Peptide array

The use of arrays of synthetically generated peptides to screen for SUMO interacting peptides was suc- cessful and the models used to generate those peptides produced a good balance of interacting and non- interacting peptides, with about 20% or 10% of the total number of peptides interacting with AtSUM1

or HsSUM1 respectively. Had the peptide generation models been too specific, too few non-interacting peptides would have been identified; conversely, had the models been too general, too few interacting peptides would have been identified.

There were a number of issues with the peptide arrays and the methods used to collect interaction data. There was a large variance in the amount of peptide produced in each spot and the lack of a suit- able method to accurately measure the peptide amount resulted in uncertainty in the amount of peptide present which prevented any quantitative analysis of intensity signals for the peptide spots. Peptide amounts could not be accurately quantified due to the limitation of using dye-based methods for short peptides, where the intensity of the staining varies greatly depending on the amino acid composition of the peptides. A similar difficulty is experienced with spectroscopic methods, where the aromatic amino acids have a higher absorbance in the UV spectrum used for quantification (Hansenet al., 2013). For larger protein molecules, dyes or spectroscopy are accurate methods of quantification as the propor- tions of amino acids between proteins is usually similar. However, for short peptides the proportions of amino acids vary significantly between peptides. It may be possible to develop a more accurate method of quantification if the sequence of the peptides being measured is known, which is the case for peptide arrays. For UV spectroscopy the absorbance of the different amino acids is well characterised. The se- quence of amino acids in a peptide could be used to estimate an absorbance per quantity of peptide. Such a method could then be used to accurately quantify the amount of peptide in array spots and differences in interaction strength could be estimated with higher accuracy.

A confounding issue with synthetic peptide arrays is the purity of the resulting peptides as the peptide spots are synthesised on the array, leaving any mis-formed or truncated polymers and often, as was the case with this work, the purity of the peptides is not assessed. However, work by Frank (2002) has shown that generally peptide spots on arrays are of high purity. Given the large number of peptides in an array, it is likely that at least a small number will be of low purity due to variation in the synthesis method and synthesis of difficult peptide sequences. Data generated for sequences with low purity will be inaccurate as impurities may inhibit a true interaction or produce a false interaction; also if the amino acid sequence is used to normalise peptide amount measurements, low purity will lead to inaccurate results. Given the likelihood that at least some peptides in the arrays used in this work will have low purity, it is likely that a number of the interaction results are false though as long as the number of these false results is low, both analysis of the binding peptides and the prediction models should only be affected to a limited degree since RF models have been shown to be very tolerant to misclassified data (Breiman, 2001).

Western blotting was used to detect peptide interactions and significant difficulty was encountered with HsSUM1 as the antibody used had high cross-reactivity resulting in a large number of the peptide spots having to be excluded from analysis. The source of the cross-reactivity is not known and may

have come from either the control GST protein binding to the peptide spots or the antibodies them- selves binding directly to the peptides. Although different antibodies were used to detect the HsSUM1 interaction, no combination was found that did not exhibit this cross-reactivity issue. Cross-reactivity in far-western blotting is a frequently occurring issue and arises due to the large number of different steps in the procedure that can generate non-specific binding. These include offsite binding of the probe protein, binding of a protein tag on the probe protein if present and binding of either the primary of secondary antibody to the peptides (Katzet al., 2011). Using different methods, with fewer steps, can alleviate this issue with fluorescently labelled probe proteins being a suitable alternative. Traditionally probe proteins were labelled by non-specifically attaching a fluorescent molecule. This method had two major drawbacks though: the sites of probe labelling varied between protein molecules and it was dif- ficult to achieve consistent labelling between labelling reactions. The stochastic nature of the labelling can also lead to important sites of the probe protein being modified inhibiting the interaction between the probe and target. Newer methods such as using fluorescently labelled amino acids during protein translation or using site specific labelling methods have alleviated these issues. The HaloTag®system from Promega can be used to specifically label a target protein. A HaloTag®plasmid is used to express the probe protein fused to a HaloTag® protein. The HaloTag® protein can then be modified with a number of fluorescent ligands that the tag specifically recognises and covalently attaches to, leaving the probe protein unmodified. The HaloTag®however is very large and there is the possibility that its large size may interfere with the interaction between the probe and its target, though fusing the tag to the opposite terminal may alleviate this problem (Hurstet al., 2009).

Fluorescently labelled probe methods additionally have the advantage that they are quantitative, allowing the strength of a protein interaction to be measured. Far western blotting on the other hand is generally semi-quantitative due the non-linear response of the chemiluminescent detection and the number of binding steps, though methods described by Weiseret al.(2005) have been able to produce quantitative results from peptide arrays using far-western blotting by calibrating the method with a series of interactors with known interaction strengths. Overall though, future work would probably benefit from using a fluorescently labelled probe to accurately measure the relative interaction strength combined with accurate measurement of peptide amounts in the arrays to normalise the intensity results. Having accurate interaction strength results would allow more accurate models of SIM binding to be constructed so a distinction between strong and weakly binding SIM peptides could be made. Actual interaction strength results would allow the use of regression models rather than the classification models used in this work, which would allow the prediction of actual interaction affinities.