intbio/SimChrom: v1 - SimChrom Initial Release

1 (Re)defining the human chromatome: an integrated meta-analysis of localization, function, abundance, physical properties and domain composition of chromatin proteins Anna K. Gribkova1,2, Grigoriy A. Armeev1,2, Mikhail P. Kirpichnikov1,3, Alexey K. Shaytan1,2,4* 1 Department of Biology, Lomonosov Moscow State University, Moscow, Russia 2 Vavilov Institute of General Genetics, Moscow, Russia 3 Shemyakin–Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, Moscow, Russia 4 International Laboratory of Bioinformatics, AI and Digital Sciences Institute, Faculty of Computer Science, HSE University, Moscow, Russia * To whom correspondence should be addressed. E-mail: shaytan[email protected]io.msu.ru GRAPHICAL ABSTRACT ABSTRACT The full complement of chromatin-associated proteins—collectively referred to as the chromatome—enables genome functioning in eukaryotes by participating in a wide range of physicochemical processes. These include mediating diverse specific and non-specific intermolecular interactions, catalyzing in situ synthesis and modification of macromolecules, facilitating ATPdependent chromatin remodeling, etc. Despite considerable progress in epigenomics and the structural characterization of many nuclear proteins and their complexes, our understanding of chromatin organization at the proteome scale remains incomplete. This gap hinders the development of a holistic view of genome regulation. In this study, we present a state-of-the-art characterization of the human 2 chromatome based on an integrative meta-analysis of diverse data sources describing the composition, abundance, and sub-nuclear localization of chromatin proteins. This effort is complemented by original analyses of their physico-chemical properties, domain architectures, and interaction patterns. To support and streamline these analyses, we developed a reference dataset of chromatin proteins, integrated with an empirical, function-based classification ontology and an associated interactive web resource — SimChrom — accessible at https://simchrom.intbio.org/. The reference dataset was carefully curated by reconciling data among protein databases, localization, and mass spectrometry-based experimental studies. Sequence-based and AI-assisted structural analyses revealed previously unannotated domains within chromatin proteins that warrant experimental validation, as well as the widespread use of multivalent interaction strategies that underpin chromatin organization. Together, our findings establish a robust framework for future studies aimed at elucidating genome function through detailed analysis of protein–protein and protein–nucleic acid interactions within chromatin. KEY POINTS ● The first comprehensive meta-analysis of human chromatin proteins that bridges diverse data types ● Established an interactive SimChrom framework for chromatome research available for the community ● Identified functionally relevant hallmarks of chromatin protein organization Keywords: chromatin, chromatome, epigenomics, proteomics, meta-analysis, genome functioning, protein domains, AI-based protein structure prediction, multivalent protein-protein interactions, intrinsically disordered proteins 1. INTRODUCTION Chromatin, according to the generally accepted definition, is the complex of DNA, proteins, and associated RNA molecules found in the nuclei of eukaryotic cells [1,2] (Fig. 1A, Interactive Fig. 1 at https://simchrom.intbio.org/#nucleus). However, in the crowded nuclear environment, it is challenging to establish stringent criteria that clearly distinguish between macromolecules that form a complex and those that do not, leaving room for interpretation of this definition. Chromatin proteins, collectively called the chromatome [3,4], enable genome functioning in space and time through active ATP-dependent processes and passive protein-DNA/RNA interactions. This functioning employs nontrivial physical phenomena such as liquid-liquid phase separation [5,6], topological constraints on the DNA, DNA looping and loop extrusion [7,8], diffusion in the crowded macromolecular environment [9], multivalent cooperative interactions [10,11], etc., all of which are regulated by the chromatome composition at specific locations and the properties of individual proteins including their posttranslational modifications (PTM), domain architecture and intrinsically disordered regions (IDR). 3 After the discovery of nucleosomes in 1970-ies chromatin research focused on elucidating molecular underpinnings of the genome organization and function at the nucleosome and supranucleosome levels [1]. During recent decades through the contributions of cryo-EM, epigenomics and 3D genomics much more details on the organization of large macromolecular assemblies [12], protein-DNA interactions and DNA topology [13,14] within chromatin have become available. We are at a point when holistic quantitative or at least qualitative models of the genome functioning based on integrating our knowledge about numerous molecular interactions and processes may seem to be within reach [15,16]. The scope of the data required for such models would lie beyond the one provided within the typical frameworks of genomics and epigenomics, and should also rely on what is sometimes referred to as “chromatomics” [3] - the systematic study of the entire content of the eukaryotic nucleus, including chromatin proteins, their spatio-temporal distribution and interactions. However, our understanding of the protein content of chromatin and its functioning at the “omics”-level is currently lagging behind our ability to probe DNA sequence, its epigenetic markup and 3D contacts. It faces certain challenges, which we detail below, with the human chromatome in mind. The first set of challenges lies in precisely defining the set of proteins that make up the chromatome. Historically, the definition was operational in nature relying on experimental chromatin extraction followed by the analysis of the protein content via physico-chemical methods (see a historical account by K.E. van Holde [1]) and later by various flavours of mass spectrometry analysis combined with different chromatin extraction and treatment techniques (review by van Mierlo and Vermeulen 2021 [17]). Unsurprisingly, the results of such studies depend on several factors – the details of the chromatin extraction techniques (e.g., non-strongly associating proteins may be not extracted), alternatively cytoplasmic proteins may contaminate the sample [18,19], the sensitivity and the resolution of the analysis method (e.g., low abundant proteins may be not detected, variations in posttranslational modifications, alternative splice isoforms) [20–22], and the transient nature of the expression of some nuclear proteins. An additional complication is the heterogeneous and dynamic composition of chromatin (sometimes called a fuzzy organel) – it depends on the cell type, cell cycle phase, as well as on the conditions experienced by the cell [23]. One has to keep in mind also that many proteins shuttle between nucleus and cytoplasm. Recent proteomics studies have estimated the number of chromatin proteins to be around 200 - 3800 [4,23–33]. Despite the above mentioned challenges, for many chromatin analysis tasks having a list of chromatin-associated proteins in the starting point. Since the human chromatome contains at least several thousand entries, any attempt at the rational understanding and description of its functioning requires some dimensionality reduction approaches. Hence, certain grouping or classification of chromatin proteins that considers their functional properties is desirable. Yet, obtaining such a classification is currently challenging. There are certain historically established classes of chromatin proteins that can be clearly defined (e.g., 4 histones, high mobility group proteins, etc.) [1], however, others have become obsolete (e.g., nuclear matrix proteins [34] or cannot be easily defined (e.g., nucleosol proteins). GeneOntology (GO) currently provides the most comprehensive set of annotations related to different aspects of gene products and is routinely used to interpret large-scale biological data, such as transcriptomics and proteomics results [35,36]. However, it cannot per se provide a straightforward and easy to comprehend classification of chromatin proteins due to the presence of a large number of chromatin related GO terms connected into a complex cumbersome hierarchy, which may be incomplete in some cases (e.g., lacks a term for histones) or include obsolete terms in certain cases (e.g., nuclear matrix). While ultimate functional classification of chromatin proteins may likely not be possible (due to the complexity of the genome functioning, different proteins contributing to many different functional processes, etc.), some approximation is at least needed to establish a framework for a rational reductionist-wise understanding of chromatin by us humans. The third set of challenges, in our mind, relates to the need for developing systems biology approaches to describe and study chromatin in a holistic way as a complex functioning system [37,38]. Considerable advances in methodology are currently needed to move from studying the structure of individual macromolecular complexes and analyzing sequence-level (albeit genome-wide) epigenomic data towards the quantitative models of chromatin operation that can grasp the emergence of complex organismal functions. Chromatin functioning relies on complex dynamics networks of multivalent interactions between macromolecules. These interactions depend on the abundance of chromatin proteins in a given compartment, their physico-chemical properties, and domain architectures that mediate specific or non-specific interactions. Understanding these issues at the chromatome-wide scale is a prerequisite for building holistic functional chromatin models. Motivated by the above mentioned challenges and the overall need to build complex models of chromatin functioning, in this work we attempted to provide the state-of-the-art meta-analysis of what is known about chromatin proteins, their localization, abundance, and properties. The uniqueness of this study is in cross-comparison of different data sources including database information, mass spectrometry data, and protein localization data. Our analysis was challenged by a common problem – the limited congruence between different data sources. To address it we developed several reference datasets for chromatin and nuclear proteins based on cross-comparison of different datasets and manual curation. Next we developed a relatively simple empirical function-based hierarchical classification of chromatin proteins (SimChrom classification) which was instrumental for all downstream analyses by allowing to compare properties between different groups of chromatin proteins. Using this framework we: 1) systematically analyzed the abundance of chromatin proteins, identified potential pitfalls in MSbased datasets, and using whole cell proteomics data quantified the presence of different chromatin proteins and chromatin protein groups in the cell; 2) characterized the interplay between amino acid composition of chromatin proteins, the prevalence of intrinsically disordered regions and specific 5 distribution of charged amino acids in their sequences; 3) analyzed the current state of structural characterization and domain annotation of chromatin proteins and based on novel AI-enabled protein structure prediction tools identified more than 200 domains in chromatin proteins that belong to currently unknown structural superfamilies and await experimental characterization, 4) characterized typical patterns of multivalent interactions employed by chromatin regulator proteins mainly engaging combinations of histone methylation, acetylation and DNA binding modes. Finally, we supplement our analyses with an interactive web resource — SimChrom — accessible at https://simchrom.intbio.org/. SimChrom harmonizes data from different sources and may be used for exploration of different chromatin protein groups and properties of individual proteins. The overall scheme of our work is outlined in Fig. 1B. Although our analysis in fact required many iterations to achieve self-consistency (e.g., development of SimChrom classification was performed concomitantly with the development of reference chromatin protein sets, and classification was employed in analysis of the quality and content of different data sources) below we present the logic of our analysis as a series of consecutive steps to the extent possible, offloading detailed consideration of certain aspects that may require the familiarization with the whole manuscript to the Supplementary Results and Discussion (Suppl. R&D) sections. Figure 1. (A) The structure of the nucleus with details of chromatin organisation at the levels of chromatin domains and chromatin regulatory proteins. (B) The overview of this study shows sources of information about chromatin proteins (e.g., their functions, subcellular localization, identification by MS-based methods or specific protein functional domains) and their use in the current study. 6 2. MATERIAL AND METHODS For the purpose of all analyses in this study we used the set of human protein identifiers and corresponding amino acid sequences representing the canonical protein isoforms (usually corresponding to the major splice isoforms) as provided by the UniProtKB/Swiss-Prot database (also known as the reviewed section of the UniProt Knowledgebase) (UniProt proteome ID UP000005640, release 2022_2) [39]. The set contained 20,272 gene entries corresponding to 20,225 unique protein IDs (some genes code for identical protein sequences). Wherever needed the original datasets were mapped to the above described set of protein UniProt IDs. 2.1. Collection and processing of data about chromatin and nuclear protein repertoires, as well as other protein groups, from databases and MS-based studies 2.1.1. Protein localization data sources (UniProt, HPA, OpenCell) Protein subcellular localization annotations were obtained from UniProtKB [39] (release 2022_2, subcellular location section), Human Protein Atlas (HPA) [40] (version 22, https://www.proteinatlas.org/download/subcellular_location.tsv.zip, accessed on 09.06.2022), the OpenCell dataset [41] was retrieved from the website (https://opencell.czbiohub.org/, accessed on 10.06.2022). To ensure high-confidence subcellular localization annotations, localization terms were filtered according to database-specific reliability criteria. In UniProt, only annotations supported by at least one evidence tag were retained. In HPA, only annotations with a reliability score exceeding the 'uncertain' threshold were included. In OpenCell, only annotations scoring above the lowest quality grade were retained. For the analysis of protein multiple localization, localization annotations from HPA and UniProt were grouped into the following generalized categories: Nucleus, Cytoplasm, Endomembrane system, Other (including chromosome, secretory, and extracellular proteins in UniProt). Detailed grouping information is provided in Suppl. Table ST3. To estimate overrepresentation or underrepresentation of a particular localization for a list of proteins enrichment analysis based on hypergeometric test (also known as Fisher’s exact test) was used with significance threshold p-value < 0.05. To estimate the congruence of subnuclear localization annotations between UniProt (protein set denoted by A) and HPA (protein set denoted by B) following metrics were used: TP = |A ∩ B|; FP = |B ∩ ¬A|; FN = |A ∩ ¬B|; Union = |A ∪ B|; Jaccard similarity coefficient = TP / Union. Performance measures: Precision = TP / (TP + FP); Recall = TP / (TP + FN); and F1-score = 2 × Precision × Recall / (Precision + Recall). 7 2.1.2. GO and function-oriented databases In Section 3.1, for a selected set of Gene Ontology (GO) terms, lists of associated human proteins were retrieved from the QuickGO database [42] (GO annotation set created on 2025-03-06) using its REST API. To capture all relevant proteins, annotations were obtained also for all their descendant terms with the relationships "is_a", "part_of", and "occurs_in". Information on human histone and non-histone epigenetic regulators was obtained from the EpiFactors database [43] (version 2.1, released September 10, 2024). DNA-binding transcription factors (dbTFs) were obtained from https://www.ebi.ac.uk/QuickGO/targetset/dbTF, accessed on 01.09.2024) [44]. 2.1.3. MS-based studies We conducted a comprehensive search on PubMed using keywords "chromatome", "chromatin proteins", "experimentally obtained chromatin proteins", "nuclear proteins", and "nuclear proteome" to gather relevant informational resources containing data on nuclear and chromatin proteins. All protein entries identified in MS-based studies were mapped to the UniProt human reference proteome (release 2022_02), protein isoforms were collapsed to canonical entries. Gene names were used as secondary identifiers to facilitate mapping for records without a direct match. Outdated UniProt entry identifiers were updated to their current counterparts in the used proteome. Below the specific details of obtaining the protein lists from respective MS-studies are provided (see Table 1). From Kustatscher et al. (2014) study [23], which provides interphase chromatin probability scores (ICP) for 7,635 human proteins, proteins with ICP > 0.5 were selected. From Alabert et al. (2014) study [26] entries with missing UniProt ID, gene name, or intensity were excluded, unmapped entries were matched by gene name. No filtering by nascent enrichment or chromatin probability was applied. From Ginno et al. (2018) study [25] proteins quantified with at least two unique peptides and consistent signal across all three replicates of at least one cell cycle stage (G1, S, or M) were selected. Although mitotic chromatin was included, the protein composition across stages was nearly identical, and excluding Mphase would not significantly alter the dataset. From Shi et al. (2021) study [27], protein entries were taken from the experimental run that used conditions involving HaeIII digestion and no 1,6-hexanediol treatment ("condition 2" in the study), representing chromatin-associated proteins extracted under native conditions using the Hi-MS protocol. Only proteins with non-zero iBAQ values in all three replicates were retained. From Torrente et al. (2011) study [4] chromatin-associated protein lists obtained by three three extraction methods were combined (total chromatin extraction, high-salt extraction, and micrococcal nuclease digestion). Protein GI accession numbers were converted to UniProt reviewed entries. From Itzhak et al. (2016) study [45] all entries annotated as "mostly nuclear" by the "Global classifier 2" were taken. From Alvarez et al. (2023) study [28] a combined list of proteins obtained by NCC (Nascent Chromatin Capture) in HeLa S3 cells and iPOND (isolation of Proteins On 8 Nascent DNA) in TIG-3 fibroblasts was taken. From the Ugur et al. (2023) study [24] proteins with non-missing raw Log2 intensity values in all three chromatin replicates for human embryonic stem cells were taken. 2.1.4. Other sources Housekeeping proteins were defined as proteins detected in all analyzed tissues by RNA-seq and downloaded from HPA (version 23) [46] (in total 8899 proteins). Manually curated protein complexes were downloaded from Complex Portal (accessed on 7 January 2025) [47]. Only complexes exclusively containing chromatin proteins were selected for domain co-occurrence analysis. 2.2 Construction of reference chromatin and nuclear protein datasets 2.2.1. The SimChrom ontology, SimChrom protein dataset, and SimChrom/SimChrom-SL classification The SimChrom chromatin proteins classification ontology was developed simultaneously with the corresponding set of human chromatin proteins that were collected according to the developed classification. To this end, for almost every SimChrom classification term we attributed specific terms from GO classification that were manually selected to represent molecular functions and biological processes that happened exclusively inside the cell nucleus and were related to the respective SimChrom term. In certain cases cellular component GO terms were also used when they were directly related to the respective SimChrom term (e.g., complexes of chromatin remodelers). The list of the GO terms attributed to every SimChrom term is given in Suppl. Table ST4. The dataset was then supplemented by proteins defined as chromatin proteins by several databases, review papers, and original studies (e.g., HistoneDB 2.0 - histone proteins, Histome2 - for PTM writers, see details and corresponding data sources in Suppl. Table ST4). At the last step for certain SimChrom categories the contents of the dataset were additionally filtered to either remove the potentially non-nuclear proteins (in the case of RNA-binding proteins) or histones from the categories belonging to the “non-histone proteins” group (see details in Suppl. Table ST4). To ensure unambiguous categorization, we constructed a SimChrom-SL (single labeled) classification, where each protein was assigned to exactly one category of the same SimChrom ontology. Assignment followed a predefined priority order: "Molecular function" and "Physicochemical properties" categories were prioritized, followed by others ("Biological processes" and "Genomic location"). Within these groups, categories containing fewer proteins were ordered first (see priority hierarchy in Suppl. Fig. SF3_2). Each protein was labeled with its first eligible category in this sequence. 9 To assess the quality of the SimChrom dataset, Gene Ontology enrichment analysis was performed. Two groups of proteins were analyzed separately: (1) proteins present in SimChrom but absent from NULOC_CS (see below), and (2) proteins present in NULOC_CS but absent from SimChrom. GO enrichment was assessed using g:Profiler [48], with multiple testing corrections applied (Bonferroni correction, significance threshold 0.05). Only driver GO terms (defined by g:Profiler) were selected for further interpretation. 2.2.2. Construction of reference localization datasets Reference datasets of nuclear (abbreviated as NULOC), non-nuclear (NON_NULOC) and cytoplasmic (CYTLOC) protein entries at different levels of confidence and uniqueness of localization were constructed (see set of localization terms in Suppl. Table ST3). The datasets were created using the combination of localization information from UniProt and HPA (full details are provided in Suppl. Table ST6). Datasets of proteins having only one specific localization (i.e. no other localization reported in the source databases) are denoted by the UL suffix. Datasets, where localization is simultaneously supported both by UniProt and HPA are denoted by CS (consensus) suffix, alternatively, if localization is supported by at least one database the JT (joint) suffix is given. The NECF (“no evidence code filtration”) suffix denotes datasets where no preliminary filtration of localization information provided by the databases based on evidence codes or confidence levels were applied. We noticed that many histone protein entries in UniProt lack evidence codes for their localization (in release 2022_2). This is apparently due to the fact that these entries appeared in UniProt before the manually curation of evidence attribution was introduced, and they are still awaiting manual review and retrofitting. We manually added histone proteins in all constructed nuclear reference datasets. 