Proteomic Mapping and Targeting of Mitotic Pericentriolar Material in Tumors Bearing Centrosome Amplification

Centrosome amplification is one of the most significant features of solid tumors (1–3). Through advances in cell imaging and clinical resource analysis, centrosome amplification has been observed at all stages of tumor development (4, 5). When cells enter mitosis, the two centrosomes assemble microtubules on each side, forming a two pole-containing spindle that drives chromosome segregation for cell division (6). It is well known that multiple centrioles cause the formation of extra spindle poles and dramatic chromosomal errors, which usually result in mitotic catastrophe (7, 8). To withstand the risk of cell death induced by mitotic catastrophe, tumor cells evolve an enhanced ability to cluster extra centrioles into the spindle pole, which restores order to mitosis, enabling cell survival and proliferation (2, 9–12). Thus, deciphering the systematic factors required for centrosome clustering in tumor cells will be of a great value not only for understanding tumor evolution, but also for providing accurate targets for killing tumors.

The centrosome is an organelle with a complicated structure whose center contains a pair of orthogonal centrioles that are approximately 250 nm in diameter, 200–400 nm in length, and characterized by a radial arrangement of microtubule triplets (13). The centrioles are surrounded by a proteinaceous matrix of pericentriolar material (PCM), which is an electron-dense material and obviously grows during mitosis (14, 15). Given that PCM is insoluble and highly dense (3, 6, 16), it is easy to imagine that the proteinaceous matrix contributes to extra centrosome clustering. This idea is supported by several clues from previous studies; for example, researchers have found that PCM exists as a gel-like condensate (16, 17), PCM proteins such as centrosomal protein 215 (CEP215) promote the clustering of extra centrosomes into pseudobipolar spindles, thereby ensuring viable cell division (18), and there is an excessive increase of PCM in centrosome-amplified cancer cells (19). Through mass spectrometry (MS) analyses of isolated centrosomes from Drosophila and human cells, previous studies identified around 250 and 500 centrosome proteins, respectively, without distinguishing between the components of centrioles and PCM (20, 21). With the development of high-resolution imaging, scientists revealed the higher-order organization of PCM, which is characterized by a complex molecular composition and hierarchical architecture (22, 23). However, proteomic profiling of PCM components in live cells has remained a challenge, largely due to its close connection to the centriole, its nonmembranous structure, and highly dynamic nature.

Over the past years, proteomic labeling methods have been developed previously (24–27). Among these methods, the engineered ascorbate peroxidase APEX2 offers the highest reaction kinetics, requiring only 1-minute labeling. Upon the addition of H₂O₂, APEX2 catalyzed the oxidation of biotin-phenol (BP) substrate into highly reactive phenoxyl-free radical, which quickly reacts with electron-rich amino acid residues (e.g., Tyr, Trp, His, and Cys) in surrounding proteins (25, 28, 29). In our attempts to label centrosomes, we were surprised to observe that when APEX2-EGFP was fused to SAS-6 centriolar assembly protein (SAS6), a coiled-coil protein in the center of centrosome (3), the APEX reaction was specifically blocked by the highly dense PCM during mitosis. In contrast, when APEX2-EGFP was fused to pericentrin (PCNT), which is localized in PCM (30), the APEX reaction occurred during both interphase and mitosis as expected. This difference in labeling inspired us to design a strategy for profiling the mitotic PCM proteome. By subtracting the proteome obtained using the SAS6-APEX2-EGFP reaction from that obtained using the PCNT-APEX2-EGFP reaction, we obtained a mitotic PCM proteome consisting of 392 components from live and undisturbed cells without any synchronization treatment. The PCM proteome shows the highest overlap with known centrosome proteins among existing centrosome proteomes, and also consists of numerous previously unidentified components. Of note, the genes encoding these PCM proteins tended to be highly expressed in a large population of patients with a wide variety of cancers. Distinct correlations of PCM protein sets with diverse types of tumors including breast, ovarian, colon, lung cancers were demonstrated. Importantly, we found that a considerable portion of cancer cells containing extra centrioles as a result of spontaneous centriole amplification or PKL4 overexpression can survive due to their intrinsic centrosome clustering ability. However, these cancer cells die upon disruption of the PCM gene kinesin family member 20A (KIF20A) in vitro and in vivo. Mechanically, KIF20A maintained PCM size for centrosome clustering by driving ubiquitination promotion proteins away from PCM. Thus, our study establishes the proteome of PCM distinguishable from that of the whole centrosome, reveals a strategy for killing tumors with centrosome amplification by targeting PCM components, and provides a proteomic resource for antitumor drug development.

Human materials used in this study were approved by the Medical Science Research Ethic Committee of Peking University Third Hospital (IRB00006761-M2019471). We obtained signed informed consents from all patients participating in the study. Animal experiments were approved by the Institutional Animal Care and Use Committee of Peking University Health Science Center (LA2018256).

Except for those specifically mentioned, all chemicals were purchased from Sigma. Anti-KLC2 (NBP1-83723, RRID: AB_11015689) and Ki67 (NB500-170, RRID: AB_10001977) antibodies were purchased from Novus. Anti-CKAP4 (16686-1-AP, RRID: AB_2276275), CRMP2 (14686-1-AP, RRID: AB_10638925), HSC70 (10654-1-AP, RRID: AB_2120153), HSPBP1 (10211-1-AP, RRID: AB_2121242), KIF20A (67190-1-Ig, RRID: AB_2882485), RIPK1 (17519-1-AP, RRID: AB_10642833), VAPB (14477-1-AP, RRID: AB_2288297), PLK4 (12952-1-AP, RRID: AB_2284150), and β-actin (66009-1-lg, RRID: AB_2782959) antibodies were purchased from Proteintech. Anti-PCM1 (5213S, RRID: AB_10556960), ubiquitin (3933S, RRID: AB_2180538), and cleaved caspase-3 (Asp175; 9664, RRID: AB_2070042) antibodies were purchased from Cell Signaling Technology. Anti-WDR1 (sc-393130, RRID: AB_2910154) and SAS6 (sc-81431, RRID: AB_1128357) antibodies were purchased from Santa Cruz Biotechnology. Anti-USP15 (ab97533, RRID: AB_10678830), H2B (ab1790, RRID: AB_302612), UBA1 (ab180125, RRID: AB_2910149), PSMC5 (ab178681, RRID: AB_2910151) antibodies were purchased from Abcam. Anti-CETN3 (HPA063704, RRID: AB_2685097) was purchased from Sigma-Aldrich. Alexa Fluor 555 streptavidin (S21381, RRID: AB_2307336), horseradish peroxidase (HRP) Streptavidin Protein (21127, RRID: AB_141596), anti-PDLIM5 (38-8800, RRID: AB_2533388) antibody, FITC anti-alpha tubulin mAb (MA1-19581, RRID: AB_1070276), Alexa FluorTM 633 goat anti-mouse IgG (A-21126, RRID: AB_2535768), Alexa Fluor 633 goat anti-Rabbit IgG (A-21071, RRID: AB_141419), Alexa Fluor 488 goat anti-mouse IgG (A-11001, RRID: AB_2534069), Alexa Fluor 488 goat anti-Rabbit IgG (A-11008, RRID: AB_143165), HRP goat anti-mouse IgG (H+L) secondary antibody (32430, RRID: AB_1185566), and HRP goat anti-rabbit IgG(H+L) secondary antibody (31466, RRID: AB_10960844) were purchased from Thermo Fisher Scientific.

