Human cell lines LX-2 (stellate liver cells; cat. no. SCC064; MilliporeSigma) and HepG2 (liver cancer cells; American Type Culture Collection; cat. no. HB-8065; LGC Limited) were cultured in standard conditions (37°C and 5% CO₂), using 1X DMEM (cat. no. 41966-029), supplemented with 10% FBS (cat. no. 16000044), 1% penicillin/streptomycin (cat. no. 10378016; all Gibco; Thermo Fisher Scientific, Inc.) and 1% L-Glutamine (cat. no. BEBP17-605E; Lonza Group, Ltd.). Both LX-2 and HepG2 cell lines have been authenticated by advanced transcriptomics, proteomics, metabolomics and functional kinomics (whole-kinome activity) profiling.

LX-2 cells (6×10⁴; ~80% confluency) were transiently transfected with the expression plasmids EGFP-hAgo2 (cat. no. 21981; 0.8 µg) and RNT1-GFP (cat. no. 17708; both Addgene, Inc.) (0.8 µg), using Lipofectamine^® 2000 (cat. no. 11668027; Invitrogen; Thermo Fisher Scientific, Inc.; 1.5 µl), according to manufacturer's instructions for 6 h at 37°C. The pcDNA3.1(+) (cat. no. V790-20; Addgene, Inc.) empty vector served as a negative control. Cells were incubated at 37°C and 5% CO₂, for 24 h post-transfection for immunofluorescence.

LX-2 and HepG2 cells (6×10⁴; ~80% confluency) were cultured on 10 mm round coverslips and fixed in 4% paraformaldehyde (cat. no. P6148; Merck KGaA) solution for 10 min at room temperature. The cells were membrane-permeabilized by incubation in 0.1% Triton-X 100 (cat. no. 1086031000; Merck KGaA), followed by blocking with 5% BSA (cat. no. A7906; Merck KGaA) for 1 h at room temperature. Cells were exposed to anti-AGO2 (cat. no. ab57113; Abcam) (1:100), anti-DICER (double-stranded RNA endoribonuclease) (cat. no. NBP1-06520; Novus Biologicals, LLC) (1:100), anti-TRBP (transactivation response element RNA-binding protein; cat. no. ab180947; Abcam) (1:100), anti-UPF1 (up-frameshift protein 1) (cat. no. 12040, Cell Signalling Technology, Inc.; 1:100), anti-centrin-2 (cat. no. 2091, Cell Signalling Technology, Inc.) (1:200), anti-peri-centriolar-material (PCM)-1 (cat. no. 5213, Cell Signalling Technology, Inc) (1:200), anti-γ-tubulin (cat. no. T5192; Merck KGaA) (1:200) and anti-α-tubulin (cat. no. 3873, Cell Signalling Technology, Inc.) (1:100) primary antibodies, overnight at 4°C. Cells were incubated with secondary antibodies [goat anti-mouse (cat. no. A-11004; Invitrogen; Thermo Fisher Scientific, Inc.) and anti-rabbit IgG (H+L)-Alexa Fluor^™ 568 (cat. no. A-11011; Invitrogen, Thermo Fisher Scientific, Inc.) and anti-mouse IgG (H+L)-Alexa Fluor 488 (cat. no. A-28175; Invitrogen; Thermo Fisher Scientific, Inc.) and donkey anti-rabbit IgG (H+L)-Alexa Fluor 488 (cat. no. A-21206, Invitrogen; Thermo Fisher Scientific, Inc.)] (1:500) for 1 h at room temperature. Vectashield^® Mounting Medium, with DAPI (Vector Laboratories, Inc.), was used to visualize cell nuclei at 405 nm excitation wavelength, using confocal laser scanning microscopy.

LX-2 cells (6×10⁴; ~80% confluency) were treated with 0.4 µg/µl demecolcine (cat. no. D7385; Merck KGaA), an inhibitor of tubulin polymerization, for 6 h at 37°C, and processed (fixation in 4% paraformaldehyde solution for 10 min at room temperature, followed by blocking in 5% BSA for 1 h at room temperature) for immunofluorescence. DMSO (cat. no. A3672; Merck KGaA) was used as the control (reference condition).

