Field

This invention relates to the diagnosis, prognosis and treatment of cancer associated with centrosome amplification (CA).

Background

The centrosome, an organelle composed of two centrioles surrounded by a pericentriolar protein matrix, is the major microtubule-organising centre of animal cells, hence being pivotal for several fundamental cellular processes, including signalling, cell polarity, division and migration (1-4). Each centrosome duplicates once per cell cycle to ensure bipolar spindle assembly and successful chromosome segregation (5,6). Centrosomes are thus implicated in the maintenance of genome stability.

Centrosome amplification (CA) - the presence of more than two centrosomes - is a common feature in cancer (7). Supernumerary centrosomes generate multipolar mitosis and consequent genome instability (6,8-10), they can accelerate and promote tumourigenesis in vivo (11-13) and promote cellular invasion and metastatic behaviour (14-17). However, CA’s pan-cancer prevalence, molecular role in tumourigenesis and therapeutic value remain poorly understood, largely due to the technical challenges associated with profiling such small cellular structures in human cancer tissues. For instance, quantifying centrosome numbers and abnormalities is often hampered by the limited thickness of formalin-fixed and paraffin-embedded tissue sections, preventing the imaging of entire cells (18). In addition, three-dimensional imaging and analysis are mandatory, but cumbersome and time consuming (19).

Summary

Through the analysis of drug sensitivity and centrosome amplification (CA) associated gene expression in human cancer cell lines, the present inventors have unexpectedly found that centrosome amplification (CA) renders cancer cells more sensitive to certain anti-cancer compounds. CA-associated gene panels have been shown to be useful in identifying candidate compounds that selectively target cancer cells exhibiting transcriptomic evidence for CA.

A first aspect of the invention provides a method of treatment of a cancer that displays centrosome amplification (CA) comprising administering a CA selective agent to an individual in need thereof.

Related aspects provide a CA selective agent for use in a method of the first aspect and the use of a CA selective agent in the manufacture of a medicament for use in a method of the first aspect.

A second aspect of the invention provides a method of screening for a CA selective agent comprising determining the effect of a test compound on the expression of a panel of CA associated genes in the cell. A decrease in expression of the panel in the presence relative to the absence of the test compound is indicative that the test compound is CA selective agent.

A CA selective agent identified by a method of the second aspect may display anti-cancer activity and may be useful for example in the development of therapeutics for the treatment of cancer, for example cancer that displays centrosome amplification (CA).

A third aspect of the invention provides a method of determining the sensitivity of a cancer cell to a CA selective agent comprising

determining the expression of a panel of CA associated genes in the cancer cell, wherein expression of the panel above a threshold value is indicative the cancer cell is sensitive to the CA selective agent.

A CA selective agent for use in any one of the first to third aspects may be selected from the group consisting of 3-CI-AHPC, CD-437, STF-31 , methotrexate, BI-2536, clofarabine, CH-79797,

NSC632839, etoposide, PHA-793887, manumycin A, CCB02, doxorubicin and compounds #1 to #351 of Table 2.

A fourth aspect of the invention provides a method of assessing genomic instability in a cancer cell comprising;

determining the expression of a panel of CA associated genes in the cell,

wherein expression of the panel above a threshold value is indicative the cancer cell has genomic instability.

Genomic instability may include one or more of tumour aneuploidy, mutation burden, number of somatic Copy Number Alterations (CNA), genomic instability, alteration of specific chromosomal arms and C>T mutations.

Genomic instability may for example be indicative of malignancy and responsiveness to cancer therapy, such as therapy targeting DNA damage.

A fifth aspect of the invention provides a method of assessing a cancer condition in an individual comprising determining the expression of a panel of CA associated genes in one or more cancer cells obtained from the individual.

Expression of the panel in the one or more cancer cells may be indicative of the status, subtype, molecular features, clinical features, diagnosis and/or prognosis of the cancer condition in the individual. In some embodiments, expression of the panel may be determined in one or more cancer cells obtained at first and second time points and the change in expression between the time points determined. Monitoring expression of the panel may be indicative of the aggressiveness and metastatic potential of the cancer condition and the effectiveness of treatments administered to the individual.

In some embodiments of the above aspects, the cancer condition may be a cancer other than breast cancer.

Other aspects and embodiments of the invention are described in more detail below.

Brief Description of Figures

Figure 1 shows the pan-cancer landscape of centrosome amplification-associated gene expression. (1a) For each sample, the CA20 score was calculated as the sum of the normalized (log2 median- centred) expression levels of the 20 signature genes. (1 b) CA20 score distribution across tumour samples of all TCGA cancer types. Cohorts are ordered by their median CA20 score. Black points and lines represent the median +/- upper/lower quartiles. (1c) Tumour samples have higher CA20 levels in all 15 cancers with both tumour and matched-normal samples available (at least 10 samples per sample type; False Discovery Rate (FDR) < 0.0001 , Wilcoxon rank-sum test). CA20 score distributions of tumour and normal samples are represented in red and blue, respectively.

ACC: adrenocortical carcinoma; BLCA: bladder urothelial carcinoma; BRCA: breast invasive carcinoma; CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; CHOL: cholangiocarcinoma; COADREAD: colon and rectum adenocarcinoma; DLBC: lymphoid neoplasm diffuse large B-cell lymphoma; ESCA: oesophageal carcinoma; GBM: glioblastoma multiforme; HNSC: head and neck squamous cell carcinoma; KICH: kidney chromophobe; KIRC: kidney renal clear cell carcinoma; KIRP: kidney renal papillary cell carcinoma; LAML: acute myeloid leukemia; LGG: low-grade glioma; LIHC: liver hepatocellular carcinoma; LUAD: lung adenocarcinoma; LUSC: lung squamous cell carcinoma; MESO: mesothelioma; OV: ovarian serous cystadenocarcinoma; PAAD: pancreatic adenocarcinoma; PCPG: pheochromocytoma and paraganglioma; PRAD: prostate adenocarcinoma; SARC: sarcoma; SKCM: skin cutaneous melanoma; STAD: stomach

adenocarcinoma; TGCT: testicular germ cell tumours; THCA: thyroid carcinoma; THYM: thymoma; UCEC: uterine corpus endometrial carcinoma; UCS: uterine carcinosarcoma; UVM: uveal melanoma.

Figure 2 shows that CA20 is associated with different breast cancer clinical and molecular features. (2a-h) CA20 score distribution per (a,e) sample type, (b,f) histological and (c,g) PAM50 molecular subtype, and (d,h) tumour stage for (a-d) TCGA breast cancer and (e-h) METABRIC samples. Black points and lines represent the median +/- upper/lower quartiles. Number of samples used in each violin is shown within brackets. *** p-value < 0.001 , **** p-value < 0.0001 and n.s. non-significant (Wilcoxon rank-sum test). (2i-j) Luminal B and basal-like human breast carcinomas display higher levels of centrosome amplification (CA). (i) Illustration of the procedure to quantify CA in patient samples (j) Percentage of cells displaying CA in breast tumours from the different PAM50 molecular subtypes (29 luminal A, 3 luminal B, 3 HER2 and 13 basal-like). Between 5 and 107 cells were analysed for each patient. * p-value < 0.05, *** p-value < 0.001 and n.s. non-significant (Wilcoxon rank-sum test).

