged after heat shock treatment ( Supplementary Fig. 9). Thus, we assume that the fraction ℎ of cells stop transcription of HSP genes right after the heat shock at = HS and gradually recover their transcription activity at a constant rate : ( ) = j ⋅ -,0 ( ) for ≤ () ⋅ b1 − ℎ ⋅ .7⋅(6.6 HS ) c ⋅ -,0 ( ) for > HS . (Eq. 9) Because the ORF length of HSP22 is 588 bp, we choose = 3 for the following analyses. In our formalization, we assume that the order of histone modification at different nucleosomes is not random, but there is a specific order (e.g. a nucleosome closest to the transcription start site is the first to be repressively modified.). For parameter fitting using wild type, the sum of squared errors between log-transformed experimental and simulation data was considered as a cost function to be minimized. The present model has as many as 12 free parameters, in which case it is generally difficult to specify the single optimal solution. Therefore, we used 144 different sets of initial parameter values that were randomly chosen from fixed ranges for each parameter (Supplementary Table 1) and obtained the best fit set of parameters (Supplementary Table 2). For parameter fitting, time unit is set to an hour. We also performed parameter fitting using experimental data from jmjq mutant. We chose the wild-type best-fit parameters as initial values for this analysis on the basis that a mutant would not be drastically different from wild type. Comparison of best fit parameters between wild type and jmjq mutant showed that the jmjq mutant has notably greater values for the strength of positive feedback for the transition from A to U ( %→$ ) and temperature dependence during heat shock ( () ). On the contrary, the jmjq mutant has smaller values for recovery rate of transcription activity after heat damage ( ) and the basic transition rate from R to U ( &→$ ). The greater positive feedback effect