Here you can find out about some of the latest happenings

To more comprehensively understand the ensemble of interactions, we integrated our data into a network of biochemical reactions to calculate a dynamic model based on ordinary differential equations (fig. S4). The model was parameterized on the basis of the measured reaction constants from this work and data from previous literature, with additional kinetic parameters approximated based on diffusion limits (table S2). Simulations of the model indicated that the expression of both RNA and Cyp33 is needed to switch the chromatin to a “repressive” state. This is indicated by an increase of free and RNA-bound H3K4me3 tail (in red) and a strong decrease (in blue) of MLL1-PHD3 bound to the H3 tail (Fig. 5B). Presence of RNA or Cyp33 alone is insufficient to cause such rebalancing (fig. S4). This rebalancing upon simultaneous presence of Cyp33 and RNA occurs even with a conservative modeling assumption that RNA contains only a single H3K4me3 binding site. Furthermore, the slight dominance of the repressive state of the system (H3K4me3 tail not bound to MLL1-PHD3) remains even if RNA-H3K4me3 affinity is assumed very weak (500 μM) (fig. S4F). The last two points suggest that the switch of this system to the repressive state can robustly tolerate certain changes in key reaction constants of the target network in the cell nucleus environment.

In term of structural plasticity, the three positions adopted by the conserved α helix downstream of the RRM in free Cyp33, bound to RNA, and bound to MLL1 PHD3 is unprecedented (Fig. 4A). Change of position upon RNA binding of a C-terminal helix or folding upon RNA binding has previously been reported for U1A N-terminal RRM and in Polypyrimidine Tract Binding Protein 1 (PTBP1) RRM1, respectively (39, 40). In both cases, the RNA induced repositioning or folding of the C-terminal helix is functionally important, as it allows U1A dimerization and stem-loop recognition, respectively (39, 40). Here, the three-helix positions found in Cyp33 allow interactions with both RNA and a protein, resulting in a cascade of binding events (Fig. 5). RNA binding to Cyp33 triggers a first switch of the helix that facilitates binding to MLL1 PHD3-H3K4me3. Then, a second positional switch triggers the release of both the RNA and the histone mark from the MLL1-Cyp33 complex. A third switch leads to the dissociation of the proteins. Overall, Cyp33 senses the presence of RNA and transduces this signal toward a chromatin structure change (Fig. 5).

Although the structure of MLL1 PHD3 and Cyp33 had been solved and their interaction had been studied (4, 13, 28), how Cyp33 binding to MLL1 leads to transcription repression remained a mystery, and contradictory mechanistic models have emerged. Notably, past structural works studied the RRM in isolation ignoring the evolutionary conserved C-terminal region of Cyp33 RRM. Our structural work revealed that this conserved region folds into an α helix that is critical to the function and the mechanism of action of Cyp33. By combining previous structural work with the five structures presented here, we can now propose a full mechanistic path explaining how Cyp33 when stimulated by RNA binding could change the chromatin from a transcriptionally active state to a repressive state (Fig. 5).