Observation of immunofluorescence stained, or transiently transfected fluorescent cells was performed using the Confocal Laser Scanning Microscope Leica SP5 (Leica Microsystems GmbH) and LAS-AF (version 2.7.3.9723) software (Leica Microsystems GmbH).

For modelling, the experimental structure with the highest resolution available was selected for AGO2 [Protein Databank (PDB) ID: 4Z4D chain: A (4Z4D.A)] (36). Structure was retrieved from PDB-REDO (37) and refined using the Protein Preparation Wizard of the Schrödinger Maestro Suite (version 2023-2) (38). The ClusPro Server (39) was used to predict docking positions for AGO2-UPF1 (PDB ID: 6EJ5.A) (40) and AGO2-centrin-2 (PDB ID: 8EBX.J) (41). Constrained docking between AGO2 and tubulin-A1A was performed, as previously described (42), using 6WSL.A (43). Interacting residues of tubulin-G1, within the γ-tubulin ring (PDB ID: 6V6S) (44), were excluded from docking between AGO2 and tubulin-G1 (PDB ID: 3CB2.A) (45). Structures 6EJ5.A, 8EBX.J and 3CB2.A were retrieved from Research Collaboratory for Structural Bioinformatics (RCSB) PDB (46), and refined with PDBFixer (version 1.9) (47) and APBS-PDB2PQR (version 3.6.2) (48).

DICER structure (PDB ID: 7XW3.A) (49) from RCSB PDB was processed with EasyModel Server (50) via Basic Modelling/Loop Refining (51) (1,450–1,470 residue range) and APBS-PDB2PQR (48). As DICER is a large protein, docking between AGO2 PIWI and DICER RNAse III A was performed, since these domains are known to interact, as based on experimental evidence (52). Each detection of interfacing residues relied on the PDBSum Server (53), while the dissociation constant (K_d) of the hypothetical complex in each docking model was predicted using the Prodigy Server (54). Selection of docking models and heatmap production for the computed K_d values were generated, as previously described (42). Predictions of whole-structure DICER-AGO2 and CEP250-AGO2 putative complexes were performed by the AlphaFold Server (55), using protein sequences from UniProt (56) (seed=42). Output files were converted to PDB format using Gemmi (version 0.7.0) (57), for compatibility with PDBsum and Prodigy Server. As AlphaFold Server provides five outputs per complex, the predicted DICER-AGO2 complex that features the PIWI domain at the interphase between DICER and AGO2 was selected, as previously described (51). Concerning the CEP250-AGO2 presumable complex, selection of the output was made according to the predicted K_d value by Prodigy Server. A comparative heatmap, including the K_d values of all putative complexes, was generated, using the Seaborn (/joss.theoj.org/papers/10.21105/joss.03021) (version 0.12) Python Package. VMD (version 1.9.3) (58) and ChimeraX (version 1.9) (59) were used for visualizing the docking models and predicted complexes, respectively.

Molecular mapping of AGO2 protein-protein interactions was conducted via the IntAct Molecular Interaction Database (EMBL-EBI; Wellcome Genome Campus) (60).

Specificity of the anti-AGO2 antibody used in the present study was verified by transfecting human stellate liver (LX-2) cells with a plasmid (EGFP-hAgo2) responsible for the overexpression of human AGO2 protein fused to the (enhanced) green fluorescent protein (GFP). GFP served as an imaging reporter for its sub-cellular visualization and detection. Following GFP-AGO2 overexpression, LX-2 cells were processed for targeted immuno-localization of the endogenous AGO2 protein (red). GFP-AGO2 overlapped with the red AGO2-signals (Fig. 1), demonstrating that the antibody used in the present study specifically bound human AGO2 protein, in liver cells. Furthermore, GFP did not alter AGO2 ability to be recognized by the, herein, used antibody, thereby indicating that, besides antigenicity, both locality and interactivity of GFP-AGO2 have to remain unaffected, presenting highly similar profiles to the endogenous AGO2-specific ones. The AGO2-specific (overlap; yellow) punctate pattern in the LX-2 cytoplasm suggests key essential roles of the protein in the control of local homeostasis during interphase of human liver cells. Αn AGO2-dependent operation of the RNAi machinery in multiple cytoplasmic foci/bodies (such as P-bodies) (61) at the interphase stage (Fig. 1) may indicate the locality-dependent regulation of sub-cellular homeostasis at the mRNA stability (or translation) level.