To guide APEX2 to centrosome, its sequence was fused to the CDS/CDS fragment of SAS6 and PCNT, and subcloned into pcDNA3 (RRID: Addgene_13031) with the tag of EGFP. For establishment of stable cell line expressing SAS6-APEX2-EGFP or PCNT-APEX2-EGFP, CRISPR/Cas9 was used to insert a sequence of APEX2-EGFP into the endogenous SAS6 or PCNT locus. The guide RNAs for SAS6 gene (AACTGTTTGGTAACTGCCC) and PCNT gene (TGTTTAATCATCGGGTGGC) were designed at https://zlab.bio/guide-design-resources and cloned into pSpCas9(BB) vector (RRID: Addgene_62988) for coexpression with Cas9. APEX2-EGFP sequence flanked by approximately 1,000-bp SAS6- or PCNT-homology arms is present in the donor plasmid. Knockdown of target genes (KIF20A, CKAP4) was achieved by cloning oligos encoding short hairpin RNA (shRNA) into pLKO.1 vector (RRID: Addgene_8453). Supplementary Table S1 provides sequences of shRNAs.

Double thymidine block was used to synchronize HEK293T cells (RRID: CVCL_0063) to early S-phase. After growing to approximately 40% confluence, cells were treated with 2 mmol/L thymidine for 15 hours, released for 9 hours, and blocked again for 15 hours. Cells were washed out of the second thymidine block by extensive washing (three times) with warm DMEM and allowed to progress for the specified time after being released. Afterward, the cells were incubated with 500 μmol/L BP dissolved in DMEM under 5% CO₂ for 30 minutes and used for the APEX reaction. All cells used were cultivated for less than 30 passages, and routinely tested negative for Mycoplasma by PCR.

Cells stably expressing SAS6-APEX2-EGFP or PCNT-APEX2-EGFP were incubated with 500 μmol/L BP dissolved in DMEM for 60 minutes at 37°C under 5% CO₂. To initiate the reaction, H₂O₂ was added at a concentration of 1 mmol/L for 1 minute at room temperature. The reaction was quenched three times by replacing the medium with an equal volume of “quencher solution” (10 mmol/L sodium ascorbate, 10 mmol/L sodium azide, and 5 mmol/L Trolox in DPBS). Cell pellets were then immediately lysed in lysis buffer [50 mmol/L Tris-HCl (pH 7.4), 150 mmol/L NaCl, 0.5% SDS, 0.5% sodium deoxycholate, 1% Triton X-100, 1× protease inhibitor cocktail, 10 mmol/L sodium azide, 10 mmol/L sodium ascorbate, 5 mmol/L Trolox, and 20 mmol/L DTT]. Lysates were gently sonicated before being centrifuged at 12,000 rpm for 5 minutes at 4°C, then added ice-cold methanol and precipitated at −80°C for 6 hours. After centrifuging at 4,000 rpm for 30 minutes at 4°C, the proteins were redissolved in 1% SDS RIPA buffer. A 4 mg protein sample (in 0.2% SDS RIPA buffer) was incubated separately for 2 hours with 150 μL of streptavidin-coated magnetic beads slurry. The beads were then washed twice with 1 mL of RIPA lysis buffer, once with 1 mL of 1 mol/L KCl, once with 1 mL of 0.1 mol/L Na₂CO₃, once with 1 mL of 2 mol/L urea in 10 mmol/L Tris-HCl (pH 8.0), and twice with 1 mL RIPA lysis buffer. After this, biotinylated proteins were eluted by boiling the beads in protein loading buffer containing 20 mmol/L DTT and 2 mmol/L biotin.

The sample for MS was prepared according to previous studies (25, 31). Briefly, biotinylated proteins eluted from streptavidin beads were analyzed on SDS-PAGE. After staining with Coomassie G-250 and destaining with water, each lane of each sample was manually cut into six bands. The six gel fractions were combined into three injections and dried with a vacuum concentrator after in-gel digestion. The dried peptides were reconstituted in 0.1% formic acid and loaded on to C18 StageTips conditioned with 50% acetonitrile/0.1% formic acid. Following two washes with 0.1% formic acid, peptides were eluted with 50% acetonitrile/0.1% formic acid and dried in a vacuum concentrator.

An Orbitrap Fusion Lumos mass spectrometer coupled online to an Easy-nLC 1200 UPLC (Thermo Fisher Scientific) was used for analysis. The extracted peptides were dissolved with 25 μL of solvent A (0.1% formic acid in water), and loaded to a homemade trap column (100 μm × 2 cm) packed with C18 reverse-phase resin (particle size, 3 μm; pore size, 120 Å; SunChrom) at a maximum pressure of 220 bar with 12 μL of solvent A, then separated on a 150 μm × 15 cm silica microcolumn (homemade, particle size, 1.9 μm; pore size, 120 Å; SunChrom) with a gradient of 11%–95% mobile phase B (80% acetonitrile and 0.1% formic acid) at a flow rate of 600 nL/minute for 30 minutes. The gradient elution conditions (30 minutes) were: 11% to 13% mobile phase B for 2 minutes; 13% to 32% for 16 minutes; 32% to 42% for 7 minutes; 42% to 95% for 1 minute; 95% for 4 minutes. An Orbitrap mass analyzer was used to acquire full scans (m/z 350–1,550) with a mass resolution of 120,000 for the MS analysis in a data-dependent manner. The most intense ions selected under top-speed mode were isolated in Quadrupole with a 1.6 m/z window and fragmented by higher energy collisional dissociation with a normalized collision energy of 32%, then detected in the Orbitrap at a mass resolution of 15,000. With maximum ion injection times of 50 and 22 ms, the AGC Targets (automatic gain control) for full MS and MS-MS were set to 4e5 and 5e4, respectively. Dynamic exclusion time was 30 seconds, and peptide match and isotope exclusion were enabled.

Using Proteome Discoverer (Thermo Fisher Scientific, version 2.1, RRID: SCR_014477), raw files were searched against the human RefSeq protein database (November 1, 2017; RRID: SCR_003496). For Q Exactive, the mass tolerance for precursor ions was set to 20 ppm, while the mass tolerance for fragment ions was 0.06 Da. Oxidation of methionine, carbamidomethylation of cysteine, and acetylation of protein N-terminal were included as variable modifications. The number of missed cleavages is limited to two. All peptides are filtered by a false discovery rate of 1%. The peak areas of all identified peptides were calculated on the basis of their MS1 intensity. We used intensity-based absolute quantification (iBAQ) to quantify proteins (32). To normalize loading samples, fraction of total (FOT) was calculated as iBAQ value divided by total iBAQ value of proteins in each sample and multiplied by 10⁵. We conducted three replicates of PCNT-APEX2-EGFP (reaction A) and SAS6-APEX2-EGFP (reaction B). We kept only protein identifications with at least two unique peptides in at least one replicate, as well as FOT values that are positive in at least two replicates. Proteins with their mean FOT value in reaction A groups significantly higher than those in reaction B groups were screened out (P < 0.05). Minus-versus-Add (MA) plot was conducted for visual representation of identified proteins by R packages (ggplot2, RRID: SCR_014601).