2.3. Protein abundance data processing 2.3.1. PaxDB data Two dataset from PaxDB [49] version 4.2 with protein abundance data were used: the dataset with the highest proteome coverage (“H. sapiens - Whole organism (Integrated)” - covers 99% of human proteome according to PaxDb, referred to as “PaxDb_INT'' in this paper) and the dataset with the highest interaction consistency score (“Whole organism, SC (Peptideatlas,aug,2014)” - covers 84% of human proteome according to PaxDb, referred to as “PaxDb_PA'' in this paper), see also Suppl. R&D Sec. 3.1. An abundance unit is protein per million, ppm, which describes protein abundance relative to all expressed molecules in the proteome. Protein abundance was obtained by aggregating abundance of individual genes with similar protein sequences (e.g., in the case of canonical histones). Cumulative abundance was defined as the sum of protein abundances for a group of protein entries; cumulative weight was calculated as abundance multiplied by the protein molecular weight. 16 Ginno et al., 2018 Total chromatin: time-resolved (G1, S, M) Density-based enrichment for mass spectrometry analysis of chromatin (DEMAC): formaldehyde-fixed cells were sonicated, subjected to cesium chloride (CsCl) gradient ultracentrifugation to isolate DNA–protein complexes by buoyant density (1.39 g/cm³). Chromatin fractions were collected, dialyzed, decrosslinked, digested with DNase I. Human T98G (glioblastoma) 3065 (chromatome); 6242 (total proteome) 3051 Shi et al., 2021 Promoterproximal chromatin Hi-MS (Hi-C-based proteomics, adapted from BL-Hi-C): cells crosslinked with 1% formaldehyde; genomic DNA digested with HaeIII (GGCC sites); ligated with biotinylated bridge linkers; nuclei lysed in 0.2% SDS; chromatin sonicated; chromatin-DNA complexes captured on streptavidin beads. Quantified sensitivity to 1,6-hexanediol evaluated via AICAP index (Anti-1,6Hexanediol Index of Chromatin-Associated Proteins). K562 3228 2848 Ugur et al., 2023 Total chromatin Chromatin Aggregation Capture (ChAC): nuclei fixed with 1% formaldehyde, lysed with SDS and urea, sonicated, and purified by protein aggregation capture (PAC) on magnetic beads. DIA-MS with DIA-NN used for quantification. Human ESCs (H9) 2487 1730 Alvarez et al., 2023 Time-resolved (nascent, G2/M, early and late G1) chromatin Nascent Chromatin Capture (NCC) method, which relies on pulse-labeling newly replicated DNA with biotin-dUTP, followed by formaldehyde crosslinking and sonication-based chromatin fragmentation. Biotinylated DNA-protein complexes were affinity-purified using streptavidin magnetic beads. HeLa S3 cells were synchronized and harvested at five post-replication time points (Nasc, Late S, G2/M, early G1, late G1) across six biological replicates. HeLa S3 1454 (present at all time points in all 6 replicates; from total of 5770) 1478 (2894 total) Isolation of Proteins On Nascent DNA (iPOND): formaldehyde crosslinking (1%), EdU labeling for 15 minutes, click chemistry with biotin-azide, chromatin fragmentation, streptavidin bead enrichment. TIG-3 fibroblasts 2351 (detected in ≥4 of 5 replicates) 2397 (2894 total) Among the protein/gene annotation databases the GeneOntology (GO) database stands out as a comprehensive attempt in describing the functions of gene products in an ever growing number of organisms [35,36]. Within the GO framework genes are annotated according to their involvement in certain molecular functions, biological processes, and cellular components. The ontology itself forms a complex interlinked hierarchy with more than 40,000 GO terms and offers annotations to nearly the entire human reference proteome. However, despite its apparent comprehensiveness the GO database could not per se provide answers to the questions that were instrumental to this study, namely, to provide a set of chromatin genes/proteins and a relatively simple functional classification of these proteins that could be used for further analysis. The GO cellular component term "chromatin" is defined broadly as "the ordered and organized complex of DNA, protein, and sometimes RNA, that forms the chromosome" and encompasses around 2000 proteins. Comparisons with other databases suggest that this number is a rather conservative estimate. For instance, up to 528 proteins listed in specialized databases of epigenetic factors (EpiFactors) and transcription factors (GO catalogue of TFs [44]) are missing from this set (see Suppl. Fig. SF2_1A); concomitantly the HPA protein localization database suggests that there are around 6000 proteins located in nucleoplasm (see Suppl. Fig. SF2_2A). Furthermore many proteins annotated by GO terms that are bona fide related to chromatin (e.g., chromatin binding) are missing from those annotated by the GO term “chromatin” (see Suppl. Fig. SF2_1C). This stems in part from the complexity of the relations between the GO terms belonging to different annotation aspects, in this example the terms “chromatin organization” and "chromatin remodeling" are not connected to the term “chromatin” within the ontology tree. The manual search and identification of the chromatin related 17 GO terms and relevant proteins is challenging because (1) of the shear number of GO terms (e.g., the word “chromatin” is found in the names of more than 60 terms, see Suppl. Fig. SF2_1C), (2) the fact that apparently relevant terms may include also non-chromatin associated entries (e.g., transcription may also include mitochondrial transcription), (3) the fact that terms describing certain historically established chromatin protein groups may be missing (e.g., histones, HMG proteins), (4) the fact that GO database is not chromatin-specific and may not be up-to-date in certain aspects (e.g., contain obsolete terms such as “nuclear matrix” or lack annotations for proteins that are available in recent literature reviews). Suppl. R&D Sec. 1.1 provides further details and examples from our analysis. A number of epigenetic/chromatin regulators/factors databases (e.g., EpiFactors [43], CRdb [57]) provide carefully curated information about chromatin proteins that are involved in what is historically assumed to be molecular mechanisms of epigenetic regulation (Fig. 2A, Suppl. Table ST1). However, they annotate only around 400-800 chromatin proteins, which is much less than is expected to be in chromatin (see Fig. 2D). Unlike GO, these databases introduce a rather simple classification of proteins (the respective categories are highlighted in Fig. 3), but lack many essential chromatin categories (e.g., histones, histone chaperones, etc.). Protein class specific databases and datasets available in published papers provided an even more trustworthy but narrow sets of information about certain classes of chromatin proteins. Of particular impact here by the number of provided entries are the databases of transcription factors. Recent databases (e.g., The Human Transcription Factors [55], GO catalogue of TFs [44]) comprise around 1500 transcription factors. Additionally, several chromatinrelated protein classes have been reviewed in the literature but lack dedicated database resources [58– 64]. The other alternative and powerful source of information about nuclear/chromatin proteins is localization databases and proteome-wide studies – UniPot, HPA and OpenCell projects are currently regarded as the most comprehensive and trusted resources on protein intracellular localization (see Fig. 2B). From each resource we extracted the sets of proteins whose localization was annotated (only annotation with sufficient confidence levels was considered for our analysis - see Methods Section 2.1.1) as belonging to the nucleus or sub-nuclear compartments according to the localization ontologies specific to each resource. The detailed cross-comparison of the datasets is presented in Suppl. Fig. SF2_2, its interactive version (Interactive Fig. 2 available at https://simchrom.intbio.org/#localization), and at length discussed in Suppl. R&D Sec. 1.2 using Suppl. Fig. SF2_2, SF2_3, SF2_4. In summary there is a considerable degree of variation both between the sets of nuclear proteins, their sub-nuclear localization annotation and the annotation ontologies themselves between the resources. It has to be kept in mind in the first place that localization annotation coverage is not complete – collectively the three resources cover 86% of the human reference proteome (with sufficient 18 confidence - see above), while only 44% of proteins are simultaneously annotated by UniProt and HPA. The resources also differ by the number of localization annotations they provide on average for each protein (median number is two and one, for HPA and UniProt, respectively), suggesting HPA is more complete with respect to annotating multilocalization of proteins. Hence, although the protein space coverage by UniProt is larger compared to HPA (70% vs 60%), the nucleome provided by the former is considerably smaller (~4700 vs 6500 proteins). Together the three resources annotate ~8000 proteins as having nuclear localization (see Venn diagrams in Fig. 2B), which amounts to 47% of proteins that have localization information according to at least one resource. Hence, as a rough estimate it is tempting to conclude that current protein localization databases suggest that around half of human proteins have some evidence of nuclear localization. However, it has to be kept in mind that the congruency between the resources remains mediocre. Among ~9000 proteins whose localization is simultaneously available in UniProt and HPA, ~3300 are annotated as nuclear by both HPA and UniProt, while another ~2000 are annotated as such only by one of the two resources (~600 by UniProt and ~1400 by HPA) (see Fig. 2B, lower left Venn diagram). The discrepancies are in part due to (1) incomplete annotation of multiple localization possibilities by the databases (among the ~2000 proteins, ~60% have a matching localization annotation between the databases other than nuclear), (2) potential biases in localization annotations (HPA tends to label nuclear proteins as vesicular proteins and UniProt tends to label nuclear proteins as secreted proteins and proteins of the extracellular matrix) (see Suppl. R&D Sec. 1.2, Suppl. Fig. SF2_3). Extrapolating the above estimates to the whole proteome (with all caveats in mind about the non-uniform annotation coverage of different protein groups) one can suggest that between 35 to 60% gene protein products may be ascribed nuclear localization depending on the chosen degree of certainty. Hence, a combination of the datasets provided by the resources may be used to construct reference nucleome datasets of varying confidence (see Results Sec. 3.2). Many nuclear proteins were found to have multiple localization annotations belonging to different cellular compartments (see Suppl. Fig. SF2_2B, Methods Section 2.1.1, and Suppl. R&D Sec. 1.2). On one hand this reflects the functionally important property of nuclear proteins to shuttle between compartments. For example, many transcription factors and coactivators (e.g., NF-κB, STAT, p53, TAF7, YAP/TAZ) regulate their action through cytoplasm/nucleus shuttling [65–67], even some histones, such as H2B, may relocalize to cytoplasm under stress and perform unconventional functions [68,69]. On the other hand, functionally irrelevant multiple localization annotations may arise due to experimental artefacts or suboptimal signal-to-noise thresholds, keeping in mind that all nuclear proteins are in fact synthesized in cytoplasm and imported to the nucleus. According to our analysis UniProt and HPA estimate separately that ~50% of proteins with nuclear localization may be also localized in other compartments (~40% in cytoplasm, 12%–22% in the endomembrane system), annotating around 48%-50% to be localized solely in the nucleus. However, once the annotations of UniProt and HPA are compared within the shared common set of proteins (nuclear localization 19 annotation is available in both databases) it turns out that only for ~40% of proteins the two databases reach consensus for their unique nuclear localization (see Fig. 2B, lower right Venn diagram). In other words (see Suppl. R&D Sec. 1.2), approximately for every five proteins identified as uniquely localized in the nucleus by one database, it is likely that two of them will have non-nuclear localization annotation in the other database (per se or in addition to the nuclear localization). The same tendency was observed for the annotations of uniquely localized cytoplasmic proteins. The obtained estimates likely reflect the suboptimal specificity of the localization information provided by the databases (additional localizations of nuclear proteins may not be always captured) and potential presence of spurious localization annotations (artifacts or incorrect localization assignments). It is, however, non-trivial to deconvolute between these two types of errors. Our analysis of sub-nuclear localization ontologies showed that the one of Uniprot is more diverse comprising 20 terms (this number includes chromosome localization which is distinct from the nuclear localization according to UniProt, although the majority – 84% – of chromosome proteins are also annotated as nuclear), while HPA and OpenCell comprize, 9 and 6 terms, respectively. However, in terms of annotation specificity only 19% of nuclear proteins in UniProt are annotated with subnuclear localization terms, while for HPA all of the nuclear proteins have some sub-nuclear localization (although 92% are considered a part of nucleoplasm, 33% bear localization annotations other than nucleoplasm). There are certain parts of the ontologies that do not match between the resources or to the state-of-the-art knowledge. For instance, HPA considers mitotic chromosomes as a part of nucleoplasm, while UniProt uses outdated “nucleus matrix” term. OpenCell is the only resource that explicitly considers “chromatin” as the possible localization for nuclear proteins, while UniProt explicitly lists “Chromosome” as the possible localization, which was in turn inherited from GO cellular compartment ontology where “chromatin” has a child-parent relation with the term “chromosome”. The above-mentioned discrepancies reflect the dynamic complexity of cellular organization, our constantly evolving understanding of nuclear organization, and the resulting difficulty in describing subcellular localization in a form of a simple hierarchical tree-like ontology. While the exact names may differ, all resources converge on the presence of the following localization terms: Nucleoplasm; Nuclear bodies; Nucleolus; Nuclear envelope. Among these terms the nucleoplasm localization is the one most related to chromatin proteins (according to HPA nucleoplasm is what is found within the nuclear membrane, but excludes nucleoli according to the respective localization ontology). If one interprets the definition of chromatin broadly (treats proteins that localize with the interphase chromosomes to be part of the chromatin “complex”) the set of proteins with nucleoplasm localization is a direct source of information about chromatin proteins. HPA lists around six thousand nucleoplasm proteins (Fig. 2B). The analysis of subnuclear multilocalization is available in Suppl. R&D Sec. 1.2 and Suppl. Fig. SF2_4. MS-based studies of chromatin extracts are another key source of information about the protein content of chromatin. Despite being the ultimate direct source of data about the composition of 20 chromatin it unfortunately has certain limitations (see Introduction, and relevant reviews [17,70]). To gain quantitative understanding into the utility of MS-based studies for our goals we have selected data from several studies in human cell lines (see Table 1 and Methods Section 2.1.3) for analysis. The selected datasets included five studies that aimed at total interphase chromatin characterization using different methods of chromatin purification and post-MS data analysis, two studies characterizing nascent chromatin, and one study characterizing total nuclear proteome. The more than twofold variation (from 1.5 to 3.5 thousand entries) in the number of detected chromatin proteins in various MSbased studies highlights the varying sensitivities of different chromatin purification/MS-detection setups (Table 1). The pairwise comparison of different chromatin datasets of comparable size (having around 3000 proteins) suggests that for any given set its fraction overlapping with any other set does not exceed 68% (see Fig. 2C). The number of chromatin proteins present simultaneously in all total chromatin datasets is 179 (Suppl. Fig. SF2_5B). These facts highlight considerable variation of MSbased data due to different sample sources and chromatin extraction techniques. We next thoroughly analyzed these protein datasets through cross-comparison between themselves, comparison with protein localization data, and tested enrichment of different chromatin protein categories (according to SimChrom classification described in Results Section 3.2). The detailed analysis is provided as Suppl. R&D Sec. 1.3, and we only succinctly summarize our conclusions below. From 10% to 38% of proteins identified in MS-based chromatin datasets currently do not have any support from localization databases through their annotated nuclear localization (see Suppl. Fig. SF2_5A,C), suggesting that even for chromatin purification protocols based on proteinDNA cross-linking there still might be a certain degree of contamination with non-nuclear proteins, mainly cytoplasmic ones (see Suppl. Fig. SF2_6). Yet, MS-based techniques may have predictive power to identify new chromatin proteins that are not annotated in the localization databases. For example, among 195 proteins reported simultaneously by at least five out of seven chromatin MS-based studies we estimated that around ~30% of proteins may have indications in the literature supporting their nuclear localization. MS-based studies are biased towards identifying the housekeeping proteins - more than 80% of nuclear/chromatin proteins reported by the MS-based studies were from the housekeeping pool, while the average expected fraction of nuclear housekeeping proteins is around 62% (see Suppl. Fig. SF2_7A,B). This is expected since many non-housekeeping proteins are conditionally expressed. However, MS-based studies tend to miss the housekeeping transcription factors too (and even to a greater extent non-housekeeping TF) apparently due to their low abundance and dynamic nature of interactions (Suppl. Fig. SF2_7C, SF2_8A, see also Results Section 3.3.1 for discussion of chromatin protein abundance). MS-based studies also struggle to recover as separate gene products proteins with very similar sequences, e.g. canonical histone isoforms (see Suppl. Fig. SF2_7C, SF2_8B). 21 To finalize our analysis we compared the datasets from three types of data sources about chromatin proteins examined above (Fig. 2D, Suppl. Fig. SF2_9). One can see that localization databases are currently leading by the number of proteins that may be considered as chromatin proteins in the broad sense (e.g., the proteins of the nucleoplasm). However, there is still limited congruence with the other data sources. For instance, 25% of GO "Chromatin" proteins are not localized in the nucleus according to HPA, moreover, of these 500 proteins, only 115 have any localization information in HPA. Notably, 42% (2699) of the proteins identified in MS-based chromatome and nucleome studies lack nuclear localization annotations in both UniProt and the HPA, whereas only 254 proteins remain entirely unannotated for subcellular localization in these databases. Taken together our analysis of different chromatome data sources revealed considerable heterogeneity of information and limited congruence between the available datasets. The available functional databases while providing functionally supported data are either limited in scope or suffer from historically-contingent complexity and sometimes discrepancies in their classification ontologies that are not tailored to provide comprehensive straightforward information about interphase chromatin proteins. The localization databases is a powerful alternative source of information that can give an upper bound for the set of chromatin proteins (since they should have nuclear/nucleoplasm localization), provide a relatively reliable estimate of the lower bound for the number of nuclear proteins, however, they suffer from an incomplete coverage of the proteome-localization space and hence difficulties in estimating false-positive and false-negative annotation rates (keeping in mind the multi-localization of proteins) and limited congruence of subnuclear localization ontologies. The MS-based studies of chromatin extracts are the most direct source of information about chromatome, they may identify new chromatin proteins not annotated currently into the databases, however, they are limited in scope (many proteins are conditionally expressed or have low expression levels) and suffer from contamination with non-nuclear proteins. 