Sub-cellular localization profiles of the AGO2 protein in LX-2 cells undergoing mitosis were investigated. Despite its cytoplasmic distribution in the form of punctate bodies at the interphase stage (Figs. 1 and 2), AGO2 protein presented novel localization patterns at the mitotic stages of dividing LX-2 cells. AGO2 was immuno-detected in the centrosomes, mitotic spindle and cytokinetic bridge, also exhibiting similarities to the α-tubulin-dependent cytoskeletal network during mitotic, but not interphase, stages (Figs. 2 and S1A-D), thereby indicating a novel, non-canonical, RNAi-dependent, activity of AGO2 in mitotic apparatus, to likely ensure local homeostasis.

To examine the role of oncogenicity in AGO2-specific patterning during the cell cycle, AGO2 sub-cellular distribution in HepG2 human liver cancer cells undergoing mitosis was immuno-profiled. Similarly to LX-2, HepG2 dividing cells were characterized by AGO2-localization patterns in the centrosomes, mitotic spindle and cytokinetic bridges of mitotic cells (Fig. 3). A cytoplasmic, punctate pattern of AGO2 topology was observed in HepG2 cells during interphase, demonstrating compartmentalization similar to LX-2 cells, both at the interphase and mitosis stages. Liver oncogenicity did not alter AGO2 patterning during cell division, thereby indicating the malignancy-independent role of AGO2 protein in controlling local homeostasis via canonical, RNAi-dependent (at interphase), or non-canonical, RNAi-dependent (at mitosis), mechanisms.

Given the novel immuno-detection profiling of AGO2 protein in liver centrosomes, its co-localization landscape was mapped with key centrosomal (centriolar or PCM) proteins. Immunofluorescence imaging of LX-2 mitotic cells revealed co-localization pattern of AGO2 with centrin-2 (Fig. 4A-D), a key structural component of centrioles (62–64). Notably, AGO2 was also co-localized with γ-tubulin, a protein that serves as a template for the initiation of α- and β-tubulin heterodimer polymerization at the PCM area (64–66), in several of the examined cells (Fig. 4I-L). PCM-1 protein, a component of centriolar satellites/PCM mass (64,67,68), was not observed in proximity to AGO2 (Fig. 4E-H), indicating a novel, non-canonical, RNAi-dependent, role for the AGO2 protein in γ-tubulin-linked, but not PCM-1-associated, peri-centriolar material, in liver cells. Of note, the PCM area comprised distinct structural ‘niches’ (or layers) with different protein compositions among them; for example, one niche contained γ-tubulin and AGO2 in proximity to each other (γ-tubulin/AGO2 co-localization) (Fig. 4I-L), whereas another contained PCM-1 in the absence of AGO2 (lack of PCM-1/AGO2 co-localization; Fig. 4E-H).

To determine AGO2-specific localization profiles throughout cell division, the sub-cellular topology of AGO2 was observed at each stage of mitosis and interphase in LX-2 dividing cells. AGO2 was immuno-detected in LX-2 centrosomes at the interphase and early prophase stages, while AGO2 distribution was similar to α-tubulin patterning during mitotic spindle formation at the metaphase stage, without alteration until late anaphase (Figs. 5 and S1A). During the subsequent cell cycle stages of telophase and cytokinesis, AGO2 was observed in the cytokinetic bridge between the daughter cells (Fig. 5). AGO2 was localized in an interior area of the cytokinetic bridge, forming fibrillar-like structures (Figs. 5 and S1C), which may stabilize the integrity of the bridge in a non-canonical, RNAi-dependent, manner. Taken together, the present data suggest novel sub-cellular topologies of AGO2, which may be associated with its role in the efficient operation of cell division machinery.