A transcript expression analysis was conducted for 392 PCM genes in 9,018 tumor samples (28 cancer types analyzable) and 5,526 paired normal controls (Supplementary Table S2) from the databases of The Cancer Genome Atlas (TCGA; https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga, RRID: SCR_003193) and GTEx (Genotype Tissue Expression, https://commonfund.nih.gov/GTEx/, RRID: SCR_013042) and run in R package (ggplot2). We calculate the mean value of gene expressional medians in all tumor samples and fold change between tumor and normal samples. For overall survival analysis from databases of TCGA on 392 PCM genes, 2,376 tumor patients (across 33 analyzable types of cancers) who relatively highly expressed the 392 genes were compared with a parallel population of 2,376 tumor patients who relatively lowly expressed the 392 genes by Kaplan–Meier curve. Log-rank test compares the survival curves of the two groups, and HRs calculates the HR based on Cox PH Model. The genome alterations of 20 representative PCM genes were annotated by Genome Nexus and standardized by https://cbioportal.org/. Gene Ontology (GO) enrichment analysis of PCM proteins were performed by the R-clusterProfiler package (RRID: SCR_016884) and a Cytoscape plugin, CluGO. The Benjamini–Hochberg procedure was used to adjust the P values for the representative GO terms shown in the current study (33).

The protein–protein interaction (PPI) network of PCM proteins was constructed and visualized by STRING database (RRID: SCR_005223) and Cytoscape software (RRID: SCR_003032), respectively. The crucial genes in PPI network were identified using a Cytoscape plugin, CytoHubba (RRID: SCR_017677). Liquid-liquid phase separation propensity (pLLPS) adopted from the FuzDrop database (https://www.pnas.org/content/117/52/33254). Set 0.6 as pLLPS threshold, proteins with pLLPS larger than 0.6 were regarded as high LLPS propensity proteins. The pharmacologic data of centrosome declustering drug noscapine were obtained from CellMiner (https://discover.nci.nih.gov/cellminer/home.do), a freely available tool that organizes and stores raw and normalized data that represent multiple types of molecular characterizations and pharmacologic levels.

Cells were fixed in 100% ice-cold methanol at −20°C for 10 minutes. The samples were then blocked with 1% BSA-supplemented PBS for 1 hour and incubated with the indicated primary antibodies (1:200–1:500) overnight at 4°C. After washing three times in PBS containing 0.1% Tween 20 and 0.01% Triton-X 100, cells were incubated with an appropriate fluorescent secondary antibody for 1 hour at room temperature. After washing three times, cells were stained with Hoechst 33342 (10 μg/mL) for 15 minutes. Finally, samples were mounted on glass slides and observed under a confocal laser scanning microscope at 63×/1.40 (Carl Zeiss 710). A Z-stack scan of the centrosome was performed at 0.25-μm intervals to obtain the images for three-dimensional (3D) reconstruction. The Imaris program (Bitplane, RRID: SCR_007370) was used to analyze different channels of PCNT/SAS6-APEX2-EGFP, biotinylated proteins, PCM1, and chromosomes imported from CZI files (34, 35). We used 3D surface rendering to reconstruct centrosome components in cells.

HEK293T cells were cultured in DMEM with 10% FBS. MDA-MB-231 cells (RRID: CVCL_0062) were cultured in RPMI1640 Medium with 10% FBS. All cells used were cultivated for less than 30 passages, and routinely tested negative for Mycoplasma by PCR. Live imaging was conducted by growing cells in imaging culture dishes (NEST, 801001) and observing them in an UltraVIEW VoX (PerkinElmer) live cell workstation at 37°C with 5% CO₂ for the indicated time. Volocity (Universal 3D Image, RRID: SCR_002668) was used to analyze the images.

Tumor samples were collected from patients with breast cancer. Supplementary Table S3 contains detailed information about the tumor samples. Following washing with PBS, 0.1–0.2 cm³ of tumor specimen was cut into smaller than 1 mm³ pieces, then digested with collagenase/hyaluronidase buffer (STEMCELL Technologies) for 6 hours at 37 °C with agitation. For further digestion, gently pipette in trypsin (0.25%) and then in a solution of Dispase (5 units/mL) and DNase I (0.05 mg/mL; STEMCELL Technologies) for 5–10 minutes. We obtained single-cell suspensions by filtration through a 40-μm filter, which were then seeded into 6-well plates coated with collagen I. The culture medium composed of DMEM:F12 supplemented with 5% FBS, penicillin/streptomycin (1%), gentamycin (0.2%), EGF (10 ng/mL), adenine (20 μg/mL), cholera toxin (10 ng/mL), HEPES (15 mmol/L), insulin (5 μg/mL), hydrocortisone (0.32 μg/mL), and ROCK inhibitor (5 μmol/L). Following stable adhesion, cells could be cultured and passaged or frozen in liquid nitrogen in DMSO/FBS (9:1) with ROCK inhibitors of 5 μmol/L.

The pLKO.1 plasmid containing shRNA was cotransfected with the packaging plasmids psPAX2 (RRID: Addgene_12260) and pMD2.G (RRID: Addgene_12259) into HEK293T cells using Lipofectamine 3000 according to the manufacturer's protocol. The cells were washed 6 hours after transfection and then changed to fresh growth culture media and allowed to incubate for an additional 48 hours. The media containing viral particles was subsequently centrifuged for 5 minutes at 3,000 × g to remove cell debris and filtered by a 0.45-μm filter. We then concentrated the viral supernatant with a Centricon Plus-20 centrifugal filter at 4,000 × g. Lentivirus supernatants were aliquoted and stored at −80°C until use. For knockdown of interest genes in MDA-MB-231 cells, 10⁵ cells were seeded in a 6-well plate at 37°C with 5% CO₂, and incubated until 30%–40% confluence was reached. We added 20 multiplicities of infection of the viral supernatant to the culture medium. Puromycin was added to the medium 72 hours later at 1 μg/mL for stable knockdown selection.

Cell proliferation was measured using xCELLigence RTCA system (Acea Bioscience, distributed by Roche Diagnostics) that allows a noninvasive, long-term monitoring of live cells. The proliferation of cells was monitored for 40–70 hours at 37°C in the incubator after seeding 5,000–10,000 cells in each well of E-16-well plates (Roche). Microelectrodes were used on the bottom of the plate to monitor impedance changes in proportion to the number of adherent cells. Real-Time Cell Analyzer (RTCA) software automatically recorded the impedance values of each well. For each sample, two parallel wells were included in one replicate, and three independent duplicates were performed.

G₀–G₁, S, and G₂–M cell-cycle profiles were assessed using propidium iodide (PI) staining. Cells were trypsinized and washed with 1% BSA in PBS (1,500 rpm, 5 minutes) before being fixed in 70% ethanol. For total DNA staining, including those used for determination of sub-G₁ population and ploidy analyses, a 20-minute incubation at 37°C in a solution of PI/RNaseA (10 μg/mL and 0.1 mg/mL, respectively) in PBS was performed. Samples were analyzed using an Attune NxT flow cytometer (Life Technologies) and the data were processed by FlowJo software (RRID: SCR_008520).

A total of 2 × 10⁶ primary tumor cells in a volume of 100 μL (1:1 mixture of PBS and Matrigel) were injected into the flanks of 5 to 6 weeks old female NOD-SCID mice (RRID: IMSR_JAX:001303), followed by caliper measurements of tumor volume every week. The tumor was injected with 5 × 10⁷ copies of shRNA-targeting lentiviruses or scrambled shRNA lentiviruses when it reached 200–250 mm³. The tumor growth was monitored for 40–60 days after reaching 2,500 mm³ as an ethical limit. Each group had 6 mice and the animals were randomized. Tumor volume was calculated from digital caliper raw data by using the formula: V = (shortest diameter)² × longest diameter × 0.5.