In the initial transcriptionally active state, MLL1 is bound near the transcription start sites of HOX genes. MLL1 binds H3K4me3 via its PHD3 domain and maintains a high level of this modification via its catalytic SET domain (Fig. 5). In the active state, the MLL1 PHD3 and BRD domains interact and are tightly bound to H3K4me3 (Kd of 4 μM) (13). Several lincRNAs are expressed in the vicinity of the HOX genes and regulate their expression (41). Among those, we show that NC3 and NC4 are bound by Cyp33 sequence specifically (Fig. 2) probably due to the presence of multiple copies of the YAAUNY RNA binding consensus sequence, which is an optimal binding sequence for the RRM of Cyp33. Although the binding affinity of Cyp33 for a single RNA motif is weak (Kd of 300 μM; table S1), the affinity is increased by avidity due to the presence of multiple copies of this motif. The lincRNAs NC3 and NC4 could therefore recruit Cyp33 to the site of transcription of the HOX gene and in proximity to MLL1 using a very sophisticated mode of regulation. The interaction of the α3 helix with the RNA binding interface of Cyp33 in its free form prevents a premature recruitment of the protein at the transcription site. A minimal amount of transcribed lincRNA will be needed to compete out the α3 helix from the β sheet surface and initiate the repressive mode of action of Cyp33. The Cyp33-MLL1 interaction happens then in two steps. First, Cyp33 PPIase binds to MLL1 and induces the isomerization of the Pro1629 from cis to trans, which weakens the interaction between the BRD domain and the PHD3 but still maintains the H3K4me3 bound to the PHD domain (Fig. 5A, steps 1 and 2) (13). The interaction with RNA also translocates the Cyp33 α3 helix on the side of the RRM (toward β4; Fig. 4A, middle), preparing the β sheet surface for subsequent interaction with MLL1 PHD3. Now, both the PHD3 domain and Cyp33 RRM are in a conformation that is optimal for them to interact. Although RNA is bound to the Cyp33 RRM, the affinity of PHD3 for the RRM is much stronger (60-fold, i.e., Kd of 5 versus 300 μM). So contrary to what was anticipated, RNA binding does not inhibit Cyp33-MLL1 interaction but would rather stimulate it by recruiting Cyp33 to the site of transcription and by repositioning the α3 helix to facilitate its binding to MLL1. Our data suggest that the complex in step 3 is only transient (Fig. 5). The dissociation constant of H3K4me3 from the MLL1/Cyp33 RRM complex is higher (Kd of 70 μM) than for MLL1/Cyp33 RRMΔα3 (Kd of 24 μM) due to the interaction of Cyp33 α3 with the α helix of MLL1 PHD3 (Fig. 3C). Therefore, the interaction of Cyp33 α3 with the MLL1 PHD3 results in a squeezing of the histone binding pocket and dissociation of H3K4me3 (Fig. 5A, step 4). We proved this step experimentally when mixing at a stoichiometric ratio Cyp33/RNA with MLL1 PHD3/H3K4me3, as it resulted in the formation of a Cyp33/MLL1 PHD3 complex and the release of both the RNA and H3K4me3 (fig. S3). The fact that the RNA and H3K4me3 interact, further pushes the equilibrium toward almost full dissociation of the histone mark from MLL1 (Fig. 5). This interaction between the RNA and the histone tail is further supported by recent publications, indicating that the nucleosome histone tails, and in particular H3K4, do interact with RNA (42, 43). The H3K4me3 mark is now accessible to histone demethylases and histone deacetylases, ultimately leading to a repressive transcriptional state of the chromatin (Fig. 5, step 4). With the decrease in RNA concentration, Cyp33 and MLL1 should ultimately dissociate via their intramolecular interactions (between α3 and the RRM in Cyp33 and between the PHD3 and the BRD domains in MLL1).

In summary, the proposed mechanistic path derived from our structural and biochemical work now explains why Cyp33 can repress MLL1-derived transcription and how this is triggered by RNA binding of Cyp33 (most probably lincRNAs). Our results reveal a very sophisticated mechanism of negative feedback regulation of transcription mediated by Cyp33, RNA, and MLL1 (Fig. 5).

We propose here that MLL1 could promote the transcription of lincRNAs NC3 and NC4, which are bound by Cyp33 RRM when the expression level of these RNAs increases. This interaction induces a change in position of Cyp33 α3-helix that allows its interaction with MLL1, forces the release of the protein from H3K4me3 and represses transcription by negative feedback (Fig. 5). The involvement of MLL1 in inducing transcription of lincRNAs is not an isolated case. MLL1 has a key role in inducing transcription of the lincRNA HOTAIR under hypoxia in several types of cancer cells (44). Previous studies reported interactions of MLL1 associated to other proteins with lincRNAs. For example, Fendrr lincRNA can interact with the TrxG/MLL complex and was shown to form a double-stranded DNA/RNA triplex, allowing the recruitment of the polycomb repressive complex 2 and subsequent H3K27 trimethylation, a repressive histone mark, at specific target sites (45). Conversely, the lincRNA HoxBlinc was shown to recruit the Setd1a/MLL1 complex to activate transcription of HoxB genes (46). In addition, it was shown that a chromosomal looping could bring the WDR5/MLL complex across the HOXA gene to promote gene transcription (41). On the basis of a protein mutant that affects the interaction of WDR5 with RNA but not with the MLL complex, it was proposed that lincRNAs could bind to WDR5 to stabilize its interaction on chromatin, which facilitates the subsequent assembly of the MLL complex and gene activation (47). All these data suggest that lincRNAs can regulate gene expression by attracting positive or negative epigenetic regulators to specific chromatin sites bound by MLL1. These other regulatory pathways use other parts of MLL1 than the PHD3 domain. Although the mode of binding of these large regulatory complexes bound to a single nucleosome were recently solved by cryo–electron microscopy (48, 49), the mode of action of RNA in these regulatory processes remain elusive. Last, the mode of action of Cyp33 in transcription repression resembles the mode of action of RNA Binding Fox-1 (RBFOX-1) (50) and of the RNA-induced silencing complex (RISC) in transcription regulation (51) except that Cyp33 recruitment to chromatin results in the release of MLL1, while both RBFOX-1 and RISC lead to the recruitment of a repressor complex.

