Dot1 histone methyltransferases share a distributive mechanism but have highly diverged catalytic properties

The conserved histone methyltransferase Dot1 establishes an H3K79 methylation pattern consisting of mono-, di- and trimethylation states on histone H3 via a distributive mechanism. This mechanism has been shown to be important for the regulation of the different H3K79 methylation states in yeast. Dot1 enzymes in yeast, Trypanosoma brucei (TbDot1A and TbDot1B, which methylate H3K76) and human (hDot1L) generate very divergent methylation patterns. To understand how these species-specific methylation patterns are generated, the methylation output of the Dot1 enzymes was compared by expressing them in yeast at various expression levels. Computational simulations based on these data showed that the Dot1 enzymes have highly distinct catalytic properties, but share a distributive mechanism. The mechanism of methylation and the distinct rate constants have implications for the regulation of H3K79/K76 methylation. A mathematical model of H3K76 methylation during the trypanosome cell cycle suggests that temporally-regulated consecutive action of TbDot1A and TbDot1B is required for the observed regulation of H3K76 methylation states.


Supplemental Figure S2
Quantified western-blot data using an Odyssey scanner of at least two replicates of the yeaststrain series shown in Figure 1C to determine the linearity of the H3K79me1, -me2 and -me3 home-made antibodies 1 and the H3K79me2 antibody from Millipore. Western-blot analysis of H3K79 methylation states of TbDot1B expressed from a galactoseinducible GAL1 promoter (G1) or the yDOT1 promoter (D) at the yDOT1 locus in S. cerevisiae. TbDot1B copy-number expression was determined by detection of the c-terminal 9xMyc tag and comparison to yDot1-9xMyc expression (not shown). H3 and Pgk1 were used as loading controls. H3K79 methylation states were compared to a wild-type yeast strain (WT), and a noninduced stain (G1, 0%) was included to determine antibody specificity. TbDot1B expression from the GAL1pr using 1% galactose (and 1% glucose) in the medium resulted in an H3K79 methylation pattern in between that of the H3K79 methylation patterns generated when TbDot1B was expressed from the DOT1pr (D) or the TEFpr + yDOT1 5'UTR (T*). H3K79 methylation patterns of TbDot1B induced by 1% galactose are shown in Figure S3 as the second data points on the x-axis. Figure S3. Quantified H3K79 methylation patterns of yeast strains expressing yDot1, TbDot1A, TbDot1B and hDot1L. H3K79 methylation patterns shown in Figures 1, 2, and S2 were quantified using the validated antibodies described in Figure 1C and 1D or measured by MS. A data quality check was performed as described in the methods and non-reproducible H3K79 methylation patterns were removed from the dataset. To visualize the correlation between Dot1 expression and H3K79 methylation patterns, the H3K79 methylation states were plotted against the Dot1 expression. Data points indicated by  represent H3K79 methylation states measured by mass spectrometry.

Figure S4. Simulations for hDot1L
A) Simulations of the H3K79 methylation pattern with dot1Δ, intermediate (single integration) hDot1L expression, and high hDot1L expression (2µ plasmid) in minimal medium using 2% galactose. Simulations (second panel) could not fit the experimental data (first panel) properly.
Likely, the estimated copy number is not representative for the active hDot1L fraction in the cell since the estimated enzyme concentration is ~5 fold higher than the substrate (H3) concentration. Fortunately, the H3K79 methylation patterns contained sufficient information to simultaneously fit the methylation rate constants for hDot1L and the hDot1L copy number of the multi-copy plasmid with the integrated hDot1L copy number as a reference. In the absence of any observed H3K79me3, the k 2 parameter was set to zero for numeric stability. This approach resulted in an improved fit to the experimental data (third panel) and yielded an estimate of the (effective) hDot1L copy number in the multi-copy plasmid expressing strain of about 20 fold lower than the estimate based on the western blot. The distributive model performed better than a processive model with the same adjustments (fourth panel).