22 3.2. The SimChrom chromatin protein classification, the SimChrom dataset and other reference datasets Figure 3. The SimChrom empirical chromatin classification ontology and the SimChrom chromatin proteins dataset. The hierarchical tree-like classification organizes 39 SimChrom categories. For each category the respective number of proteins from the SimChrom dataset is given in parentheses (proteins can simultaneously belong to more than one category with the exception of histones). The pictograms on the left of each category name provide the information about the presence of similar categories in the ontologies of other databases (see legend). The colored bars show whether the specific category was derived from grouping the proteins according to certain aspects: the similarity of their molecular functions, physico-chemical properties, involvement in similar biological processes or localization in similar genomic locations (see legend). Note: the latter annotations may deviate from GO annotation aspects. Taking into account the advantages and disadvantage of different sources of information about chromatin proteins presented above, we aimed at constructing a reference set of chromatin proteins together with a classification ontology and several supplementary nuclear localization protein datasets that can be later used in analyzing the repertoire, abundance, functional, structural, and physicochemical properties of chromatin proteins. Our aim was to create a relatively simple classification ontology that while potentially sacrificing the details will enable a holistic human-understandable overview of the chromatome (see Suppl. R&D Section 1.1 for the discussion of GO complexity and ensuing challenges). The current version of SimChrom classification focuses on classification of chromatin/nucleoplasm proteins leaving aside the classification of the nuclear envelope proteins, which are historically not considered to be a part of chromatin. The SimChrom classification ontology was created by manually analyzing, critically evaluating, selecting and combining into a tree-like classification scheme information from (1) the historically established consensus on chromatin proteins classification (e.g., histone, non-histone proteins, HMG-proteins [1]), (2) classification used in major eterochromatin associated chromosome inacti ation uclear binding roteins uchromatin associated reribosome associated Centromere associated metabolic rocesses tem lated transcri tion ethylated binding binding s istone writers category is in i actors category is in C Cistrome 13, C 1 , C Cancer 1 , C db 9 3 similar category is resent in chromatin/e igenetic regulators hysico chemical ro erties 23 databases of chromatin and epigenetic regulators (e.g., EpiFactors, FACER), (3) classification used in the three aspects of Gene Ontology, (Suppl. Fig. SF3_1). Our hierarchical SimChrom classification is presented in Fig. 3. The majority of classification terms used in SimChrom was inspired by GO-based classification, yet only a small subset of terms was used. The main focus of the classification was to classify chromatin proteins according to their functions and biological processes that they are involved in, but genomic-location (which is also indirectly related to function) and physical properties (e.g., highmobility group proteins of A, B and N families) were also considered (Fig. 3 highlights what classificational aspects are most relevant for each term using a color bar). The SimChrom ontology was developed simultaneously with the SimChrom dataset in an iterative manner by obtaining sets of proteins annotated by various GO terms, extracting them from literature and domain specific databases, manually curating, validating and filtering (see Methods Section 2.2). Only major splice isoforms of genes are included in SimChrom. The resulting SimChrom dataset contains 3045 proteins, is available as a Suppl. Table ST5 and viewable in the Interactive Fig. 3 at the SimChrom web-site (https://simchrom.intbio.org#classification). The descriptive details about the SimChrom dataset are available in Suppl. R&D Sec. 2. In the default SimChrom classification (depicted in Fig. 3) every protein from the SimChrom dataset may belong to more than one SimChrom ontology category. This provides the needed degree of flexibility since many proteins indeed may bona fide belong to several categories due to their complex functional, physico-chemical or structural properties. However, in certain cases of holistic analysis an even simpler classification may be useful, which ascribes every protein to only one category. Such single label classification (SimChrom-SL) based on the same SimChrom ontology was also developed (see Methods Section 2.2, Suppl. Fig. SF3_2). Briefly, if the protein belonged to several categories by default it was ascribed to the category with the least number of other proteins (i.e. the most specific category for this protein) with functional categories taking priority (see Suppl. Fig. SF3_2 for category priority order). As auxiliary datasets based on the results of Section 3.1 we have compiled several reference datasets of nuclear and non-nuclear proteins at different levels of support (depending on whether nuclear localization is supported by one or several localization databases), confidence (depending on the evidence codes and reliability scores provided by the databases), and also whether proteins are uniquely localized in the nucleus or have multiple localization in the nucleus and other cellular compartments (see Methods Section 2.2.2). The list of the datasets and their definition is presented in Suppl. Table ST6, the datasets are available for download in the Interactive Table 2 at https://simchrom.intbio.org/#download. Instrumental to our further analysis will be the “nuclear localization consensus” (NULOC_CS) dataset – the set of nuclear proteins, whose nuclear localization is supported (with sufficiently good confidence levels) both by UniProt and HPA and does not contradict the data from OpenCell, and the “nuclear localization joint dataset with no evidence code 24 filtering” (NULOC_JT_NECF) dataset - the maximally broad set of nuclear proteins, which includes proteins whose nuclear localization is supported by any of the localization databases at any levels of confidence. The NULOC_CS dataset contains 3296 entries, while NULOC_JT_NECF contains 8912 entries. To evaluate the contents of our SimChrom dataset we performed its cross-comparison to the localization based datasets described above (NULOC_CS and NULOC_JT_NECF) (see Suppl. Fig. SF3_3). Detailed discussion of the results is provided in Suppl. R&D Sec. 2. Briefly, almost all SimChrom proteins had some evidence of nuclear localization (95% were present in NULOC_JT_NECF dataset, 60% in NULOC_CS dataset, see Suppl. Fig. SF3_3). For the SimChrom proteins that did not have high confidence support of nuclear localization (non present in NULOC_CS) GO enrichment analysis of SimChrom-exclusive proteins revealed minimal association with nonnuclear functions, with only a minor subset (~10 centromere-associated proteins) linked to such categories (Suppl. Fig. SF3_4, Suppl. Table ST8). Moreover, no additional chromatin-related GO categories were found to be underrepresented in SimChrom, indicating its broad coverage of chromatinassociated functions (Suppl. Table ST9). 60% of SimChrom proteins were successfully identified in MS-based chromatomes and nucleomes (see Suppl. R&D Sec. 1.3, Suppl. Fig. SF2_7). The remaining 40%, predominantly low-abundant transcription factors, were likely undetected due to their transient nature and dynamic interaction properties, which pose challenges for MS-based detection. Furthermore GO enrichment analysis of MS-derived proteins absent in SimChrom or nuclear reference sets did not reveal a bona fide chromatin-associated category (see Suppl. Table ST7). Together, these results support the quality of the SimChrom dataset, suggesting that SimChrom is sufficiently comprehensive in its coverage of chromatin-related proteins and its categories. 3.3. Analysis of the human chromatome Equipped with the datasets described above we aimed at a comprehensive characterization of the chromatome, including characterization of it composition (numbers of proteins belonging to different chromatin categories), abundance (the number of individual protein molecules present in the cells), physico-chemical properties of the amino acid sequences of the proteins, their domain architectures and interaction patterns (including engagement in multivalent interactions). The full discussion of the results is presented in Suppl. R&D Sec. 3 and Suppl. Fig. SF4_1 - SF 4_4, SF5_1 - SF5_5, SF6_1 - SF6_3, SF8_1, SF8_2. The sections below summarize our analysis. 25 3.3.1. The chromatome composition and abundance of chromatin proteins 32 resulting 2D projections onto the main UMAP components revealed that (1) chromatin and cytoplasmic proteins occupied overlapping domains on the 2D map, but with a visible shift between their centers, suggesting there is an overall difference in the average amino acid composition, (2) certain chromatin protein groups formed dedicated clusters on the map, suggesting significant distinctness in their composition (see cluster 1, 2 and outliers shown by the arrows in Fig. 5K,L). Further analysis revealed that in the 2D UMAP map transcription factors, containing zinc finger domains and homedomains formed distinct clusters (see Suppl. Fig. SF5_3A,B). The most distinct group (cluster 1) was almost exclusively (415 out of 422) composed of zinc-finger containing DNA-binding transcription factors (240 housekeeping and 175 non-housekeeping) with the median number of zinc-finger domains (ZFD) of around 10 (Suppl. Fig. SF5_3C). Zinc-finger containing DNA-binding transcription factors were also present in cluster 2, but the median number of zinc-finger domains (ZFD) in that cluster was only three, hence containing a lower proportion of amino acids specific to ZFD (Suppl. Fig. SF5_3D). ZFD are enriched in histidine and cytosine (see Fig. 5J and discussion below). Other protein groups that occupied distinct positions on the UMAP map, included (1) histones, (2) serine/arginine-rich splicing factors (enriched in serine and arginine), and (3) reverse transcriptases of endogenous retroviruses (enriched in isoleucine and threonine) (see Fig. 5K, Suppl. Fig. SF5_3D). The detailed analysis of amino acids composition of different chromatin protein groups is presented in Suppl. R&D Sec. 3.2. Briefly, among the top four enriched amino acids in chromatin proteins are serine, cysteine, proline, and histidine (Fig. 5J). The enrichment of cysteine and histidine is solely contributed by the ZFD of transcription factors (Suppl. Table ST13, Suppl. Fig. SF5_4B, Suppl. Fig. SF5_5A). The total enrichment of serine and proline in chromatin proteins is attributed due to their enrichment in the non-IDR regions (relative to IDR and non-IDR regions of cytoplasmic proteins), and more importantly due to the higher proportion of IDR regions in chromatin proteins (46% vs 23%) that in turn have a considerably higher proportion of these amino acids than non-IDRs (Suppl. Table ST13). Serine was also enriched in non-IDR regions globally, while the enrichment of proline in non-IDRs was observed only in a few categories (e.g., HMG-proteins) (Suppl. Fig. SF5_4H). The enrichment of positively charged amino acids is only statistically significant for lysine, but not for arginine, and the enrichment is relatively moderate (1.03 in chromatin) (Suppl. Fig. SF5_4A). Arginine is highly enriched in non-IDRs, but it is the most depleted amino acid in IDRs of chromatin proteins versus the respective regions of the cytoplasmic ones. The depletion of negatively charged amino acids in chromatin/nuclear proteins is statistically significant for aspartate (fold enrichment is around 0.9), while the depletion of glutamate is statistically non-significant. Interestingly, aspartate is enriched in IDRs and significantly depleted in non-IDRs. This suggests that the increased positive charge of chromatin/nuclear proteins has its main contributions in the depletion of aspartate and enrichment of arginine in non-IDRs, and moderate global enrichment of lysine. Among the most relatively depleted amino acids in chromatin/nucleus are hydrophobic aliphatic amino acids, they are relatively rare in IDRs and hence the large proportion of IDRs in 33 chromatin proteins accounts for their lower total fraction (Suppl. Fig. SF5_4F,I). Tryptophan, which is the rarest amino acid (~1% in proteins), is the most depleted amino acid on average in chromatin/nuclear proteins and almost in all chromatin categories, except for a few. 34 3.3.3. Domain composition of chromatin proteins and identification of new structural domains 35 Figure 6. Domain composition of chromatin proteins and identification of new structural domains. (A) A schematic overview of chromatin proteins’ domain annotation analysis and identification of uncharacterized new domains. Sources of annotation and typical annotation patterns for an abstract protein are schematically outlined. A structure with a novel domain identified using AIbased annotation pipeline implemented in TED resource is shown on the right. (B) Cumulative annotation coverage of all chromatin protein sequences combined at the amino acid level via different resources. Annotation coverage with experimental structures in the PDB database, the AlphaFold database, and three domain annotation databases (Pfam, CATH, TED) is presented. For AlphaFold and PDB additional information about the fraction of annotated amino acids belonging to IDRs and nonIDRs is depicted (see Methods Section 2.4). For all annotations additionally the fraction of amino acids belonging to annotated Pfam domain models in the Pfam database is also depicted, for Pfam and TED additionally the fraction of amino acids resolved in PDB is also depicted. (C) Analysis of the structural domains in chromatin proteins identified by the TED resource via AlphaFold-based algorithm. The number and fractions of structural domains that have matching structures in the PDB database at various levels of sequence identity are depicted. The structural matches were identified via FoldSeek (see Methods Section 2.5). For those domains that were not matched to PDB structures directly a few were annotated by CATH (depicted in orange), the remaining fraction (depicted in magenta) represent novel structural domains present only in TED. (D) Analysis of functional domain diversity in chromatin proteins as identified by the Pfam database. 11147 domains belonging to 1753 Pfam domain models were identified. The plot characterizes domain models with respect to the availability of a matching structure in PDB (the median sequence identities of the matches between the chromatin proteins’ domains belonging to the respective Pfam model and their best structural match in the PDB database as identified by FoldSeek are shown), an annotated TED domain, or otherwise the absence of structural characterization (see Methods Section 2.5). (E) Analysis of functional domain diversity in chromatin proteins as identified by the Pfam database for proteins belonging to different chromatin categories according to SimChrom-SL classification. Subpanels 1-5 represent various characteristics. 36 Figure 7. The most representative protein domains/families (according to Pfam) in proteins belonging to functional SimChrom categories. The dashed rectangles highlight the presence of particular groups of domains in certain categories of chromatin proteins (left - transcription factors and similar DNA-binding proteins, right - various histone interacting and modifying proteins). The plot is based on SimChrom-SL chromatin protein classification; only Pfam domain models present in more than five proteins were considered. Only datapoints with the size of more than 5% are displayed. The full size plots based on both SimChrom and SimChrom-SL classifications are available as Interactive Fig. 4 (https://simchrom.intbio.org/#domain_composition). Next we set out to systematically analyze the available data on structural characterization, domain annotation and domain composition of chromatin proteins. We specifically explored the structurally uncharacterized portion of the chromatome (the “dark” proteome) and identified potential new structural domains that are predicted by AI-based protein structure prediction tools (see Fig. 6A). Historically, protein domains are loosely defined as evolutionary conserved units with similarities at functional, structural and/or sequence levels [72]. Related individual protein domains may be grouped and aligned to produce domain models, catalogued and annotated by a number of resources/databases such as PFAM [73], CDD [74], CATH [75], InterPro [76, p20], etc (see Suppl. R&D Sec. 3.3 for a thorough discussion). The ultimate experimental structural characterization of chromatin proteins is available in the PDB database, however, recent progress in protein structure prediction spurred by AlphaFold resulted in new approaches to the structural characterization and discovery of new structural domains (e.g., as implemented in the TED database used below [52]) (Fig. 6A). Fig. 6B shows the fractions of the aggregate number of amino acids in all human chromatin proteins (referred below to as “aggregate chromatome sequence”, or ACS) which are structurally characterized or have domain annotations in different databases. Detailed discussion is available in 37 Suppl. R&D Sec. 3.3. Briefly, despite recent tremendous progress in structural biology many human chromatin proteins still lack direct structural characterization. On one hand only 25% of ACS can be mapped directly to PDB structures and 25% can be mapped to known structural protein superfamilies (through CATH). On the other hand, AlphaFold 2 identifies 53% of ACS as belonging to non-IDRs, and TED predicts that 35% of ACS belong to domains having well-defined 3D structures. The latter is a conservative estimate of structurally characterizable ACS, since both partially ordered and disordered regions can become ordered in protein-protein complexes. For example, 9% of ACS is available in PDB (where protein complexes are present) while not being annotated with TED domains (which rely on single protein chain structure predictions). This is in line with the fact that among 6246 TED domains found in chromatin proteins almost half of them (42%) are directly covered by the PDB database. However, the majority of other domains (56%) can be matched to a PDB structure of a homologous protein at various levels of sequence identity (from 99% to 5%, see Fig. 6C). The majority of these homologous domains are in fact different paralogous sequences found within human genes (even for domains with sequence identity of 35-50% the fraction of human sequences among the matches was 51%), for matches with sequence identity above 35% the second largest contribution came from structures of mammalian homologues, for matches with sequence identity below 35% significant contributions were from structures derived from proteins of fungi, protostomia and bacteria (see Suppl. Fig. SF6_1A for details). Additionally, 6% of TED domains that lacked direct hits among the PDB structures were mapped to protein structural superfamilies in the CATH database. The remaining 4% (241) represented domains could not be matched to any known protein structure or protein structure superfamilies and potentially represent new types of structural superfamilies/folds. These domains are presented in Suppl. Table ST14 (see also Interactive Table 3 at https://simchrom.intbio.org/#novel_structural_domains), ranked via their structural complexity by the number of their secondary structure elements. Among these domains, 123 domains have annotations in Pfam or other domain annotation databases present in InterPro, leaving 118 domains that are completely without annotations. The latter domains belong to 106 chromatin proteins, which may be considered as prospective new targets for experimental studies of their function and structure. Among such proteins are, for example, (1) a protein encoded by the GTF3C1 gene (it has a previously unannotated and uncharacterized structural domain with a length of 233 amino acids, see detailed characterization in Suppl. Fig. SF6_2A), (2) the globular domain of the testis specific linker histone H1.7, which has a quite different sequence from other H1 proteins resulting in a predicted structure that has a different topology (the “wing” of the globular domain consists of three beta-sheets rather than two [77], see Suppl. Fig. SF6_2B and Suppl. R&D Sec. 3.3). We used the sequence-based Pfam domain annotation to characterize the diversity of different types of evolutionary related protein domains (hereafter referred to as Pfam domain models or Pfam domain types) found in chromatin proteins and typical domain composition thereof. In total 1753 38 different Pfam domain models matched various parts of chromatin proteins (Fig. 6D). 42% of these were considered fully structurally characterized, i.e., every individual domain in chromatin proteins belonging to these models can be found in PDB. 34% of domain models are partially characterized – their domains could be matched to a PDB structure of a homolog (using FoldSeek, see Methods Section 2.5). 14% of these Pfam domain models were not matched by FoldSeek to PDB structures with our strict criteria (see Methods Section 2.5), but could be still identified in PDB via sequence search methods – these represented more flexible domains with IDR regions, repeats and coiled-coils (34 Pfam models), DNA-binding motifs, etc. 3% (55 domain models) could be matched to structural domains predicted by AlphaFold and found in the TED database. These represent prospective targets for validation with structural biology methods and further investigation of their interactions. For instance, among these domain models are domains, potentially associated with chromatin remodeling (SANTA, zf-C3Hc3H), histone PTM writing (DUF7030, COMPASS-Shg1), zinc fingers (zf_CCCH_4, zfLITAF-like, zf-WIZ, SWIM), etc. 7% of Pfam domain models currently have no structural information that can be assigned either through the PDB or TED databases. We next analyzed the diversity of Pfam domain models in various SimChrom-SL protein categories (Fig. 6E, subpanels 1,2) and the domain content of individual proteins belonging to these categories (Fig. 6E, subpanels 3-5). Detailed discussion is available in Suppl. R&D Sec. 3.3. Briefly, the number of distinct Pfam domain models found in chromatin proteins (~1700) is comparable to the number of chromatin proteins (~3000), at the same time an average chromatin protein usually contains two Pfam domains representing two different domain models. The majority of Pfam domain models are present only in a single chromatin protein, but there are also those that are present in dozens or even hundreds of proteins (Suppl. Fig. SF6_1B). Certain chromatin groups stand out in terms of their domain composition in some aspects: the number of individual domains is high in housekeeping TF (due to ZFDs); transcription factors, histones and HMG proteins are relatively poor in their domain diversity (i.e., the proteins in these categories harbour a limited number of distinct Pfam domain models); histone PTM writers on average have domains belonging to three different domain models (while this number is one or two for all others). Still a considerable number of chromatin proteins may harbor domains belonging to several domain models. DNA-acting enzymes, histone PTM writers, chaperones, remodelers, transcription factors may have as much as 8-9 Pfam domain models present in their sequence (see Suppl. Table ST15). There are 118 chromatin proteins harboring at least five different domain types (see Suppl. Fig. SF6_1C, leftpanel). This highlights the multivalency of protein interactions in chromatin, keeping in mind that many proteins further form protein-protein complexes increasing their interaction potential (see next section). The average individual domain length in chromatin proteins is around 65 amino acids (the median is 28 aa), however, this number is biased by the presence of many zinc-finger domains (around 22 aa in length). Subpanel 5 in Fig. 6E gives a more balanced view for each SimChrom category. For the majority of protein groups the median domain 39 length in protein is around 100 amino acids (mean is 137, median is 134). Only 70 chromatin proteins had no domain annotation at all. The birds-eye view of the most frequently matched Pfam domain models in proteins of various functional SimChrom-SL categories is presented in Fig. 7. The data is presented for domain models that occur in at least five chromatin proteins and in at least 10% of proteins in a category (the threshold for data point depiction is 5%). The comprehensive interactive analysis figure with the ability to alter these thresholds and switch between SimChrom and SimChrom-SL classifications systems is available at Interactive Fig. 4 (https://simchrom.intbio.org/#domain_composition). In Fig. 7 the following categories and their respective domains can be grouped revealing their partially shared domain composition: 1) the categories containing transcription factors and their zinc finger, homeodomains and KRAB domains form the most frequently occurring entities, 2) some chromatin regulators, such as PTM writers, readers, erasers and chromatin remodelers together with their Chromo-, Bromo-, and PHD domains. 