Characterizing the molecular components of the basic unit of life; the cell, is crucial for a complete understanding of human biology. The cell is divided into compartments to create a suitable environment for the resident proteins to fulfill their functions. Therefore, spatial mapping of the human proteome is essential to understand protein function in health and disease.

Spatial proteomics is most commonly investigated using mass spectrometry or imaging, combined with machine learning for the data analysis. Until now, studies have been limited to high abundant proteins and relied on the purification of organelle fractions from a bulk of cells. Within the scope of this thesis, we were able to systematically localize proteins in their native cellular environment using antibody-based imaging techniques, and to investigate protein subcellular localization and dynamics on a single cell level, introducing a major advance within the field of spatial proteomics.

Paper I of this thesis presents a subcellular map of the human proteome, where the spatial distribution of 12,003 human proteins was mapped into 30 subcellular structures, half of which were not previously localized. Besides providing a valuable dataset for cell biology, this study is the first to reveal the spatial complexity of human cells with proteins localizing to multiple compartments and pronounced single cell variations. Paper II reports on the systematic temporal dissection of these single cell variations and the identification of cell cycle correlated variations. We identified 258 novel cell cycle regulated proteins and showed that several of these proteins may be connected to proliferative diseases. A key finding of Paper II is that proteins showing non-cell cycle dependent variations are significantly enriched in mitochondria, whereas cell cycle dependent proteins are enriched in nucleoli. In Paper III and IV, we spatiotemporally characterized the proteomes of these two organelles, mitochondria and nucleoli, in greater detail.

In Paper III, we expanded the mitochondrial proteome with 560 novel proteins. As many as 20% of the mitochondrial proteome showed variations in their expression pattern at the single cell level, most often independent of the cell cycle. Paper IV provides a complete characterization of the nucleolar proteome. Nucleoli are not only important for ribosome synthesis and assembly, but are also crucial for cell cycle regulation through the recruitment of its proteins to the chromosomal periphery during cell division. Here, we presented the first proteome-wide spatiotemporal analysis of the nucleolus with its sub-compartments, and identified 69 nucleolar proteins that relocated to the chromosomes periphery during mitosis.

In conclusion, this thesis unravels the spatiotemporal proteome organization of the human cell over the course of a cell cycle and offers a valuable starting point for a better understanding of human cell biology in health and disease.

This thesis focuses on the spatiotemporal mapping of proteins at a subcellular level. In other words, determining the cellular location of proteins over time. From a biological point of view, knowledge about protein location is fundamental to understand protein function. In the longer run, this also means better understanding of cells in the context of health and disease, since protein malfunction and mislocalization are two important factors during disease development. 

Using an antibody-based imaging approach, Paper I contains a subcellular map of 12 003 protein in 30 different cellular structures, freely accessible as part of the Human Protein Atlas (www.proteinatlas.org). Apart from enabling exploration of the organellar proteomes, we conclude that half of the human proteins localize to multiple compartments, and that almost one fifth display cell-to-cell variations in terms of protein expression. Paper II aimed to decrease the cumbersome work of manual protein location annotations by leveraging the power of the crowd through citizen science. By integrating the image-classification task into a video game with a massive player base, EVE online, protein location labels could be efficiently and rapidly assessed compared to manual curation from a few experts. To compare the performance of the players, a deep learning classifier was developed. The algorithm was capable of classifying protein location in images containing several challenging localization problems, such as multilocalizing proteins, cell line variations and rare classes. Using the protein location data from Paper I and Paper II, Paper III presents an image-based characterization of the nucleolar proteome. In total, 1 318 nucleolar proteins are included, of which 157 localizes to a fourth nucleolar compartment, the nucleolar rim. Additionally, 65 proteins were detected on the chromosomal periphery during mitosis, and these could be further divided into two recruitment phenotypes with different temporal profiles. Also, the mitotic chromosome proteins are enriched for intrinsically disordered domains, suggesting liquid-like properties of the perichromosomal layer. Paper IV presents a systematic dissection of the variable proteome drafted in Paper I. We show evidence for 539 proteins being correlated to cell cycle variations, of which a minority are also cycling at a transcriptional level, suggesting protein regulation at a translational or post-translational level. Additionally, we detected hundreds of proteins with previously unknown relations to mitosis and the cell cycle, many being linked to proliferation and oncogenic functions.

In conclusion, Paper I and Paper II provide a basis for further in-depth studies of proteins at a subcellular level, while Paper III and Paper IV show how this resource can be used to study proteins in space and time. The results enable system-level investigation of protein dynamics, as well as provide exciting insights into organellar organization, such as the nucleolus.

Den här avhandlingen fokuserar på spatiotemporal karakterisering av proteiner på subcellulär nivå. Med andra ord beskrivs proteiners lokalisation i cellen över tid. Kunskap om proteiners lokalisation är grundläggande för att bättre förstå proteiners funktion. I ett längre perspektiv ger det också bättre förståelse av både friska och sjuka celler, eftersom felfungerande proteiner och felaktig proteinlokalisation är två faktorer som driver sjukdomsutveckling framåt.

Som en del av Human Protein Atlas-projektet (www.proteinatlas.org), innehåller Artikel I en subcellulär karta över 12 003 proteiner i 30 stycken olika organeller som kartlagts med hjälp av antikroppsbaserad mikroskopi. Utöver att ge en överblick över de olika organellproteomen, kunde vi också konstatera att hälften av alla proteiner lokaliserar till mer än en cellstruktur, samt att nästan en femtedel uppvisar skillnader i uttryck mellan celler. Målet i Artikel II var att minska arbetstiden för manuell annotering av proteinlokalisation genom att låta tusentals datorspelare i ett medborgarforskningsprojekt bidra. En bildklassificeringuppgift integrerades i datorspelet EVE online. Tack vare detta kunde proteinernas lokalisation snabbt och effektivt bestämmas, till skillnad från manuell annotering utförd av ett fåtal experter. För att utvärdera hur spelarna presterade, utvecklade vi samtidigt en klassificeringsalgoritm med hjälp av djup maskininlärning. Algoritmen visade sig kunna predicera proteinlokalisation i bilder trots att de innehöll flera försvårande omständigheter, såsom multilokaliserande proteiner, cellinjevariationer och ovanliga klasser. Med hjälp av proteinlokalisationerna från Artikel I och Artikel II, presenterar vi i Artikel III en bildbaserad karakterisering av nukleolens proteom. Sammanlagt visar vi data för 1 318 nukleolproteiner, varav 157 lokaliserar till en fjärde substruktur i gränsen mellan nukleolen och nukleoplasman. Vi visar även att 65 nukleolproteiner omlokaliseras till kromosomernas yta under celldelningen. Uttrycket av dessa proteiner kunde delas upp i två olika fenotyper baserat på när de tidsmässigt rekryteras till kromosomerna under mitosen. Förekomsten av oordnade proteindomäner var överrepresenterade i denna proteingrupp, något som öppnar upp för spekulationer om potentiella vätskeliknande egenskaper av den mitotiska kromosomytan. Artikel IV beskriver en systematisk genomgång av de varierande proteinerna som påträffats i Artikel I. 539 proteiners uttrycksvariation kunde kopplas till cellcykeln, varav en minoritet också varierar på transkriptnivå. Detta resultat indikerar att proteinreglering sker translationellt eller post-translationellt. Vi presenterar även data för hundratals proteiner som tidigare inte haft en känd koppling till cellcykeln, varav många kan relateras till proliferation och onkogenetiska funktioner.