Background

Knowledge of the subcellular localization of a protein is crucial in defining its functions within the cell and regulation of the spatial organization of the proteome is essential for cellular homeostasis. Subcellular protein localization allows proteins to act as members of higher-order complexes with binding partners such as proteins, metabolites and co-factors, and contribute to the functions of organelles. In some cases, the sequestration of enzymes within specific subcellular compartments is critical to ensure that intermediates are not exchanged between different metabolic pathways or that potentially detrimental reactions (e.g., proteolysis) are confined to a ‘safe’ space. Protein localization can be influenced by post-translational modification, such as the phosphorylation of proteins within signaling cascades that ultimately allow proteins such as transcription factors (TFs) to localize within the nucleus to exercise their gene regulatory functions, or allowing other proteins to traverse the nuclear membrane and regulate nuclear-localized transcription factors. One example is the case of the osmostress response pathway in yeast, where a signaling cascade is activated that results in the phosphorylation of the protein Hog1p, whereupon it can translocate to the nucleus and phosphorylate the TF Hot1p which, in turn, can recruit RNA polymerase II to the promoters of responsive genes (Alepuz et al., 2003).

A growing body of evidence suggests that differential protein localization, even within the same cell type, may allow the protein to fulfil different roles depending on its respective localization within a cell, sometimes referred to as “moonlighting” (Gancedo et al., 2016). In Saccharomyces cerevisiae, examples of such proteins include Gcn4p, which is the general transcription factor that controls amino acid biosynthesis (Hope and Struhl, 1986) and also acts as a ribonuclease (Nikolaev et al., 2010); as well as Aco1p, which has roles in mitochondrial DNA stability in addition to its more widely known function as tricarboxylic acid cycle enzyme (Chen et al., 2005). Isoforms of the same protein have further been observed with different localizations in the same cell type, suggesting that this differential localization may offer another level of regulation with regard to protein function. An example of this is the case of a DNA methyltransferase (Dnmt1) in murine stem cells, in which one isoform localized to chromatin and the other isoform is found in an alternative nuclear location (Christoforou et al., 2016).

It is clear that there are many levels of complexity regarding protein subcellular localization. In order to fully understand the in vivo functions of proteins, it is vital to ascertain where, within the cell, they are located. Several methods exist that can be applied to approach this challenge, including fluorescence microscopy imaging, organelle proteomics, proximity biotinylation methods, and protein correlation profiling methods; all of which have been applied in Saccharomyces cerevisiae in different contexts to determine the subcellular localization of yeast proteins. Whole-cell, high-throughput studies based on fluorescence microscopy have been performed using GFP- or other fluorescence protein-tagged protein libraries (Huh et al., 2003; Tkach et al., 2012; Breker et al., 2013; Denervaud et al., 2013; Chong et al., 2015; Weill et al., 2018; Yofe et al., 2016). Although powerful, these methods have documented problems in that the fluorescence signal of a particular protein of interest may not exceed the background level of cellular autofluorescence, particularly in yeast, and therefore it may not be possible to determine the location of such proteins. Additionally, some proteins may require their unmodified C- or N-terminus for correct localization (Weill et al., 2018; Yofe et al., 2016). As such, performing protein localization studies where proteins are tagged with a fluorescent protein either N- or C-terminally to the ORF in question may result in a given protein being reported with different localizations within the same biological system, depending on which strategy was used (Stadler et al., 2013).

Organelle enrichment strategies have also been employed to define organelle proteomes, thereby contributing to our knowledge on protein localization. These strategies often exploit the characteristic buoyant densities of organelles in their enrichment. The relative ease with which mitochondria can be highly enriched, for example, has been exploited in many studies that have attempted to define the mitochondrial proteome, with varying levels of specificity, in S. cerevisiae. This includes some studies that used more crude approaches to attempt to define the mitochondrial proteome (Sickmann et al., 2003; Reinders et al., 2006) or specific sub-mitochondrial proteomes (Zahedi et al., 2006; Vogtle et al., 2012) in some cases without control for other organelle proteins that were co-enriched along with the mitochondrial proteins. Improvements have been made in recent studies that mapped the entirety of the mitochondrial sub-organellar proteome using more complex fractionation and proteome quantitation methods (Morgenstern et al., 2017; Vogtle et al., 2017). For organelles of more heterogeneous density, however, the difficulty of such studies can be compounded by the presence of contaminating membranes from other organelles that co-enrich with the organelle of interest (Wiederhold et al., 2010). In focusing on a specific organelle, such methods fail to report on proteins that may be present at multiple locations in vivo. Further, there is no control for the proportion of an organelle that is lost upon enrichment.