Figure 3 shows that CA20 is associated with genomic instability features in cancer. (3a) CA20 score is associated with genome doubling. Box plots of CA20 score per whole genome doubling status. **** p-value < 0.0001 (Wilcoxon rank-sum test). (3b, 3d-f) CA20 is associated with different genomic instability features. Smooth scatter plots showing correlation between CA20 score and (3b) aneuploidy score (measured as the total number of altered chromosome arms), (3d) number of mutations per Mb, (3e) number of CNAs and (3f) clones per tumour across TCGA tumour samples (Spearman’s correlation coefficient, r = 0.44, 0.48, 0.47 and 0.43, respectively, p-value < 2.2e-16 for all). Chromosome arm alterations associated with CA20 score. Volcano plot shows the results of linear regression analyses comparing CA20 score between samples with deletion or amplification of each chromosome arm. Arms whose deletions or amplifications are associated with higher CA20 (FDR < 0.05) are represented in blue and red, respectively. Chromosome arms with FDR < 1e-5 are highlighted and box plots of CA20 score per chromosome arm alteration are shown for 5q, 16p and 7p arms. **** p-value < 0.0001 (linear regression). (3g) Hierarchical clustering of TCGA cancer types based on the independent association between the different genomic instability features and CA20 score. Unsupervised hierarchical clustering using Euclidean distances calculated based on multiple linear regression p-values of association with CA20 of aneuploidy score, number of mutations per Mb, number of CNAs and clones per tumour, per TCGA cohort and with all cohorts together (PanCancer). Heatmap colour scale according with -log 10 of linear regression p-values. Main clusters are highlighted with different shades of grey.BLCA: bladder urothelial carcinoma; CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; GBM: glioblastoma multiforme; HNSC: head and neck squamous cell carcinoma; KIRC: kidney renal clear cell carcinoma; LGG: low-grade glioma; LUAD: lung adenocarcinoma; LUSC: lung squamous cell carcinoma; PRAD: prostate adenocarcinoma; SKCM: skin cutaneous melanoma; STAD: stomach adenocarcinoma; THCA: thyroid carcinoma.

Figure 4 shows that CA20 is associated with cancer’s mutational spectrum. (4a) Somatic mutations pan-cancer-wide associated with the CA20 score. The volcano plot shows the results of linear regression analyses comparing the CA20 score between mutated and wild-type samples for 14,589 genes (at least 20 mutated samples). Genes whose mutations are associated with higher and lower CA20 (FDR < 0.05) are represented in red and blue, respectively. The top 10 genes are highlighted. (4b) TP53 mutations are associated with CA20 in different cancer types. Linear regression coefficients, representing CA20 score differences between TP53 mutated and wild-type tumour samples, across TCGA cohorts with at least 20 mutated samples. Significant associations (FDR < 0.05) are coloured as in (a). (4c) Mutational signatures pan-cancer-wide associated with CA20, independently of other types of genomic instability. Left: Significance of linear regression analyses (- Iog10 p-value) between CA20 and contribution of each mutational signatures including, as independent variables, the four genomic instability features: aneuploidy, mutation burden, CNA and number of clones per tumour. Positive and negative significant associations (FDR < 0.05) are coloured in red and blue, respectively. Right: Smooth scatter plots showing correlations between CA20 and the contribution of mutational signature 1 (linked with ageing) in 3 TCGA cohorts. Linear regression p-values are shown. (4d) Causal effect of CA20-associated mutations on CA20 levels. Scatter plot of linear regression’s coefficient from (a) versus Connectivity Map (CMap)’s knock-down score, ranging from 100 (CA20 up-regulation) to -100 (CA20 down-regulation), for each gene in common. Genes are coloured as in (a) and the ones with both significant linear regression associations and absolute knock-down score higher than 80 are highlighted. (4e and 4f) Gene Set Enrichment Analysis (GSEA) of genes ranked by their CMap’s knock-down score using (e) a manually curated list of centriole duplication factors and (f) MSigDB’s Hallmark Gene Sets (unfolded protein response and mitotic spindle were significantly associated, FDR < 0.05). GSEA p-values are shown. BLCA: bladder urothelial carcinoma; BRCA: breast invasive carcinoma; CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; ESCA: oesophageal carcinoma; GBM: glioblastoma multiforme; HNSC: head and neck squamous cell carcinoma; KICH: kidney chromophobe; LGG: low- grade glioma; LIHC: liver hepatocellular carcinoma; LUAD: lung adenocarcinoma; LUSC: lung squamous cell carcinoma; OV: ovarian serous cystadenocarcinoma; PAAD: pancreatic

adenocarcinoma; PRAD: prostate adenocarcinoma; SARC: sarcoma; SKCM: skin cutaneous melanoma; STAD: stomach adenocarcinoma; UCS: uterine carcinosarcoma.

Figure 5 shows that high CA20 is associated with poor patient prognosis, hypoxia and lower stromal infiltration in cancer. (5a) Kaplan-Meier plots for patient stratification based on CA20 score (patients divided by CA20 median: lower CA20 in blue and higher CA20 in red) in 8 different cancer types. Numbers at risk every 2.5 years (tables) and 5-year survival rates (points and dashed lines) are shown. P-values for log-rank tests for differences in survival are shown. (5b) CA20 upregulation is associated with hypoxia. Smooth scatter plot showing correlation between the hypoxia and the CA20 scores across TCGA tumour samples (Spearman’s correlation coefficient, r = 0.61 , p-value < 2.2e- 16). (5c) CA20 upregulation is associated with hypoxia in different cancer types. Linear regression coefficients, representing the CA20 score dependence on hypoxia score, independently of genomic instability, across the TCGA cohorts with information for all covariates. Significant associations (FDR < 0.05) are coloured. (5d) CA20 is associated with lower stromal cell infiltration. Smooth scatter plot showing correlation between the CA20 and the stromal scores across TCGA tumour samples (Spearman’s correlation coefficient, r = -0.52, p-value < 2.2e-16). (5e) CA20 is associated with lower stromal cell infiltration in head and neck squamous cell carcinoma and lung adenocarcinoma. Linear regression coefficients, representing the CA20 score dependence on stromal score, independently of genomic instability, across the TCGA cohorts with information for all covariates. Significant associations (FDR < 0.05) are coloured. ACC: adrenocortical carcinoma; BLCA: bladder urothelial carcinoma; BRCA: breast invasive carcinoma; CESC: cervical squamous cell carcinoma and endocervical adenocarcinoma; GBM: glioblastoma multiforme; HNSC: head and neck squamous cell carcinoma; KICH: kidney chromophobe; KIRC: kidney renal clear cell carcinoma; LGG: low-grade glioma; LUAD: lung adenocarcinoma; LUSC: lung squamous cell carcinoma; MESO: mesothelioma; PAAD: pancreatic adenocarcinoma; PRAD: prostate adenocarcinoma; SKCM: skin cutaneous melanoma; STAD: stomach adenocarcinoma; THCA: thyroid carcinoma; UVM: uveal melanoma.

Figure 6 shows the identification of compounds that selectively kill cancer cells with high CA20.

