Introduction

Cells constantly modulate their proteomes in response to physiological and environmental changes. The timely removal and turnover of cellular proteins is integral to protein homeostasis1. In eukaryotes, individual proteins, complexes, and large assemblies are degraded via either autophagy or the ubiquitin-proteasome system (UPS)2. In mammalian cells, ~80% of the cellular proteome is degraded through the UPS1. In this pathway, the designated protein cargo is tagged with ubiquitin (Ub) molecules through a series of enzymatic reactions, marking them for degradation by the proteasome3. Following the action of E1 and E2 enzymes, the E3 ligase brings both the E2–ubiquitin complex and the substrate protein in proximity, allowing the transfer of Ub from the E2 enzyme to a lysine residue on the target protein4,5. This process is often repeated (poly-ubiquitination), resulting in substrates with distinct types of Ub-chains. In UPS, for instance, K48-linked Ub-chains are recognized by Ub-binding domains (UBDs) on 19S proteasomal particles, initiating the degradation of substrates1. In autophagy, ubiquitination often serves as a necessary condition for identifying substrates, conferring specificity6. Cargo components, damaged organelles, and intracellular pathogens targeted for degradation are often ubiquitinated. Further, autophagy receptors are enriched in UBDs to recognize modified cargo components7 or themselves strongly ubiquitinated to trigger aggregation of protein assemblies in the cytosol and organellar membranes8,9, thus enhancing autophagic flux.

E3 ubiquitin ligases confer substrate specificity for ubiquitination. They recognize distinct targets, operate in diverse cellular locations, and exert spatial control of protein turnover10,11. In addition to controlling homeostatic processes, E3 ligases regulate immunity and inflammation pathways12,13. Given their tissue-specific expressions and association with developmental and metabolic syndromes, including cancer progression, E3 ligases have emerged as promising candidates, particularly for drugging previously undruggable targets14. In stark contrast to E1 (~10) and E2 enzymes (~50), a substantial number of E3 ligases (~600) have been recognized in humans15,16. This count of putative E3s stems from various investigations: Li et al.17 identified ~617 potential human E3-encoding genes by conducting a genome-wide search to detect RING (Really Interesting New Gene) finger catalytic domains using hidden Markov models. Subsequently, Deshaies and Joazeiro18 characterized ~300 RING and U-box E3 ligases, while Medvar et al.19 documented ~377 E3 ligases, with a primary focus on confirmed catalytic activity. Despite these efforts, many human E3 ligases have been only partially characterized. A significant fraction remains unexplored and hypothetical or unknown20. To date, those studied exhibit extensive heterogeneity in their sequence, domain composition, 3D structure, subcellular localization, and tissue expression, establishing them as one of the most diverse classes of enzymes. Furthermore, several E3 ligases function as multi-subunit complexes with varied substrate specificities modulated by specific receptors, adaptors, and scaffold proteins21. The extensive variety and large numbers of E3 ubiquitin ligases create a bottleneck for pattern recognition and large-scale study. Therefore, detailed characterization and analysis of the human E3 ligome—the complete set of E3 ubiquitin ligases encoded by the human genome—is essential for a comprehensive understanding.

The current classification of the E3 ligases—based on the ubiquitin-transfer mechanism—categorizes them into three main classes: RING (Really Interesting New Gene), HECT (Homologous to the E6AP Carboxyl Terminus), and RBR (RING-Between-RING) classes15. This classification drastically oversimplifies the mechanistic diversity of E3 ligases, compels the grouping of enzymes with hybrid characteristics, and fails to accommodate emerging information on recent and atypical ligases, limiting its overall utility18. A multi-scale classification of the human E3 ligome offers a unique solution to tackle the complexity and remarkable diversity inherent in these enzymes at various scales. This organized approach can provide more accurate and functional groupings, crucial for a nuanced understanding of different E3 ligase families. Further, emerging patterns detected help trace evolutionary relationships more effectively, revealing conserved elements and adaptive changes that are not evident. Furthermore, mapping essential information such as functional diversity, substrate-specificities, and druggability onto the classification provides a global view, guiding specific and directed investigations to fill in the missing information.