40 3.3.4. Multivalent interactions in chromatin protein Figure 8. Analysis of multivalent interactions in chromatin proteins. (A) Schematic illustration of multivalent interactions. (B) Distribution of chromatin proteins from the chromatin/epigenetic regulators group with respect to the total number of Pfam domains (right) and distinct Pfam domain models (left), red lines indicate median values. The distributions of all chromatin proteins are shown in Suppl. Fig. SF6_2C. (C) Co-occurrence of domains corresponding to different Pfam domain models in chromatin proteins. Only domains found in proteins belonging to chromatin/epigenetic regulator groups are depicted (see Methods Section 2.5 and Fig. 3). Domains are grouped into several functional classes (see description at the top of the plot). The values indicate the conditional probabilities of a domain in column (A) occurring alongside a domain in row (B) in chromatin proteins. Along the diagonal, data belonging to individual domain groups are highlighted with shading, dashed lines highlight groups associated with histone methylation or acetylation. The following abbreviations are used in the domain subgroup names of the latter: W - writers, R - readers, and E - erasers. (D) Cooccurrence of domains from different functional classes in chromatin proteins and protein complexes. Only combinations that are present in more than one protein or complex are shown, see full version in Suppl. Fig. SF8_2. (E) Examples of domain architectures in chromatin proteins containing the largest number of chromatin/epigenetic regulator domains. The top shows domains from 3D structures colored by their main function; the links between domains are not shown. The bottom shows the order of domains at the sequence level. 41 The presence of multiple domains (belonging to the same or different domain models) in chromatin proteins is a known feature contributing to their ability to engage in multivalent interactions (Fig. 8A) [10]. Below we present the analysis of such domains engaged in multivalent interactions (referred to as EMVI-domains hereafter) that are found in chromatin/epigenetics regulator proteins (see Fig. 3 for definition of this group). These proteins often contain many domains (Fig. 8B). The median total number of domains found in chromatin proteins and in chromatin/epigenetic regulators is two. Nevertheless, many chromatin proteins contain more (16% - have three domains, 10% - four domains, 7% - five, six to fourteen - 10%). There are 409 Pfam domain models that are found in combination with other models or in multiple copies in at least one chromatin regulator protein. To limit our analysis to a manageable set of EMVI-domains, we selected those that were found in multiple copies or in combination with another Pfam domain in at least three chromatin regulator proteins (94 Pfam domain models in total), and from those we selected 59 domain models that we were able to manually classify based on the information currently available in the literature according to their functional binding modes. The following functional groups of domains were used: histone methylation/acetylation/phosphorylation, chromatin remodeling, histone binding, DNA binding, DNA methylation, protein dimerization/oligomerization, PPI, RNA binding. Histone post-translational modifications were further subdivided into readers, writers and erasers functional subgroups (see Fig. 8C and Suppl. Table ST16 for the list of domains and their detailed classification). Domains involved in histone methylation are most present in chromatin regulators, followed by DNA binding, Histone acetylation, Histone phosphorylation and Chromatin remodeling associated domains (Suppl. Fig. SF8_1A). We analyzed the co-occurrence of selected EMVI-domains in all chromatin proteins. There were in total 851 chromatin proteins (589 of these are transcription factors) that had more than one EMVI-domain. The conditional probability of finding a corresponding domain A in a chromatin protein given that another domain B is already present was estimated and is presented in Fig. 8C (columns and rows correspond to domains A and B, respectively). The Interactive Fig. 5 is available at https://simchrom.intbio.org/#domain_co-occurrence (also extends the analysis to unclassified potential EMVI-domains found in at least two chromatin regulator proteins). The matrix in Fig. 8C allows to trace the interplay between different domains employed in architectures of chromatin proteins. The largest groups of domains in Fig. 8C are those involved in histone methylation and DNA binding, suggesting that these mechanisms are the most represented and employed in chromatin functioning regulation. See Suppl. R&D Sec. 3.4 for detailed discussion of the results. Briefly, in certain cases one can see 100% association between the presence of various domains in chromatin proteins. This may be due to direct structural interactions between the domains or likely due to functional reasons. Among the Pfam domains that co-occur with the most number of other different Pfam domains is the PHD domain 48 Taken together we hope that our work establishes a holistic framework for further advances in the field of chromatin research which will help to understand genome functioning though deeper appreciation of the complex role played by the chromatome. ACKNOWLEDGEMENTS We thank A.L. Sivkina, D.K. Malinina, N.S. Gerasimova, A.V. Lyubitelev, S.V. Ulianov, and A.A. Gavrilov for valuable discussions that helped to improve this work. AUTHOR CONTRIBUTIONS AKG: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Visualization. GAA: Resources, Software. MPK: Conceptualization. Writing – review & editing. AKS: Conceptualization, Formal analysis, Funding acquisition, Methodology, Supervision, Validation, Writing – original draft, Writing – review & editing. SUPPLEMENTARY DATA Supplementary material is available online, including supplementary figures, tables, supplementary results and discussion. CONFLICT OF INTEREST Non declared. FUNDING This work was funded by the Russian Science Foundation grant #25-14-00046 (https://rscf.ru/en/project/25-14-00046/) (construction of chromatin protein classification, analysis of chromatin proteins domain composition, AI-based prediction of new structural domains), the Russian Science Foundation grant #23-74-10012 (https://rscf.ru/en/project/23-74-10012/) (analysis of physicochemical properties of chromatin proteins), and within the framework of the Ministry of Science and Higher Education of the Russian Federation project “Whole-Genome Epigenetic Analysis as the Basis for the Development of Genetic Technologies for the Prevention and Treatment of COVID” (FFRW-2023-0007), no. 123120500032-9 (analysis of multivalent interactions of chromatin protein domains). A.K.S. was supported by the HSE University Basic Research Program (structural 49 characterization of chromatin proteins) and A.K.G. was supported by the Gennady Komissarov Foundation (construction of reference datasets about protein localization). DATA AVAILABILITY The SimChrom database including interactive supplementary materials about chromatin proteins’ classification, localization, functions, domain composition are freely available at a GitHub hosted web site https://simchrom.intbio.org/. The SimChrom source code is available at GitHub https://github.com/intbio/SimChrom and archived via Zenodo. REFERENCES 1. Van Holde KE. Chromatin. Springer; 1989. doi:10.1007/978-1-4612-3490-6 2. Bernstein E, Allis CD. RNA meets chromatin. Genes Dev. 2005;19(14):1635-1655. doi:10.1101/gad.1324305 3. Imhof A, Bonaldi T. “Chromatomics” the analysis of the chromatome. Mol BioSyst. 2005;1(2):112-116. doi:10.1039/B502845K 4. Torrente MP, Zee BM, Young NL, et al. Proteomic Interrogation of Human Chromatin. PLOS ONE. 2011;6(9):e24747. doi:10.1371/journal.pone.0024747 5. Ulianov SV, Velichko AK, Magnitov MD, et al. Suppression of liquid-liquid phase separation by 1,6hexanediol partially compromises the 3D genome organization in living cells. Nucleic Acids Res. 2021;49(18):10524-10541. doi:10.1093/nar/gkab249 6. Rippe K. Liquid–Liquid Phase Separation in Chromatin. Cold Spring Harb Perspect Biol. 2022;14(2):a040683. doi:10.1101/cshperspect.a040683 7. Ulianov SV, Khrameeva EE, Gavrilov AA, et al. Active chromatin and transcription play a key role in chromosome partitioning into topologically associating domains. Genome Res. 2016;26(1):70-84. doi:10.1101/gr.196006.115 8. Davidson IF, Peters JM. Genome folding through loop extrusion by SMC complexes. Nat Rev Mol Cell Biol. 2021;22(7):445-464. doi:10.1038/s41580-021-00349-7 9. Kanada R, Terakawa T, Kenzaki H, Takada S. Nucleosome Crowding in Chromatin Slows the Diffusion but Can Promote Target Search of Proteins. Biophys J. 2019;116(12):2285-2295. doi:10.1016/j.bpj.2019.05.007 10. Ruthenburg AJ, Li H, Patel DJ, Allis CD. Multivalent engagement of chromatin modifications by linked binding modules. Nat Rev Mol Cell Biol. 2007;8(12):983-994. doi:10.1038/nrm2298 11. Armeev GA, Gribkova AK, Shaytan AK. NucleosomeDB - a database of 3D nucleosome structures and their complexes with comparative analysis toolkit. bioRxiv. Preprint posted online April 18, 2023:2023.04.17.537230. doi:10.1101/2023.04.17.537230 50 12. Armeev GA, Gribkova AK, Shaytan AK. Nucleosomes and their complexes in the cryoEM era: Trends and limitations. Front Mol Biosci. 2022;9:1070489. doi:10.3389/fmolb.2022.1070489 13. Doronin SA, Ilyin AA, Kononkova AD, et al. Nucleoporin Elys attaches peripheral chromatin to the nuclear pores in interphase nuclei. Commun Biol. 2024;7(1):1-18. doi:10.1038/s42003-024-06495-w 14. Pletenev IA, Bazarevich M, Zagirova DR, et al. Extensive long-range polycomb interactions and weak compartmentalization are hallmarks of human neuronal 3D genome. Nucleic Acids Research. 2024;52(11):6234-6252. doi:10.1093/nar/gkae271 15. Consens ME, Dufault C, Wainberg M, et al. Transformers and genome language models. Nat Mach Intell. 2025;7(3):346-362. doi:10.1038/s42256-025-01007-9 16. Hwang Y, Cornman AL, Kellogg EH, Ovchinnikov S, Girguis PR. Genomic language model predicts protein co-regulation and function. Nat Commun. 2024;15(1):2880. doi:10.1038/s41467-024-46947-9 17. van Mierlo G, Vermeulen M. Chromatin Proteomics to Study Epigenetics - Challenges and Opportunities. Mol Cell Proteomics. 2021;20:100056. doi:10.1074/mcp.R120.002208 18. Kustatscher G, Grabowski P, Rappsilber J. Multiclassifier combinatorial proteomics of organelle shadows at the example of mitochondria in chromatin data. Proteomics. 2016;16(3):393-401. doi:10.1002/pmic.201500267 19. Ohta S, Bukowski-Wills JC, Sanchez-Pulido L, et al. The Protein Composition of Mitotic Chromosomes Determined Using Multiclassifier Combinatorial Proteomics. Cell. 2010;142(5):810-821. doi:10.1016/j.cell.2010.07.047 20. Sinitcyn P, Richards AL, Weatheritt RJ, et al. Global detection of human variants and isoforms by deep proteome sequencing. Nat Biotechnol. 2023;41(12):1776-1786. doi:10.1038/s41587-023-01714-x 21. Guo T, Steen JA, Mann M. Mass-spectrometry-based proteomics: from single cells to clinical applications. Nature. 2025;638(8052):901-911. doi:10.1038/s41586-025-08584-0 22. Wierer M, Mann M. Proteomics to study DNA-bound and chromatin-associated gene regulatory complexes. Hum Mol Genet. 2016;25(R2):R106-R114. doi:10.1093/hmg/ddw208 23. Kustatscher G, Hégarat N, Wills KLH, et al. Proteomics of a fuzzy organelle: interphase chromatin. EMBO J. 2014;33(6):648-664. doi:10.1002/embj.201387614 24. Ugur E, de la Porte A, Qin W, et al. Comprehensive chromatin proteomics resolves functional phases of pluripotency and identifies changes in regulatory components. Nucleic Acids Research. 2023;51(6):26712690. doi:10.1093/nar/gkad058 25. Ginno PA, Burger L, Seebacher J, Iesmantavicius V, Schübeler D. Cell cycle-resolved chromatin proteomics reveals the extent of mitotic preservation of the genomic regulatory landscape. Nat Commun. 2018;9(1):4048. doi:10.1038/s41467-018-06007-5 26. Alabert C, Bukowski-Wills JC, Lee SB, et al. Nascent chromatin capture proteomics determines chromatin dynamics during DNA replication and identifies unknown fork components. Nat Cell Biol. 2014;16(3):281-291. doi:10.1038/ncb2918 27. Shi M, You K, Chen T, et al. Quantifying the phase separation property of chromatin-associated proteins under physiological conditions using an anti-1,6-hexanediol index. Genome Biology. 2021;22(1):229. doi:10.1186/s13059-021-02456-2 28. Alvarez V, Bandau S, Jiang H, et al. Proteomic profiling reveals distinct phases to the restoration of chromatin following DNA replication. Cell Reports. 2023;42(1). doi:10.1016/j.celrep.2023.111996 51 29. Chou DM, Adamson B, Dephoure NE, et al. A chromatin localization screen reveals poly (ADP ribose)- regulated recruitment of the repressive polycomb and NuRD complexes to sites of DNA damage. Proceedings of the National Academy of Sciences. 2010;107(43):18475-18480. doi:10.1073/pnas.1012946107 30. Federation AJ, Nandakumar V, Searle BC, et al. Highly Parallel Quantification and Compartment Localization of Transcription Factors and Nuclear Proteins. Cell Reports. 2020;30(8):2463-2471.e5. doi:10.1016/j.celrep.2020.01.096 31. Dutta B, Ren Y, Hao P, et al. Profiling of the Chromatin-associated Proteome Identifies HP1BP3 as a Novel Regulator of Cell Cycle Progression. Mol Cell Proteomics. 2014;13(9):2183-2197. doi:10.1074/mcp.M113.034975 32. Geladaki A, Kočevar Britovšek N, Breckels LM, et al. Combining LOPIT with differential ultracentrifugation for high-resolution spatial proteomics. Nat Commun. 2019;10(1):331. doi:10.1038/s41467-018-08191-w 33. Wang H, Syed AA, Krijgsveld J, Sigismondo G. Isolation of Proteins on Chromatin Reveals Signaling Pathway–Dependent Alterations in the DNA-Bound Proteome. Molecular & Cellular Proteomics. 2025;24(3). doi:10.1016/j.mcpro.2025.100908 34. Razin SV, Iarovaia OV, Vassetzky YS. A requiem to the nuclear matrix: from a controversial concept to 3D organization of the nucleus. Chromosoma. 2014;123(3):217-224. doi:10.1007/s00412-014-0459-8 35. Ashburner M, Ball CA, Blake JA, et al. Gene Ontology: tool for the unification of biology. Nat Genet. 2000;25(1):25-29. doi:10.1038/75556 36. The Gene Ontology Consortium, Aleksander SA, Balhoff J, et al. The Gene Ontology knowledgebase in 2023. Genetics. 2023;224(1):iyad031. doi:10.1093/genetics/iyad031 37. Gorski S, Misteli T. Systems biology in the cell nucleus. Journal of Cell Science. 2005;118(18):40834092. doi:10.1242/jcs.02596 38. Johnstone CP, Wang NB, Sevier SA, Galloway KE. Understanding and Engineering Chromatin as a Dynamical System across Length and Timescales. Cell Systems. 2020;11(5):424-448. doi:10.1016/j.cels.2020.09.011 39. The UniProt Consortium. UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Research. 2025;53(D1):D609-D617. doi:10.1093/nar/gkae1010 40. Thul PJ, Åkesson L, Wiking M, et al. A subcellular map of the human proteome. Science. 2017;356(6340):eaal3321. doi:10.1126/science.aal3321 41. Cho NH, Cheveralls KC, Brunner AD, et al. OpenCell: Endogenous tagging for the cartography of human cellular organization. Science. 2022;375(6585):eabi6983. doi:10.1126/science.abi6983 42. Binns D, Dimmer E, Huntley R, Barrell D, O’Donovan C, Apweiler R. QuickGO: a web-based tool for Gene Ontology searching. Bioinformatics. 2009;25(22):3045-3046. doi:10.1093/bioinformatics/btp536 43. Marakulina D, Vorontsov IE, Kulakovskiy IV, Lennartsson A, Drabløs F, Medvedeva YA. EpiFactors 2022: expansion and enhancement of a curated database of human epigenetic factors and complexes. Nucleic Acids Research. 2023;51(D1):D564-D570. doi:10.1093/nar/gkac989 44. Lovering RC, Gaudet P, Acencio ML, et al. A GO catalogue of human DNA-binding transcription factors. Biochimica et Biophysica Acta (BBA) - Gene Regulatory Mechanisms. 2021;1864(11):194765. doi:10.1016/j.bbagrm.2021.194765 45. Itzhak DN, Tyanova S, Cox J, Borner GH. Global, quantitative and dynamic mapping of protein subcellular localization. Hegde RS, ed. eLife. 2016;5:e16950. doi:10.7554/eLife.16950 52 46. Uhlén M, Fagerberg L, Hallström BM, et al. Tissue-based map of the human proteome. Science. 2015;347(6220):1260419. doi:10.1126/science.1260419 47. Balu S, Huget S, Medina Reyes JJ, et al. Complex portal 2025: predicted human complexes and enhanced visualisation tools for the comparison of orthologous and paralogous complexes. Nucleic Acids Res. 2025;53(D1):D644-D650. doi:10.1093/nar/gkae1085 48. Raudvere U, Kolberg L, Kuzmin I, et al. g:Profiler: a web server for functional enrichment analysis and conversions of gene lists (2019 update). Nucleic Acids Res. 2019;47(W1):W191-W198. doi:10.1093/nar/gkz369 49. Wang M, Herrmann CJ, Simonovic M, Szklarczyk D, von Mering C. Version 4.0 of PaxDb: Protein abundance data, integrated across model organisms, tissues, and cell‐lines. Proteomics. 2015;15(18):3163-3168. doi:10.1002/pmic.201400441 50. Akdel M, Pires DEV, Pardo EP, et al. A structural biology community assessment of AlphaFold2 applications. Nat Struct Mol Biol. 2022;29(11):1056-1067. doi:10.1038/s41594-022-00849-w 51. Bordin N, Sillitoe I, Nallapareddy V, et al. AlphaFold2 reveals commonalities and novelties in protein structure space for 21 model organisms. Commun Biol. 2023;6(1):160. doi:10.1038/s42003-023-04488-9 52. Lau AM, Bordin N, Kandathil SM, et al. Exploring structural diversity across the protein universe with The Encyclopedia of Domains. Science. Published online November 1, 2024. doi:10.1126/science.adq4946 53. van Kempen M, Kim SS, Tumescheit C, et al. Fast and accurate protein structure search with Foldseek. Nat Biotechnol. Published online May 8, 2023:1-4. doi:10.1038/s41587-023-01773-0 54. Gligorijević V, Renfrew PD, Kosciolek T, et al. Structure-based protein function prediction using graph convolutional networks. Nat Commun. 2021;12(1):3168. doi:10.1038/s41467-021-23303-9 55. Lambert SA, Jolma A, Campitelli LF, et al. The Human Transcription Factors. Cell. 2018;172(4):650665. doi:10.1016/j.cell.2018.01.029 56. Draizen EJ, Shaytan AK, Mariño-Ramírez L, Talbert PB, Landsman D, Panchenko AR. HistoneDB 2.0: a histone database with variants—an integrated resource to explore histones and their variants. Database. 2016;2016:baw014. doi:10.1093/database/baw014 57. Zhang Y, Zhang Y, Song C, et al. CRdb: a comprehensive resource for deciphering chromatin regulators in human. Nucleic Acids Research. 2023;51(D1):D88-D100. doi:10.1093/nar/gkac960 58. Hammond CM, Strømme CB, Huang H, Patel DJ, Groth A. Histone chaperone networks shaping chromatin function. Nat Rev Mol Cell Biol. 2017;18(3):141-158. doi:10.1038/nrm.2016.159 59. Reeves R. High mobility group (HMG) proteins: Modulators of chromatin structure and DNA repair in mammalian cells. DNA Repair. 2015;36:122-136. doi:10.1016/j.dnarep.2015.09.015 60. Mayran A, Drouin J. Pioneer transcription factors shape the epigenetic landscape. J Biol Chem. 2018;293(36):13795-13804. doi:10.1074/jbc.R117.001232 61. Sun H, Fu B, Qian X, Xu P, Qin W. Nuclear and cytoplasmic specific RNA binding proteome enrichment and its changes upon ferroptosis induction. Nat Commun. 2024;15(1):852. doi:10.1038/s41467-02444987-9 62. Van Nostrand EL, Freese P, Pratt GA, et al. A large-scale binding and functional map of human RNAbinding proteins. Nature. 2020;583(7818):711-719. doi:10.1038/s41586-020-2077-3 63. Azad GK, Swagatika S, Kumawat M, Kumawat R, Tomar RS. Modifying Chromatin by Histone Tail Clipping. Journal of Molecular Biology. 2018;430(18, Part B):3051-3067. doi:10.1016/j.jmb.2018.07.013 53 64. Lee H, Noh H, Ryu JK. Structure-function relationships of SMC protein complexes for DNA loop extrusion. BioDesign. 2021;9(1):1-13. doi:10.34184/kssb.2021.9.1.1 65. Cartwright P, Helin K. Nucleocytoplasmic shuttling of transcription factors. Cell Mol Life Sci. 2000;57(89):1193-1206. doi:10.1007/pl00000759 66. Cheng D, Semmens K, McManus E, et al. The nuclear transcription factor, TAF7, is a cytoplasmic regulator of protein synthesis. Science Advances. Published online December 2021. doi:10.1126/sciadv.abi5751 67. Shreberk-Shaked M, Oren M. New insights into YAP/TAZ nucleo-cytoplasmic shuttling: new cancer therapeutic opportunities? Mol Oncol. 2019;13(6):1335-1341. doi:10.1002/1878-0261.12498 68. Kobiyama K, Kawashima A, Jounai N, et al. Role of Extrachromosomal Histone H2B on Recognition of DNA Viruses and Cell Damage. Front Genet. 2013;4:91. doi:10.3389/fgene.2013.00091 69. Zeng Z, Chen L, Luo H, Xiao H, Gao S, Zeng Y. Progress on H2B as a multifunctional protein related to pathogens. Life Sciences. 2024;347:122654. doi:10.1016/j.lfs.2024.122654 70. Sigismondo G, Papageorgiou DN, Krijgsveld J. Cracking chromatin with proteomics: From chromatome to histone modifications. PROTEOMICS. 2022;22(15-16):2100206. doi:10.1002/pmic.202100206 71. Huang Q, Szklarczyk D, Wang M, Simonovic M, von Mering C. PaxDb 5.0: curated protein quantification data suggests adaptive proteome changes in yeasts. Molecular & Cellular Proteomics. Published online August 31, 2023:100640. doi:10.1016/j.mcpro.2023.100640 72. Vogel C, Bashton M, Kerrison ND, Chothia C, Teichmann SA. Structure, function and evolution of multidomain proteins. Current Opinion in Structural Biology. 2004;14(2):208-216. doi:10.1016/j.sbi.2004.03.011 73. Paysan-Lafosse T, Andreeva A, Blum M, et al. The Pfam protein families database: embracing AI/ML. Nucleic Acids Res. 2025;53(D1):D523-D534. doi:10.1093/nar/gkae997 74. Wang J, Chitsaz F, Derbyshire MK, et al. The conserved domain database in 2023. Nucleic Acids Res. 2023;51(D1):D384-D388. doi:10.1093/nar/gkac1096 75. Waman VP, Bordin N, Alcraft R, et al. CATH 2024: CATH-AlphaFlow Doubles the Number of Structures in CATH and Reveals Nearly 200 New Folds. Journal of Molecular Biology. 2024;436(17):168551. doi:10.1016/j.jmb.2024.168551 76. Blum M, Andreeva A, Florentino LC, et al. InterPro: the protein sequence classification resource in 2025. Nucleic Acids Res. 2025;53(D1):D444-D456. doi:10.1093/nar/gkae1082 77. Lyubitelev AV, Nikitin DV, Shaytan AK, Studitsky VM, Kirpichnikov MP. Structure and functions of linker histones. Biochemistry (Moscow). 2016;81(3):213-223. doi:10.1134/S0006297916030032 78. Wiener N. Cybernetics. Scientific American. 1948;179(5):14-19. 79. Shah SG, Mandloi T, Kunte P, et al. HISTome2: a database of histone proteins, modifiers for multiple organisms and epidrugs. Epigenetics & Chromatin. 2020;13(1):31. doi:10.1186/s13072-020-00354-8 80. Wiśniewski JR, Hein MY, Cox J, Mann M. A “Proteomic Ruler” for Protein Copy Number and Concentration Estimation without Spike-in Standards*. Molecular & Cellular Proteomics. 2014;13(12):3497-3506. doi:10.1074/mcp.M113.037309 81. Palii CG, Cheng Q, Gillespie MA, et al. Single-Cell Proteomics Reveal that Quantitative Changes in Coexpressed Lineage-Specific Transcription Factors Determine Cell Fate. Cell Stem Cell. 2019;24(5):812820.e5. doi:10.1016/j.stem.2019.02.006 54 82. Baskar R, Chen AF, Favaro P, et al. Integrating transcription-factor abundance with chromatin accessibility in human erythroid lineage commitment. Cell Rep Methods. 