Recently, new proximity labeling methods based on promiscuous biotinylation have shown promise in determining protein subcellular localization in organisms other than yeast. This includes the more detailed study of sub-organellar proteomes (Rhee et al., 2013; Hung et al., 2014), including cytoplasmic faces of organelles (Hung et al., 2017). The BirA-based BioID method (Roux et al., 2012) has been applied in S. cerevisiae to define the subcellular proteins that interact with a ribosomal scaffold protein (Opitz et al., 2017) and the APEX2 method (Lam et al., 2015) has been demonstrated to traverse the cell wall in proof-of-principle experiments in both S. cerevisiae and the fission yeast Schizosaccharomyces pombe (Hwang and Espenshade, 2016). Methods such as these focus only on the proximal interactors of a protein of interest, that are currently in the range of tens of nanometres (Rhee et al., 2013; Kim et al., 2014) and do not provide a whole-cell picture of the spatial organization of the proteome.

To overcome some of the inherent limitations of the methods described above, it is imperative to look at protein subcellular localization in the context of the whole cell. This has been commonly studied using whole-cell protein correlation profiling methods. Such methods are predicated on the observation that, when cell lysates are fractionated by some means, such as differential or density gradient centrifugation, proteins that localize to the same organelle will exhibit similar distributions when the protein distributions across the entire fractionation scheme are monitored (de Duve, 1971). This was originally monitored using enzyme assays that are representative of the organelles in question, but now employs quantitative mass spectrometry methods. Several such methods have been published, which offer differing levels of proteome coverage and subcellular resolution, employing different fractionation and mass spectrometry quantitation methods (Christoforou et al., 2016; Itzhak et al., 2016; Jean Beltran et al., 2016; Mulvey et al., 2017; Jadot et al., 2017; Geladaki et al., 2019). The hyperplexed Localization of Organelle Proteins by Isotope Tagging (hyperLOPIT) method uses density gradient centrifugation for subcellular fractionation, followed by multiplexed proteome quantitation using 10-plex TMT isobaric tags (Thompson et al., 2003) and SPS-MS3 quantitation on an Orbitrap Fusion Lumos mass spectrometer. The exquisite subcellular resolution offered by the density gradient coupled with multiplexed quantitation, followed by sophisticated machine learning analysis methods means that hyperLOPIT allows unbiased protein localization determination of thousands of proteins in parallel in a single experiment. Indeed the method has afforded the highest resolution spatial map of any whole-cell correlation profile-based spatial proteome study to date (Thul et al., 2017; Gatto et al., 2019).

We recently published a yeast spatial proteome mapping study in which we mapped the proteome in high throughput using hyperLOPIT, under nitrogen-sufficient conditions (Nightingale et al., 2019). In our study, we produced a dataset that provided localization data for 2,846 proteins. Only 936 of these proteins were found to map to a unique subcellular location, a number which corresponded to 32% of our observable spatial proteome. This suggests, in common with previous studies (Christoforou et al., 2016; Thul et al., 2017), that a large proportion of the proteome is dynamic or resides at multiple locations.

We provide here a protocol that describes how to perform nitrogen cavitation lysis (Hunter and Commerford, 1961; Simpson, 2010; Wang et al., 2014) subcellular fractionation, and a version of the hyperLOPIT protocol that has been modified for use with the yeast S. cerevisiae (see Figure 1 for an overview of the protocol). We further draw the reader’s attention to the fact that the subcellular fractionation and Western blotting portion of this protocol may be performed as a standalone experiment without TMT labeling. Further to this, in cases where a mass spectrometer capable of resolution of TMT 10- or 11-plex tags (such as the Thermo Fusion series) is not available, isobaric tags with lower multiplexing capability may instead be used. This may be achieved using iTRAQ (Ross et al., 2004) 4- or 8-plex, or TMT 6-plex, but, importantly, will result in reduced overall subcellular resolution relative to TMT 10- or 11-plex, as less of the subcellular fractionation gradient will be sampled in the experiment. HyperLOPIT and its predecessor, LOPIT, have been employed to great effect to map the subcellular proteomes of multiple different species and under numerous conditions (Sadowski et al., 2006; Nikolovski et al., 2012; Groen et al., 2014; Christoforou et al., 2016; Thul et al., 2017). We envisage that this protocol will therefore be applicable and modifiable to study the spatial proteomes of other yeasts, including the industrial yeast Komagataella phaffii (syn. Pichia pastoris) and pathogenic yeasts such as Candida albicans.

We recently published a protocol in Methods in Molecular Biology (Nightingale et al., 2018) that describes how to perform hyperLOPIT in S. cerevisiae. The lysis and subcellular fractionation portions of this protocol are substantially updated from a previous protocol by Wang and colleagues (2014). Here we provide a protocol with updated mass spectrometry data processing parameters using Proteome Discoverer 2.1 (Thermo Fisher Scientific, www.thermofisher.com), and with more focus on post-mass spectrometry informatics analysis. For a more in-depth and detailed informatics protocol than is presented here, we direct the reader to the publication of Breckels and colleagues (Breckels et al., 2016b). The subcellular fractionation portion of the hyperLOPIT protocol is technically challenging, requiring an unbroken stretch of experimental work lasting from cell harvest to gradient fraction collection. We strongly recommend that experimental work not be paused until the gradient fractions have been collected.


Figure 1. Overview of the hyperLOPIT method for Saccharomyces cerevisiae. Also included in parentheses are approximate timescales for each step, and potential stopping points.