H3K79 methylation pattern analysis
For analysis of H3K79 methylation and Dot1 protein expression cells were harvested in mid-log phase. Proteins were extracted using the SUMEB protocol 5 . H3K79 methylation signals were detected using home-made antibodies against H3K79me1, -me2 and -me3 1 and H3K79me2 (04-835, Millipore), while antibodies for histone H3 1 and Pgk1 (A-6457, Invitrogen) were used as loading controls. To determine Dot1 copy numbers TbDot1A and TbDot1B were C-terminally 9xMyc-tagged as described above and expression levels were compared to yDot1-9xMyc expression levels using a Myc antibody (sc-789, Santa Cruz). H3K79 methylation patterns were identical for TbDot1A and -B with and without 9xMyc tag as expected 6 . yDot1 expression levels were detected using a yDot1 home-made antibody 7 . hDot1L copy numbers in yeast were estimated using the 9xMyc tag and 3xFLAG tag detected with Flag-M2 (F3165, Sigma). As the full length hDot1L enzyme was not more active than the shorter variant (lacking the nonconserved C-terminal domain), the shorter hDot1L 1-430 variant was used in this study (results not shown and 8 ). Quantitative western blot analysis was performed using the LI-COR Odyssey IRDye® IR imager (Biosciences) and the Odyssey LI-COR software. For details about the normalization of the H3K79 methylation signals based on the linearity of the antibodies, see below. Histone H3 purification for mass spectrometry and determination of growth rates was performed as described previously 1 . Data were obtained of at least two replicates.

H3K79 methylation signal normalization and antibody validation
To determine H3 antibody specificity, H3 was detected in a dilution series of a yeast protein extract. This revealed that our H3 antibody 1 does not have a linear detection range but showed the relationship y=x 0.5 , in which x is the measured H3 signal and y the corrected signal. The relationship was used to normalize the H3 immunoblot signals on each blot. H3K79 methylation signals were normalized over H3. To validate the linearity of the H3K79 methylation antibodies, quantitative western blot signals were plotted against absolute H3K79 methylation levels determined by mass spectrometry (Figure 1C-D). In GraphPad Prism 6.0 software, a non-linear regression fit was applied using the option 'straight line'. The correlation between western blot and MS data for the H3K79me1 and -me3 home-made antibodies and the H3K79me2 antibody from Millipore was good based on the R squares, and these antibodies were used for the analysis described in this study. The H3K79 methylation signals were normalized using the polynomial function that resulted from the correlation analysis (H3K79me1 y=0.084x; H3K79me2 (MP) y=0.064x; H3K79me3 y=0.016x+0.061 in which x and y are respectively the detected and normalized H3K79 methylation signals). To detect the H3K79me2 profile generated by hDot1L, the H3K79me2 homo made antibody was used as the non-linearity is caused by cross-reactivity with H3K79me3, which is not generated by hDot1L.

H3K79 methylation, Dot1 expression and growth rates for simulation.
For computational simulations, the H3K79 methylation data from quantitative western-blots (Table SIIIB) were replaced by MS data (Table SIIIC) when available, to generate a high quality quantitative dataset (Table SIIID). Since western-blot data are semi-quantitative, we put a few constrains to the data to select only the most reliable data for modeling purposes (Table SIIIA).
First, negative values for H3K79 methylation were set to zero (H3K79me0 drops as a consequence). Second, a quality check was performed to determine whether individual strains behaved as outliers. Therefore, the total methyl number (TMN) was calculated (maximum is 100% H3K79me3, which would be TMN=300). Per Dot1 enzyme, strains were ranked in order of increasing Dot1 copy number. When the TMN dropped >10% compared to the preceding strain or when TMN>330, strains were considered as outliers and removed from the dataset (Table  SIIIA). Growth rates of strains expressing low and high levels of each Dot1 protein were determined by cell counting. Growth rates of strains expressing intermediate levels of each Dot1 protein were estimated based on relative colony growth of the strains compared to strains with low ahd high expression of the respective Dot1 protein.