2022;2(3):100188. doi:10.1016/j.crmeth.2022.100188 83. Shinohara K, Toné S, Ejima T, Ohigashi T, Ito A. Quantitative Distribution of DNA, RNA, Histone and Proteins Other than Histone in Mammalian Cells, Nuclei and a Chromosome at High Resolution Observed by Scanning Transmission Soft X-Ray Microscopy (STXM). Cells. 2019;8(2):164. doi:10.3390/cells8020164 84. Hock R, Furusawa T, Ueda T, Bustin M. HMG chromosomal proteins in development and disease. Trends in Cell Biology. 2007;17(2):72-79. doi:10.1016/j.tcb.2006.12.001 85. Holehouse AS, Alberti S. Molecular determinants of condensate composition. Molecular Cell. 2025;85(2):290-308. doi:10.1016/j.molcel.2024.12.021 86. Miao J, Chong S. Roles of intrinsically disordered protein regions in transcriptional regulation and genome organization. Current Opinion in Genetics & Development. 2025;90:102285. doi:10.1016/j.gde.2024.102285 87. Zanzoni A, Ribeiro DM, Brun C. Understanding protein multifunctionality: from short linear motifs to cellular functions. Cell Mol Life Sci. 2019;76(22):4407-4412. doi:10.1007/s00018-019-03273-4 88. Kumar M, Michael S, Alvarado-Valverde J, et al. ELM—the Eukaryotic Linear Motif resource—2024 update. Nucleic Acids Res. 2024;52(D1):D442-D455. doi:10.1093/nar/gkad1058 89. Ghitti M, Colley LS, Mantonico MV, Musco G, Bianchi ME. Intrinsic disorder and fuzzy interactions drive multiple functions of HMGB1. Trends in Biochemical Sciences. Published online September 1, 2025. doi:10.1016/j.tibs.2025.08.001 90. Hatos A, Monzon AM, Tosatto SCE, Piovesan D, Fuxreiter M. FuzDB: a new phase in understanding fuzzy interactions. Nucleic Acids Res. 2022;50(D1):D509-D517. doi:10.1093/nar/gkab1060 91. Jonas F, Navon Y, Barkai N. Intrinsically disordered regions as facilitators of the transcription factor target search. Nat Rev Genet. 2025;26(6):424-435. doi:10.1038/s41576-025-00816-3 92. Már M, Nitsenko K, Heidarsson PO. Multifunctional Intrinsically Disordered Regions in Transcription Factors. Chemistry. 2023;29(21):e202203369. doi:10.1002/chem.202203369 93. Sabari BR, Dall’Agnese A, Young RA. Biomolecular Condensates in the Nucleus. Trends in Biochemical Sciences. 2020;45(11):961-977. doi:10.1016/j.tibs.2020.06.007 94. Requião RD, Fernandes L, Souza HJA de, Rossetto S, Domitrovic T, Palhano FL. Protein charge distribution in proteomes and its impact on translation. PLOS Computational Biology. 2017;13(5):e1005549. doi:10.1371/journal.pcbi.1005549 95. Fisher RS, Elbaum-Garfinkle S. Tunable multiphase dynamics of arginine and lysine liquid condensates. Nat Commun. 2020;11(1):4628. doi:10.1038/s41467-020-18224-y 96. Hong Y, Najafi S, Casey T, Shea JE, Han SI, Hwang DS. Hydrophobicity of arginine leads to reentrant liquid-liquid phase separation behaviors of arginine-rich proteins. Nat Commun. 2022;13(1):7326. doi:10.1038/s41467-022-35001-1 97. Dang M, Li T, Zhou S, Song J. Arg/Lys-containing IDRs are cryptic binding domains for ATP and nucleic acids that interplay to modulate LLPS. Commun Biol. 2022;5(1):1315. doi:10.1038/s42003-022-04293-w 98. Amaro RE, Åqvist J, Bahar I, et al. The need to implement FAIR principles in biomolecular simulations. Nat Methods. 2025;22(4):641-645. doi:10.1038/s41592-025-02635-0 55 99. Armeev GA, Kniazeva AS, Komarova GA, Kirpichnikov MP, Shaytan AK. Histone dynamics mediate DNA unwrapping and sliding in nucleosomes. Nat Commun. 2021;12. doi:10.1038/s41467-021-22636-9 100. Fedulova AS, Armeev GA, Romanova TA, et al. Molecular dynamics simulations of nucleosomes are coming of age. WIREs Computational Molecular Science. 2024;14(4):e1728. doi:10.1002/wcms.1728 101. Kilgore HR, Chinn I, Mikhael PG, et al. Protein codes promote selective subcellular compartmentalization. Science. Published online March 7, 2025. doi:10.1126/science.adq2634 102. Yang X, Zhu H, Shi L, et al. AlphaFold-guided structural analyses of nucleosome binding proteins. Nucleic Acids Res. 2025;53(14):gkaf735. doi:10.1093/nar/gkaf735 103. Lim Y, Tamayo-Orrego L, Schmid E, et al. In silico protein interaction screening uncovers DONSON’s role in replication initiation. Science. 2023;381(6664):eadi3448. doi:10.1126/science.adi3448 104. Ruthenburg AJ, Li H, Patel DJ, David Allis C. Multivalent engagement of chromatin modifications by linked binding modules. Nat Rev Mol Cell Biol. 2007;8(12):983-994. doi:10.1038/nrm2298 Supplementary materials (Re)defining the human chromatome: an integrated meta-analysis of localization, function, abundance, physical properties and domain composition of chromatin proteins Anna K. Gribkova1,2, Grigoriy A. Armeev1,2, Mikhail P. Kirpichnikov1,3, Alexey K. Shaytan1,2,4* 1 Department of Biology, Lomonosov Moscow State University, Moscow, Russia 2 Vavilov Institute of General Genetics, Moscow, Russia 3 Shemyakin–Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences, Moscow, Russia 4 International Laboratory of Bioinformatics, AI and Digital Sciences Institute, Faculty of Computer Science, HSE University, Moscow, Russia * To whom correspondence should be addressed. Email: [email protected] 2 Table of contents Supplementary Tables ............................................................................................................................................ 3 Supplementary Figures ........................................................................................................................................... 5 1. Sources of information about chromatin and nuclear proteins and their critical evaluation ......................... 5 2. The SimChrom chromatin protein classification, the SimChrom dataset and other reference datasets ...... 14 3. Analysis of the human chromatome ............................................................................................................. 18 3.1. The chromatome composition and abundance of chromatin proteins ....................................................... 18 3.2. Physico-chemical properties and amino acid composition ........................................................................ 22 3.3. Domain composition of chromatin proteins and identification of novel structural domains ..................... 27 3.4. Multivalent interactions in chromatin proteins .......................................................................................... 30 Supplementary Results and Discussion ................................................................................................................ 32 1. Sources of information about chromatin and nuclear proteins and their critical evaluation ............................. 32 1.1. Analysis of chromatin proteins’ representation in the GO database and other protein-function oriented databases .......................................................................................................................................................... 33 1.2. Detailed comparative analysis of nuclear proteins subcellular localization between UniProt, HPA, and OpenCell .......................................................................................................................................................... 35 1.3. Detailed comparative analysis of sets of chromatin proteins identified in MS-based studies ................... 39 2. The SimChrom chromatin protein classification, the interactive SimChrom database and other reference datasets ................................................................................................................................................................. 44 3. Analysis of the human chromatome ................................................................................................................. 46 3.1. The chromatome composition and abundance of chromatin proteins ....................................................... 46 3.2. Detailed analysis of the physico-chemical properties and amino acid composition .................................. 52 3.3. Detailed analysis of the domain composition of chromatin proteins and identification of new structural domains ............................................................................................................................................................ 58 3.4. Detailed analysis of the multivalent interactions in chromatin proteins .................................................... 63 References ............................................................................................................................................................ 68 9 Supplementary Figure SF2_5. The MS-based human chromatome (and nucleome) datasets examined through the lens of annotations provided by the localization databases and the SimChrom chromatin protein classification. (A) The list of MS-based datasets of chromatin/nuclear proteins from the respective studies analyzed in this work together with a short description of the experimental/analysis workflow (left) and the plots (right) showing the size of the datasets and the fractions of the datasets that overlap with the SimChrom dataset or nuclear localization datasets (NULOC_CS and NULOC_JT_NECF). The median values are shown by dotted in the plots. (B) Cumulative fraction of proteins identified by at least N MS-based chromatin studies relative to the total number of chromatin proteins identified in at least one MS-based study. (C) Overlap of protein entries identified in MS-based studies that lack nuclear localization according to the database annotations (NULOC_JT_NECF dataset). (D) Number of protein entries identified in MS-based studies that have no annotations in the databases. Left: proteins lacking localization data in both UniProt and HPA. Right: proteins lacking both localization data and GO annotation. 10 Supplementary Figure SF2_6. GO enrichment analysis for protein entries from MS-based chromatome datasets that are absent from both the NULOC_JT_NECF dataset and the SimChrom classification (n = 2232). 11 Supplementary Figure SF2_7. (A) The Venn diagram showing overlaps between the set of SimChrom proteins, the housekeeping proteome and the combined protein set of MS-based chromatome datasets (union of protein content from eight MS-based studies analyzed in this work). (B) The percentage of housekeeping (HK) proteins in MS-based chromatomes and nucleome (range: 76% - 90%). (C) The percentage of SimChrom proteins by SimChrom category in MS-based chromatomes and nucleome. 12 Supplementary Figure SF2_8. (A) Fold enrichment of chromatin-associated proteins identified in MS-based studies for every SimChrom category (fold enrichment is calculated with respect to the distribution of proteins among the categories in SimChrom). Only statistically significant values (pvalue of Fisher exact test with Benjamini correction < 0.05) are shown. (B) Number of histone proteins detected in MS-based studies compared to the reference counts from MS_HistoneDB and HistoneDB 2.0. 13 Supplementary Figure SF2_9. The overlap of chromatin/nuclear protein entries from different types of sources: protein function databases (GO "Chromatin"), protein localization DBs (Uniprot Nucleus, HPA Nucleus or Nucleoplasm), MS-based studies of chromatome and nucleome proteins (two protein sets are used - see legends: 1. union of protein entries from five total chromatin studies, two nascent chromatin and one nucleome; 2. protein entries that are present in three out of seven MS-based chromatin datasets). The “background” set of proteins for each panel is shown in italic. 14 2. The SimChrom chromatin protein classification, the SimChrom dataset and other reference datasets Supplementary Figure SF3_1. The scheme of creation of the SimChrom classification ontology and SimChrom protein dataset is shown in panels (A) and (B), respectively. 15 Supplementary Figure SF3_2. The order of SimChrom categories (from top to bottom) used to create the single-label SimChrom-SL classification of chromatin proteins. The categories were ordered as follows: molecular function and physicochemical properties were placed first, followed by the others. Among them, categories containing fewer proteins were ordered earlier. The number of proteins belonging to the respective SimChrom and SimChrom-SL categories is also shown. 16 Supplementary Figure SF3_3. SimChrom protein analysis using protein localization information and MS-based chromatomes and nucleome. (A-B) A Venn diagram showing the overlap between the protein sets: SimChrom, proteins identified in MS-based studies and the reference nuclear protein sets NULOC_CS (A) or NULOC_JT_NECF (B). (C) The percentage of proteins from SimChrom categories that are found in NULOC_CS and NULOC_JT_NECF datasets. 17 Supplementary Figure SF3_4. SimChrom protein dataset analysis using protein localization information. (A) The number of SimChrom-SL classified proteins without nuclear localization according to NULOC_JT_NECF (the broadest dataset that combined all nuclear protein entries from all protein localization databases at any level of confidence). (B) The unexpected enriched GO terms for proteins were identified for the SimChrom proteins that are absent in NULOC_CS. 18 3. Analysis of the human chromatome 3.1. The chromatome composition and abundance of chromatin proteins Supplementary Figure SF4_1. Chromatin proteins abundance analysis. (A, B) Distribution of proteins from PaxDb_INT and PaxDb_PA datasets according to their relative abundance values. The distribution for housekeeping (HK) and non-housekeeping (non-HK) are also shown (see legend). The distribution was constructed by taking the logarithm of the abundance values in ppm, making a histogram (bin size of 0.15) and smoothing it with a gaussian kernel for visual clarity. (C) Fraction distribution of lowabundant (LA) and high-abundant (HA) housekeeping (HK) and non-housekeeping (non-HK) proteins in the whole proteome (PaxDb_INT), protein localization datasets (NULOC_CS and NULOC_JT), and SimChrom and MS-based chromatomes. (D) The distribution of low-abundant (LA) housekeeping (HK) proteins among SimChrom-SL categories. 25 Supplementary Figures SF5_4. Comparison of amino acid composition between chromatin and uniquely localized nuclear proteins relative to cytoplasmic proteins (SimChrom, NULOC_CS_UL, CYLOC_CS_UL datasets). The comparison is done separately for the total protein sequence (A,B,C), IDRs (D,E,F), and non-IDRs (G,H,I). Subplots (A, D, G) presents the median fractions of amino acids for chromatin and nuclear proteins (subpanel 1 on each plot), the fold enrichment (FE) of these fractions relative to the cytoplasmic proteins (subpanel 2 on each plot), the black line indicates FE = 1. The adjusted p-value is shown for the statistical tests (Mann-Whitney test) comparing the median values of amino acid fractions for chromatin and nuclear proteins with the cytoplasmic ones (subpanel 3 on each plot). Gray highlights indicate a lack of statistical significance (adj. p-value > 0.05). Detailed analysis of the distribution of the selected amino acids in the total sequence of proteins belonging to respective SimChrom-SL protein categories is presented in panels (B-I): enriched amino acids in chromatin proteins are shown in panels (B,E,H), depleted - in panels (C,F,I). In the top of each plot (the first three rows) the following datapoints for the fold enrichment are given: “Total” - for all proteins from SimChrom or NULOC_CS_UL datasets (the latter also depicted by dashed line), “Common” – for common proteins among SimChrom and NULOC_CS_UL datasets, “Not common” – for proteins not present in the partner dataset (e.g., for SimChrom those present in SimChrom but absent in NULOC_CS_UL will be depicted, and vice versa for Nuclear_UL dataset). 26 Supplementary Figures SF5_5. Additional comparisons of amino acids composition for different protein groups. (A) Comparison of amino acid composition in chromatin proteins and chromatin proteins without zf-C2H2 containing proteins relative to cytoplasmic proteins. The median fraction of amino acids for protein subsets (subpanel 1), the fold enrichment (FE) relative to the cytoplasmic proteins (subpanel 2), where the black line indicates FE = 1. The adjusted p-value is shown for the statistical tests comparing the median values of chromatin and chromatin proteins that lack zinc-finger domains with the cytoplasmic ones (subpanel 3). Gray shading indicates values that lack statistical significance (adj. p-value > 0.05). (B) Fold enrichment of amino acids’ median fractions in chromatin proteins vs uniquely localized cytoplasmic ones, total sequences, IDRs and non-IDRs were analyzed separately. (C) Median value of amino acids’ fractions in chromatin proteins for total protein sequences, IDRs and non-IDRs. 27 3.3. Domain composition of chromatin proteins and identification of novel structural domains Supplementary Figures SF6_1. (A) Taxonomic distribution of source organisms for PDB structures with domains homologous to chromatin proteins (taxon of the best match to the structural domains identified by the TED resource, see Figure 6, Methods). (B) The histogram showing how many Pfam domain models (Y-axis) are found in exactly N (X-axis) chromatin proteins. One can see that the majority of Pfam domain models are represented only by domains found in one chromatin protein. The red line indicates median values. (C) The distribution of chromatin proteins according to the total number of Pfam domains identified in proteins (see also Supplementary Table ST15). (D) The distribution of proteins according to the number of zf-C2H2 domains in Housekeeping and Nonhousekeeping DNA-binding transcription factors (HK TFs and Non-HK TFs, respectively). The lines indicate median values (7 and 9). (E) Analysis of functional domain diversity in chromatin proteins as identified by the Pfam database for proteins belonging to different chromatin categories according to SimChrom classification. Subpanel 1-5 represent various characteristics. This is the same as Figure 6E but SimChrom classification instead of SimChrom-SL classification is used. 28 Supplementary Figures SF6_2. The examples of novel structural domains identified in chromatin proteins: structures, colored by TED domain annotation and AlphaFold2 pLDDT score, and its annotation in InterPro (screenshot). (A) General transcription factor 3C polypeptide 1 (gene GTF3C1, protein Q12789). (B) Testis-specific H1 histone (gene H1-7, protein Q75WM6). 29 Supplementary Figures SF6_3. Predictions of GO molecular function (MF) (panel A) and biological processes (BP) terms (panel B) for novel structural domains without information in other DBs according to InterPro. 30 3.4. Multivalent interactions in chromatin proteins Supplementary Figures SF8_1. (A) The number of chromatin regulator proteins that contain EMVIdomains of certain groups. (B) The number of chromatin proteins with different numbers of EMVIdomains belonging to different groups ('DNA binding' domain functional group is not shown). (C) Cooccurrence of EMVI-domains belonging to different functional groups in chromatin proteins. The values indicate the estimated conditional probabilities to find in a chromatin protein a domain specified in the column name given that a domain specified in the row name is already present. 31 Supplementary Figures SF8_2. The UpSet plot shows combinations of EMVI domains classified by their functional groups/subgroups in chromatin proteins (panel A) and protein complexes that exclusively contain chromatin proteins (panel B). 32 Supplementary Results and Discussion 1. Sources of information about chromatin and nuclear proteins and their critical evaluation This section includes supplementary results and discussion to section 3.1. Sources of information about chromatin and nuclear proteins and their critical evaluation in the main text. A note on the distinction between and definition of nuclear proteome and chromatome Nuclear proteome and chromatome are two terms that are historically used to describe the protein content of the nucleus and the proteins associated with genome packaging, maintenance and functioning (see Figure 1A for modern view of nucleus structure). The exact distinction between these two terms may be fuzzy and is often based on different consensus (protein localization or functional classification ontologies) or operational (experimental based extraction techniques) definitions. During interphase when the nucleus envelope is intact the chromatin proteins obviously reside inside the nucleus and are a part of the nuclear proteome. Hence, terminologically the nucleome seems to be more straightforwardly defined just by the protein contents of the nucleus. However, during mitosis and meiosis once the nucleus disintegrates as a distinct organelle, the situation becomes more complex. During these stages of the cell cycle there are no nuclear proteins per se while chromatin proteins can still be defined as those associated with the DNA in chromosomes. Another debatable question is whether all of the proteins inside the nucleus can be considered chromatin proteins (even if nuclear envelope proteins are set aside). According to one modern view apart from the chromatin compartment the nucleus contains also interchromatin compartments [2] and nuclear bodies enriched in RNA and protein complexes (e.g., nucleolus, nuclear speckles). Historically the soluble fraction of nuclear proteins was attributed to nucleosol or nuclear sap. However, to say that proteins localized in these compartments do not interact with genomic DNA at least transiently would be an oversimplification. Nucleoplasm proteins mostly also interact with genomes, some parts of the genome interact with the nucleolus (the so-called, nucleolar associated domains or NADs). Even proteins of the nuclear envelope – lamins do interact with the genomic DNA (e.g., forming laminaassociated domains, LADs). 