(6a) Compounds with selective activity on cell lines with high or low CA20 score. The volcano plot shows the results of Spearman’s correlation analyses between CA20 scores and compound Area Under the dose-response Curve (AUC) across Cancer Therapeutics Response Portal (CTRP) human cancer cell lines. Note that lower AUC means higher drug activity. The compounds whose activity was associated with high and low CA20 (FDR < 0.05) are represented in blue and red, respectively. The top 6 compounds are highlighted. (6b) Top 6 compounds targeting cells with higher CA20 score. Smooth scatter plots showing correlation between CA20 score and compound AUC across CTRP cell lines for the top 6 compounds from (a). Spearman’s correlation coefficients, r, and respective p-values are shown. (6c) Compounds that down-regulate the CA20 gene set. Heatmap of CMap’s drug score, ranging from 100 (maximum CA20 up-regulation) to -100 (maximum CA20 down-regulation) per cell line. Drug average score (last column) is the mean of drug scores across cell lines. The 20 compounds with the lowest drug average score are shown and ranked accordingly. Tissue of origin of human cancer cell lines: PC3: prostate; VCAP: prostate; A375: melanoma; A549: lung; HA1 E: kidney; HCC515: lung; HT29: colon; MCF7: breast; HEPG2: liver. (6d) compounds selectively targeting cells with higher CA20 also down-regulate these genes. Scatter plot showing correlation between CTRP’s Spearman’s correlation coefficient from (a) and CMap’s drug average score from (c) for the 164 compounds tested in both datasets (Spearman’s correlation coefficient, r = 0.26, p-value = 8.3e-4). Points are coloured as in (a) and the predicted protein targets of compounds with both significant Spearman’s correlations and drug average score lower than -90 are shown.

Figure 7 shows a quantile-quantile (Q-Q) plot of observed versus expected—log 10 of Spearman’s correlation p-values between CA20 and drug-sensitivity (in AUC), with positive or negative sign if the correlation is positive or negative, respectively, across the Cancer Therapeutics Response Portal (CTRP) human cancer cell lines for 354 compounds. The solid line in the Q-Q plot indicates the distribution of compounds under the null hypothesis of no correlation. The compounds whose activity was associated with high and low CA20 (FDR < 0.05; Fig 6a) are represented in blue and red, respectively.

Figure 8 shows a heatmap of CMap’s drug score, ranging from 100 (maximum CA20 up-regulation) to -100 (maximum CA20 down-regulation) per cell line. Drug average score (last column) is the mean of drug scores across cell lines. The 20 compounds with the highest drug average score are shown and ranked accordingly. Tissue of origin of human cancer cell lines: PC3: prostate; VCAP: prostate; A375: melanoma; A549: lung; HA1 E: kidney; HCC515: lung; HT29: colon; MCF7: breast; HEPG2: liver. Figure 9 shows association between compounds’ targets and cell proliferation of TCGA samples. We used linear regression (gene expression ~ /¾ + /^‘proliferation rate + fc*cohort) to calculate the association between expression of each compound’s predicted target gene (we merged compound target annotations from the CTRP and CMap datasets) and proliferation rates across TCGA primary tumour samples. (9a and 9b) Scatter plots showing correlations between linear regression coefficient and (9a) CMap’s average scores or (9b) CTRP’s Spearman correlation coefficients of the respective compounds (Spearman’s correlation coefficient, r = 0.016 and -0.26, p-value = 0.84 and 9e-04, respectively). (9c) As in Fig 6d, but with compounds coloured by the linear regression coefficient of the predicted target gene (using the strongest association when a compound has more than one target gene). The two compounds with no annotated target gene are represented in grey. (9d) Example for gene RARG. Smooth scatter plot showing correlation between RARG gene expression and predicted proliferation rates of TCGA primary tumour samples. The linear regression p-value is shown.

Table 1 shows CA associated genes whose expression is very highly correlated (Rho > 0.65) with the expression of each of the CA20 genes of Ogden A et al. Sci Rep. 2017;7(1 ):262 across both the 11688 samples, representing 54 different human tissues, of the Genotype-Tissue Expression (GTEx) project (Broad Inst, MA, USA) and the 9749 samples, representing 27 types of cancer and matched normal tissues, of The Cancer Genome Atlas (TCGA) (NCI, USA).

Table 2 shows 351 compounds identified in the Connectivity Map (CMAP) (Broad Inst, MA USA) that down-regulate the expression of one or more of the CA associated genes of the CA20 panel and Table 2 in human cells.

Detailed Description

This invention relates to the unexpected finding that centrosome amplification renders cancer cells more sensitive to certain anti-cancer compounds (i.e. the compounds are CA selective). These compounds may be useful, for example in the treatment of cancer and the development of cancer therapeutics.

Aspects of the invention relate to methods of treating a cancer that displays centrosome amplification (CA) in an individual by administering a CA selective agent to an individual in need thereof.

A cancer that displays centrosome amplification (CA) comprises a high proportion of cancer cells which contain more than two centrosomes. For example, at least 10%, at least 20%, at least 30%, at least 40%, at least 50% or at least 60% of the cancer cells may contain more than two centrosomes.

Cancers that display centrosome amplification may be identified by any convenient technique. For example, immunofluorescence microscopy may be used to quantify centrosomes in cancer cells. Quantification may be performed manually or using cell image analysis software (e.g. CellProfiler™, Broad Institute MA, USA). In some preferred embodiments, a CA displaying cancer may be identified by the presence of a CA transcriptome signature, for example, a signature comprising expression of a panel of CA associated genes. CA associated genes may include genes that encode centrosome structural proteins, such as TUBG1. CA associated genes may also include genes whose

dysregulation induces CA, such as AURKA, CCNA2, CCND1, CCNE2, CDK1, CEP63, CEP152,

E2F1, E2F2, LM04, MDM2, MYCN, NDRG1, NEK2, PIN1, PLK1, PLK4, SASS6, STIL

CA associated genes may also include genes whose expression is highly correlated (e.g. Rho > 0.85) with the expression of genes encoding centrosome structural proteins or genes whose dysregulation induces CA. Suitable CA associated genes are shown in Table 1 (genes #1 to #119). Preferred CA associated genes may include genes #1 to #88 of Table 1 , genes #1 to #70, genes #1 to 47, genes #1 to #25 most preferably genes #1 to #8 of Table 1.

The panel may comprise 5 or more, 10 or more 15 or more, preferably 20 or more CA associated genes. Suitable panels of CA associated genes are known in the art (e.g. CA20; Ogden A et al. Sci Rep. 2017;7(1 ):262) or may be composed of 5 or more, 10 or more 15 or more, preferably 20 or more of CA associated genes #1 to #1 19 of Table 1 , more preferably genes #1 to #88, genes #1 to #70, genes #1 to 47, genes #1 to #25 or genes #1 to #8.

A CA panel score may be calculated as the sum of the normalized (log2 median-centred) expression levels of the CA associated genes in the panel. A cancer which displays CA may be identified from the CA panel score. For example, a panel score above a threshold value may be indicative of CA in a cancer cell. Suitable threshold values are dependent on the technique used to measure the gene expression and may be determined using routine experimentation. For example, suitable threshold value may for a CA panel score, such as a CA20 panel score, may be in the range -10 to 1 (see Figure 3).