Here, we systematically catalog all E3 ubiquitin ligases to build a comprehensive and manually curated human E3 ligome. We then encode the relationships between high-confidence E3 ligases using multiple distance measures at various granular layers spanning the molecular- and the systems-level organization. By amalgamating selected distance measures from multiple layers into an optimized emergent distance metric, we group all human E3 ligases into distinct families and subfamilies. Our classification delineates features and patterns specific to E3 ligase families, providing insights into their organization. By combining CRISPR-Cas9 dropout screens and proteomic analysis with functional enrichment analysis, we identify essential E3s differentially regulated under stress conditions. We demonstrate the utility of this unbiased classification by mapping the existing state of knowledge on E3 ligase domain architecture, 3D structure, function, substrate networks, and small molecule interactions to gain generic and family-specific insights. The multiscale classification framework developed here embodies canonical and atypical E3 mechanisms largely reflecting the ubiquitin code, offering a comprehensive roadmap to navigate the vast landscape of E3 ligase biology, laying the groundwork for future therapeutic applications.

Results

Assembly of the human E3 ligome

To comprehensively identify all E3 ligases in the human genome, we conducted a census using datasets from previously published studies and public repositories. By visualizing their overlaps, we found that all existing datasets were largely inconsistent (Fig. 1a and Supplementary Fig. 1a). Most strikingly, only 99 proteins were consistently categorized as human E3 ligases from all eight datasets. The low overlap in these datasets reflects the diverse approaches and often variable definitions used to collate E3 systems (Supplementary Table 1). We resolved these conflicts by explicitly defining the catalytic components of E3 systems, i.e., polypeptide sequences containing one or more catalytic domains (C ={d**c}, see “Methods”). Using this objective criterion (({{X}_{i}\in {\bigcup }_{n=1}^{{8}| \exists {d}_{i}\in C}); Supplementary Table 2) facilitated proper annotation and targeted analysis of E3s. We found that 462 polypeptide sequences, across all datasets (({\bigcup }_{n=1}}{8}{A}_{n}=1448)), contain at least one catalytic domain constituting the curated E3 ligome (Fig. 1b and Supplementary Fig. 1b).

Fig. 1: Diversity of the human E3 ligome.

a A visualization showing the intersections of eight E3 ligases datasets (A1, ⋯ , A8) obtained from existing literature and public repositories. The matrix layout for all intersections of individual datasets is sorted by size. Filled circles and their corresponding bars indicate sets that are part of the intersection and their sizes, respectively. Individual proteins (X**i) from the all eight datasets ({\bigcup }_{n=1}^{{8}{A}_{n}=1448) annotated with one or more domains, d**i, belonging to a set of well-studied catalytic components of E3 enzymes (C = {d**c}) were compiled to form the high-confidence E3 ligome, ({{X}_{i}\in {\bigcup }_{n=1}}{8}| \exists {d}_{i}\in C}). b Pie chart showing the extent of protein annotations and filtering to identify the catalytic and non-catalytic components of the human E3 ligome. c Distribution of consensus scores for all annotated protein classes reflects cross-dataset reproducibility on E3 ligase catalytic components. The distribution of (d) protein lengths and annotation coverage for (e) all domains and (f) catalytic domains highlights the heterogeneity of the E3 ligome. g Distribution of structural coverage of the E3 ligome at class-level. Barplots (left axis) display the number of available PDB structures for n = 208 RING, n = 21 HECT, and n = 8 RBR proteins. Violin plots (right axis, min, max, median, and mean values with mirrored density estimates on either side) represent distributions of fractional coverage for n = 2001 RING, n = 168 HECT, and n = 90 RBR structures. h The total number of unique GO terms associated with E3 classes indicates their functional vista under biological process BP, cellular component CC, and molecular function MF ontologies. n-values on bars indicate unique proteins with GO terms.