Immunoprecipitation was performed with the indicated antibodies according to the protocol of ProFound Mammalian Co-Immunoprecipitation Kit (Pierce). Western blots were performed using RIPA buffer to extract total protein from cell lysates. We separated proteins using PAGE (SDS-PAGE) and then electrotransferred them to polyvinylidene fluoride membranes. The membranes were then blocked in Tris buffered saline with Tween 20 (TBST) containing 5% skimmed milk for 2 hours, and then incubated with primary antibodies overnight at 4°C using the indicated dilutions (1:500–1:1,000). Following three washes in TBST, the membranes were incubated for 1 hour with 1:1,000 dilution of HRP-conjugated secondary antibody. For biotin-labeled proteins, streptavidin-HRP conjugate was used. Finally, protein bands were visualized by enhanced chemiluminescence detection system (Amersham Biosciences).

Microscale thermophoresis (MST) was performed according to the previous work as described previously (36). In brief, purified recombinant protein was labeled with a RED-NHS protein labeling kit (NanoTemper) according to standard protocol. The protein was then incubated at a constant concentration (10–50 nmol/L) with 2-fold serial dilutions of indicated protein in MST optimized buffer (50 mmol/L Tris-HCl pH 7.4, 150 mmol/L NaCl, 10 mmol/L MgCl₂, 0.05% Tween-20). Equal volumes of binding reactions were mixed by pipetting and incubated for 15 minutes at room temperature. Mixtures were enclosed in standard-treated or premium-coated glass capillaries and loaded into the instrument (Monolith NT.115, NanoTemper). Measurement protocol times were as follows: fluorescence before 5 seconds, MST on 30 seconds, fluorescence after 5 seconds, delay 25 seconds. For all the measurements, 200–1,000 counts were obtained for the fluorescence intensity. The measurement was performed at 20% and 40% MST power. F_norm = F₁/F₀ (F_norm: normalized fluorescence; F₁: fluorescence after thermodiffusion; F₀: initial fluorescence or fluorescence after T-jump). Kd values were determined with the NanoTemper analysis tool.

Cultured cells were fixed in 2.5% glutaraldehyde and 4% PFA, treated with 1% OsO₄, washed and progressively dehydrated. The samples were then incubated in 1% uranyl acetate in 70% methanol, before final dehydration, preimpregnation with ethanol/epon (2/1, 1/1, 1/2) and impregnation with epon resin. After mounting in epon blocks for 48 hours at 60°C to ensure polymerization, ultrathin sections (70 nm) were cut on an ultramicrotome (Ultracut E, Leica) and analyzed using a Philips Technai 12 transmission electron microscope.

The in vitro motor motility assay followed a previously described protocol (37). Briefly, the polarity of microtubules was marked using different fluorescent labels to distinguish plus and minus ends. The GMPCPP-stabilized (Roche, NU-405L) microtubule seeds (5% Alexa-647 labeled and 10% biotin labeled) were immobilized on coverslips using NeutrAvidin protein. A total of 8 μmol/L 10% rhodamine-labeled tubulin was then added to the dynamic assay buffer (BRB80 supplemented with 1% β-mercaptoethanol, 80 mmol/L d-glucose, 80 μg/mL glucose oxidase, 32 μg/mL catalase, 160 μg/mL casein, 0.001% tween, 1 mmol/L MgCl₂, 2 mmol/L GTP). After incubating for 5 minutes at 35˚C, we washed out free tubulin dimers with BRB80 containing 20 μmol/L taxol, which stopped the polymerization and stabilized the “polarity-marking” microtubules. Next, GFP-UBA1 was added to the taxol-stabilized microtubule assay buffer (BRB80 supplemented with 1% β-mercaptoethanol, 80 mmol/L d-glucose, 80 μg/mL glucose oxidase, 32 μg/mL catalase, 160 μg/mL casein, 0.001% tween, 1 mmol/L MgCl₂, 20 μmol/L taxol, 1 mmol/L ATP) in the presence or absence of KIF20A. Images were collected every 0.1 seconds using a total internal reflection fluorescence microscope (Olympus IX83-ZDC, objective: 100× 1.49 N.A. Olympus) equipped with an Andor 897 Ultra EMCCD camera (Andor).

Histologic staining was done at Peking University's Third Hospital's Immunohistochemistry Core. For 12–16 hours, tissues were fixed in 10% neutral-buffered formalin solution before being transferred to 70% ethanol. The tissues were embedded in paraffin and cut into 5 μm sections on polylysine-coated slides and stained with hematoxylin and eosin, or indicated antibodies. Anti-Ki67 and anti-cleaved caspase-3 were diluted 1:200 and 1:500, respectively. Images were taken and analyzed using an Olympus BX51 microscope and DP73 CCD camera.

Unless otherwise specified, all experiments were conducted in three biologically independent replicates. Means and SDs were plotted. Student t test was used for statistical analyses. P < 0.05 (*) was considered statistically significant. Statistical details are included in figure legends.

All raw files and search results for MS of PCM proteomics have been deposited in ProteomeXchange (RRID: SCR_004055) via iProX (www.iprox.org) with the identification no. PXD020754 (for ProteomeXchange) and IPX0003115000 (for iProX).

To achieve centrosomal targeting, we inserted APEX2-EGFP into the endogenous loci of PCNT and SAS6 in human embryonic kidney 293T cells (Supplementary Fig. S1A and S1B). PCNT is reported to be retained in PCM throughout the cell cycle, while SAS6 is the core scaffold of the centriole that is tightly surrounded by PCM (3, 30). When monitoring the behaviors of PCNT-APEX2-EGFP and SAS6-APEX2-EGFP in 293T cells, we found that both fusion proteins were specifically retained at centrosomes during the entire cell cycle, and they were colocalized with endogenous pericentriolar material 1 (PCM1), a representative PCM marker (Fig. 1A; Supplementary Videos S1 and S2). Accordingly, we observed that a biotin-labeled reaction triggered by PCNT-APEX2-EGFP (reaction A) specifically occurred at centrosomes during both interphase and mitosis (Fig. 1B), and the reaction signal remained stable throughout the cell cycle (Fig. 1C). However, the reaction triggered by SAS6-APEX2-EGFP (reaction B) occurred only during interphase but not mitosis, although the fluorescent signal of SAS6-APEX2-EGFP was always detectable (Fig. 1B and C). By performing 3D reconstruction based on a series of high-resolution scans of the centrosome using the Imaris program, we verified that PCNT-APEX2-EGFP was colocalized with, but not covered by, endogenous PCM1, and that the APEX labeling reaction occurred normally. In contrast, SAS6-APEX2-EGFP was tightly enclosed by endogenous PCM1 and no reaction signal was detected (Fig. 1D and E). This inspired us to design a strategy for acquiring the mitotic PCM proteome in living cells by subtracting the proteome set obtained from SAS6-APEX2-EGFP labeling (set B) from that obtained from PCNT-APEX2-EGFP labeling (set A; Fig. 1F). This strategy avoids the use of any synchronization treatments that would impair the centrosome and the spindle, and provides a natural protein profile of mitotic PCM in cycling live cells.