In some embodiments, a cancer that displays CA may be a cancer that has high CA. CA may be determined to be high or low in a cancer cell obtained from an individual with cancer. A high CA cancer may display a CA panel score above the median CA panel score of the samples of the respective cancer type. A low CA cancer may display a CA panel score below the median CA panel score of the samples of the respective cancer type.

A method of treatment of a cancer displaying centrosome amplification (CA) as described herein may comprise administering to an individual in need thereof a therapeutically effective amount of a CA selective anti-cancer agent.

A CA selective anti-cancer agent is a compound that selectively kills cancer cells with CA i.e. the anti- cancer agent exerts an increased cytotoxic effects on cell with centrosome amplification relative to control cells which do not display centrosome amplification. The drug sensitivity of a CA selective anti- cancer agent (e.g. Area Under the dose-response Curve, AUC) may negatively correlate with the expression of a panel of CA associated genes, such as CA20 i.e. higher CA gene expression may be associated with lower drug AUC and, therefore, higher drug activity.

In preferred embodiments, a CA selective agent may be selected from the group consisting of (E)-4- [3-(1-adamantyl)-4-hydroxyphenyl]-3-chlorocinnamic acid (3-CI-AHPC; MM002; PubChem 9866186), 6-(4-Hydroxy-3-tricyclo[3.3.1.13,7]dec-1-ylphenyl)-2-naphtha lenecarboxylic acid (AHPN; CD-437; PubChem ID 135411 ), 4-[[[[4-(1 ,1-Dimethylethyl)phenyl]sulfonyl]amino]methyl]-/V-3- pyridinylbenzamide (STF-31 ; PubChem ID 984333), methotrexate (amethopterin; PubChem CID 126941 ), polo-like kinase 1 (PLK1 ) inhibitors, such as 4-[[(7R)-8-cyclopentyl-7-ethyl-5-methyl-6-oxo- 7H-pteridin-2-yl]amino]-3-methoxy-N-(1-methylpiperidin-4-yl) benzamide (BI-2536; PubChem CID 11364421 ), clofarabine (PubChem CID 119182), A^-Cyclopropyl^-^-il-methylethy phenyllmethyl]- 7H-pyrrolo[3,2-/]quinazoline-1 , 3-diamine dihydrochloride (CH-79797; PubChem ID 45073452), (3E,5E)-3,5-bis[(4-methylphenyl)methylidene]piperidin-4-one (NSC632839; PubChem ID 5351362), etoposide (PubChem ID 36462), 3-Methyl-N-[1 ,4,5,6-tetrahydro-6,6-dimethyl-5-[(1-methyl-4- piperidinyl)carbonyl]pyrrolo[3,4-c]pyrazol-3-yl]butanamide (PHA-793887, PubChem ID 46191454), N- [5-hydroxy-5-[7-[(2-hydroxy-5-oxocyclopenten-1-yl)amino]-7-o xohepta-1 ,3,5-trienyl]-2-oxo-7- oxabicyclo[4.1.0]hept-3-en-3-yl]-2,4,6-trimethyldeca-2,4-die namide (manumycin A; PubChem 4010) and (7S,9S)-7-[(2R,4S,5S,6S)-4-amino-5-hydroxy-6-methyloxan-2-yl ]oxy-6,9,11-trihydroxy-9-(2- hydroxyacetyl)-4-methoxy-8,10-dihydro-7H-tetracene-5,12-dion e (doxorubicin, PubChem 31703).

A CA selective agent may reduce the expression of genes associated with CA (i.e. CA antagonists).

CA antagonists may also include cyclin-dependent kinase (CDK) inhibitors, such as purvalanol-a and aminopurvalanol-a, JAK3-inhibitor-VI, etoposide, RARG agonists, such as CD-437, coagulation factor II (F2R) inhibitors, farnesyltransferase (FNTA and FNTB) inhibitors, ubiquitin isopeptidase (USP13 and USP5) inhibitors, DNA topoisomerase II alpha (TOP2A) inhibitors and tubulin inhibitors, such as CCB02 (3-methoxybenzo[b][1 ,6]naphthyridine-4-carbonitrile) (Mariappan et al EMBO (2019) 38 e99876).

CA antagonists may also include any one of compounds #1 to #351 of Table 2, which have been shown to down-regulate the expression of CA associated genes, for example any one of compounds #1 to #281 , any one of compounds #1 to #207, or any one of compounds #1 to #3 of Table 2.

The terms“CA antagonist” and“CA selective agent” as used herein, cover pharmaceutically acceptable salts and solvates of these compounds.

Cancers suitable for treatment as described herein may include cancers that display CA i.e. a high proportion of cancer cells in the patient contain more than two centrosomes. Cancer is a

physiological condition in mammals that is typically characterized by unregulated or abnormal cell growth or proliferation. Cancers may include lymphomas, such as Hodgkin lymphoma, non-Hodgkin lymphoma, and solid cancers such as carcinomas, sarcomas, skin cancer, melanoma, bladder cancer, brain cancer, breast cancer, uterus cancer, ovary cancer, prostate cancer, kidney cancer, lung cancer, colorectal cancer, gastrointestinal cancer, cervical cancer, liver cancer, head and neck cancer, oesophageal cancer, pancreas cancer, renal cancer, adrenal cancer, stomach cancer, testicular cancer, cancer of the gall bladder and biliary tracts, thyroid cancer, thymus cancer, cancer of bone, and cerebral cancer, as well as cancer of unknown primary (CUP). In some embodiments, the breast cancer may be luminal B breast cancer or ductal breast cancer, In some embodiments, the cancer is not breast cancer.

Preferred cancers that display CA (i.e. a high CA panel score) may include cervical, testicular and oesophageal cancer, subtypes of breast cancer, such as luminal B breast cancer or ductal breast cancer, and subtypes of lung cancer, such as lung squamous cell carcinoma. Other suitable cancers that display high CA (for example a CA20 panel score >0) are shown in Figure 1 B.

A cancer may be located at its primary location, such as the brain in the case of cerebral cancer, or at a distant location in the case of metastases. A tumour may result from of any of the cancers mentioned above. A tumour which is the result of a particular cancer may include both a primary tumour and tumour metastases of said cancer. Thus, a tumour which is the result of cancer of a first tissue, for example, includes both a primary tumour in the first tissue and metastases of the cancer found in other tissues of a patient’s body.

An individual suitable for treatment as described above may be a mammal, such as a rodent (e.g. a guinea pig, a hamster, a rat, a mouse), murine (e.g. a mouse), canine (e.g. a dog), feline (e.g. a cat), equine (e.g. a horse), a primate, simian (e.g. a monkey or ape), a monkey (e.g. marmoset, baboon), an ape (e.g. gorilla, chimpanzee, orang-utan, gibbon), or a human.

In some preferred embodiments, the individual is a human. In other preferred embodiments, nonhuman mammals, especially mammals that are conventionally used as models for demonstrating therapeutic efficacy in humans {e.g. murine, primate, porcine, canine, or leporid) may be employed.