To substantiate our curation process, we defined a consensus score for each protein based on its presence in various source datasets (Fig. 1c). We found that the HECT and RBR classes of E3 ligases showed high agreement across datasets (confidence score ≥ 0.6; orange and purple bars). The RING class (green bars) had a broad distribution of consensus scores indicative of annotation challenges. However, the most significant discrepancy among the datasets (confidence score ≤ 0.25) was due to misannotated proteins. E1, E2, and other non-catalytic components of E3 systems, such as receptors, scaffolds, and adaptor proteins, were often merged with E3 ligases, accounting for an additional 298 proteins, leaving 514 unclassified and 174 with no domain annotations (Fig. 1b). Furthermore, several proteins obtained from UniProt and BioGRID using keyword-based searches (Supplementary Fig. 1c) were not included by others, have low consensus scores, and remain unclassified and unannotated, excluding 688 proteins from the curated E3 ligome (Fig. 1c, black bars). Our approach thus minimized false positives and true negatives, includes high-confidence catalytically active E3s and accounts for pseudo-E3 ligases and other E3s with untested catalytic activities, providing a detailed assessment of the completeness of the human E3 ligome (see Supplementary Note 1).

To quantify the diversity of the human E3 ligome, we mapped the sequence, structure, and functional features of individual E3s corresponding to well-known E3 classes (RING, HECT, and RBR). We found that the length distribution of the E3s is broad, ranging from 100 to 5000 residues (mean size = 635 residues; Fig. 1d). The average fractional coverage of E3s annotated with unique domains is 37%, 42%, and 53% for RING, HECT, and RBR classes, respectively (Fig. 1e). Furthermore, on average, the RING, HECT, and RBR domains span 23%, 31%, and 39% of their total lengths, respectively (Fig. 1f). By mapping information from the Protein Data Bank (PDB), we found 2259 distinct structures representing RING, HECT, and RBR-containing proteins (2001+168+90), providing partial structural information for 51% (208+21+8 unique UniProt records) of the E3 ligome (Fig. 1g). Analysis of AlphaFold models revealed that for most E3s, the coverage of structured domains is high, and the amount of intrinsic disorder is generally low (pLDDT ≤ 50 covering only ≤ 10% E3 length; Supplementary Fig. 1d). We quantified the functional diversity of the E3 ligome by retrieving the unique Gene Ontology (GO) annotations corresponding to Biological Processes (BP), Cellular Component (CC), and Molecular Function (MF). We annotated 96−100% of the E3s with unique GO terms (Fig. 1h). The number of distinct GO terms captured the diversity of functional assignments attributed to the three E3 classes.

Metric learning for classification of the human E3 ligome

To study the organization and relationships of proteins within the human E3 ligome, we attempted to classify these enzymes using multiple sequence alignment (MSA) followed by phylogenetic tree construction. However, we obtained a low-quality MSA with numerous gaps (Supplementary Fig. 2a), primarily due to (i) high sequence divergence, (ii) numerous proteins with uneven length distributions, (iii) inadequate alignment of conserved, catalytic domains, and (iv) an extensive repertoire of domain architectures (Supplementary Fig. 2b).

To capture the complex relationships within the human E3 ligome, we used a machine-learning approach to learn an emergent distance measure. Using a linear sum model, we combined multiple distance measures with optimal weights to reproduce class-level organization (partial ground truth) in hierarchical clustering (Fig. 2a). We first computed twelve pairwise distance matrices for all E3 ligase pairs (({d}_{PQ}^{{i}) where i = {1, ⋯ , 12}, for all P and Q ∈ E3 ligome; (12\times}{462}C_{2}) distances) across distinct granular layers: primary sequence, domain architecture, 3D structure, function, subcellular localization and expressions (see “Methods”). These distances between ligase pairs are widely distributed and capture their relationships across distinct molecular- and systems-level hierarchies (Fig. 2b). Interestingly, most distance measurements showed low correlations (Fig. 2c), suggesting that they capture largely orthogonal information from the distinct granularity layers. Only the three domain architecture-based distances which quantify domain composition (({d}_{{{\rm{PQ}}}}^{{{{\rm{Jac}}}})), domain order (({d}_{{{\rm{PQ}}}}}{{{\rm{GK}}}\gamma})), and domain duplication (({d}_{{{\rm{PQ}}}}^{{{{\rm{Dup}}}})) are highly correlated (Pearson r ≥ 0.5). Further, the 3D structure-based distance measure (({d}_{{{\rm{PQ}}}}}{{{\rm{Str}}}})) is also positively correlated with domain composition and duplication distances (Pearson r ≥ 0.5).

