Post-translational modifications such as phosphorylation, acetylation and methylation are central to many biological processes. The ProtoArray® platform has been used previously to characterize the enzymatic activity of enzymes such as kinases and ubiquitin ligases [8, 9]. However, similar approaches for profiling PKMT activity on a proteome-wide scale have not been described. The protein arrays described here serve as a powerful tool to investigate the specificity of PKMTs, because they allow screening of more than 9,500 different substrates in a single experiment. In this study, we used the protein array platform and developed two independent screening methods for the identification of new substrates of PKMTs.
Each of the two detection methods used in the study has certain advantages, but also has limitations on sensitivity, and this motivated the complementary approach that we used. Antibody specificity is a crucial parameter for the success of the fluorescent detection method. As many commercially available pan-methyl antibodies crossreact with unmodified or non-methyl-modified sequences, we used the HEMP platform  to carefully characterize the antibody that was used for the protein array experiments. Although the antibody we used did not crossreact with unmethylated peptides, it also failed to detect all methylated peptides (Figure 2D). Thus, detection with the pan-methyl antibody probably missed a subset of biologically important targets. Improved methyl-specific antibodies should overcome this limitation in the future. In using GST as a negative control in these experiments, incorporating it in the substrate-candidate filtering method (Figure 3A), we increased the reliability of the results by eliminating potential targets that crossreact with the pan-methyl antibody in a PKMT-independent manner. Furthermore, using an antibody as a detection method makes the procedure fast, inexpensive and convenient.
We also used a second detection method with radioactively labeled SAM to screen the protein array for new PKMT substrates. Radioactively labeled SAM has been successfully used for in vitro PKMT assays in candidate-based approaches, and has led to the characterization of the activity of novel PKMTs and the identification of new methylated substrates, mainly histones [3, 14, 19, 20]. One of the main advantages of this radiolabeled SAM detection method is that it exhibits very low background signal when incubated with GST, making it a very sensitive method for detection and identification of methylation events, reducing the likelihood of false-positive results. Furthermore, in detecting new methylation events, it does not rely on new methylation events being recognized by existing antibodies. However, under our current conditions, this radioactivity assay has limitations in its sensitivity relating to the activity of the methyltransferase and the amount of 3H-methyl donor from SAM (used at a much lower stoichiometric ratio), as well as the inherent limitations of signal and background in the detection and imaging of radioactivity exposed on film.
Despite these limitations, the overlap between the methods was highly significant, supporting reproducible enrichment of targets of such modification as detected by both methods (Figure 4B). Further, although each assay did indeed identify distinct candidates, all of the six candidate targets tested, including all three of those detected by one (but not both) of the two methods, were experimentally validated (Figure 4D). Accordingly, this observation suggests that each of these two complementary methods has a different sensitivity profile, allowing identification of distinct groups of proteins that are targets of methyltransferases and that are experimentally intractable by the alternative method. Moreover, when considering the union of PKMT targets identified by both methods, significant enrichment was seen for proteins with common subcellular localization, molecular functions and roles in biological processes (see Additional file 3, Figure S3; see Additional file 6, Figure S4), consistent with our findings. Thus, these two methods exhibit both overlapping and complementary detection of candidate methyltransferase protein substrates.
Despite the strength of this protein array system as a proteomic platform, there are still several limitations that have to be taken into account when using this system. First, the latest generation protein microarrays used here (version 5.0) contain approximately 9,500 immobilized human proteins, representing only around on-third of the proteome. In many cases, proteins that were reported to be a target for a specific PKMT were not present on the array, including, p53 , DNMT1  and TAF10 , which have all reported to be methylated by SETD7. Custom arrays have been previously used for global analysis of protein phosphorylation in yeast [23, 24]; however, the yeast proteome is significantly smaller than the human proteome, and an array displaying the entire human proteome is not currently available. Second, for many proteins on the protein array used, the spotted protein does not cover the full-length sequence, so some positive hits may be missed. For example, we recently found that SETD6 methylates RelA on lysine 310 [14, 17]; although the array does contain RelA, the sequence covers only amino acids 1 to 221, and lacks lysine 310. Third, in some cases, the signal intensity for a specific known substrate on the array was not strong enough to pass the stringent threshold we used in order to reduce the number of false positives. Finally, it is likely that subsets of recombinant proteins are poor in vitro substrates for PKMTs because they are not properly folded, or are lacking crucial cofactors as part of larger macromolecular complexes.
In total, we identified 118 and 114 SETD6 new candidate substrates using the fluorescent and the radioactive detection methods respectively, and all six that were tested from both methods were confirmed to be genuine in vitro substrates. Two serine/threonine kinases were discovered: PAK4 regulates cytoskeletal architecture, cell proliferation, and the cell cycle, and is required for embryonic viability [25, 26], while PLK1 is involved in regulation of mitosis, including centrosome maturation and spindle assembly .
We also validated the methylation of the ribosomal protein RPS27L by SETD6; RPS27L is overexpressed in multiple human cancers, including colon , prostate , breast , liver , and head and neck carcinomas , and was recently shown to be a p53 target gene that regulates p53 protein levels . We found that SETD6 methylates two splicing factors, DNAJC8 and SRSF2. Although the function of DNAJC8 is still unknown, SRSF2 has been identified as a serine/arginine-rich protein belonging to the family of SR proteins that are crucial regulators of constitutive and alternative pre-mRNA splicing, and is also involved in regulating apoptosis in response to genotoxic stress . Finally, as a substrate for SETD6, we also identified the elongation factor transcription elongation factor A protein 1 (TCEA1) which is necessary for efficient RNA polymerase II transcriptional elongation . Together, these six proteins are involved in diverse biological processes, and future work is needed to elucidate the mechanistic and biological consequences of these SETD6-mediated methylation events.