An individual with cancer may display at least one identifiable sign, symptom, or laboratory finding that is sufficient to make a diagnosis of cancer in accordance with clinical standards known in the art. Examples of such clinical standards can be found in textbooks of medicine such as Harrison’s Principles of Internal Medicine, 15th Ed., Fauci AS et al., eds., McGraw-Hill, New York, 2001. In some instances, a diagnosis of a cancer in an individual may include identification of a particular cell type (e.g. a cancer cell) in a sample of a body fluid or tissue obtained from the individual. In some embodiments, the individual may have been previously identified or diagnosed with cancer or a method of the invention may comprise identifying or diagnosing cancer in the individual for example by determining the presence of an identifiable sign, symptom, or laboratory finding indicative of cancer in the individual.

Treatment may be any treatment or therapy, whether of a human or an animal (e.g. in veterinary applications), in which some desired therapeutic effect is achieved, for example, the inhibition or delay of the progress of the condition, and includes a reduction in the rate of progress, a halt in the rate of progress, amelioration of the condition, cure or remission (whether partial or total) of the condition, preventing, delaying, abating or arresting one or more symptoms and/or signs of the condition or prolonging survival of a subject or patient beyond that expected in the absence of treatment.

In particular, treatment may include inhibiting cancer growth, including complete cancer remission, and/or inhibiting cancer metastasis. Cancer growth generally refers to any one of a number of indices that indicate change within the cancer to a more developed form. Thus, indices for measuring an inhibition of cancer growth include a decrease in cancer cell survival, a decrease in tumour volume or morphology (for example, as determined using computed tomographic (CT), sonography, or other imaging method), a delayed tumour growth, a destruction of tumour vasculature, and a decrease in levels of tumour-specific antigens. Administration of CA selective agents as described herein may improve the capacity of the individual to resist cancer growth, in particular growth of a cancer already present the subject and/or decrease the propensity for cancer growth in the individual.

A suitable individual for treatment may be non-responsive to first line therapy for the cancer.

Whilst a CA selective agent may be administered alone, it is preferable to present it as a

pharmaceutical composition (e.g. formulation) which comprises the CA selective agent, together with one or more pharmaceutically acceptable carriers, adjuvants, excipients, diluents, fillers, buffers, stabilisers, preservatives, lubricants, or other materials well known to those skilled in the art and, optionally, other therapeutic or prophylactic agents. Such materials should be non-toxic and should not interfere with the efficacy of the active compound. The precise nature of the carrier or other material will depend on the route of administration, which may be by bolus, infusion, injection or any other suitable route, as discussed below. Suitable materials will be sterile and pyrogen-free, with a suitable isotonicity and stability. Examples include sterile saline (e.g. 0.9% NaCI), water, dextrose, glycerol, ethanol or the like or combinations thereof. The composition may further contain auxiliary substances such as wetting agents, emulsifying agents, pH buffering agents or the like.

Methods of the invention may therefore comprise the step of formulating a CA selective agent as described herein with a pharmaceutically acceptable carrier, adjuvant or excipient. In some embodiments, the CA selective agent may be the only active ingredient in the pharmaceutical composition.

The term“pharmaceutically acceptable” as used herein pertains to compounds, materials, compositions, and/or dosage forms which are, within the scope of sound medical judgement, suitable for use in contact with the tissues of a subject (e.g., human) without excessive toxicity, irritation, allergic response, or other problem or complication, commensurate with a reasonable benefit/risk ratio. Each carrier, excipient, etc. must also be“acceptable” in the sense of being compatible with the other ingredients of the formulation. The precise nature of the carrier or other material will depend on the route of administration, which may be any non-oral route, for example by injection, e.g. cutaneous, subcutaneous, or intravenous.

The CA selective agent may be administered to a subject by any convenient route of administration, whether systemically/peripherally or at the site of desired action, including but not limited to parenteral, for example, by injection, including subcutaneous, intradermal, intramuscular, intravenous, intraarterial, intracardiac, intrathecal, intraspinal, intracapsular, subcapsular, intraorbital,

intraperitoneal, intratracheal, subcuticular, intraarticular, subarachnoid, and intrasternal, preferably subcutaneous; by implant of a depot, for example, subcutaneously or intramuscularly. For surface cancers, such as skin cancer, the CA selective agent may be applied topically in solution, for example with an organic solvent, such as DMSO.

For intravenous, cutaneous or subcutaneous injection, or injection at the site of affliction, the CA selective agent will be in the form of a parenterally acceptable aqueous solution which is pyrogen-free and has suitable pH, isotonicity and stability. Those of relevant skill in the art are well able to prepare suitable solutions using, for example, isotonic vehicles such as Sodium Chloride Injection, Ringer's Injection, or Lactated Ringer's Injection. Preservatives, stabilisers, buffers, antioxidants and/or other additives including buffers such as phosphate, citrate and other organic acids; antioxidants, such as ascorbic acid and methionine; preservatives (such as octadecyldimethylbenzyl ammonium chloride; hexamethonium chloride; benzalkonium chloride; benzethonium chloride; phenol, butyl or benzyl alcohol; alkyl parabens, such as methyl or propyl paraben; catechol; resorcinol; cyclohexanol; 3’- pentanol; and m-cresol); low molecular weight polypeptides; proteins, such as serum albumin, gelatin or immunoglobulins; hydrophilic polymers, such as polyvinylpyrrolidone; amino acids, such as glycine, glutamine, asparagines, histidine, arginine, or lysine; monosaccharides, disaccharides and other carbohydrates including glucose, mannose or dextrins; chelating agents, such as EDTA; sugars, such as sucrose, mannitol, trehalose or sorbitol; salt-forming counter-ions, such as sodium; metal complexes (e.g. Zn-protein complexes); and/or non-ionic surfactants, such as TWEEN™,

PLURONICS™ or polyethylene glycol (PEG) may be included, as required. Suitable carriers, excipients, etc. can be found in standard pharmaceutical texts, for example, Remington’s

Pharmaceutical Sciences, 18th edition, Mack Publishing Company, Easton, Pa., 1990.

The pharmaceutical compositions and formulations may conveniently be presented in unit dosage form and may be prepared by any methods well known in the art of pharmacy. Such methods include the step of bringing into association the CA selective agent with the carrier which constitutes one or more accessory ingredients. In general, the compositions are prepared by uniformly and intimately bringing into association the active compound with liquid carriers. Formulations may for example be in the form of liquids or solutions.

A CA selective agent may be administered as described herein in therapeutically-effective amounts.

A“therapeutically-effective amount" is the amount of an active compound, or a combination, material, composition or dosage form comprising an active compound, which is effective for producing some desired therapeutic effect, commensurate with a reasonable benefit/risk ratio. The appropriate dosage of an active compound may vary from individual to individual. Determining the optimal dosage will generally involve the balancing of the level of therapeutic benefit against any risk or deleterious side effects of the administration. The selected dosage level will depend on a variety of factors including, but not limited to, the route of administration, the time of administration, the rate of excretion of the active compound, other drugs, compounds, and/or materials used in combination, and the age, sex, weight, condition, general health, and prior medical history of the individual. The amount of active compounds and route of administration will ultimately be at the discretion of the physician, although generally the dosage will be to achieve therapeutic plasma concentrations of the active compound without causing substantial harmful or deleterious side-effects.

In general, a suitable dose of the CA selective agent may be in the range of about 10 pg to about 400 mg per kilogram body weight of the subject per day, preferably 200 pg to about 200 mg per kilogram body weight of the subject per day, for example 5-10 mg/kg/day. Where the active compound is a salt, an ester, prodrug, or the like, the amount administered is calculated on the basis of the parent compound and so the actual weight to be used is increased proportionately.