Fig. 2: Metric learning for E3 ligases.

a Schematic of the metric learning process. b Distribution of various pairwise distance measures spanning the molecular and systems level organization. c Pearson correlation of distance measures indicate orthogonality, mostly r ∈ (− 0.3, 0.3). Distances based on sequence alignment, domain composition, 3D structure (catalytic), and molecular function (marked in blue) are combined into an emergent distance (DPQ) with appropriate weights. d By maximizing element-centric similarity, a measure of the overlap of emergent hierarchical clusters (right) with the ground truth (left) (e) evaluates individual metrics and their linear combinations. f Regression weights (mean± S.D.) corresponding to the four relevant distances as a function of fractional tree cutoff h. 100 clusters with largest SEC were sampled at each value of h to estimate the mean and S.D.

Next, to learn an emergent distance measure, DPQ, we combined four individual distances (({d}_{{{\rm{PQ}}}}^{i})), representative of E3 sequence, domain composition, structural, and functional level organization, with their appropriate weights (w**i ∈ {0.05, ⋯ , 0.95} in 0.1 intervals). These distances capture intrinsic molecular attributes and their relationships spanning the molecular scale. By uniformly sampling the weights, we constructed 105 combination measures as a function of the hyperparameter (fractional tree cutoff, h, between 0.05 and 0.95). By simultaneously maximizing element-centric similarity22 of the emergent hierarchical clusters resulting from combined measures, with partial ground truth (weakly-supervised scheme, Fig. 2d), we optimized an emergent distance measure (DPQ) with appropriate weights (({\widehat{w}}_{i})). We found that the linear combination of distances provided clusters with high element-centric similarity SEC compared to clusters obtained from individual distances (Fig. 2e, black curve vs. colored).

Normalized Mutual Information (NMI) and Fowlkes–Mallows Index (FMI) compare clustering assignments (various distance-based vs. ground truth), but they are sensitive to cluster count (determined by tree cutoff, h; Supplementary Fig. 3a). Therefore, optimized weights ({\widehat{w}}_{i}) were obtained by averaging one hundred realizations of hierarchical clustering with maximum SEC22. The weights corresponding to maximum SEC initially varied and then plateaued (at h ≥ 0.75; Fig. 2f), resulting in the construction of an optimized emergent distance measure, DPQ (Eq. (1)). We found that the relative influence of 3D structure, domain composition, and sequence alignment ((\widehat{{w}}_{i}\ge 0.5)) was more significant on the final learned metric and its ability to reproduce class labels accurately. Compared to the emergent distance measure, we found variable tree topologies with poor overlap and highly entangled trees for all four individual distances (Supplementary Figs. 3b–e).

$${D}_{{{\rm{PQ}}}}=0.43{d}_{{{\rm{PQ}}}}^{{{{\rm{MF}}}}+0.55{d}_{{{\rm{PQ}}}}}{\gamma }+0.60{d}_{{{\rm{PQ}}}}^{{{{\rm{Jac}}}}+0.70{d}_{{{\rm{PQ}}}}}{{{\rm{Str}}}}.$$

(1)