Computational simulation of H3K79 methylation patterns.
We made use of a kinetic model reported previously to predict experimental H3K79 methylation data for the distributive yeast Dot1 enzyme from S. cerevisiae 9 . The model described a basic distributive enzyme mechanism in combination with constant histone production (  ) and first order growth dilution of all histones (specific growth rate  ) as a result of the cell volume increase 10,11 . The state variables corresponding to the nuclear concentrations of unmethylated H3K79 (H3K79me0), mono-(H3K79me1), di-(H3K79me2) and trimethylated (H3K79me3) were described by differential equations, respectively: with sequential methylation steps represented by mass-action kinetics.
The proportionality with the Dot1 concentration ('dot' in the equations) is in accordance with the assumption that the enzyme (Dot1) levels are significantly smaller than the protein substrate H3 levels.
The Dot1 concentration is determined by the estimations from immunoblot experiments. As previous demonstrated 9 , more complicated enzyme kinetic equations (describing substrate saturation or competition or product inhibition) have typically too many parameters to allow a unique fit to the available experimental data. Moreover, the Dot1 substrate Sadenosylmethionine (AdoMet) was assumed to be constant. The kinetic constants in this study therefore represent apparent rate constants. The constant  is set by the steady state assumption that histone production equals loss by growth dilution, i.e..  =  *(me0 + me1 + me2 + me3) =  *H tot . This assumption disregards the possible involvement of replication-independent histone turn-over in yeast [12][13][14][15] and may lead to an underestimation of the histone production and the rate constants. Whereas further model extensions may be possible in the future, the experimental validation in this study and De Vos et al. 9 have nevertheless affirmed the predictive power of our current approach. To evaluate whether the methylation data agree more strongly with either a distributive or a processive enzyme mechanism the following basic model was used (see also 9 ). It describes a processive enzyme mechanism with the unmethylated histone as only substrate and the subsequent methylated forms as (leaky) intermediates or products and was defined as follows: In this mechanism the unmethylated H3K79 yields either the trimethylated form, without release from the enzyme, or leads to mono-or di-methylated H3K79 intermediates through irreversible dissociation from Dot1 (hence it leakiness). Like for the distributive model the enzyme activity is represented by three mass-action kinetic equations ( 0 v 's) with corresponding rate constants ( 0 k 's). The other parameters are equivalent between the models enabling straight-forward comparison and justifying our conclusion that the processive model was generally not as successful in simulating the experimental data for all enzymes analyzed.
The comparative test to estimate the rate constants consisted of a parameter estimation by ordinary least squares optimization using non-linear regression according to the procedure described 9 . The input from experiments consisted of the fractions of the different methylation states for different growth rates and Dot1 concentrations (Table SIII) and the simulation output determining the objective function to be minimized consisted of the predicted unique steadystate methylation fractions. The values for the Dot1 concentration (dot), the specific growth rate (), and the histone synthesis rate (, depending on H tot as explained above) were considered to be fixed in this procedure. The amounts of histone H3 was taken to be 200,000 molecules/cell 16 . These values were converted into concentrations using an approximate cell nucleus volume of 2.9 μm 3 in wild type cells grown in glucose media 17 . Briefly, numerical integration runs via the NDSolve function of Mathematica TM (7.0) (http://www.wolfram.com/) were used to evaluate the least squares objective function as a function of the Dot1 kinetic constants with the NMinimize function with the 'Automatic' search method within realistic parameter bounds (on a MacBook with Intel Core i7 CPU). The sum of the squares (SS) is a direct measure of the accuracy of the fit for a specific dataset (as also can be visually assessed via the corresponding prediction of the methylation states). Moreover, the influence of data uncertainty on the resulting parameter estimates was investigated via a Monte Carlo-type approach with multiple estimation runs (until convergence of the mean parameter estimates) with artificial Gaussian noise (amounting to a mean coefficient of variation of 10 % superposed on the data, yielding parameter error estimates (Table SIV). First, the distributive model was evaluated against an extended dataset from yeast (Table SIIIE) with resulted in parameter estimates very close to the original ones (Table I). The improved parameter values will submitted to the JWS database (jjj.biochem.sun.ac.za/database/) upon publication to update the deposited model. Then the presented distributive and processive models were used to simulate the H3K79 methylation patterns determined by western blot and mass spectrometry of TbDot1A, TbDot1B and hDot1L. For hDot1L, in the absence of any significant trimethylated fraction, the trimethylation reaction rate constant could not be determined and was set to zero. The experimental estimate of the hDot1L for the highest copy number data point (Fig. S4, middle panel) led to an underestimation of the methylation fractions of the second data point. This indicated that the very high copy number did not represent an equivalent enzyme activity level (see Results). Simultaneously estimating the copy number for the third data point resolved this problem (Fig. S4). The distributive model successfully fitted the experimental data of yDot1, TbDot1A, TbDot1B and hDot1L.

Single cell models for methylation dynamics during the cell cycle
A model was constructed to describe dynamic changes in histone methylation during the Trypanosoma cell cycle. This model accounts for one cell cycle of 13 hours 2 in which the G2/M, S and G1 phases take up respectively 40%, 20% and 40% of the time 3 . Histone synthesis was assumed to take place strictly in S phase 15 , whereas growth dilution is assumed to be continuous 10,11 . The value of the production rate () was set such that the value averaged over the complete cell cycle equals the growth dilution rate. No explicit cell division was taken into account assuming that this does not affect the nuclear concentrations. Starting from a random set of initial conditions the model was simulated and the final methylation states replaced the initial values until starting and final states matched close enough (within 0.1% methylation) in order to represent a stable cycle. The values from the parameter estimation runs were used for the reaction rate constants (Table I). To establish a model for H3K76 methylation during the trypanosome cell cycle, TbDot1A and -B expression levels and activity window were manually fit to recapitulate the published H3K76 methylation pattern in trypanosomes (see below). The total histone concentration was taken to be 840,000 molecules/cell (based on dividing the (diploid) genome of 70 Mb by 166 bp/nucleosome or 170 µM for an approximate nuclear volume of 8.