The pharmaceutical compositions comprising the active compounds may be formulated in a dosage unit formulation that is appropriate for the intended route of administration.

Administration in vivo can be effected in one dose, continuously or intermittently (e.g., in divided doses at appropriate intervals). Methods of determining the most effective means and dosage of administration are well known in the art and will vary with the formulation used for therapy, the purpose of the therapy, the target cell being treated, and the subject being treated. Single or multiple administrations can be carried out with the dose level and pattern being selected by the physician. Multiple doses of the CA selective agent may be administered, for example 2, 3, 4, 5 or more than 5 doses may be administered. The administration of the CA selective agent and concomitant radiotherapy may continue for sustained periods of time. For example treatment with the CA selective agent may be continued for at least 1 week, at least 2 weeks, at least 3 weeks, at least 1 month or at least 2 months. Treatment with the CA selective agent may be continued for as long as is necessary to reduce cancer symptoms or achieve complete remission.

The CA selective agent may be administered alone or in combination with other treatments, either simultaneously or sequentially dependent upon the individual circumstances. For example, a CA selective agent may be administered in combination with one or more additional active compounds, for example chemotherapeutic agents, checkpoint inhibitors or other anti-cancer drugs.

An individual may be successfully "treated" according to the methods of described herein if the individual shows one or more of the following: a reduction in the number of or complete absence of cancer cells; a reduction in the tumor size; or retardation or reversal of tumor growth, inhibition, e.g., suppression, prevention, retardation, shrinkage, or reversal of metastases, e.g., of cancer cell infiltration into peripheral organs including, for example, the spread of cancer into soft tissue and bone; inhibition of, e.g., suppression of, retardation of, prevention of, shrinkage of, reversal of or an absence of tumor metastases; inhibition of, e.g., suppression of, retardation of, prevention of, shrinkage of, reversal of or an absence of tumor growth; relief of one or more symptoms associated with the specific cancer; reduced morbidity and mortality; improvement in quality of life; or some combination of effects. Beneficial or desired clinical results include, but are not limited to, alleviation of symptoms, diminishment of extent of disease, stabilized (/.e., not worsening) state of disease, delay or slowing of disease progression, amelioration or palliation of the disease state, and remission (whether partial or total), whether detectable or undetectable. "Treatment" can also mean prolonging survival as compared to expected survival if not receiving treatment. Those in need of treatment include those already with the condition or disorder as well as those prone to have the condition or disorder or those in which the condition or disorder is to be prevented

The present inventors have also recognised that compounds that reduce the expression of CA associated genes exert an anti-cancer effect on cancers that display CA, Other aspects of the invention relate to methods of screening to identify CA selective agents or CA antagonists that are potentially useful in the treatment of cancer. For example, a method of screening for a compound useful in the treatment of cancer may comprise; determining the effect of a test compound on the expression of a panel of CA associated genes in a cell.

In some embodiments, the expression of a panel of CA associated genes may be determined in the presence or absence of the test compound. A decrease in expression in the presence relative to the absence of the test compound may be indicative that the compound is a CA antagonist that is potentially useful in the treatment of cancer. For example, a CA panel score may be calculated as the sum of the normalized (log2 median-centred) expression levels of the CA associated genes in the panel. A decrease in the CA panel score in the presence relative to the absence of the test compound may be indicative that the compound is a CA antagonist that is potentially useful in the treatment of cancer.

Cells for use in screening may include cancer cells, for example primary cancer cells derived from patients or immortalised cancer cell lines, including cell lines derived from leukaemia (e.g. HL_60, K_562), lung cancer (e.g. NCI_H23, HOP_92) or renal cancer (e.g. UO-31 ). Cells for use in screening, for example as controls, may also include non-cancer cells, such as non-cancerous mammary epithelium cells (e.g. HB2). The precise format of any of the screening or assay methods of the present invention may be varied by those of skill in the art using routine skill and knowledge. The skilled person is well aware of the need to employ appropriate control experiments. For example, in some embodiments, the expression of the panel and/or the CA panel score may be determined in control cells.

A test compound may be an isolated molecule or may be comprised in a sample, mixture or extract, for example, a biological sample. Compounds which may be screened using the methods described herein may be natural or synthetic chemical compounds used in drug screening programmes.

Extracts of plants, microbes or other organisms, which contain several characterised or

uncharacterised components may also be used. Suitable test compounds for screening include cyclin-dependent kinase (CDK) inhibitors, such as purvalanol-a and aminopurvalanol-a, JAK3- inhibitor-VI and analogues and derivatives thereof, etoposide and analogues and derivatives thereof, RARG agonists, such as CD-437, coagulation factor II (F2R) inhibitors, farnesyltransferase (FNTA and FNTB) inhibitors, ubiquitin isopeptidase (USP13 and USP5) inhibitors, and DNA topoisomerase II alpha (TOP2A) inhibitors. Other suitable test compounds include variants, analogues and derivatives of the CA selective compounds listed above.

Suitable test compounds also include analogues, derivatives, variants and mimetics of any of the compounds listed above, for example compounds produced using rational drug design to provide test candidate compounds with particular molecular shape, size and charge characteristics suitable for modulating CA expression.

Combinatorial library technology provides an efficient way of testing a potentially vast number of different compounds for ability to modulate CA gene expression. Such libraries and their use are known in the art, for all manner of natural products, small molecules and peptides, among others.

The use of peptide libraries may be preferred in certain circumstances.

The amount of test compound which may be added to an assay of the invention will normally be determined by trial and error depending upon the type of compound used. Typically, from about 0.001 nM to 1 mM or more concentrations of putative inhibitor compound may be used, for example from 0.01 nM to 100mM, e.g. 0.1 to 50 mM, such as about 10 mM. Even a compound which has a weak effect may be a useful lead compound for further investigation and development.

A test compound identified as displaying selective activity against cells with CA may be investigated further.

For example, the effect of the test compound in normal non-cancer cells may be determined.

Preferred compounds do not exert cytotoxic effects on non-cancer cells that do not display CA. A test compound identified as a CA antagonist or CA selective anti-cancer agent may be isolated and/or purified or alternatively, it may be synthesised using conventional techniques of recombinant expression or chemical synthesis. Furthermore, it may be manufactured and/or used in preparation, i.e. manufacture or formulation, of a composition such as a medicament, pharmaceutical composition or drug. Methods described herein may thus comprise formulating the test compound in a pharmaceutical composition with a pharmaceutically acceptable excipient, vehicle or carrier for therapeutic application.

Following identification of a CA selective anti-cancer agent that is potentially useful in the treatment of cancer or the development of therapeutics for the treatment of cancer, for example cancer associated with centrosome amplification (CA), a method may further comprise modifying the compound to optimise its pharmaceutical properties. Suitable methods of optimisation, for example by structural modelling, are well known in the art. Further optimisation or modification can then be carried out to arrive at one or more final compounds for in vivo or clinical testing.

Other aspects of the invention relate to methods of determining the sensitivity of a cell to CA selective agents, such as CA antagonists. For example, the sensitivity of a cell, such as a cancer cell, to a CA selective agent may comprise determining the expression of a panel of CA associated genes in the cell. Expression of the panel above a threshold value may be indicative the cancer cell is sensitive to the CA selective agent.