Organization of the human E3 ligome

Using the optimized emergent distance metric, DPQ (Eq. (1)), we constructed a scaled hierarchical tree classifying the human E3 ligome (Fig. 3 and Supplementary Fig. 4a). To assess the validity of nodes, branch stability, and the robustness of our classification, we resampled the emergent distance matrix (n = 500) and assigned bootstrap support at each branch point (Fig. 3, grey circles). The bootstrap support for all nodes beyond tree cutoff, h > 0.15, is 95−100%, indicating a stable branch pattern (Supplementary Fig. 4b) with a fixed tree topology. At *h *≤ 0.15, the bootstrap support for the nodes dropped drastically. This allowed us to use a tree cutoff threshold, h = 0.25, to parse the dendrogram and obtain robust and stable clusters with clear family and subfamily patterns while preserving RING-, HECT-, and RBR-class segregation.

Fig. 3: Classification of the human E3 ligome.

Unrooted hierarchical tree computed using the optimized emergent distance metric DPQ (scaled branch lengths). The RBR (purple), HECT (orange), and RING classes (blue/green/yellow) are partitioned at h = 0.25 into 1, 2, and 10 families, respectively. Each cluster is defined by shared sequence, domain-architectural (mapped), structural, and functional elements. Boxes show family information, i.e., family name, size, and subfamilies, with representative examples. Grey-filled circles denote bifurcation nodes with ≥ 95% bootstrap support, and * denotes families with a few class-level outliers (3/13).

We identified thirteen distinct clusters or E3 families (h = 0.25). At the class level, the E3 ligome is well segregated into ten RING families (Fig. 3, blue to green colors; clockwise arrangement from RING1 to RING10), two HECT (Fig. 3, top-branch; orange), and one RBR family (Fig. 3, bottom-branch; purple). Each E3 family is subdivided into one or more subfamilies (Fig. 3, boxes) with distinct patterns. Mapping domain architecture information onto the individual leaves aids recognition of well-preserved sequence and domain features, consistent with family and subfamily grouping, a pattern more evident in the unscaled circular dendrogram of the E3 ligome (Supplementary Fig. 4a). Further, few heterogeneous families are grouped more closely and emerge from single branches (bootstrap support ≈ 90−95%; Supplementary Fig. 4b) hinting at divergence of plausible superfamilies: (i) RBR and RING1–3 branch (small E3s), (ii) RING7–9 branch (medium E3s), and (iii) HECT2–RING10 branch (large E3s). This organization stems from the central node that bifurcates the E3 ligome into two groups characterized by average protein size (Fig. 3). The bottom branch displays six families with smaller E3s, while the top branch groups seven larger E3 families.

The E3 family organization reflects mechanistic differences (Supplementary Fig. 4c and Supplementary Table 3). The RING E3s mediate the direct transfer of Ub to the substrate, while the RBR and HECT E3s enable ubiquitin transfer via a two-step mechanism. The RBR-containing E3s form a homogeneous cluster, highlighting their conserved sequence and the TRIAD supra domain. Similarly, HECT-domain-containing E3s are organized into two clusters/families, HECT1 and HECT2. The HECT1 family is homogeneous and includes three subfamilies: NEDD4-like, HERC, and other HECT E3s. The HECT2 family contains a pure HECT E3 subfamily and an outlier subfamily containing large multi-domain RING-type E3s ( > 2000 residues) often with atypical mechanisms (e.g., MYCBP2, RNF213, see Supplementary Note 2). The most abundant RING-domain-containing E3s are organized into 10 families, each characterized by further grouping related proteins into distinct subfamilies with shared sequence elements, domain architectures, and structural features. For instance, the RING2 family comprises membrane-associated RING-CH-type domain (MARCH) E3 ligases (Fig. 3, bottom-right). This family includes all small MARCH E3 ligases characterized by their transmembrane domains and sequence lengths below 500 amino acids. TRIM E3 ligases are exclusively limited to two distinct families, RING5 and RING8, and feature the SPRY domain (Fig. 3, bottom-left). E3 ligases containing BTB/POZ and Zn-finger domain repeats are grouped into the RING6 family (Fig. 3, upper-left).

