The RNA helicase DHX34 functions as a scaffold for SMG1-mediated UPF1 phosphorylation

Nonsense-mediated decay (NMD) is a messenger RNA quality-control pathway triggered by SMG1-mediated phosphorylation of the NMD factor UPF1. In recent times, the RNA helicase DHX34 was found to promote mRNP remodelling, leading to activation of NMD. Here we demonstrate the mechanism by which DHX34 functions in concert with SMG1. DHX34 comprises two distinct structural units, a core that binds UPF1 and a protruding carboxy-terminal domain (CTD) that binds the SMG1 kinase, as shown using truncated forms of DHX34 and electron microscopy of the SMG1–DHX34 complex. Truncation of the DHX34 CTD does not affect binding to UPF1; however, it compromises DHX34 binding to SMG1 to affect UPF1 phosphorylation and hence abrogate NMD. Altogether, these data suggest the existence of a complex comprising SMG1, UPF1 and DHX34, with DHX34 functioning as a scaffold for UPF1 and SMG1. This complex promotes UPF1 phosphorylation leading to functional NMD.

predictions we displayed the RONN results with the command-line graphing program gnuplot embedded into a user-made Linux script. Each protein is represented with a different color. DHX34 is represented using a thicker dark blue line. The predictions suggest that disorder propensity in DHX34 accumulates at the C-terminal end of the protein. Representative reference-free averages of DHX34 monomers. Averages of monomers are also shown in Fig. 1c. (b) A 3D template for angular refinement was obtained without any initial bias, using reference-free averages of DHX34 and the volume generator in EMAN2 (command used: e2initialmodel.py) 4 . The method tentatively assigns a certain orientation (Euler angles) to each reference-free average provided, and a 3D average volume is computed. Inspecting the similarity between projections of the 3D template in several orientations and the reference-free averages monitors the quality of the prediction. (c) Initial 3D template obtained for DHX34 using e2initialmodel.py in and SMG1C alone, analyzed before mixing. (a) The whole data set (60,070 images) was subjected to 2D reference-free classification methods. A first round of classification was used to remove those images that did not correspond to molecule images but were incorrectly selected by the unsupervised automatic particle picking. These were easily identified, as they appeared as noisy and featureless average images. 45,000 images assigned to classes showing features, therefore corresponding to images of molecules, were then split and classified again (b). Subsequently, the new 2D averages obtained were first compared with those averages obtained for DHX34 and SMG1C as controls.

Supplementary
Images of DHX34, either monomers or dimers, were very different to those of SMG1C, and similar to the averages obtained for DHX34 alone ( Fig. 1 and Supplementary Fig.   2). These images could then be readily identified and split from the data set. Similarly, images corresponding to SMG1C, showing or not, an additional density attached, were clearly identified by comparison with our previous work 5 and these were split to generate a new independent dataset. (c) Averages assigned to either DHX34 monomers, oligomers, SMG1C or SMG1C-DHX34 were used to generate unbiased low-resolution 3D templates for refinement and classification using the volume generator from EMAN2 4 , Maximum-likelihood 3D classification methods as implemented in XMIPP 6 , and using these four templates as seeds, were employed to re-classify the 45,000 images. Following this strategy, 13,080 images were sub-classified in a group of images similar to SMG1C but showing a density attached to SMG1C, which was considered to correspond to the SMG1C-DHX34 complex. (d) Images of SMG1C-DHX34 were then refined using as initial reference a low-pass filtered version of the SMG1C or SMG1C-DHX34 templates, using EMAN. Supporting the robustness of the methodology applied, images classified as SMG1C-DHX34 generated a similar result in both refinements, using either SMG1C alone or SMG1C-DHX34 as templates. Additionally, although the information about DHX34 was never used for the refinement of the images of SMG1C-DHX34 at any stage, the 3D structure of DHX34 bound to SMG1C is closely similar to that obtained from DHX34 alone (Fig. 3c), supporting the classification strategy used.

Supplementary Figure 5 | Image processing and 3D structure of SMG1C-DHX34.
(a) Representative reference-free averages of SMG1C-DHX34 complexes. Averages of these complexes are also shown in Fig. 3a. (b) A 3D template for angular refinement was obtained without any initial bias, using reference-free averages of SMG1C-DHX34 (images of the complex split from the total dataset after the image classification strategy described in Supplementary Fig. 4), and the volume generator in EMAN2 (command used: e2initialmodel.py) 4 . The method tentatively assigns a certain orientation (Euler angles) to each reference-free average provided, and a 3D average volume is computed.
Inspecting the similarity between projections of the 3D template in several orientations and the reference-free averages monitors the quality of the prediction. (c) Initial 3D template obtained for SMG1C-DHX34 using e2initialmodel.py in EMAN2  DHX34 mRNA levels in NMD complementation assay using TCR β PTC reporter DHX34 mRNA levels in NMD complementation assay using TCR β WT reporter

Supplementary Table I
Depletion and mRNA expression levels of DHX34 in NMD complementation assay shown in Figure 6. Average levels determined by quantitative qRT-PCR from three independent experiments ± standard deviations are presented.