Since AGO2 forms the protein core of the RISC complex, it was hypothesized that other RISC components may also reside in key mitotic structures of dividing cells. Hence, the overlapping topology and distribution patterns of AGO2, DICER and TRBP2 proteins in LX-2 cells were examined. At the interphase stage, AGO2 co-localized with DICER in several cytoplasmic foci/bodies (potentially P-bodies; Fig. 6), directly reflecting their locality-dependent joint action in the RNAi mechanism. During mitosis, a co-localization pattern of AGO2 and DICER proteins was detected in liver centrosomes (Figs. 6 and S1E), whereas at the cytokinesis stage, AGO2 and DICER presented distinct sub-cellular compartmentalization profiles; AGO2 resided in the cytokinetic bridge, while DICER was located in the midbody structure (Figs. 6 and S1F). Despite their different distribution, overlap suggested the locality-dependent functionality of RNAi machinery at the cytokinetic bridge/midbody ‘borders’.

The TRBP2 protein, a key component of the RNAi machinery that undergoes cell cycle regulation (42), was not co-localized with AGO2 at any mitotic apparatus (Fig. 7), suggesting the non-canonical, RNAi-dependent, role of AGO2 and DICER joint actions in the LX-2 centrosome and cytokinetic bridge. AGO2 and DICER may function together, in the absence of TRBP2, in a non-canonical, RNAi-dependent, manner, to control local homeostasis at the centrosomal and cytokinetic bridge level in liver cells.

To investigate RNA binding proteins (RBPs) other than AGO2 and their capacity to compartmentalize in the centrosome, mitotic spindle and/or cytokinetic bridge mitotic machinery, UPF1, a key RBP family member and AGO2 interactor (69), was next examined. UPF1 is a monomeric RNA helicase that co-localizes with AGO2 in processing bodies (potentially P-bodies) and contributes to the RNA silencing program (70–72). Hence, localization of both proteins was imaged in control and LX-2 cells transfected with the EGFP-hAgo2 and RNT1-GFP plasmids, responsible for the overexpression of GFP-AGO2 and GFP-UPF1 chimeric proteins, respectively. In control [transiently transfected with the empty vector pcDNA3.1(+)] cells, the endogenous AGO2 and UPF1 proteins presented weak co-localization in the cytoplasm, but not in the nucleus, of LX-2 interphase cells (Fig. 8A). Similarly, the overexpression of AGO2 (GFP-AGO2) protein did not cause notable co-localization with endogenous UPF1, whereas the UPF1 (GFP-UPF1) overexpression resulted in a strong, punctate co-localization profile with endogenous AGO2, with the two proteins being immuno-detected in the same sub-population of cytoplasmic foci/bodies of LX-2 interphase cells (Fig. 8B and C). To ensure technical integrity, target specificity of the anti-UPF1 antibody was checked by transiently transfecting LX-2 cells with the RNT1-GFP expression plasmid, and transfected cells were stained with the anti-UPF1 antibody to obtain overlapping (yellow) signals (Fig. S2). To confirm reliability of the data, the distribution of both AGO2 and UPF1 proteins were separately examined in LX-2 cells, transiently transfected with the empty vector pcDNA3.1(+) (Fig. S3).

The present data suggest GFP did not disturb UPF1 capacity to be specifically recognized by the antibody, likely indicating unimpaired antigenicity that is mechanistically coupled with undisturbed locality and interactivity, in the GFP-UPF1 molecular context and LX-2 cytoplasm setting. UPF1 (GFP-UPF1) overexpression induced cytoplasmic accumulation of UPF1-containing foci/bodies, with a distinct sub-population carrying both UPF1 (GFP-UPF1) and AGO2 (endogenous) proteins (yellow), suggesting their cooperative actions in the RNA silencing process. Concentration-dependent formation of UPF1-bodies may induce AGO2 recruitment to generate UPF1/AGO2-bodies in a locality-specific manner.