33 1.1. Analysis of chromatin proteins’ representation in the GO database and other proteinfunction oriented databases The most comprehensive gene and gene products classification resource to date is GeneOntology (GO), which classifies proteins according to the three interrelated ontologies describing molecular function, biological processes, and cellular components (called aspects). GO annotates 97% of all proteins of the human reference proteome (as provided by UniProt), but this classification also has several drawbacks. GO combines many categories (currently around 42 thousand terms) of various scope describing various aspects of gene functioning connected via different types of relationships (such as “A is B”, “A is part of B”, “A regulates B”, “A occurs in B”, etc.) in a non-treelike structure (directed acyclic graph). This complex intertwined hierarchy of GO makes it difficult to get a holistic picture of various chromatin protein groups, and apply a reductionist way of thinking while interpreting the results of bioinformatics analysis of protein sets made using GO classification. Another drawback is that GO omits categories that are historically well established in the community of chromatin researchers (e.g., such categories as “histone proteins”, “high-mobility group proteins”, etc.), again hampering interpretation of GO-based data analysis using the established knowledge (see discussion below). The GO cellular component term "chromatin" is defined broadly as "the ordered and organized complex of DNA, protein, and sometimes RNA, that forms the chromosome." Consequently, functionally relevant chromatin terms — such as "nucleosomal DNA binding", "DNA-binding transcription factor activity', 'Histone H3K27 DNA-binding transcription factor activity', 'Nucleus', 'Histone H3K27 monomethyltransferase activity" — may not be linked hierarchically to the chromatin GO node, resulting in incomplete overlaps between protein lists. Based on our analysis, over 500 functionally defined chromatin proteins (inferred from literature and other sources) are absent from GO annotations (see Supplementary Figure SF2_1A). While the GO provides nearly comprehensive coverage of the proteome, its classification structure presents challenges for extracting or comparing specific protein sets. The GO hierarchy is complex, with heterogeneous relationships between nodes, overlapping protein annotations across terms, and varying levels of detail and completeness. For instance, manually inspecting GO terms associated with chromatin-related keywords (e.g., "DNA", "transcription", "histone", "RNA polymerase") is impractical due to their sheer volume (>100 terms; see Supplementary Figure SF2_1B). Furthermore, GO terms inherently include proteins from all child terms, which can lead to unintended inclusions. For example: the term "DNA-templated" transcription incorporates "mitochondrial transcription"; "gene expression" encompasses functionally distinct processes like "protein maturation" and "translation". 34 When compared with the EpiFactors database [3] (containing epigenetic regulator protein entries obtained by text mining) 46% of entries in EpiFactors are missing from the list of GO 'chromatin' proteins, see Figure 2D. Moreover, while a specialized review by Hammond et al., 2017 [1] lists 35 proteins of histone chaperone category, the GO term "histone chaperone activity" includes only 14, with just 6 overlapping entries (see Supplementary Figure SF2_1D). Several functionally important but small chromatin protein categories are entirely missing from GO, including: HMG proteins, Histone tail cleavage proteins, general TFs. Even for well-annotated classes like histones, inconsistencies persist. Most are classified under "structural constituent of chromatin", but this term also includes two non-histone proteins (HMGA1 and LMNTD2). The Gene Ontology offers the most comprehensive coverage, annotating nearly the entire human reference proteome. In contrast, chromatin/epigenetic regulator databases typically include 400– 800 proteins, while protein class-specific databases vary significantly in scope, ranging from as few as 30 proteins (e.g., chromatin remodelers of the SWI/SNF family) to over 1500 (e.g., transcription factors). Despite their utility, none of these resources fully captures the complexity of chromatinassociated proteins, either in terms of protein coverage or functional classification. While chromatin/epigenetic regulator databases encompass key cofactors for certain protein complexes, they exclude critical categories such as transcription factors, RNA polymerase subunits, DNA-modifying enzymes, DNA repair machinery, and HMG proteins. Conversely, protein class-specific databases are limited to at most six functional categories in total, including histones, chromatin remodelers (from select families), histone post-translational modification (PTM) writers/readers/erasers, and transcription factors. Additionally, several chromatin-related protein classes have been reviewed as the gene group in HGNC (e.g., 'High mobility group', 'DNA polymerases') or in the literature but lack dedicated database resources. Examples include histone chaperones [1], SMC complexes [4], HMG proteins [5], pioneer transcription factors [6], nuclear RNA-binding proteins [7,8], and histone tail cleavage enzymes [9]. The full list of chromatin-associated protein class specific sources are available in Supplementary Table ST1. The absence of centralized repositories for these protein classes highlights a critical gap in current bioinformatics resources. The revealed problems for using GO directly for extracting a set of chromatin proteins arise from several factors (see also [10] for a broader discussion of GO applicability): 1) protein multifunctionality: many proteins participate in diverse processes or localize to multiple compartments, 2) ambiguous term definitions: brief GO descriptions may lead to inconsistent interpretations, 3) annotation delays and errors: lag times in updates and propagation of errors in curated datasets, 4) curation bias: well-studied proteins are annotated more thoroughly than niche categories. 41 The MS-based datasets showed high variability, only 179 proteins were in common between the chromatin datasets only (see Supplementary Figure SF2_5B). The largest MS-based datasets of total chromatin (Shi et al., 2021; Ginno et al., 2018) and nascent chromatin (Alabert et al, 2014; Alvarez et al., 2023) contained around three thousand proteins – approximately the same amount as the number of proteins in SimChrom. However, surprisingly the number of proteins for these datasets that were present in SimChrom was small (25-35%). This low consistency with SimChrom was also observed in the Torrenete et al., 2011 dataset. These three datasets (Shi et al., 2021; Ginno et al., 2018; Torrente et al., 2011) were based purely on experimental chromatin extraction techniques (Supplementary Figure SF2_5A). Alternatively, the consistency with SimChrom was twice as high (50-65%) for the Kustatscher et al., 2014 and Itzhak et al., 2016 datasets. These datasets stand out from the other datasets. Kustatscher et al., 2014 used a machine learning classification approach based on MS-data signals with a manually provided chromatin proteins training dataset, which was based on literature and database mining. In Itzhak et al., 2016 the proteins were considered nuclear if their MS-measured intensity in the crude nuclear extract exceeded 85% of the global intensity. We hypothesize that these two latter datasets have better consistency with SimChrom dataset in part because by their design they are initially biased either by the information already available in the literature and various databases (in the case of Kustatscher et al., 2014) or by the selection of proteins that are preferentially localized in the nucleus and hence having higher chances to be described in the literature and databases (in the case of Itzhack et al., 2016). When MS-based datasets were compared to the nuclear localization datasets, we observed the same tendency. Around 95% of proteins in Kustatscher et al., 2014 and Itzhak et al., 2016 datasets may be found in our broad nuclear localization dataset (NULOC_JT_NECF), while for other MS-based datasets the proportion was around 60-70% (Torrente et al., 2011; Alabert et al., 2014; Ginno et al., 2018; Shi et al., 2021) and 75-80% (Ugur et al., 2023; Alvarez et al., 2023). The same tendency was observed for the consensus NULOC_CS dataset (the portion of the MS-based datasets present in NULOC_CS varied between 30% and 73%). To further understand the origins of these discrepancies we analyzed proteins that were found in MS-based experimental chromatin datasets but not present in nuclear localization datasets or SimChrom (Supplementary Figure SF2_5C). The Itzhak et al., 2016 dataset had a 98% overlap with NULOC_JT_NECF and its inclusion would not affect the results presented below. In total there were 2232 of such proteins, and only a minor fraction of these (94 proteins) did not have localization annotation in the databases (see Supplementary Figure SF2_5D). Five proteins were present in four total chromatin MS-based datasets (see Supplementary Figure SF2_5C), they included proteins encoded by the FLNB (Filamin B, actin-binding protein), GMPS (Guanine Monophosphate Synthase), CHERP (Calcium Homeostasis Endoplasmic Reticulum Protein), ILKAP (ILK Associated 42 Serine/Threonine Phosphatase), PLEC (Plectin) genes. However, the Ugur et al., 2023 dataset includes only PLEC and CHERP. While according to UniProt and HPA all of these are non-nuclear proteins predominantly localized in cytoplasm (with additional localizations in cytoskeleton, intermediate filaments, endoplasmic reticulum, Golgi apparatus), manual literature mining confirmed experimental evidence supporting the presence of these proteins in the nucleus (e.g. [16]). We additionally randomly selected 20 proteins from a subset of proteins that were reported by at least five out of seven chromatin MS-based studies (there were 195 such protein coding genes) and manually performed literature searches. From those 15 for 5 genes literature evidence was found suggesting their nuclear localization (CALR [17], PDIA4 [18], ABCF2 [19], SEC23B [20], EIF3D [21]). Hence, it may be stated that MSbase studies currently have predictive power to identify new chromatin proteins that are not annotated as such by the localization and functional databases. It is not straightforward to estimate the potential contamination of MS-based datasets with nonnuclear proteins since one cannot come up with an ultimate reference set of non-nuclear proteins. Even for well studied proteins there are still chances that they may have axillary functionality in the nucleus that have not yet been experimentally characterized. Still to address this problem we relied on analyzing the above mentioned set of 2232 proteins using information available in GO. GO enrichment analysis revealed that these proteins were mainly associated with a diverse set of GO-terms related to nonnuclear organelles/compartments, cellular metabolism, protein translation and maturation suggesting that no important chromatin associated categories were missed during the construction of SimChrom dataset that could account for this discrepancy (Supplementary Figure SF2_6). Previous studies suggest that the results of MS-based studies may be contaminated by cytoplasmic and mitochondrial proteins [12,22]. In our analysis 2025 proteins were related to cytoplasm, 115 to Golgi vesicle transport according to GO. No enriched terms include mitochondria proteins. It is still possible that among 2232 proteins there are still nuclear proteins, whose annotation by GO does not account for their additional, moonlighting functions in the nucleus. For instance, 30 proteins were associated with translation ("translation", “translation initiation factor activity”) according to GO, many proteins of the translational apparatus are known to be moonlighting proteins with functional roles in the nucleus [23]. Similarly, Golgi apparatus cooperating with nucleus and ER in vesicular transport. In a different type of analysis we looked at chromatin proteins that were not identified by the MS-studies but were included in our SimChrom dataset. There were 1246 such proteins (or 41% of SimChrom). According to the HPA classification of housekeeping proteins, 67% (839 proteins) of these were not housekeeping, consistent with the idea that they were missed by MS-based studies, because they were not expressed in the cell lines. However, it was also found that MS-based studies are biased towards identifying the housekeeping proteins. More than 75% of nuclear/chromatin proteins reported by the MS-based studies were from the housekeeping pool, while the average expected fraction of 43 nuclear housekeeping proteins is around 62% (Supplementary Figure SF2_7B). Among the set of 1246 proteins not found in the MS-based studies the dominant SimChrom category was related to DNAbinding transcription factors (see Supplementary Figure S2_7C). 1148 DNA-binding TF were missed by MS-based studies, including 394 housekeeping TFs. This highlights another potential source of discrepancy - housekeeping TFs may be present in small amounts or be washed away during chromatin/nucleome extractions and thus be missed by MS analysis. To further understand the discrepancies between the MS-based datasets and SimChrom we performed enrichment analysis for the SimChrom categories in the experimental datasets (Supplementary Figure SF2_8A). It can be seen that categories related to DNA-binding transcription factors were mainly depleted in the MS-based datasets, consistent with the analysis described above. The separate analysis of housekeeping TF suggests that different experimental methods show high variability in their ability to recover TF in chromatin extracts. The Torrente et al., 2011 dataset includes only 8 housekeeping TF, while the Kustatscher et al., 2014 and Itzhak et al., 2016 datasets each contain 158. The dynamic nature of transcription factors’ interactions with chromatin likely explains these facts. The most enriched categories were related to protein involved in RNA binding and metabolism, histone chaperone, remodelers and heterochromatin associated factors. This is likely related to the higher chances of these proteins to be detected due to their universal presence in the cells and high expression levels. Detailed analysis of the SimChrom categories representation in MS-based datasets further revealed some details of the differences between the datasets (Supplementary Figure SF2_7C). For instance, the ML-based Kustatscher et al., 2014 dataset was able to recover twice as many DNA transcription factors than Ginno et al., 2018 and Shi et al., 2021 datasets. The ratio of housekeeping and non-housekeeping TFs for Ginno et al., 2018 and Shi et al., 2021 datasets is the same, and is higher for the Kustatscher et al., 2014; Itzhak et al. 2016; Alabert et. al., 2014 datasets. Interestingly the nucleome Itzhak et al., 2016 dataset was also able to recover the same amount of TF as the Kustatscher et al., 2014 dataset, although the size of the dataset was three times smaller than Shi et al., 2021 and Ginno et al., 2018 datasets. This again points to the fact that TF may be lost during chromatin extraction. The representation of some other SimChrom categories had significant differences between the MS-based datasets likely attributed to the differential extraction probability. For instance, only 29% and 25% of RNA polymerases and histones, respectively, were present in the Ginno et al., 2018 dataset, while 50-60% were present in the Shi et al., 2021 dataset. A closer look at the histone proteins (for which currently information about all 62 expressed proteins is known [15,24,25] revealed that while certain tissue specific histone variants were missed as expected, all MS-based studies were not able to recover all canonical histones variants, especially for the H2B-histone (Supplementary Figure SF2_8B). For instance, from 14 canonical H2B proteins isoforms the Shi et al., 2021 dataset was able to recover six proteins (corresponding to the genes products of H2BC1, H2BC3, H2BC4, H2BC13, H2BC18, H2BC26), while the Itzhak et al., 2016 dataset four proteins (corresponding to genes products of H2BC4, H2BC11, H2BC12, H2BC13). While 44 the canonical histones are likely to be expressed simultaneously in the cell, these discrepancies may be due to the sensitivity of MS-based analysis or expression variation in cell lines. Taken together, our analysis indicates that MS-based chromatin datasets exhibit inconsistencies with subcellular localization or functional annotations available in UniProt, HPA, or GO. Up to 35% of chromatin proteins identified by MS-based techniques lack known nuclear localization according to the databases. Among these, approximately 30% of proteins identified simultaneously by at least three experimental studies may possess an additional nuclear localization that is not currently captured by the databases, but may have been reported in research papers. For other proteins, the most parsimonious explanation of their presence is the considerable contamination of chromatin extracts by mainly cytoplasmic proteins. From another point of view, many chromatin proteins annotated in the databases are not present in the MS-based datasets. While this is partially due to the limited number of genes expressed in the analyzed cell lines, our analysis also suggests that many proteins (such as housekeeping transcription factors) are lost during chromatin extraction, likely due to the dynamic nature of their interactions and their low expression. Finally, we showed that MS-based studies that filter their results by selecting the proteins that are highly enriched in the nucleus with respect to other cellular compartments, or use ML-assisted classification based on database/literature data, show better consistency with the information in the databases. However, this comes at the expense of decreasing the size of their datasets and likely limiting their ability to identify new proteins associated with the nucleus/chromatin. 2. The SimChrom chromatin protein classification, the interactive SimChrom database and other reference datasets This section supplements section 3.2. The SimChrom chromatin protein classification, the SimChrom dataset and other reference datasets in the main text. Note: Interactive Figure 3 (https://simchrom.intbio.org/#classification) is the interactive version of Figure 3, which is the key source of information for the analysis presented below. The largest subgroup of the SimChrom “non-histone proteins'' category is the “DNA-templated transcription” group (1547 proteins), which consists of the proteins belonging to the “Regulation of transcription” subgroup (1514 proteins) and "RNA polymerases" subgroup (34 proteins). The "DNA metabolic processes'' form the second largest protein subgroup (495 proteins) and include DNA replication, repair, and recombination machinery. “Nuclear RNA binding proteins'' is another major group of proteins (309 proteins) present in SimChrom. There are a lot of RNA binding proteins in the cell (more than 1500 [26]), in the “Nuclear RNA binding proteins'' category we aimed at including only 45 those that are found inside the nucleus (see Methods Section 2.2 and Supplementary Table ST4), including those involved in preribosome formation, RNA processing and modification inside the nucleus. Other major subgroups of our classification included “Histone modification” (257 proteins), "DNA-acting enzymes'' (258 proteins including DNA methylation and demethylation enzymes), “Centromere-associated” (241 proteins) (see Figure 3). Several specific subgroups of various scope containing proteins important for chromatin functioning were also included in our classification, such as “Histone chaperones'' (32 proteins), ATP-dependent "Chromatin remodelers'' complexes (114 proteins), “SMC complexes” (28 proteins, including cohesins implicated in chromatin loop extrusion). To evaluate the contents of our SimChrom dataset we performed its cross-comparison to the localization-based datasets described above (NULOC_JT_NECF and NULOC_CS) (see Supplementary Figure SF3_3). The vast majority of SimChrom entries was also present in the NULOC_JT_NECF dataset (the broadest dataset that combined all nuclear protein entries from all protein localization databases at any level of confidence), which is consistent with chromatin proteins being only a subset of nuclear proteins. However, a minor subset of SimChrom (156 proteins) was not classified as nuclear by the localization databases (Supplementary Figure SF3_3A). To further understand the nature of this minor discrepancy we analyzed the SimChrom categories which contributed the most entries to this subset (Supplementary Figure SF3_3C) or where a significant proportion of entries in the respective category was absent in the localization databases (Supplementary Figure SF3_4A). In this subset 32 entries out of 156 were not annotated by the localization databases at all, 27 were considered by UniProt as only chromosomal (this is consistent with “Centromere-associated” category of SimChrom having the most entries not present in NULOC_JT_NECF), and 124 had only non-nuclear localization according to the localization databases. A manual review of the latter entries suggested that they included both bona fide nuclear proteins (such as histone acetyl and methyltransferases), other proteins such as ribosomal proteins, various kinases and proteins involved in mitochondrial DNA processing. Further research and data would be needed to clarify the localization status of the latter entries. The SimChrom category having the largest proportion of proteins absent from the localization databases was “Histone tail cleavage” (Supplementary Figure SF3_3C). This is consistent with the fact that for many of the histone tail cleavage enzymes (e.g., metalloproteinases, cathepsins, neutrophil elastase) the histone tail cleavage activity in the nucleus is not their primary function and manifests only in specific conditions and cell development stages [9]. The comparison of SimChrom with the NULOC_CS (our stringent high confidence consensus dataset of nuclear proteins) showed a sufficiently higher number of SimChrom proteins that were not included in the NULOC_CS dataset (1208) (Supplementary Figure SF3_3A). This is not unexpected since only 44% of the human proteome has simultaneous localization annotations at sufficient levels of confidence from UniProt and HPA (see Results Section 3.1). The intersection of SimChrom and NULOC_CS datasets included 1837 proteins (Supplementary Figure SF3_3A), and may be 46 considered as a set of chromatin proteins with a high level of confidence. To further validate that the subset of SimChrom that was not present in NULOC_CS (1208 proteins) represented nuclear proteins we performed GO enrichment analysis of this subset against a list of all GO-terms and then select nonnuclear associated terms for further analysis (see Methods Section 2.2, Supplementary Figure SF3_4B, Supplementary Table ST8). The analysis confirmed the low number of SimChrom proteins that were associated with bona fide non-nuclear GO categories (the “Centromere associated proteins” having the highest number – around a dozen out of 156 – of proteins that had “non-nuclear” GOannotation terms, that belongs to charged multivesicular body proteins and dynactin complex subunits). Finally, as a byproduct of NULOC_CS and SimChrom comparison we find that 1459 proteins were present in NULOC_CS but not in SimChrom - this set may be considered as both a high-confidence set of nuclear non-chromatin proteins and a set of proteins that should be added to SimChrom. The GO analysis of this dataset did not reveal any clear GO categories that should have been included in SimChrom as categories related to chromatin functioning (see Supplementary Table ST9). Limitations of the proposed chromatin classification SimChrom and its contents include the following: the absence of cell cycle control proteins and checkpoint signaling proteins, the lack of detailed classification for proteins involved in reading, writing and erasing DNA and RNA modifications. The 'Genomic location' categories require additional curation supported by experimental evidence to enhance the accuracy and reliability of their protein content. In addition, we did not consider the protein components of nonmembrane nuclear organelles whose proteins may also functionally interact with nucleic acids (directly or through phase separation). The classification does not include protein isoforms. Also, SimChrom is limited currently to human proteins only. These limitations will be addressed in the future versions of SimChrom. 