Although our emergent metric largely maximizes pure and homogeneous clusters (e.g., RBR, RING2, RING5, RING6, RING8, and HECT1), heterogeneity often arises at the subfamily level, resulting in sub-groupings of E3s with varied and unique domain architectures. Isolated proteins (singletons) in the RING1, RING7, RING8, and RING9 families form distinct subfamily groupings, complicating pattern detection. Only RING1, RING7, and HECT2 families display occasional class-level outliers (Supplementary Table 3). Supplementary Notes 3 to 15 describe each family structure in detail with information on subfamily branching, characteristic features, and distinct patterns, along with outliers providing a nuanced description (Supplementary Figs. 5–17 and Supplementary Notes 3–15).

Functional segregation of the human E3 ligome

To evaluate the human E3 ligome, we conducted a CRISPR-Cas9 dropout screen of UPS genes, using cellular fitness as the main phenotype. We identified 53 catalytic and 32 non-catalytic E3 components to be essential for cell fitness (FDR ≤ 0.05 and (| {\log }_{2}(FC)| \ge 1.0); Fig. 4a, b). Notably, these essential E3s were predominantly enriched in RING1/4/7, and RING9 families, suggesting critical biological roles (Fig. 4a). Several E2 enzymes and adaptors were also essential, reinforcing the importance of the ubiquitin conjugation and multi-subunit E3s (Fig. 4b). Overall, our CRISPR screen replicates correlated well with DepMap data (Pearson r ≥ 0.5; Supplementary Fig. 18a). GO analysis of 53 essential E3s showed significant enrichment for nuclear components and DNA damage, replication, and repair processes (Fig. 4c), indicating their roles in genome integrity and nuclear regulation fundamental to cell survival. These findings point to essential E3 components crucial for cell viability.

Fig. 4: Functional segregation of the E3 ligome.

Volcano plots of Gene essentiality analysis derived from CRISPR screens for (a) catalytic and (b) non-catalytic components of the E3 ligome. c GO enrichment analysis for essential catalytic E3s. d The functional landscape of the E3 ligome (biological processes) is captured by the network of GO annotation clusters. Individual nodes representing GO clusters (20 labeled) are drawn as pie charts (sizeproportional to # of E3s; colored by family enrichment) connected by distinct edges (κ-similarity ≥ 0.3). e The heatmap displays all functional clusters corresponding to family-specific enrichment of E3s (p value estimated using hypergeometric test (two-sided), discrete color scale for p value ≤0.01; white otherwise). Colored triangles show examples of family specific enrichment for (f) K6-linked ubiquitination (purple) and antiviral innate immune response (green), (g) starvation response under 6h EBSS treatment (blue), and (h) DNA damage response under 4h 100 nM CPT treatment (orange). For panels f–h gene essentiality data ({\log }_{2}(FC)) or DepMap Gene Effect scores (*) are plotted for individual E3s. The ratio denotes the fraction of E3s with experimental evidence (PMIDs) for GO functions directly. g, h panels also show volcano plots of proteomic analysis, highlighting significantly up-regulated and down-regulated proteins (red scatter; adjusted p values were obtained using Benjamini-Hochberg method in two-sided moderated t-tests) with overlapping E3s (colored) and control proteins (blue filled circles).

To understand the functional diversity of the human E3 ligome, we filtered high-confidence GO terms and mapped them onto our classification, enabling us to draw functional clusters and visualize their networks across all three ontologies. This allowed recognition of generic and family-specific functions (enriched, (-{\log }_{10}(p)\ge 2)). At the BP level, as expected, the network analysis revealed prominent core functional sub-clusters associated with all terms containing “ubiquitination (Ub)” (Fig. 4e, Top). These BPs are shared across all families, indicating their generality. Additional core clusters relate to innate immunity, host-driven viral restriction, NF-κB regulation, and IL-17 signaling (Fig. 4d). Further, cooperative diverse non-degradative functions such as regulation of gene expression, protein stability, cell growth, and ERAD pathway are enriched within the E3 ligome (see Supplementary Note 16).