To obtain the proteomes that are specifically labeled by PCNT-APEX2-EGFP or SAS6-APEX2-EGFP from live cells, 293T cells stably expressing PCNT-APEX2-EGFP and SAS6-APEX2-EGFP, respectively (Supplementary Fig. S2A), were incubated with BP for half an hour, and the labeling reactions were triggered by H₂O₂ for 1 minute. After quenching, cells were lysed immediately and biotinylated proteins were affinity enriched with streptavidin-coated beads. Enriched proteins were resolved by gel electrophoresis, digested into tryptic peptides, and quantified with LC/MS-MS. As shown in Fig. 2A, both PCNT-APEX2-EGFP (reaction A) and SAS6-APEX2-EGFP (reaction B) biotinylated a large number of proteins, while very few proteins were biotinylated in the negative control samples (wild-type 293T cells), indicating that proteomic capture in our system was successful. We generated proteomic data from three biological replicates for each of the reaction A and B. Protein abundance is displayed in violin plots and the correlation analysis shows good consistency between replicates (Pearson correlation coefficient, 0.88–0.94; Supplementary Fig. S2B and S2C). A total of 2,401 (for set A) and 2,379 (for set B) proteins were detected (Supplementary Table S4). The intensity-based absolute quantification (iBAQ) algorithm was used to quantify proteins, which divides the sum of all precursor-peptide intensities by the number of theoretically observable peptides. We normalized each protein's iBAQ value to the sum of all iBAQ values, generating a normalized iBAQ value for each protein (32, 38, 39). By comparing average normalized iBAQ values, 392 proteins from set A that were significantly more abundant than those in set B (i.e., set A minus set B) were identified as mitotic PCM components (Fig. 2B; Supplementary Table S5). To verify the reliability of MS data, the proteomic data were also searched against the Human Peptide Atlas database and neXtProt human proteins database, respectively. The protein expression levels analyzed by the three databases are strongly correlated with each other in all samples (Supplementary Fig. S2D–S2F). And most PCM proteins (99.2%) can be recognized in the other two databases (Supplementary Fig. S2G), and their peptides were validated by the uniqueness examination of the neXtProt peptide uniqueness checker (Supplementary Fig. S2H–S2I). These results indicate that the identified proteins in our study are reliable.

To assess the specificity of the PCM list, we combined three well-known databases, that is, the Human Protein Atlas (https://www.proteinatlas.org/, 548 centrosome proteins), GO database (http://geneontology.org/, 908 centrosome proteins), and MiCroKit (Midbody, Centrosome, Kinetochore, Telomere, and Spindle) database (http://microkit.biocuckoo.org/, 540 centrosome proteins), and the proteomics data from a study of isolated human centrosomes (40) to generate a comprehensive reference set containing a total of 1,478 annotated centrosome proteins. A total of 40.4% of our PCM proteins were found in this comprehensive reference, and the percentage is higher than that in existing reports of centrosome-related proteomics (20, 40, 41). Through GO analysis, we found that these PCM proteins were mainly annotated to centrosome-related processes such as “centrosome cycle,” “membrane docking,” “cilium assembly,” and “G₂–M transition of mitotic cell cycle” (Fig. 2C), further confirming the nature of the PCM proteome. In particular, this list contained representative PCM proteins such as PCM1, γ-tubulin ring complex, CEP164, CEP152, CEP192, cytoskeleton associated protein 5 (CKAP5), Ninein, PCNT, CDK5RAP2, and microtubule-associated protein 4. In addition, many previously unidentified PCM proteins such as CKAP4, heat shock cognate 71 kDa protein (HSC70), KIF20A, PDZ, and LIM domain 5 (PDLIM5), ubiquitin specific peptidase 15 (USP15), and VAMP associated protein B and C (VAPB) were identified (Fig. 2D). Of note, polo like kinase 1 (PLK1) was found in the PCM pool, while its family member PLK4, as well as SAS4, SAS6, and centrin3, which are core centriole components but not mitotic PCM components (3), were absent, validating the relative accuracy of the current PCM list.

In addition to the biological processes shown in Fig. 2C, several interesting cellular events such as “stress granule assembly,” “mRNA metabolism,” “posttranscriptional gene silencing,” and “cell shape” were found to be involved in PCM proteins (Fig. 2E; Supplementary Table S6). Intriguingly, deubiquitination-related proteins such as USP15, USP 9 X-linked (USP9X), ubiquitin C-terminal hydrolase L1 (UCHL1), and OTU domain-containing protein 4 (OTUD4) were found enriched in the PCM matrix (Fig. 2E). Given the disparate sizes of interphase PCM and mitosis PCM, it is possible that activating deubiquitination in the mitotic PCM may be a major mechanism for maintaining the large size of PCM. Because of the high density of PCM, we suspect that there must be abundant protein interactions among the PCM components that lay the foundation for the nature and functions of the centrosome. We thus built a PPI network using STRING and Cytoscape. Almost all the PCM proteins (373/392) formed a network consisting of 3,364 known or predicted interactions (Supplementary Fig. S2J), suggesting the tight association of proteins within the PCM matrix. Among the top 100 proteins that were ranked by the number of interactions formed, 17 belong to the CEP family proteins and five are HAUS augmin like complex subunit (HAUS) family proteins (Fig. 2F). These findings are consistent with the roles of the CEP and HAUS proteins, which are essential for centrosome integrity (42–45). It is also worth noting that among the proteins with the most interactions, cyclin dependent kinase 1 (CDK1), PLK1, and protein phosphatase 2 regulatory subunit A (PPP2R1A) ranked in the top three, indicating that phosphorylation and dephosphorylation frequently occur in the PCM microenvironment (Fig. 2F).

Increasing evidence has shown that membraneless organelles can assemble from their constituent molecules into biomolecular condensates through phase separation (46). However, whether the formation of mitosis PCM is mediated by phase separation is controversial (47). To explore the potential role of phase separation in PCM formation, we asked whether our identified PCM proteins are enriched for proteins with high droplet-forming propensities. We used a recently published database, FuzDrop, which quantifies the propensity for formation of liquid-like droplets across the human proteome. We found that PCM proteins displayed an overall higher propensity to form droplets compared with non-PCM proteins (Fig. 2G; Supplementary Fig. S2K). Intriguingly, the PCM proteins with high propensity scores are enriched for distinct cellular components, with the top two terms relating to non–membrane-bounded organelles (Fig. 2H). In line with the picture that intrinsic disordered regions (IDR) within proteins may contribute to phase separation (48), some of these enriched PCM proteins such as treacle ribosome biogenesis factor 1, ataxin 2 like, LIM domain containing preferred translocation partner in lipoma, splicing factor 1, RNA binding motif protein X-linked and nuclear autoantigenic sperm protein indeed contain IDRs (Fig. 2I; Supplementary Fig. S2L), as estimated by PONDR. These results are consistent with the hypothesis that proteins with phase separation capacity may contribute to the formation of the mitotic PCM.