To determine the UPF1 sub-cellular patterning towards mitosis, endogenous UPF1 immuno-detection profiles were obtained from LX-2 dividing cells. UPF1 and α-tubulin (a structural component of microtubule networks) distribution profiles were not similar to each other, either in interphase or in mitotic LX-2 cells, since the UPF1 protein was missing from the centrosome/mitotic spindle/cytokinetic bridge pathway that mechanistically typifies mitosis (Fig. 9). Taken together, the present data demonstrated the capacity of AGO2, but not other RBPs, such as UPF1, to specifically control structural architecture and/or functional integrity of the centrosome, mitotic spindle and cytokinetic bridge assemblies, in a non-canonical, RNAi-dependent mode to provide spatial and temporal homeostasis.

AGO2 and α-tubulin exhibited similar sub-cellular distribution patterns, especially during centrosome/mitotic spindle and cytokinetic bridge formation, leading to the hypothesis that the two proteins may physically interact and exhibit crosstalk in a locality-dependent manner. A clinically relevant concentration of demecolcine (0.4 µg/µl), a potent microtubule-polymerization inhibitor, was administered to LX-2 cells, to disintegrate the cytoskeletal system. A complete disruption of the microtubule network was observed 6 h post-administration, with AGO2 also losing its typical, canonical, immuno-localization pattern in LX-2 interphase cells (Fig. 10). During cell division, both centrosome and mitotic spindle were lost, and the AGO2-specific foci in the centrosome/mitotic spindle axis were also absent (Figs. 10 and S1D), demonstrating the ability of fibrillar α-tubulin to regulate the recruitment of AGO2 protein in the centrosome and mitotic spindle assembly of diving cells, to maintain spatial and temporal homeostasis in liver-cell normal settings.

AGO2 localization in mitotic apparatus was observed in cases of abnormal cell divisions. In rare LX-2 cell sub-populations characterized by aberrant number, geometry and topology of centrosomes and mitotic spindles (such as multi-polar spindles), AGO2 co-localized with α-tubulin in all pathogenic/pre-oncogenic assemblies (Fig. 11), demonstrating the importance of AGO2 protein in centrosome/mitotic spindle biosynthesis and structural/functional integrity in dividing liver cells undergoing mitotic stress. Removal or elimination of AGO2 protein, specifically from multi-polar spindles, may inhibit the transition from a pre-oncogenic to an oncogenic cell state, thereby opening a novel therapeutic window of locality-dependent AGO2 activity suppression against human malignancies, including liver cancer. To determine the involvement of AGO2 in tumorigenesis, the interactome landscape of AGO2 protein has been mapped (60). AGO2 is characterized by a plethora of interactions, with several of its interactors being affected by driver oncogenic mutations in human cells (Fig. S4).

AGO2 protein-protein interaction profiling was examined, through an advanced in silico molecular modelling approach. Docking tests were carried out between AGO2 and its (known or potential) interactors, including α- and γ-tubulin, centrin-2, DICER and UPF1 proteins. In silico, AGO2 was able to recognize and bind proteins that immuno-co-localized with AGO2 in liver cells (Figs. 12A and S5). DICER (first) and centrin-2 (second) displayed the highest AGO2-binding activity, according to their comparatively low K_d values, which denoted intrinsic structural stability of protein-containing bi-molecular complexes (Fig. 12B). These molecular modelling metrics corroborate the immunofluorescence-derived co-localization patterns, indicating joint actions between partners (for example, AGO2, centrin-2 and DICER), to control local (for example, centrosomal) homeostasis, during mitosis. α-tubulin had the lowest AGO2-binding capacity (highest K_d) among proteins (Fig. 12B), thus indicating the contribution of additional co-factors to strengthen the interaction. Furthermore, AGO2 is as an interactome component (interactor) of CEP250 (centrosomal protein 250), a centriole linker protein with disordered regions that, due to high plasticity levels, promote multivalent interactions at the centrosome (73). Molecular modelling tests assess the CEP250-AGO2 complex stability in silico (Fig. S6), indicating a novel role of AGO2 in controlling CEP250-specific, local, homeostasis at the centriole linker sub-compartment.