3. Analysis of the human chromatome 3.1. The chromatome composition and abundance of chromatin proteins This section supplements and expands section 3.3.1. The chromatome composition and abundance of chromatin proteins in the main text. To understand chromatin functioning it is important to know the chromatome content not only in terms of the set of proteins associated with chromatin, but also in terms of their abundance (i.e., the (relative) number of proteins per cell or organelle). Hence, we aimed at analyzing the available mass spectrometry data to address this question. The analysis of MS protein intensities from the experimental chromatome/nucleome studies discussed above, revealed a high degree of variability (see Figure 4A, Supplementary Figure SF4_1). For instance, the estimated relative mass of histone proteins varied 47 from 0.1 to 58 % depending on the study, suggesting a high degree of bias due to different experimental techniques and analysis pipelines used to process raw mass spectrometry data (see Figure 4A). Hence, for further analysis we relied on the “whole-organism” protein abundance information available in PaxDB for H. sapiens [27]. PaxDb provides high quality information on protein abundance combined from many experiments with high coverage, dynamic range and interaction consistency (estimated consistency of abundance data with data on protein functional interactions) integrated over many cell types and conditions. Among the datasets available in PaxDb we have chosen two whole-organism datasets: the dataset with the highest proteome coverage (“H.sapiens - Whole organism (Integrated)” - covers 99% of human proteome according to PaxDb, referred to as “PaxDb_INT'' in this paper) and the dataset with the highest interaction consistency score (“Whole organism, SC (Peptideatlas,aug,2014)” - covers 84% of human proteome according to PaxDb, referred to as “PaxDb_PA'' in this paper), see Supplementary Figure S4_1A. Our analysis showed that with respect to PaxDb_PA, PaxDb_INT dataset has additional abundance information for around 2700 human proteins that almost exclusively have low levels of expression (less than 1 ppm, see Supplementary Figure SF4_1B). Among these proteins there are up to around 700 nuclear/chromatin proteins, hence we opted to use PaxDb_INT for general characterization of the abundance distribution of chromatin/nuclear proteins (presented in Figure 4B). PaxDb_PA dataset showed a higher consistency with respect to the relative abundance of functionally interacting chromatin proteins. The total abundance of different types of histone proteins (H3, H4, H2A, H2B) matched their expected equimolar ratio (see Supplementary Figure SF4_2A,B). Hence, PaxDb_PA was used for a detailed analysis of chromatin protein abundance distribution between chromatin protein groups and individual proteins (Figure 4C,D). Around half of the whole-organism human proteome consists of low abundance proteins with expression levels of less than 1 ppm (~50%, see Figure 4B, Supplementary Figure SF4_1C, Supplementary Table ST10). The whole-proteome abundance distributions are positively skewed towards the low abundant proteins. In PaxDb_INT dataset this skewness is additionally supplemented by a second peak at the low abundance values (Figure 4B, Supplementary Figure SF4_1A). Among the low-abundance proteins only 25% of them correspond to the housekeeping proteins, while the proportion of housekeeping proteins among high-abundance proteins (abundance of more than 1 ppm) is 68% (see distribution in Figure 4B, Supplementary Figure SF4_1C, and Supplementary Table ST10). We used PaxDb_INT data to analyze abundance distributions of database-derived chromatome/nucleome protein sets discussed in the previous sections of the paper. The NULOC_CS, NULOC_JT, and SimChrom datasets reported above all manifested distributions mirroring that of the whole proteome for PaxDb_INT data (see Figure 4B) with a significant proportion of proteins in these datasets still represented by the low-abundance proteins (40%, 44%, and 48%, for NULOC_CS, NULOC_JT, and SimChrom, respectively, see Supplementary 48 Figure SF4_1C, Supplementary Table ST10). We next aimed at understanding the types of proteins contributing to the low and high-abundance portion of the nucleome/chromatome. The proportions of house-keeping/non-housekeeping proteins in lowand high-abundance fractions for different databasederived datasets are given in Supplementary Figure SF4_1C and show that around 60% and 40% of chromatin/nuclear proteins are housekeeping ones for the highand low-abundance fraction, respectively. As discussed above nuclear and chromatin databased-derived protein sets are on average enriched in housekeeping proteins with respect to the whole proteome (~58.5 % vs ~47%, see Supplementary Results and Discussion Section 1.3, Supplementary Figure SF4_1C, Supplementary Figure SF2_7A,B). This increase in the proportion of housekeeping proteins stems from both the increase of the number of low-abundance and high-abundance housekeeping proteins relative to the respective total numbers of low-abundance and high-abundance proteins in nucleome/chromatome datasets. A detailed analysis showed that the increase of the fraction of housekeeping proteins among the low-abundance ones was more than expected, while for the highabundance ones was less than expected (under the assumption that highand low-abundance fractions should contribute to the increase proportionally to the number of housekeeping proteins belonging to these fractions, see Supplementary Figure SF4_1C). For instance, under parsimonious considerations the overall increase in the fraction of housekeeping proteins for SimChrom with respect to the whole proteome (60% vs 47%) should imply the increase of house-keeping proteins’ fraction among the lowabundance proteins from 25% to 32% (25*60/47=32), yet an increase to 38% was observed. This highlights the important role that low-abundance housekeeping proteins play in chromatin functioning. A more detailed analysis revealed that 64% of these housekeeping low-abundance chromatin proteins belong to the housekeeping DNA-binding transcription factors group (Supplementary Figure SF4_1D). We next applied similar analysis to the sets of chromosome/nucleosome proteins identified in MS-based studies. The resulting distributions differed considerably from the distributions of databasederived protein sets discussed above, having a single maxima centered at higher values of abundance (see Figure 4B). This fact again suggests that MS-based studies of chromatin are able mainly to recover highly expressed proteins and miss low expressed proteins (low-abundance proteins are in the range of 1%-27% of the identified protein sets, Supplementary Table ST10, Supplementary Figure SF4_1C). The protein sets recovered by MS-based studies were significantly enriched in housekeeping proteins when compared to database-derived datasets (Supplementary Table ST10, Supplementary Figure SF4_1C). The abundance distributions varied between different MS-derived chromatin/nucleome protein sets. The chromatome studies based on extraction and/or cross-linking techniques (Alabert et al., 2023; Shi et al., 2021; Ginno et al., 2018; Torrento et al., 2011) had distributions shifted towards higher values of abundance, than the studies by Kustatscher et al., 2014; and Ugur et al., 2023 suggesting that the latter studies were also able to capture more chromatin proteins 49 with low abundance. The nucleome study by Itzhak et al., 2016 was also able to capture more lowerabundant proteins. We next aimed at understanding the abundance of different chromatin protein groups and individual chromatin proteins in the cell relying on our SimChrom-SL classification using PaxDb_PA abundance dataset. The resulting diagrams depicting abundance variations of chromatin proteins, belonging to different SimChrom-SL categories, the number of proteins belonging to the respective categories, and the cumulative abundances (calculated both as the total number of protein molecules and the total molecular weight of protein molecules belonging to each SimChrom-SL category) are presented in Figure 4C. To gain additional insights into the functioning of chromatin in Figure 4D we plotted the abundance values of highly expressed chromatin proteins (abundance of more than 1% of the H4 histone abundance) belonging to SimChrom-SL categories of the “Molecular function” or “Physico-chemical properties” type. Abundance data for all histone proteins and non-histone chromatin proteins with abundance more than 0.01% of Histone H4 is presented in the Supplementary Table ST12. It is important to note that many chromatin proteins have additional localization in other cellular compartments, hence the presented data reflects the overall abundance of the chromatin proteins in the cell rather than their abundance in the nucleus. To shed more light on the protein abundance in the nucleus we have also built diagrams analogous to Figure 4C only for 802 SimChrom proteins that are uniquely localized in the nucleus (according to our NULOC_CS_UL dataset) (see Supplementary Figure SF4_3A). These proteins are also highlighted in Figure 4D. As seen in panel 1 of Figure 4C chromatin categories vary substantially by their median abundance from 0.09 ppm to 570 ppm and there is still considerable variation in the abundance values within the categories. The most abundant chromatin protein is histone H4 (~11000 ppm), which is expressed by a family of genes almost exclusively coding the same protein sequence (except for H4C7, which has a negligible abundance). It is convenient to measure the abundance of all other proteins in fractions of H4 abundance (see Figure 4D). Each nucleosome contains two copies of H4 histones, therefore the numbers are also easily converted to relative abundance of chromatin proteins per nucleosome. Detailed analysis of histone protein abundance is in Supplementary Table ST12 and shown in Supplementary Figure SF4_2A,C. The total number of core nucleosomal histone types H3, H4, H2A, H2B expressed by various genes sums up to similar numbers (~10400-10900 ppm) consistent with their equimolar association within nucleosome core particles. The cumulative abundance of H1 histones (~4500 ppm) suggests that slightly less than one H1 histone is associated with each nucleosome. The most abundant core histone variants are H3.3 (23% of H4, 2530 ppm), H2A.X (6.5%, 714 ppm), H2A.Z (10%, 1140 ppm), H2A.W (3.8%, 423 ppm). The least abundant histone variants are H2A.B and H1.7 (less than 1 ppm). Despite the relatively small number of protein coding human histone genes (108), many of which code for identical sequences, the cumulative abundance of histone proteins exceeds that of all other chromatin protein categories even if proteins with multiple localization are taken into account (see panel 2,3 in 50 Figure 4C). However, when the total molecular weight of proteins belonging to different categories is compared, the relatively small size of histone proteins (median ~15 kDa) results in them yielding the first place to RNA processing proteins (see panel 4, Figure 4C). Collectively the cumulative weight of proteins belonging to “Nuclear RNA binding proteins” category (that combines Preribosomeassociated, RNA modification, and RNA processing categories) amounts to 30.4% of all SimChrom proteins weight (4.8% of whole-organism proteome weight). However, many proteins from these categories are also localized in cytoplasm, and the major contribution to their cumulative molecular weight likely comes from the cytoplasmic fraction. If the same analysis is performed only for the SimChrom proteins that are uniquely localized in the nucleus (Supplementary Figure SF4_3A), the mass fraction of histone goes up to 38% of all the chromatin proteins that are uniquely localized in the nucleus. Other functional chromatin protein groups (or groups with specific properties) with high values of median abundance and high level of individual protein abundance include HMG A/B/N, histone tail cleavage, histone chaperones, chromatin remodelers and other categories (see Figure 4D). The high mobility group proteins (HMG A/B/N) are the second group after histones ranked by their median abundance. Although grouped together due to historical reasons, they include three separate superfamilies: HMGA (contains AT-hook domains), HMGB (contains DNA binding HMG-box domain), and HMGN (contains nucleosome binding domain). Our analysis suggests that the ratio of HMG proteins to nucleosomes is 1:8, 1:2, 1:3 for HMGA, HMGB, or HMGN proteins, respectively. However, the majority of HMG proteins are not exclusively localized in the nucleus (Figure 4D). The histone tail cleavage proteins are another small group of proteins in our classification with high median abundance in the whole-organism proteome. These enzymes, however, are not exclusively specific for histone cleavage, and likely perform their main functions outside the nucleus by cleaving other proteins. Among histone chaperones the H3-H4 histone chaperone NPM1 and H2A-H2B histone chaperone NCL have the highest abundance, 32% and 13% of H4 abundance, respectively. The most abundant histone variant specific chaperone is ANP32E (specific for H2A.Z-H2B with abundance of 2%). The whole nucleosome chaperone FACT complex consisting of SSRP1 and SUPT16H gene products, has an abundance of around 1%, amounting to one FACT complex per around 50 nucleosomes. Among RNA polymerase subunits POLR2E the common subunit E of RNA polymerases I, II, and III is the most abundant protein (0.58% of histone H4 abundance or around 1 per 90 nucleosomes). The exclusive components of polymerase II (POLR2B, POLR2C, POLR2D, and others) have their abundances in the range of 0.05-0.3%. With a median human gene length of 24kb and nucleosomal repeat length of around 200 bp this gives a lower estimate of one polymerase II per approximately 10 genes. Among genes involved in chromatin remodeling actin encoding genes (ACTB, ACTA1 and actin-like ACTL6A) are leading by the abundance of their protein products. While actin is a component of some chromatin remodeling complexes (e.g., SWI/SNF) the major contribution to its abundance clearly comes from its 57 SF5_5B). Their enrichment values are in the range 1.13-1.18. By classes of amino acids chromatin/nuclear proteins are mostly enriched in polar (N, Q, T, C, G, P), small (P, G, A, S), and positive (K, R) amino acids (Supplementary Figure SF5_4A, Supplementary Figure SF5_5B). It is important to note that such an analysis should be taken with a grain of salt, because the enrichment of certain amino acids may vary across different categories of chromatin proteins, and the categories with high number of protein entries have higher contribution to the overall average. To elucidate this variability we have also performed enrichment analysis for proteins in major functional SimChrom categories (see Supplementary Figure SF5_4). Serine and proline are among amino acids that are relatively abundant in proteins (abundance of around 7-8% and 5-6% in chromatin and cytoplasmic proteins, respectively, Supplementary Table ST13). The total enrichment of serine in chromatin proteins is attributed simultaneously due to its enrichment in the IDR, non-IDR regions (relative to IDR and non-IDR regions of cytoplasmic proteins), and more importantly due to higher proportion of IDR regions in chromatin proteins (46% vs 23%) that in turn have a considerably higher proportion of serine than non-IDRs (Supplementary Table ST13). The slight enrichment of serine in IDRs was observed almost across all SimChrom categories (Supplementary Figure SF5_4E). In non-IDR regions serine showed both enrichment and depletion in certain categories, and the overall enrichment was driven mainly by transcription factors due to the large number of proteins in these categories. The total enrichment of proline in chromatin proteins is attributed due to its enrichment in IDRs (relative to IDRs of cytoplasmic proteins and more importantly due to higher proportion of IDR regions in chromatin proteins (proline is the most enriched amino acid in IDRs of both chromatin and cytoplasmic proteins versus the non-IDRs, fold enrichment 1.8-2.1, Supplementary Table ST13). The overall enrichment of proline in non-IDRs was close to one and statistically not significant. Certain small groups, such as histones and HMG proteins showed considerable deviations in proline content in their non-IDRs (Supplementary Figure SF5_4H). The enrichment of proline in IDRs was observed in many SimChrom categories (Supplementary Figure SF5_4E). In certain categories, such as Non-Housekeeping transcription factors and pioneer TFs enrichment was high (1.37 and 1.55 fold, respectively). Surprisingly, the enrichment of proline in IDRs of housekeeping TF was depleted (FE of 0.92). Suggesting that while there is still a considerable fraction of prolines in IDRs of housekeeping TF, this fraction is significantly lower than in IDRs of non-house keeping TF (7 % and 10.4 % median fractions, respectively). Cysteine and histidine are among amino acids that have a relatively low abundance in proteins (abundance of around 1-2.5%, Supplementary Figure SF5_5B). The total enrichment of cysteine and histidine is mainly driven by the prevalence of zinc fingers containing transcription factors (Supplementary Figure SF5_4B). The exclusion of these proteins from analysis resulted in the disappearance of any statistically significant enrichment Supplementary Figure SF5_5A). 58 Interestingly, the enrichment of positive amino acids is only statistically significant for lysine, but not for arginine, and the enrichment is relatively moderate (1.03 in chromatin) (Supplementary Figure SF5_4A). Lysines are enriched in IDRs and non-IDRs of chromatin proteins, while arginines are depleted in IDRs and enriched in non-IDRs (when compared with IDRs and non-IDRs of cytoplasmic proteins) (Supplementary Figure SF5_5B). The overall higher positive charge of chromatin proteins stems also from the depletion of negatively charged amino acids in their sequence. The depletion of aspartate in chromatin/nuclear proteins is statistically significant (fold enrichment is around 0.9), while the depletion of glutamate is statistically non-significant (Supplementary Figure SF5_4A). This suggests that the increased positive charge of chromatin nuclear proteins has its main contributions in the depletion of aspartate, and moderate enrichment of lysine. Within the IDR regions of chromatin proteins tyrosine and asparagine were also significantly enriched (FE of 1.23 and 1.17 versus the IDRs of cytoplasmic proteins, respectively), mainly due to the contribution of transcription factors (Supplementary Figure SF5_4F). Among the most relatively depleted amino acids in chromatin/nucleus are tryptophan and hydrophobic/aliphatic amino acids like valine, isoleucine, leucine, and methionine (Supplementary Figure SF5_4A). Tryptophan is the rarest amino acid in proteins (around 1%). Certain categories like histone and HMG proteins lack it completely (Supplementary Figure SF5_4C). It is depleted in almost all chromatin categories, except for a few small categories such as DNA (de)methylation, histone tail cleavage, and histone modification, where the enrichment comes from non-IDRs (Supplementary Figure SF5_4I). Hydrophobic/aliphatic amino acids are depleted in IDRs vs non-IDRs of proteins and hence the large proportion of IDRs in chromatin proteins accounts for a lower fraction of these amino acids in chromatin proteins (Supplementary Figure SF5_4F,I). The most enriched amino acids in the uniquely localized nuclear proteins were the same except for cysteine (this difference may be traced to the diminished number of transcription factors enriched in cysteines in the NULOC_CS_UL dataset, apparently because of their multiple localization, see Supplementary Figure SF5_4A, Supplementary Figure SF4_3B,). 