To investigate the association between the PCM proteins and clinical tumors, we mapped all 392 candidates to TCGA and GTEx databases, which store genomic information from patients with different types of cancers, including breast cancer, ovarian cancer, colon tumors, cholangio carcinoma, and brain glioma, as well as the information from paired normal controls (9,018 tumor samples and 5,526 normal controls; Supplementary Table S2). Interestingly, we found that the global expression patterns of the 392 genes could be used to classify the data between tumors and normal controls. In particular, colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), uterine carcinosarcoma, breast invasive carcinoma (BRCA), glioblastoma multiforme (GBM), testicular germ cell tumors were significantly distinguished from the control group, implying the deep involvement of PCM proteins in the development of these cancers (Fig. 3A). Among the 392 PCM genes, 280 tended to be more highly expressed in tumor samples compared with their paired normal controls (Fig. 3B). We next analyzed the top 20 most highly expressed PCM genes, namely cell division cycle 6 (CDC6), CDK1, CEP55, CKAP2, desmoglein 2 (DSG2), desmoplakin (DSP), gamma-glutamylcyclotransferase (GGCT), G₂- and S-Phase expressed 1 (GTSE1), KIF11, KIF20A, NDC80 kinetochore complex component (NDC80), NIMA related kinase 2 (NEK2), NUF2 component of NDC80 kinetochore complex (NUF2), poly(A) binding protein cytoplasmic 1 (PABPC1), PLK1, regulator of chromosome condensation 2 (RCC2), ribophorin II (RPN2), solute carrier family 2 member 1 (SLC2A1), trophinin associated protein (TROAP), TTK protein kinase (TTK). Analysis of the expression levels of the 20 genes across diverse tumor types, such as cervical squamous cell carcinoma and endocervical adenocarcinoma, COAD, ESCA, lung squamous cell carcinoma, ovarian serous cystadenocarcinoma (OV), thymoma, and uterine corpus endometrial carcinoma, revealed that all the genes were highly expressed in most of the tumor types (Fig. 3C). Also, analysis of genomic alterations revealed that the top 20 candidates tended to be amplified but not deleted, which was consistent with the aforementioned high levels of expression in these tumors (Fig. 3D). We compared the overall survival of patients, determined from the database, with the expression of the 20 PCM genes. Interestingly, patients with cancer (n = 2,376) in which PCM genes were highly expressed had significantly shorter survival times than those (n = 2,376) in which these genes were lowly expressed (Fig. 3E).

Compared with RNA expression data in TCGA, clinical proteomic data could provide more accurate information on how and why cancer develops. We thus mapped all 392 PCM proteins to the clinical proteomic tumor analysis consortium (CPTAC) database, which includes proteomic information from patients with different types of cancers as well as the information from paired normal controls (49). Correlation analysis of RNA expression and protein levels of PCM genes shows that the protein levels of most PCM genes have positive correlation with that of RNA expression significantly, for example, 86.4% in BRCA, 93.6% in liver hepatocellular carcinoma (LIHC; Supplementary Fig. S3A–S3D). Consistent with RNA expression analysis, global expression patterns of the 392 PCM proteins could also classify the data between tumors and normal controls, implying the deep involvement of PCM proteins in these cancers’ development (Supplementary Fig. S3E–S3L). PCM proteins are significantly highly expressed in different cancers (50.9% in BRCA, 55.1% in LIHC), and most of the top 20 highly expressed PCM genes in RNA expression analysis are consistently highly expressed in CPTAC database (Supplementary Fig. S3M–S3T). Therefore, the analysis of clinical proteome data from CPTAC further supports our results.

Because the PCM proteins are proposed to cluster extra centrosomes and increase tumor cell survival, we hypothesized that tumors that are resistant to centrosome declustering drugs would have more of a requirement for these PCM genes. Thus, we delved into public databases for any data on cancer cells treated with centrosome declustering drugs. Using CellMiner (https://discover.nci.nih.gov/cellminer/home.do), which is designed for the integration and study of molecular and pharmacologic data for NCI-60 cancerous cell lines screened with over 100,000 chemical compounds and natural products, we analyzed gene expression patterns in cells treated with the centrosome declustering drug noscapine (50). Across the 56 available types of cancerous cells, we compared the three most noscapine-sensitive cells (ME.MDA.N, LE.RPMI.8226, and LE.SR) with the three most noscapine-resistant cells (BR.T.47D, LC.NCI.H226, and OV.OVCAR.4). Intriguingly, the global expression level of PCM genes in noscapine-resistant cells was significantly higher than that in noscapine-sensitive cells (Fig. 3F). When looking into the detailed expression patterns of PCM genes across these tumor cells, we found that the noscapine-resistant cells commonly highly expressed a series of genes including USP15, CAND1, SPTAN1, ANKRD17, CTNNB1, SCLT1, and METAP1 (Fig. 3G and H). Moreover, the three types of noscapine-resistant tumor cells had specific sets of highly expressed genes. The breast cancer cell line BR.T.47D tended to have higher expression of HAUS3, CCDC15, SLC25A18, and MAP1A, while the lung cancer and ovarian cancer cell lines LC.NCI.H226 and OV.OVCAR.4 had elevated expression levels of CKAP2, IQGAP1, CKAP4, and SERINC1, and CCDC18, NUP153, HSPBP1, and LMTK2, respectively (Fig. 3I). These findings suggested that PCM genes contribute to the resistance to centrosome declustering drugs, and that targeting PCM-related proteins/genes might decrease the resistance to this type of drugs.

Next, we focused on breast cancer and mined potential PCM candidates that are required for the survival of breast cancer cells. We marked the 392 PCM genes from the data of breast cancer patients and the related controls from TCGA and GTEx. Of note, 52 genes were significantly more highly expressed in breast tumor tissues (Fig. 4A). We selected KIF20A and CKAP4, two newly identified PCM candidates among the 52 highly expressed genes in tumors (Fig. 4A and B; Supplementary Fig. S3U), for deeper research. In particular, KIF20A expression is significantly positively correlated with the chromosome instability(Supplementary Fig. S3V). Tumor tissues highly expressing KIF20A exhibit higher stemness score, indicating that these tumors tend to be poorly differentiated (Supplementary Fig. S3W). Consistently, we found that higher KIF20A expression was associated with poor prognosis of patients with breast cancer in disease specific survival analysis (Supplementary Fig. S3X). In addition to investigating the subcellular localization of endogenous KIF20A and CKAP4 (Fig. 2D), we costained the two proteins with centriole marker SAS6 and PCM marker PCM1 in breast cancer MDA231 cells, and reconstructed the centrosome structure by 3D modeling. As expected, both endogenous KIF20A and CKAP4 were colocalized with PCM1, occupying the outer sphere of the whole centrosome, while the centriole protein SAS6 was retained within the center of the centrosome and was surrounded by KIF20A and CKAP4 (Fig. 4C).

To explore the roles of KIF20A and CKAP4 in centrosome clustering, we established breast cancer cells bearing extra poles by stably expressing PLK4, a multiple centriole inducer (51), in MDA231 cells (MDA231-PLK4; Supplementary Fig. S4A) and monitored them using live cell imaging. When compared with normal MDA231 cells, 22.3% of which showed transient extra poles during mitosis, MDA231-PLK4 cells showed a higher rate (58.5% of cells) of transient extra pole formation, which resulted from PLK4-induced multiple centriole formation (Fig. 4D and E). Of note, a substantial portion of the extra pole-bearing cells among both MDA231 cells and MDA231-PLK4 cells did not die from mitotic errors, but overcame the final formation of extra poles and survived, producing a new generation (Fig. 4D–F; Supplementary Video S3 and S4): in particular, we observed clustering of poles in 88.6% of extra pole-bearing cells in the MDA231 group and 59.2% of those in the MDA231-PKL4 group, and these cells thus survived. When KIF20A was knocked down in MDA231 and MDA231-PLK4 cells (Supplementary Fig. S4B), the percentages of cells with transient extra poles reached 42.8% and 88.4%, respectively (Fig. 4D and E). Importantly, upon the knockdown of KIF20A, nearly all extra pole-bearing cells from both the MDA231 and MDA231-PKL4 groups were not able to cluster multiple poles, leading to mitotic catastrophe (Fig. 4D–F; Supplementary Video S5). A similar phenotype was observed in CKAP4 knockdown cells (Fig. 4D–F; Supplementary Fig. S4B; Supplemenatry Video S6).