In some embodiments, a CA panel score may be calculated as the sum of the normalized (log2 median-centred) expression levels of the CA associated genes in the panel. A CA panel score above a threshold value may be indicative that the cancer cell is sensitive to the CA selective agent. The threshold value may depend on the technique used to measure the gene expression and may be determined using routine experimentation for any particular set up. For example, a suitable threshold value may for a CA panel score, such as a CA20 panel score, may be in the range -10 to 1 (see Figure 3).

In some embodiments, the threshold value may be the median CA panel score of a population of samples of the respective cancer type. A high CA cancer that is sensitive to CA selective agents as described herein may display a CA panel score above the median CA panel score of the samples of the respective cancer type. A low CA cancer that is not sensitive to CA selective agents may display a CA panel score below the median CA panel score of the samples of the respective cancer type.

CA selective agents are described in more detail above.

Other aspects of the invention relate to methods of assessing genomic instability in cancer cells. A method of assessing genomic instability in a cancer cell may comprise determining the expression of a panel of CA associated genes in the cell. Expression of the panel is indicative of the presence, level or amount of genomic instability in the cancer cell.

Expression above a threshold value may be indicative of genomic instability. For example, expression of the panel may be indicative of one or more of tumour aneuploidy, mutation burden, number of somatic Copy Number Alterations (CNA), tumour heterogeneity, alteration of specific chromosomal arms and C>T mutations.

Threshold values are described in more detail above.

Genomic instability is associated with malignancy and the presence of genomic instability may be indicative of a malignant or aggressive cancer. The assessment of genomic instability may therefore be useful in the prognosis of a cancer condition in a patient and in assessing the effectiveness of cancer treatment.

Other aspects of the invention relate to the assessment of cancer conditions. A method of assessing a cancer condition in an individual may comprise determining the expression of a panel of CA associated genes in one or more cells obtained from the individual.

Suitable panels of CA associated genes include the CA20 panel described herein,

In some embodiments, the cells may be cancer cells or cells suspected of being cancer cells.

Expression of the panel may be indicative of the status, diagnosis and/or prognosis of the cancer condition in the individual. For example, increased expression of the panel in the test cells relative to controls may be indicative that the test cells are cancer cells.

Expression of a panel of CA associated genes above a threshold value as described herein may be indicative of a poor prognosis for the individual i.e. reduced likelihood of survival or increased likelihood of disease progression relative to individuals with the respective cancer whose expression of the panel is below the median expression for the cancer type,.

As described above, the expression of panel of CA associated genes may be expressed as a CA panel score. A CA panel score may be calculated as the sum of the normalized (log2 median-centred) expression levels of the CA associated genes in the panel. A CA panel score above a threshold value may be indicative that of a poor prognosis for the individual. The threshold value may depend on the technique used to measure the gene expression and may be determined using routine

experimentation for any particular set up. For example, a suitable threshold value may for a CA panel score, such as a CA20 panel score, may be in the range -10 to 1 (see Figure 3).

In some embodiments, a suitable threshold value may include the median CA panel score of a population of samples of the respective cancer type. A high CA cancer that has a poor prognosis may have a CA panel score above the median CA panel score of the samples of the respective cancer type. A low CA cancer that does not have a prognosis may have a CA panel score below the median CA panel score of the samples of the respective cancer type.

Suitable cancer conditions for assessment using a CA associated gene panel as described herein include Mesothelioma, Adrenocortical Carcinoma, Brain Lower Grade Glioma, Breast Invasive Carcinoma, Lung Adenocarcinoma, Pancreatic Adenocarcinoma, Kidney Chromophobe and Uveal Melanoma (see Figure 5A).

In some embodiments, expression of the panel may be determined in cancer cells obtained at more than one time point, for example, at first and second time points. The change in expression between the time points may be determined. Monitoring expression of the panel may be indicative of the aggressiveness and metastatic potential of the cancer condition and/or the responsiveness of the cancer to treatments administered to the individual.

Other aspects and embodiments of the invention provide the aspects and embodiments described above with the term“comprising” replaced by the term“consisting of and the aspects and embodiments described above with the term“comprising” replaced by the term’’consisting essentially of.

It is to be understood that the application discloses all combinations of any of the above aspects and embodiments described above with each other, unless the context demands otherwise. Similarly, the application discloses all combinations of the preferred and/or optional features either singly or together with any of the other aspects, unless the context demands otherwise.

Modifications of the above embodiments, further embodiments and modifications thereof will be apparent to the skilled person on reading this disclosure, and as such, these are within the scope of the present invention.

All documents, websites and sequence database entries mentioned in this specification are incorporated herein by reference in their entirety for all purposes.

“and/or” where used herein is to be taken as specific disclosure of each of the two specified features or components with or without the other. For example“A and/or B” is to be taken as specific disclosure of each of (i) A, (ii) B and (iii) A and B, just as if each is set out individually herein.

Certain aspects and embodiments of the invention will now be illustrated by way of example and with reference to the figures described above. CA may be estimated from the expression levels of CA-associated genes. Recently, proof-of-principle gene-expression-based CA signatures have been developed (20-23), the most comprehensive one being CA20, based on the expression of TUBG1, which encodes the most abundant centrosomal protein, and 19 other genes whose overexpression has been experimentally shown to induce CA (23). This signature was proposed to reflect CA levels in breast tumour samples and shown to have a prognostic value in two independent breast cancer cohorts (23).

In the present study, we used CA20 to estimate relative CA levels across 9,721 tumour and 725 matched-normal samples of 32 cancer types from The Cancer Genome Atlas (TCGA), thereby revealing the first pan-cancer landscape of CA-associated gene expression. We show the association of CA20 with distinct breast cancer clinical and molecular features. We also break down the independent associations of CA20 with different sorts of genomic instability across cancer types. Finally, we show that high CA20 is associated with poor clinical outcome in different cancer types, having identified candidate compounds for selectively targeting cancer cells exhibiting transcriptomic evidence for this hallmark of cancer.

Methods

TCGA dataset

Publicly available RNAseqV2 (quantified through RNA-seq by Expectation Maximization) (74) and clinical data for 9,721 tumour and 725 matched-normal samples from The Cancer Genome Atlas (TCGA; https://canceraenome.nih.gov/) were downloaded from Firebrowse (http://firebrowse.org/)· Gene expression (read counts) data were quantile-normalized using voom (75). For each sample, the CA20 score was calculated as the sum of the across-sample (including both tumours and matched- normal samples) normalized (log2 median-centred) expression levels of the CA20 published signature genes (23): AURKA, CCNA2, CCND1 , CCNE2, CDK1 , CEP63, CEP152, E2F1 , E2F2, LM04, MDM2, MYCN, NDRG1 , NEK2, PIN1 , PLK1 , PLK4, SASS6, STIL and TUBG1 (Fig 1a). Predicted proliferation rates of each TCGA tumour sample were retrieved from (24) (n = 9,568). Whole genome doubling (corresponding to 0, 1 and > 2 genome doubling events in the clonal evolution of the cancer), aneuploidy (both aneuploidy score - number of altered chromosome arms - and alterations per chromosome arm) and mutation burden characterizations were retrieved from (34) (n = 9,166). Since the chromosomal arm status was not available for TCGA normal samples, we have selected only those with no CNA in the chromosomal arms tested, to make sure they are intact. CNA (n = 8,879; copy number levels were derived with the GISTIC algorithm (76) and considered as CNA if having a score lower than -1 (loss) or higher than 1 (gain)) and mutation (n = 7,120; including classification as silent, missense, splice site or nonsense ones) processed data were downloaded from Firebrowse (http://firebrowse.org/). Mutations were classified as likely pathogenic and pathogenic based on ClinVar database’s (https://www.ncbi.nlm.nih.gov/clinvar/) variant summary annotation