In summary, we structurally and functionally addressed open questions on the mechanism of Cyp33 regulated and MLL1-mediated gene expression. Initially, it was not clear how RNA binding results in a cross-talk between the two domains of Cyp33. Furthermore, a major contradiction between two models built on the existence or not of a ternary complex among Cyp33, MLL1, and the histone H3 was existing in literature. Namely, whether MLL1 remains bound to H3K4me3 in its repressive state and Cyp33 provokes through an unknown mechanism the recruitment of co-repressors (13) or whether binding of Cyp33 to MLL1 results in histone H3 dissociation with a concomitant repression (28). Our results revealed that the RRM domain of Cyp33 has a C-terminal third α helix that plays a central role in the regulation of MLL1-mediated gene expression by Cyp33. In addition to be the molecular sensor of RNA binding to the RRM, α3 helix allosterically dictates Cyp33 interaction with MLL1 and forces the protein to be released from the specific activation marks in the histone H3 leaving them exposed for epigenetic erasers. The enigmatic role of RNA in this process seems to play a more critical function than initially anticipated. Our data indicate that RNA could potentially not only recruit Cyp33 but also help to release MLL1 from H3K4me3 by interacting with the histone tails. It opens unexpected perspectives on RNA-mediated gene regulation, which can now be investigated further in cells. Leukemogenic MLL1 variants lack the entire homeobox and all adjacent domains including the writer domain SET (22). In agreement with our conclusions, it was shown that the sole reinsertion of PHD3 restores Cyp33 recruitment and rescues the aberrant transcription caused by MLL1 oncogenic fusion proteins (26, 27). Hence, its involvement in this interaction network makes Cyp33 a key player for the understanding of the oncogenic nature of MLL1 in infant leukemia in particular and potentially the mechanism of leukemogenesis in general.

Vectors encoding the Cyp33 RRM∆α and Cyp33 RRM constructs were transformed into chemical competent BL21-CodonPlus (RIL), and the vector encoding the codon-optimized construct for MLL1 PHD3 was transformed into chemical competent BL21-(DE3) Escherichia coli cells. All proteins were expressed using the IMPACT (Intein-Mediated Purification with an Affinity Chitin-binding Tag) expression system. Expression was performed in either LB for unlabeled protein or M9 minimal medium enriched with 13C-glucose and/or 15NH4Cl for 15N and 13C or only 15N labeling schemes.

The cells were grown at 37°C until the optical density at 600 nm (OD600) reached 0.8 and were then induced with 0.5 mM isopropyl-β-d-thiogalactopyranoside at 20°C and incubated for another 24 hours. The cells were harvested, centrifuged, and resuspended in 30 ml of lysis buffer [30 mM Hepes and 0.5 M NaCl (pH 8.0)] and 3 μl of 1 M phenylmethylsulfonyl fluoride protease inhibitor. This cell suspensions were lysed using a M110S homogenizer of Microfluidics and purified on chitin beads (New England Biolabs) by washing with lysis buffer, high salt buffer [30 mM Hepes and 2 M NaCl (pH 8.0)], and again lysis buffer. In case of purification of one of the Cyp33 variants, intein autocleavage was induced with lysis buffer containing 50 mM dithiothreitol (DTT) followed by at least 12 hours of incubation at room temperature and subsequent elution using twice 20 ml of NMR buffer [40 mM KCl and 20 mM KH2PO4 (pH 7.0)]. Cleavage of MLL1 PHD3 protein could not be achieved by intein autocleavage, because DTT at high concentrations has the tendency to complex Zn2+ ions. Instead, 1 mg of sequence specific protease from Tobacco Etch Virus (TEV) protease was added onto the chitin column and incubated overnight at room temperature. Except the addition of 10 μM ZnCl to the NMR buffer, the same elution protocol was applied as described for the Cyp33 variants. The NMR buffer was selected using differential scanning fluorimetry. In this method, the melting temperature of a protein is tested in presence of a variety of buffer conditions (96 conditions) and a dye with affinity for the hydrophobic parts of the protein by monitoring fluorescence-based thermal shifts (52). The conditions that were selected in the end are both keeping the protein stable and suitable for solution NMR experiments.

The elution products were concentrated to a volume of less than 1 ml using Vivaspin 2 centrifugal concentrators with a 10-kDa cutoff for the Cyp33 variants or a 5-kDa cutoff for the MLL1 PHD3 protein and further purified by size exclusion chromatography using a Superdex 75 10/300-GL column in according NMR buffer. For Cyp33 variants to be used for addition of RNA, 10 μl of SUPERase•In Ribonuclease Inhibitor (Ambion) were supplemented to the sample before applying it to the column.