Using the xCELLigence RTCA system for real-time monitoring of long-term cell proliferation, we found that the knockdown of KIF20A or CKAP4 suppressed proliferation of MDA231 cells, and the suppression was reinforced in MDA231-PKL4 cells (Fig. 4G). It is known that mitotic catastrophe induces apoptosis (52, 53). We thus measured the occurrence of apoptosis of these cells. In a flow cytometry assay, MDA231 cells showed normal percentages of cells in each stage of the cell cycle, with few dead cells marked by sub-G₁. In contrast, KIF20A or CKAP4 knockdown caused increased percentages of both mitotic-stage cells and sub-G₁ cells. As expected, the percentages of mitotic wandering cells and sub-G₁ cells were dramatically higher in MDA231-PKL4 cells upon KIF20A or CKAP4 knockdown (Fig. 4H; Supplementary Fig. S4C). Consistent with this, fluorescence staining for active apoptosis and immunoblotting against the activated (cleaved) form of the apoptotic executioner protein caspase-3 revealed that knockdown of KIF20A or CKAP4 in both MDA231 and MDA231-PKL4 cells obviously promoted apoptosis (Fig. 4I and J). These data indicate that PCM proteins are required for the survival of cancer cells bearing extra spindle poles, and that disruption of PCM genes could be an effective strategy for killing tumors.

Because KIF20A presents a more significant trend of high expression than CKAP4 across wide types of cancers, we selected KIF20A for mechanistic investigation of centrosome cluster. For this aim, we synchronized MDA231 cells to mitosis and immunoprecipitated proteins that potentially associated with KIF20A by anti-KIF20A antibody (Fig. 5A). MS analysis identified 227 proteins from the immunoprecipitated solution. By mapping with the 392 PCM protein set, 31 proteins were filtered out that may interact with KIF20A in PCM during mitosis (Fig. 5B; Supplementary Table S7). Intriguingly, seven ubiquitination promotion proteins [ubiquitin A-52 residue ribosomal protein fusion product 1 (UBA52), ubiquitin like modifier activating enzyme 1 (UBA1), proteasome 20S subunit beta 2 (PSMB2), proteasome 26S subunit ATPase 1 (PSMC1), proteasome 26S subunit ATPase 2 (PSMC2), proteasome 26S subunit ATPase 5 (PSMC5), and proteasome 26S subunit non-ATPase 9 (PSMD9)] were found among the 31 candidates, indicating that KIF20A would regulate ubiquitination event in PCM (Fig. 5B). As KIF20A belongs to Kinesin superfamily motor proteins that functions for intracellular transport (54), we hypothesized that KIF20A drives ubiquitination promotion proteins away from PCM to decrease ubiquitination for maintaining the large size of PCM during mitosis. Thus, we expressed and purified (or purchased) recombinant proteins of GST-KIF20A, GST-UBA52, GST-UBA1, GST-PSMB2, GST-PSMC1, GST-PSMC2, GST-PSMC5, and GST-PSMD9 (Fig. 5C), and examined binding affinity of KIF20A between these ubiquitination promotion proteins by MST, a sensitive assay for biomolecules affinity. Analysis of binding curves showed that KIF20A directly bound UBA1 and PSMC5 with high affinity [dissociation constant (Kd), 0.82 μmol/L and 1.97 μmol/L, respectively], and KIF20A mildly bound PSMB2 and PSMD9 (Kd, 5.93 μmol/L and 6.92 μmol/L, respectively; Fig. 5D; Supplementary Fig. S5A–S5C), implying that the motor protein KIF20A may transport the ubiquitination promotion proteins away from PCM.

Next, we detected protein level of endogenous UBA1 and PSMC5 at the centrosome in mitotic MDA231 cells by immunofluorescent staining. Of note, compared with weak signal of UBA1 and PSMC5 in control cells, accumulated UBA1 and PSMC5 were obviously observed at the centrosome in KIF20A knockdown cells bearing clustering-failed extra spindle poles (Fig. 5E and F). As expected, these cells showed higher intensity of ubiquitination marked by ubiquitin in the presence of proteasome inhibitor MG132 (Fig. 5G). Consistently with the staining results in MDA231 shKIF20A cells, when KIF20A was depleted in the 293T cells stably expressing PCNT-APEX2-EGFP, the global ubiquitination level was elevated in the PCM proteome captured by streptavidin-coated beads (Fig. 5H). Accordingly, the size of the mitotic PCM was decreased upon KIF20A knockdown in MDA231 cells revealed by the fluorescence intensity of PCNT and electron microscope (Fig. 5I; Supplementary Fig. S5D). To validate the transport manner of the ubiquitination promotion protein by KIF20A, we tracked GFP-UBA1 behavior in motor motility assay in the microtubule assay buffer containing active “polarity-marking” microtubules. In the absence of KIF20A, GFP-UBA1 was distributed in the buffer and could not associated with microtubules. While when KIF20A was added, GFP-UBA1 was aggregated along microtubules and transported from minus ends (corresponding to centrosomes in cells) to plus ends (corresponding to chromosomes in cells; Fig. 5J–M; Supplementary Videos S7, S8, and S9). Collectively, these data suggest that PCM protein KIF20A maintains PCM integrity for centrosome clustering during mitosis by driving ubiquitination promotion proteins away from PCM.

To investigate whether disruption of PCM gene KIF20A could kill tumors in vivo, we harvested fresh tumor masses from 10 patients with breast cancer. Detailed information about the clinical samples is provided in Supplementary Table S3. Primary tumor cells isolated from each mass were lively labeled against microtubules with cell-permeable taxol-based fluorescent probes for microscopic analysis of mitosis. Three nontransformed cell lines, 293T, human fetal lung fibroblast (MRC5) cells, and retinal pigment epithelial (RPE1) cells, were used for comparison. As expected, all nontransformed 293T, MRC5, and RPE1 cells showed sporadic extra pole formation during mitosis, with 2.6%, 3.1%, and 0.8% of cells, respectively. Notably, the 10 primary tumor cell lines from patients with breast cancer (referred to as Tumor I to Tumor X) had significantly higher rates of extra pole formation (28.8%, 24.5%, 22.8%, 20.9%, 20.2%, 17.9%, 17.4%, 16.4%, 14.7%, and 14.4% of cells, respectively) than nontransformed cells (Fig. 6A). We then selected Tumor I (28.8% of cells with extra poles), Tumor V (20.2% with extra poles), and Tumor X (14.4% with extra poles) for a patient-derived xenograft (PDX) mouse model study. We monitored these PDX mice for 40–55 days when all tumor cells developed into obvious solid tumors. Tumor X grew a little faster than Tumor I and Tumor V, but there was no significant difference in growth rate (Fig. 6B, blue lines). Histochemical analysis revealed that these tumors were actively proliferating labeled by Ki67, and there was less apoptosis marked by cleaved caspase-3 (Fig. 6C and D). Intriguingly, when KIF20A was knocked down in PDX tumors using a KIF20A-shRNA lentivirus at day 20 after implantation, we observed dramatic shrinking of Tumor I and Tumor V, and mild shrinking of Tumor X. The mean reductions in volume of Tumor I, V, and X were 75.9%, 76.8%, and 51.6%, respectively (Fig. 6B, red lines). IHC analysis revealed that the shrunken tumor masses contained fewer Ki67-positive cells and showed strong signals of apoptosis, which was very different from the observations in control tumor masses (Fig. 6C and D). To confirm the declustering effect of KIF20A knockdown in vivo, we harvested tumor masses and recorded the number of mitotic cells in frozen sections by performing immunofluorescence staining against microtubules and PCM1. Most of the mitotic cells in Tumor I, Tumor V, and Tumor X showed two spindle poles. In contrast, when these tumors were treated with KIF20A shRNA, a large number of extra pole-bearing cells (64.7% of cells in Tumor I, 60.5% in Tumor V, and 48.3% in Tumor X) were found among the mitotic cells. All the extra pole-bearing cells contained multiple foci of PCM1 with disordered or unseparated chromosomes (Fig. 6E and F). These data suggest that targeting PCM proteins such as KIF20A effectively kills tumors in vivo.