Examining family-specific GO enrichment uncovered functional specialization supported by experimental evidence (Fig. 4e, colored triangles; Supplementary Table 4). For instance, the RBR family members, RNF14, RNF144A, and PRKN, demonstrated specificity for K6-linked-Ub (Fig. 4f, left). K6-linked chains flag stalled RNA-protein cross-linked complexes (RNF14), DNA-sensing adaptor STING for activation of interferon signaling (RNF144), and damaged mitochondria for clearance (PRKN)23,24,25. Similarly, TRIM E3s (RING5) were significantly enriched in antiviral innate immune response (Fig. 4f, right). They regulate pattern recognition receptor activity in cells, such as RIG-1 and MDA5-mediated responses26,27.

To test family-specific functions, we performed whole-cell proteomics in response to cellular perturbations (EBSS and CPT treatment) and monitored the differential expression of E3 ligases. The cellular responses to EBSS and CPT treatments are broad and multifaceted, impinging on a wide range of BP functional clusters (Fig. 4e, blue and orange triangles). 18/34 implicated E3s have direct evidence linking them to starvation response pathways (Supplementary Table 4). Notably, MGRN9 and BRCA1 were also essential in our CRISPR screens (Fig. 4g, left). Our analysis revealed a differential expression (p value ≤ 0.05) of the E3 ligases TRIM27/32, and UBR1 in addition to key autophagy regulators SQSTM1, CALCOCO2, GABARAPL2, and MAP1LC3B2, highlighting a coordinated modulation of the autophagic machinery during starvation (Fig. 4g, right colored vs. blue). In contrast, the observed up-regulation of EIF4EBP1, a translational repressor regulated by mTOR signaling28, and GDF15, a stress-responsive cytokine involved in metabolic adaptation29, indicates activation of complementary stress response pathways that may support cellular survival. TRIM27/32 regulate autophagy initiation and selective degradation pathways by ubiquitinating essential autophagy proteins such as ULK1 and p62, thereby promoting autophagosome formation and cargo recognition30,31,32.

Similarly, experimental evidence links 29/87 implicated E3s directly to DNA damage response (DDR) (Supplementary Table 4), of which 12 E3s turned out to be essential (e.g., TRAF3/7, MDM2; Fig. 4h, left). We found that the E3 ligases TRIM27, UHRF1, TNFAIP3, and RNF168 were significantly down-regulated in response to CPT, while UBR5 was up-regulated in addition to the control proteins (e.g., TOP1, XRCC6, DDB1; Fig. 4h, right, colored vs. blue). TOP1, targeted by CPT, forms covalent DNA-protein cross-links that cause replication-associated double-strand breaks and trigger DDR33. In response, ubiquitin signaling, via RNF8/168-mediated K63-linked histone-Ub, recruits DDR mediators like 53BP1, BRCA1, and RAD1834,35. UBR5 limits Ub-signaling by degrading RNF168, ensuring DNA repair fidelity. Further, UHRF1 promotes DDR by driving chromatin remodeling and BRCA1 recruitment36. Additionally, UHRF1, RNF168, and RNF8 function in DNA replication37. Together, these mechanisms underscore Ub-signaling to maintain cellular homeostasis and genome integrity.

At the MF level, all E3s have “ubiquitin-protein ligase” activity (Generic; Supplementary Fig. 18b). More than 20 MFs could be attributed to family-specific domain architectures (Supplementary Fig. 18d). The Zn-finger domains are also common to transcription factors. For e.g., they equip E3s for p53 binding (RING3), histone-Ub (RING9), and unmethylated CpG binding (RING7). Other non-catalytic E3 domains mediate PPI interactions with actin, SH3-domains, Kinases, and proteases. Distinct subcellular localization of E3s exerts spatial control of protein-Ub (Supplementary Fig. 18c). Most E3 ligases are cytosolic, which form an essential part of the “ubiquitin ligase complexes” (CC: Generic). We find enriched E3 families for distinct CCs: CD40 receptor, GID, and nBAF complexes (RING1); early endosomes and lytic vacuoles (RING2). In total, we identified 17 unique cellular components with distinct E3-specific enrichment patterns (Supplementary Figs. 18e, 19a).