NMR spectra were acquired at 303.15 K for the free Cyp33 RRM domain and at 310.15 K for all complexes. Triple-resonance experiments [experiment correlating atom names HN and CA (HNCA) and CBCAcoNH] for backbone assignment and three-dimensional (3D) Total Correlation SpectroscopY (TOCSY) experiments (hCccoNH and HcccoNH) for side-chain assignments (53) were collected at 500, 600, or 700 MHz using Bruker Avance III spectrometers equipped with TCI cryoprobes. Spectra dedicated for RNA resonance and nOe assignment were collected using samples in 100% D2O and 90% H2O/10% D2O for all other purposes. Homonuclear 2D, 15N- and 13C-edited 3D Nuclear Overhauser and Exchange Spectroscopy (NOESY) experiments for structure calculation and assignments of aromatic residues were all acquired on a Bruker Avance III HD 900 spectrometer equipped with a TCI cryo probe (TM = 120 ms). All spectra were processed with TopSpin 3.0 and analyzed with Sparky 3.1.1.4.

RNA resonance assignment was achieved using [1H-13C]-HSQCs (Heteronuclear Single Quantum Coherence) using the natural abundance of the 13C isotope, homonuclear 2D TOCSY (spin_lock = 50 ms), 2D NOESY (TM = 120 ms), and ω2-filtered 2D NOESY (sample with 13C-labeled protein). Assignments of nOes were based on the manual analysis of ω3-filtered 13C-resolved 3D NOESY (54) and homonuclear 2D NOESY experiments for the protein RNA complex with 13C-labeled protein and based on automated analysis (as described in next section) of standard 15N- or 13C-resolved 3D NOESY experiments for the protein-protein complex with one, the other, or both protein components, 13C and 15N labeled.

Backbone 15N-[1H] heteronuclear nOes were measured on a Bruker Avance III 750 spectrometer equipped with a TCI room temperature probe at a transmitter frequency of 750.134 MHz for the proton and 76.019 MHz for 15 N (55). For the backbone 15N-[1H]-nOe and for the reference experiment, a relaxation delay of 2 s and a water gate solvent suppression was used.

Regarding the structure calculation, initial peak picking and nOe assignments was performed using the ATNOS/CANDID package. For NOESYs of nonuniformly labeled samples, e.g., only one component was labeled, ATNOS/CANDID had to be aborted after completing the first cycle, and the consolidated shift list concatenated to the according NOESY had to be modified. Hence, if ATNOS/CANDID performed the peak picking of an 13C-resolved 3D NOESY where only MLL1 PHD3 was 13C labeled but not Cyp33-RRM, then shift assignments of all given shift lists were consolidated. Thus, the respective shift list contains 13C shifts of MLL1 PHD3 and Cyp33 RRM. This led to the problem that peaks were potentially interpreted wrong, caused by picked artifacts. To overcome this problem, the according shift list in cycle one was manually modified by deleting all resonance assignments, which were not detected by manual inspection of the respective NOESY experiment, and ATNOS/CANDID was restarted from cycle two. Because RNA cannot be interpreted by ATNOSCANDID standard library, only protein shifts were given, and nOe signals of RNA were manually picked, assigned, calibrated, and further used as distance restraints.

Peak lists of the final seventh cycle and manual derived restraints involving RNA were used as an input for the program CYANA 3.0 (56). The “noeassign” protocol of CYANA was used to reassign and calibrate the nOe signals of the given peak lists, resulting in protein-protein restraint lists. These lists were cleaned by applying a cutoff for the quality factor of 0.5 and by reviewing the peak lists and inspection of the NOESY spectra. Including all distance restraint lists and, in some cases, torsion angle restraints for the protein backbone derived by TALOS+ (38) or sugar pucker torsion angle restraints of RNA based on coupling efficiency in the homonuclear 2D TOCSY, CYANA was further used to calculated 250 structures by a simulated annealing protocol (MD, molecular dynamics steps = 20,000). On the basis of the target function, the 50 best structures were selected for refinement.

The AMBER 9 package (57) was used for structure refinement in the presence of the force field ff99SB (58) and implicit solvent [generalized Born model to mimic water as described in (59)]. Harmonic square-well penalty functions with force constants of 20 kcal mol−1 Å−2 for distance restraints and 300 kcal mol−1 rad−2 for torsion angle constraints were applied. First, a short minimization with long-range electrostatics treatment by the particle mesh Ewald method (60) using steepest descents energy minimization, followed up with conjugate gradient minimization was performed. The minimized structures were then refined using a simulated annealing protocol of 30,000 steps. For all refinements, 1-fs time steps in combination with constraint bond lengths by applying SHAKE (61) and 15-Å nonbonded cutoff were used. Scaling factor for the one to four electrostatic and one to four nonbonded van der Waals interactions were set to default values as used for the parameterization of the ff99SB force field (scee = 1.2 and scnb = 2.0). The details of applied input temperature, restraint ramping, and actual system temperature over the course of refinement can be seen in fig. S4. From the 50 structures refined in AMBER, 30 structures with the lowest AMBER energy were preselected, from which 20 structures with the lowest violation energy were selected for the final representative ensemble. The statistics for these ensembles can be seen in Table 1.