The centrosome was first found more than a hundred years ago for the formation of two spindle poles during cellular mitosis (3). The centrosome contains a pair of orthogonal centrioles localized at the core, which is surrounded by a proteinaceous matrix of PCM (3). It was recently realized that the PCM matrix plays a broad variety of roles in mitosis progression, DNA damage response, as well as protein degradation (55, 56). More importantly, given the centrosome amplification in most solid tumors (1–3) and the unconventional survival of tumor cells with extra centrosomes (2, 9–12), it is tempting to imagine that PCM, a highly dense and gel-like condensate that rapidly grows during mitosis (3, 6, 16, 17), would cluster extra centrosomes and defend against mitotic errors, enabling tumor cell survival. Thus, deciphering PCM components, especially PCM proteins that are technically inseparable from centrioles, will not only facilitate a deeper understanding of centrosome biology, but also provide a valuable resource for targeting tumors. However, because of the amorphous appearance of PCM, which lacks an encompassing membrane, its highly dynamic behavior, and physical connection to the centriole, no current strategy is capable of decoding the profile of this microscale matrix. In this study, taking advantage of differential labeling between PCNT-APEX2-EGFP (set A) and SAS6-APEX2-EGFP (set B), we were able to obtain a PCM proteome by subtracting set B proteins from set A proteins identified in living and undisturbed cells without any synchronization treatment. When mapping our PCM proteins to three well known databases, that is, the Human Protein Atlas (548 centrosome proteins), GO database (908 centrosome proteins), and MiCroKit database (540 centrosome proteins), and data from a proteomics study of isolated human centrosomes (40), we found that 40.4% of our PCM proteins were present in at least one of these databases. This percentage is higher than that in existing reports on centrosome-related proteomics (20, 40, 41), and also higher than the overlap observed when any of three databases above were compared with the others. In addition, given the peripheral localization of PCM proteins and the absence of a membrane surrounding the centrosome, conventional techniques may be not able to recover all centrosome components. We believe that the 392 proteins identified in the current work might provide a relatively full picture of PCM, especially in living cells.

By analyzing PPI networks using STRING and Cytoscape, we found that 373 of the 392 proteins participated in more than 3,000 interactions within the microscale matrix, making the amorphous PCM a unique region characterized by high-density protein binding and dynamics. Many of the proteins with a large number of interactions are involved in phosphorylation and deubiquitinination, suggesting that these processes are important in the mitotic PCM. It is well known that phosphorylation promotes the recruitment of proteins; these proteins create new expansive outer layers of the PCM, which substantially increases centrosome diameter (16, 57, 58). However, whether deubiquitination is an incidental event or a predominant event like phosphorylation is unclear. Recently, scientists discovered that USP33 deubiquitinates centrosome protein CCP110, and antagonizes SCF^{cyclin F}-mediated ubiquitination and degradation of CCP110, making it the first centrosome deubiquitinating enzyme identified whose expression regulates centrosome homeostasis (59). Considering that various deubiquitination-related proteins such as USP15, USP9X, UCHL1, and OTUD4 are significantly enriched in the PCM network generated using our proteomic data, we propose that deubiquitination is a predominant biological mechanism for maintaining PCM size during mitosis. Other lines of evidence show that spindle tension is required for the cluster of multiple centrosomes into a bipolar spindle array in tumor cells bearing extra centrosomes (12, 60, 61). Depletion of proteins involved in microtubule-kinetochore attachment, such as the chromosomal passenger complex proteins [aurora-B, inner centromere protein (INCENP), survivin, and borealin] and Ndc80 microtubule-kinetochore attachment complex proteins (NDC80, SPC24, and SPC25), induces spindle multipolarity (12). Also, centromere dysfunction can compromise mitotic spindle pole integrity (61). Thus, the combination analysis of our PCM proteins and centromere proteins (62, 63) would much valuable for understanding the fundamental events in mitosis.

In particular, we found that ubiquitination promotion proteins such as UBA1 and PSMC5 were driven away from the centrosome by PCM protein KIF20A. KIF20A belongs to kinesin superfamily proteins that mediate the transport of various cargos, including the newly synthesized protein complexes, vesicles and mRNAs along the microtubule filaments to their destinations (54). Previous work revealed that KIF20A could specifically bind to the GTP-bound form of RAB6, which controlled the motility and localization of the Golgi apparatus in interphase cells (64). Recently, KIF20A was observed involved the regulation of the chromosome passenger complex at chromosomes in mitotic cells (65). Our new recognition of KIF20A at the centrosome where microtubule minus ends are enriched defines the starting point of this kinesin in mitotic cells, highlighting the export of ubiquitination promotion proteins from the centrosome. Coordinating with the enrichment of deubiquitination proteins, this is an efficient way for maintaining PCM growth for extra pole cluster. Another attractive finding comes from database analysis of the centrosome-declustering drug noscapine. A striking difference in the expression of the 392 PCM genes was observed between the most noscapine sensitive and nonsensitive cancer cells, implying the involvement of the PCM components in centrosome-declustering drug resistance. To date, centrosome declustering drugs have not been applied widely as clinical treatments, although in principle they should kill tumors effectively. On the basis of our findings, it is easy to imagine that tumor cells could counteract centrosome-declustering drugs by overexpressing certain PCM factors, such as CEP164, KIF20A, METAP1, and ALS2. Using centrosome-declustering drugs not in isolation, but in combination with the disruption of a PCM factor(s), may substantially improve anti-cancer effect in the future.

P. Zou reports grants from National Natural Science Foundation of China and Ministry of Science and Technology of People's Republic of China during the conduct of the study. No disclosures were reported by the other authors.

B. Xie: Data curation, funding acquisition, investigation, visualization, writing–original draft. Y. Pu: Investigation. F. Yang: Investigation. W. Chen: Investigation, visualization. W. Yue: Investigation. J. Ma: Software, investigation, visualization. N. Zhang: Investigation. Y. Jiang: Investigation. J. Wu: Investigation, visualization. Y. Lin: Software, writing–original draft. X. Liang: Software, writing–original draft. C. Wang: Software, writing–original draft. P. Zou: Conceptualization, writing–original draft. M. Li: Conceptualization, supervision, funding acquisition, writing–original draft, project administration.

This work was supported by the National Natural Science Foundation of China (NSFC; 81871160 and 81521002 to M. Li; 32088101 and 21727806 to P. Zou; 32100554 to B Xie), by “Clinic + X” program (to M. Li) of Peking University, and by the Ministry of Science and Technology (2018YFA0507600 and 2017YFA0503600 to P. Zou). P. Zou is sponsored by Bayer Investigator Award. MS experiments were carried out by ZhengDa Health Company (Beijing, P.R. China). The authors thank Dr. Wenchuan Leng for the assistance with analysis of MS data.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.