3.3. Detailed analysis of the domain composition of chromatin proteins and identification of new structural domains This section supplements and expands section 3.3.3. Domain composition of chromatin proteins and identification of new structural domains in the main text. 59 Next we set out to systematically analyze the available data on structural characterization, domain annotation and domain composition of chromatin proteins. We specifically explored the structurally uncharacterized portion of the chromatome (“dark” proteome) and identified potential new structural domains that are predicted by AI-based protein structure prediction tools (see Figure 6A). Historically, protein domains are loosely defined as evolutionary conserved units with similarities at functional, structural and/or sequence levels [33]. Domains may represent single proteins or exist in a variety of various sequence contexts. Sequences of related individual protein domains may be grouped and aligned to produce domain models. Domain models are catalogued and annotated by a number of resources/databases such as PFAM [34], CDD [35], CATH [36], InterPro [37], and may be further grouped into superfamilies, clans, folds, etc [38,39]. Domain models are usually defined through multiple sequence alignments (MSA) and corresponding hidden Markov models (HMM). In structurebased approaches (e.g. CATH/Gene3D database) domain superfamilies are assigned through grouping and alignment of available experimental 3D structures. The ultimate experimental structural characterization of chromatin proteins is available in the PDB database, however, recent progress in protein structure prediction spurred by AlphaFold resulted in new approaches to the structural characterization and discovery of new structural domains (e.g., as implemented in the TED database) [40] (Figure 6A). Structure prediction algorithms combined with structure similarity search algorithms, such as FoldSeek [41], now allow to find remote homologs and assign individual domains to their respective superfamilies. Figure 6B shows the fraction of the aggregate number of amino acids in all human chromatin proteins (referred below to as “aggregate chromatome sequence”, or ACS) which are structurally characterized or have domain annotations in different databases. According to AlphaFold approximately one half of the ACS (47%) is predicted to be intrinsically disordered, or to become ordered within protein-protein complexes (the structurally uncharacterizable “dark” chromatome) (see Methods), and the rest as having distinct 3D structure. Direct experimental structural data in PDB is available for only one fourth of the ACS (20% of ACS are simultaneously considered ordered by AFDB and available in PDB). Hence, we envision that at least one third (34%) of the aggregate human chromatome sequence is amenable to characterization with structural biology methods but has not yet been characterized (constitutes the potentially structurally characterizable “dark” chromatome). The Pfam database (the largest sequence-based database of protein domains and protein families to date) has annotations for around 39% of the aggregate human chromatome sequence. The CATH database, which focuses on identifying and annotating structural domains, annotates 25% of ACS, while the automated AlphaFold-driven TED resource finds structural domains in 35% of ACS. The difference between the fraction of ACS annotated by TED and that considered ordered by AFDB was traced to at least several facts: 1) AlphaFold is known to be biased to predict long solitary alpha-helices which are 60 not considered domains by algorithms that identify structural domains, 2) the TED algorithm frequently fails to annotate repetitive regions that contain visually identifiable secondary structure elements within large multidomain proteins, 3) we identified non-IDRs as regions no less than 4 amino acids whereas median length of TED domain in human proteins were 108 aa. A caveat that has to be kept in mind, is that current automated analysis using AlphaFold is based only on predictions for single chain proteins, while in reality chromatin proteins engage in many intermolecular interactions. To some extent Pfam and CATH/TED are complimentary (see Figure 6B). In addition to 39% of ACS annotated by Pfam, TED annotates additionally 13% of ACS, and CATH adds annotations to 3% of ACS on top of it (yielding a combined annotation coverage of 55% by these three resources). Next we analyzed the structural characterization of the aggregate human chromatome sequence from the point of view of structural domains present in chromatin proteins (as identified by the most comprehensive TED database, which automatically detects structural domains) (see Figure 6C). Chromatin proteins contain in total 6246 individual TED domains. Using FoldSeek and combination of FoldSeek and CATH resources (see Methods) we matched these domains to the structurally related domains in PDB or CATH superfamilies. The remaining domains were analyzed for the presence of previously uncharacterized structural folds/superfamilies and potential functional roles of these domains. Among the 6246 predicted structural domains constituting human chromatin proteins, 34% had exact matches in PDB structures (100% sequence identity, see Methods), 56% matched PDB structures of homologues with different levels of sequence identity (from 99% to 5%, for details see Figure 6C). The majority of these homologous domains were in fact different paralogous sequences found within human genes (even for domains with sequence identity of 35-50% the fraction of human sequences among the matches was 51%), for matches with sequence identity above 35% the second largest contribution came from structures of mammalian homologues, for matches with sequence identity below 35% significant contributions were from structures derived from proteins of fungi, protostomia and bacteria (see Supplementary Figures SF6_1A for details). Additionally, 6% of TED domains that lacked direct hits among the PDB structures were mapped to protein structural superfamilies in the CATH database (the information about potential sequence variation in each homologous superfamily collected in CATH database combined with AlphaFold structural predictions allowed to identify more distant structurally characterized homologues). The remaining 4% (241 TED domains) represented domains that could not be matched to any known protein structure or protein structure superfamily, potentially representing new types of structural superfamilies of even protein folds. These domains are presented in Supplementary Table ST14 (see also Interactive Table 3 at https://simchrom.intbio.org/#novel_structural_domains) and ranked via their structural complexity by the number of their secondary structure elements. Among these domains, 123 domains have annotations in Pfam or other domain annotation databases present in InterPro, leaving 118 domains that are completely without annotations. The latter domains belong to 106 chromatin proteins, which may be 61 considered as perspective new targets for experimental studies of their function and structure. Among such proteins are, for example, a protein encoded by the GTF3C1 gene, a General transcription factor 3C polypeptide 1 (it has a previously unannotated and uncharacterized structural domain with length of 233 amino acids) (see detailed characterization in Supplementary Figure SF6_2A). Another instructive example is the globular domain of the testis specific linker histone H1.7 (product of H1-7 gene, see Supplementary Figure SF6_2B). Despite the considerable amount of studies dedicated to the elucidation of the structure of H1-linker histones [42], the H1.7 histone variant (previously, named HANP1/H1T2) has a quite different sequence resulting in a predicted structure that has a different topology than other known H1 histones (the “wing” of the globular domain consists of three beta-sheets rather than two). The relation of this domain to the H1 histone family cannot be identified with conventional sequence analysis methods (such as those implemented in the Pfam database), however, it should be noted that new deep-learning-based annotation approaches (such as Pfam-N) are able to annotate it (see Supplementary Figure SF6_2B). Next we predicted GO molecular functions and biological processes for mentioned above 118 not-annotated chromatin protein domains using DeepFRI [43], a Graph Convolutional Network for predicting protein functions by leveraging sequence features extracted from a protein language model and protein structures, see Methods Section 2.5. The top-7 common GO MF terms: ion binding, organic cyclic compound binding, heterocyclic compound binding, protein binding, cation binding, metal ion binding, nucleic acid binding. Top-10 GO BP terms: organic substance metabolic process, primary metabolic process, cellular metabolic process, nitrogen compound metabolic process, macromolecule metabolic process, organonitrogen compound metabolic process, regulation of cellular process, cellular macromolecule metabolic process, cellular nitrogen compound metabolic process, cellular response to stimulus. 9 out of 118 TED domains lacked the predicted GO molecular function by DeepFRI: two of them were in members of the regulatory factor X (RFX) family of transcription factors (encoded by genes RFX1, RFX5). Many chromatin proteins contain similar, evolutionary related individual protein domains whose kinship may be identified by matching them to the same Pfam domain sequence models. Hence, we used the Pfam domain annotation to characterize the diversity of protein domains found in chromatin proteins and typical domain composition thereof. In total 1753 different domain types (sequence models) were identified in chromatin proteins (Figure 6D). Next we analyzed the structural information available for these models. 76% of domain models had at least one individual domain among chromatin proteins that could be matched to a PDB structure using FoldSeek (bona fide structural domain in Figure 6D). To characterize the comprehensiveness of the structural characterization of each domain model we estimated the median sequence identity between all individual domains in chromatin proteins belonging to the said domain model and their best matches in PDB found via FoldSeek (see Methods 62 Section 2.5, Figure 6D). 42% of domain models were considered fully characterized, i.e. every individual domain in chromatin proteins belonging to these models can be found in PDB, 34% of domain models are partially characterized. 14% of Pfam domains were not matched by FoldSeek to PDB structures, but could be still identified in PDB via sequence search methods – these represented IDR regions, repeats, etc. 3% (55 domain models) did not match any PDB structure but could be matched to structural domains predicted by AlphaFold and found in the TED database. These represent prospective targets for validation with structural biology methods and further investigation of their interactions. For instance, among these domains are domains, potentially associated with chromatin remodeling (SANTA, zf-C3Hc3H), histone PTM writing (DUF7030, COMPASS-Shg1), zinc fingers (zf_CCCH_4, zf-LITAF-like, zf-WIZ, SWIM) etc. 7% of Pfam domain models currently have no structural information that can be assigned either through the PDB or TED databases. We next analyzed the diversity of Pfam domains in various SimChrom-SL protein categories (Figure 6E, subpanels 1,2) and the domain content of individual proteins belonging to these categories (Figure 6E, subpanels 3-5). Pfam identified 11147 individual domains in chromatin proteins belonging to 1753 domain types (Pfam models); only 70 chromatin proteins had no domain annotation at all. For the distribution of the total number of Pfam models and distinct Pfam models identified in chromatin proteins, see Supplementary Figure SF6_1B, Supplementary Figure SF6_1C. Expectedly, large SimChrom-SL categories consisting of more than one hundred proteins harbored the largest number of different domain types (e.g., DNA-acting enzymes, histone PTM writers, transcription factors, etc.), while the smaller categories had less (see Figure 6E, subpanel 1). Although Pfam may not be comprehensive in its annotation, we estimated the number of distinct domain types per protein in each category (relative domain diversity, Figure 6E, subpanel 2). The average domain diversity was around one for all categories. The categories with the considerably lower domain diversity were transcription factors categories (their variability relies on different combinations of zinc-finger domains that are described through only a few Pfam domain models), histones (their functional variability is often conferred by only small changes in the sequence), and HMG-constainting proteins (this is a very small group of proteins with only nine proteins and three corresponding Pfam models). The median number of Pfam domains in human chromatin proteins was two (which corresponds to the structure based domain analysis presented above). Certain chromatin protein categories had a higher median number of domains, including Housekeeping TF, histone PTM writers and readers (Figure 6E, subpanel 3). Interestingly the median number of domains for Non-housekeeping TF was two (they more often rely on single homeodomains than on cassettes of zinc-finger domains), although the group is diverse and proteins with as many as 32 domains were present. This, however, is again explained by the large number of zinc-finger domains that may be present in such proteins (see Supplementary Figure SF6_1D). The presence of additional domains in PTM writers and readers may be hypothesized to have evolved due to the functional necessity for multivalent binding to different chromatin structures (see 63 below for a detailed analysis). Some categories mostly consist of single domain proteins, such as Centromere-associated, DNA repair, Regulation of transcription, RNA polymerases (but this group also includes proteins with a maximum of 42 domains), DNA recombination, RNA modification, Histones, HMG_A/B/N, etc. The analysis of the number of distinct different domain types present in chromatin protein categories corroborates the above mentioned analysis (Figure 6E, subpanel 4). Proteins from PTM writers group have the median number of three distinct domain types, while all other categories have less. Still many chromatin proteins harbor many distinct domain types, DNA-acting enzymes, histone PTM writers, chaperones, remodelers, transcription factors with as much as 8-9 distinct domains are present (see Supplementary Table ST15). There are 118 chromatin proteins harboring at least five different domain types (see Supplementary Figure SF6_1C). This highlights the multivalency of protein interactions in chromatin, keeping in mind that many proteins further form protein-protein complexes increasing their interaction potential. The average individual domain length in chromatin proteins is around 65 amino acids (the median is 28 aa), however, this number is biased by the presence of many zinc-finger domains (around 22 aa in length). Subpanel 5 in Figure 6E gives a more balanced view for each SimChrom category. For the majority of protein groups the median domain length is around 100 amino acids (mean is 137, median is 134). The birds-eye view of the most frequently occurring Pfam domains’ in various functional SimChrom-SL categories is presented in Figure 7. The data is presented for domains that occur in at least five chromatin proteins and in at least 10% of proteins in a category (the threshold for data point depiction is 5%). The comprehensive interactive analysis figure with the ability to alter these thresholds and switch between SimChrom and SimChrom-SL classifications systems is available in Interactive Figure 4 (https://simchrom.intbio.org/#domain_composition). In Figure 7 the following categories and their respective domains can be grouped revealing their partially shared domain composition: 1) the categories containing transcription factors and their zinc finger, homeodomains and KRAB domains form the most frequently occurring entities, 2) some chromatin regulators, such as PTM writers, readers, erasers and chromatin remodelers together with their Chromo, Bromodomain, PHD. 3.4. Detailed analysis of the multivalent interactions in chromatin proteins This section supplements and expands section 3.3.4. Multivalent interactions in chromatin protein in the main text. The presence of multiple domains (belonging to the same or different domain models) in chromatin proteins is a known feature contributing to their ability to engage in multivalent interactions (Figure 8A) [44]. Below we present the analysis of such domains engaged in multi-valent interactions (referred to as EMVI-domains hereafter) that are found in chromatin/epigenetics regulator proteins (see 64 Figure 3 for definition of this group). To limit our analysis to a manageable set of EMVI-domains, we selected those that were found in multiple copies or in combination with another Pfam domain in at least three chromatin regulator proteins (94 Pfam domains in total), and from those we selected 59 domains that we were able to manually classify based on the information currently available in the literature according to their functional binding modes. The following functional groups of domains were used: histone methylation/acetylation/phosphorylation, chromatin remodeling, histone binding, DNA binding, DNA methylation, protein dimerization/oligomerization, PPI, RNA binding. Histone post-translational modifications were further subdivided into readers, writers and erasers functional subgroups (see Figure 8C, Interactive Figure 5 (https://simchrom.intbio.org/#domain_cooccurrence), and Supplementary Table ST16 for the list of domains and their detailed classification). We consider this subset of chromatin proteins’ domains as representative to illustrate the concept of multivalency in chromatin regulators interactions, since the selected domains are extensively characterized and their functions are known. A comprehensive analysis would require characterization of all 409 Pfam domain models that are found in combination with other models or in multiple copies in at least one chromatin regulator protein. As a compromise our online database includes the analysis of 163 Pfam domain models that are present in at least two chromatin regulator proteins (see https://simchrom.intbio.org/#domain_co-occurrence). First, we analyzed the co-occurrence of selected EMVI-domains in all chromatin proteins. There were in total 922 chromatin proteins (306 with the exclusion of transcription factors) that had more than one selected EMVI-domain (including multicopies). The conditional probability of finding a corresponding domain A in a chromatin protein given that another domain B is already present was estimated and is presented in Figure 8C (columns and rows correspond to domains A and B, respectively). The Interactive figure 5 is available at https://simchrom.intbio.org/#domain_cooccurrence (also includes unclassified potential EMVI-domains found in at least two chromatin regulator proteins). The matrix in Figure 8C allows to trace the interplay between different domains employed in architectures of chromatin proteins. The largest groups of domains in Figure 8C are those involved in histone methylation and DNA binding, suggesting that these mechanisms are the most represented and employed in chromatin functioning regulation. There were 49 cases where association between the presence of various domains in chromatin proteins was 100% (red squares in Figure 8C), among them for 18 cases (9 domain pairs) the association was reciprocal (i.e. P(A|B) = P(B|A)). In certain cases this exclusive association between domains may be traced to the fact that they form a larger structural complex with direct structural interactions between the domains as judged by the visual inspection of AlphaFold based predictions (MOZ_SAS and zf-MYST, ADD_DNMT3 and DNMT3_ADD_GATA1-like). In other cases the association is likely due to functional reasons, in our analysis in the majority of cases such domains 65 were confined to the histone methylation readers subgroup (KDM3B_Tudor and PWWP_KDM3B; C5HCH, NSD_PHD, PHD-1st_NSD, and PHDvar_NSD found in Histone-lysine Nmethyltransferases). Among the EMVI Pfam domains that co-occur with the most number of other different Pfam domains in chromatin regulator proteins is the histone methylation/acetylation domains: PHD domain (45 other domains), Bromodomain (38), SET (40) and PWWP (28) and chromatin remodeling Helicase_C (33) and SNF2-rel_dom (31). The diagonal elements in Figure 8C show that certain domains tend to be present in multiple copies, particularly often, MBT, WD40 and zinc finger (zf-C2H2) domains. However, all these are special cases of short repeat domains, where multiple copies are needed to form one functional unit. Not so often, but in a considerable number of proteins PHD and Bromodomain may be found in multiple copies. Another more general view of multivalent interactions in chromatin proteins may be obtained if we trace the relationships within or between different functional groups of domains. One can see that domains from the same functional group (e.g., histone methylation) tend to co-occur (Figure 8C) in chromatin proteins and may also be present in multiple instances in proteins (Supplementary Figure SF8_1B). For example, there may be up to nine histone methylation associated domains in chromatin proteins (Supplementary Figure SF8_1B). If a chromatin protein has a domain involved in histone methylation (either writing, reading or erasing) there is an estimated 38% chance that there will be another different functional domain from this group of domains (Supplementary Table ST17, Supplementary Figure SF8_1C). For acetylation this estimated probability is 31%, for phosphorylation 18%. Note, that domain categorization is not trivial. For example, WD40 is a repeat that folds into a higher order structure (median number in chromatin proteins are four). They can recognize both unmodified histone tails and methylated regions [45] and its classification may affect this part of study. The associations between the occurrence of domains from different functional groups can also be observed. This can be seen in Figure 8C, and the upset plot in Figure 8D (see also Supplementary Figure SF8_2A). One can see that domains involved in histone methylation (one of the most abundant group by the number of Pfam models and the number of chromatin proteins) may be in a considerable number of chromatin proteins combined with other EMVI-domains (particularly DNA binding domains and histone acetylation), association association with with histone binding domains, chromatin remodeling, DNA methylation, (di/oligo)-merization and RNA binding domains was also observed, see Supplementary Figure SF8_2C. The same can be said about domains involved in histone acetylation, although in a somewhat smaller number of cases, and with exclusion of their combination with 66 dimerization domains. Notably, domains involved in histone phosphorylation were not found in combination with domains from other functional groups in our analysis. This may reflect an evolutionary strategy whereby combinations of histone methylation and acetylation evolved to delicately regulate gene expression at the epigenetic level, while phosphorylation remained as a more general mechanism affecting a broad number proteins and pathways in the cell. DNA binding domains are found to be associated with methylation, acetylation, dimerization, and chromatin remodeling domains although the relative number of proteins harboring combinations of such domains is small compared to the number of protein (mainly transcription factor) that have DNA binding domains in their sequence. Domains associated with catalytic subunits chromatin remodeling complexes in certain cases are found together with histone acetylation, methylation or DNA binding domains. A more detailed analysis reveals that one “reader” domains for acetylation and methylation are found in these proteins. For a more comprehensive view of multivalent interactions it is reasonable to (1) analyze not only pairwise co-occurence of different functional domains, but simultaneous co-occurrence of domains from several functional groups in one proteins, (2) extend the analysis to complexes of chromatin proteins. The results of such analysis are presented in Figure 8D (see Methods Section 2.1.4 for our selection of 513 protein complexes where all proteins are chromatin proteins from Complex Portal). In our analysis at the level of individual proteins, proteins harbored domains only from up to three functional groups. Particularly, DNA binding domains may be combined with (di/oligo)merization, chromatin remodeling, histone methylation, methylation and acetylation or histone acetylation and chromatin remodeling domains. The formation of chromatin protein complexes considerably affects the available combinations of functional domains. Among the 513 analyzed protein complexes, 181 complexes contained domains EMVI-domains from the analyzed functional domain groups, 101 complexes harboured more than one domain, 80 complexes harbored domains from different functional groups. From these 80 complexes the majority were various chromatin remodeling complexes (53 complexes), others representatives included acetyltransferase complexes (13), deacetylase (4), and DNA-methylation (2). One can see that the largest number of analyzed complexes (22) simultaneously contained domains from four functional groups (DNA binding, histone methylation, histone acetylation, and chromatin remodeling), in a select number of complexes domains belonging to up to six functional groups were observed (all the above mentioned together with domains involved in DNA methylation and histone binding). These were all complexes involved in chromatin remodeling. For example, 'MBD2 or MBD3/NuRD nucleosome remodeling and deacetylase complex'. Notably, in chromatin complexes histone acetylation domains are found more often than histone methylation domains (unlike in the case when individual proteins are analyzed), this might, however, be biased by the current list of