GO annotations from author/curator statements and electronic methods—despite varying confidence levels—provide a rich, abundant dataset of key testable hypotheses on E3 systems (Supplementary Figs. 18f, 19a, and Supplementary Note 17). Publication counts for each protein-specific annotation highlight knowledge gaps and pinpoint well-studied and underexplored E3 systems (Supplementary Fig. 19b).

Interaction landscape of the human E3 ligome

E3 ligases can operate as standalone or multi-subunit enzymes. Complex E3s consist of scaffolds, adaptors, and substrate receptors that determine specificity, stability, and regulation21. For instance, RBX1 binds scaffolds (CUL1–CUL5) and anchors the E2 enzyme to form the catalytic core for Ub transfer (Fig. 5a). Its interactions with various cullins, adaptors, and receptors enable ~250 CRL configurations, providing modular regulation and substrate specificity. By contrast, standalone E3 ligases, like MDM2, c-CBL, PARKIN, or SMURF1/2, either have specialized domains or undergo specific PTMs that recognize substrates and facilitate E2 binding and ubiquitin transfer. For example, HECTD3 operates via a two-step ubiquitin transfer mechanism (Fig. 5b). However, substrate binding occurs through specific motifs within the non-HECT regions that presumably recognize particular degrons (sequence motifs, distinct PTMs, or unique structural elements).

Fig. 5: Protein–protein interactions of the E3 ligome.

Representative examples of E3 ligases functioning as a (a) multi-subunit protein complex (CRL) or (b) a standalone enzyme (HECD3). c Venn diagram of pairwise interactions of adaptors, receptors, and scaffold proteins with E3s. d Annotation of 462 E3 ligases into complex, standalone, or unclassified modes of action. e Family-wise mapping of data from (d). f Pairwise E3–substrate interactions for all E3s obtained by integrating data from known ESIs, mapped transient direct and indirect PPIs, and predicted ESIs. g Mapping of the ubiquitinated proteome with E3s (≈ 62%, n = 12464). h Schematic showing substrate categorization into E3-specific, family-specific, and promiscuous classes (left) and their relative distributions mapped onto E3 families (right). i Representative examples for the three types of ESI networks.

Previous annotations38,39 report 6 complex, 329 standalone, and several unclassified E3s. We extended this annotation by curating non-catalytic components and cataloging their direct interactions with E3s (Fig. 5c). Multi-subunit complex structures are only resolved for four E3s (RBX1/2, ARI1, and APC11). While partial complexes are resolved for 12 E3s (e.g., APC11, ARI1/2), we found several binary direct interactions between E3s and non-catalytic subunits, re-annotating 75 E3s operating as complexes (Fig. 5d, black), leaving 277 standalone38 and 110 unclassified E3s (Fig. 5d, red). RING8 family displayed many complex E3s (50%), followed by RING1 (26%), while RING2 and HECT2 families displayed entirely standalone E3s (Fig. 5e, Supplementary Table 5). Consistent with our findings, we observe that MARCH-type E3s (RING2) operate in the membrane environment primarily as standalone enzymes. HECT2 proteins (e.g., HECD3) possess multiple domains for adaptor, receptor, and scaffolding, explaining their standalone function.

Next, we assembled the E3–substrate interaction (ESI) network by integrating data from known ESIs (n = 2012), direct PPIs (n = 5844), indirect PPIs (n = 6530), and predicted ESIs (n = 64802; Fig. 5f, see Methods). Integrating these data and verifying their ubiquitination status resulted in excluding false positives (E3-associated) and improving the annotation of likely substrates (Supplementary Figs. [20](https://www.nature.com/articles/s41467-025-67450-