(ftp://ftp.ncbi. nlm.nih.gov/pub/clinvar/tab_delimited/variant_summary.txt.gz ; accessed in November 12 ^th 2018), and 5,601 likely driver mutations were obtained from the Cancer Genome Interpreter (https://www.cancergenomeinterpreter.org/mutations; accessed in November 12 ^th 2018) (45). The list of 299 cancer driver genes was retrieved from (43). Intra-tumour heterogeneity data, measured by the number of clones per sample, were retrieved from (77) (n = 1 ,080). The mutational signature profiles were retrieved from mSignatureDB (48) (n = 9,004). The predicted fraction of stromal (stromal score) and immune (immune score) cells in TCGA tumour samples (n = 2,463) was retrieved from (78). We used the scores calculated based on RNASeqV2 expression levels. Importantly, no CA20 gene was used by the authors to infer those cell proportions (78).

TCGA tumour samples were analysed for hypoxic status based on expression of 95 genes included in the hypoxia 99-metagene signature (56). The four missing genes are three ( LOC149464 , LOC56901 and TIMM23) for which expression levels were not available and NDRG1, excluded for being part of the CA20 gene signature. The hypoxia score was calculated like the CA20 score. Additional clinical information for TCGA breast tumour samples was retrieved from (27).

METABRIC dataset

Normalized gene expression data for 1992 primary breast tumours and 144 normal breast tissue samples from the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) (28) were retrieved from European Genome-Phenome Archive (EGAC00001000484). Gene expression was profiled with lllumina HT-12 v3 microarrays, with probe-level intensity values being mean- summarised per gene. The CA20 score was calculated as for the TCGA dataset. Clinical information for the same samples was downloaded from cBioPortal (http://www.cbioportal.ora/) (79).

Analyses of CA in human breast carcinoma samples

Quantification of CA in breast cancer samples was performed as described in (26). Briefly, formalin- fixed and paraffin-embedded human breast carcinoma samples were consecutively retrieved from the files of the Department of Pathology, Hospital Xeral-Cies, Vigo, Spain. This series comprises 29 luminal A, 3 luminal B, 3 HER2 and 13 basal-like tumours. Some of these samples had already been used in one of our recent studies (26). The status of the oestrogen receptor (ER), progesterone receptor (PR), epidermal growth factor receptor 2 (HER2), antigen Ki67, and the basal markers epidermal growth factor receptor, cytokeratin 5, cytokeratin 14, P-cadherin and Vimentin was previously characterized for all tumour cases. According to their immunoprofile, breast tumour samples were classified as luminal A (ER+, PR+, HER2- and KΪ67-), luminal B (ER+, PR+, HER2 overexpressing or KΪ67+), HER2 (ER-, PR-, HER2 overexpressing) or basal-like carcinomas (ER-, PR-, HER2-, basal marker+). Representative tumour areas were carefully selected and at least two tissue cores (0.6 mm in diameter) were deposited into a tissue microarray. This study was conducted under the national regulative law for the handling of biological specimens from tumour banks, with samples being exclusively used for research purposes in retrospective studies. Informed consent was obtained from all human participants.

For immunofluorescence staining, 3 pm-thick tissue sections were deparaffinised in Clear-Rite-3 (Thermo Scientific, USA, CA) and rehydrated using a series of solutions with decreasing concentrations of ethanol. High temperature (98 °C, 60 min) antigenic retrieval with Tris-EDTA pH = 9.0 (LeicaBio systems, UK) was performed, followed by incubation with UltraVision protein block (Thermo Scientific) for 30 min at room temperature. The slides were, afterwards, incubated with mouse anti-GT335 (1/800 dilution, Adipogen Ref. AG- 20B-0020-C100) and rabbit anti-pericentrin (1/250 dilution, Abeam AB4448) in UltraAb diluent (Thermo Scientific) overnight at 4 °C. The sections were then washed three times, 5 min per wash, with 1 * PBS + 0.02% Tween20 before a 1 h room temperature incubation with the secondary antibodies, anti-lgG rabbit coupled to Alexa 488 and anti- IgG mouse coupled to Alexa-594 (Invitrogen), diluted at 1/500 in PBS. Finally, sections were washed extensively with 1 * PBS + 0.02% Tween20 and then counterstained and mounted with Vectashield containing DAPI (VectorLabs, CA, USA).

Imaging was performed on a Zeiss Imager Z1 inverted microscope, equipped with an AxioCam MRm camera (Zeiss) and ApoTome (Zeiss), using the *100 1.4 NA Oil immersion objective. Images were taken as Z-stacks in a range of 10-14 pm, with a distance between planes of 0.3 pm, and were deconvolved with AxioVision 4.8.1 software (Zeiss). Only the structures positive for GT335 (centriolar marker) and pericentrin (PCM marker) were analysed and scored. Between 5 and 107 cells were analysed for each patient and cells with more than 4 centrioles were considered as having CA.

CTRP dataset

Normalized gene-level expression and drug sensitivity (n = 481 compounds) data for 823 human cancer cell lines from the Cancer Therapeutics Response Portal (CTRP) v2 were retrieved from (58). The CA20 score was calculated as for the aforementioned datasets. Compounds with more than 20% of missing data (n = 127) were removed from the analyses. Area Under the dose-response Curve (AUC) was used as the metric of cell line’s drug sensitivity, measured over a 16-point concentration range. Note that lower AUC means higher drug activity.

Connectivity Map dataset

The Connectivity Map (CMap) database of signatures (49) was interrogated using CA20 genes as an individual query in the CLUE L1000 tool (https://clue.io/H 000-query#individual, login required; CA20 genes were used as putative UP-regulated genes). For each of the 9 human cancer cell lines profiled within the Touchstone dataset (PC3, VCAP, A375, A549, HA1 E, HCC515, HT29, MCF7 and HEPG2), a connectivity score was computed per perturbation (gene knock-down, gene overexpression, small molecule administration) (49), reflecting its effect on the expression of CA20 genes (except for

SASS6, not profiled in this dataset). We calculated an average connectivity score per perturbation by averaging the 9 cell lines’ connectivity scores in order to have a more robust connectivity score that can be used across different cell types and tissues. Two types of perturbations were analysed: 3,799 gene knock-downs and 2,837 compounds. The Broad compound ID was used to match the 164 compounds tested by CMap and CTRP, so that the results of the analyses of the two datasets could be combined. Multiciliogenesis datasets

Normalized gene expression data for adult mouse airway epithelial cells during multiciliogenesis (triplicates for three different time points: days 0, 2 and 4) was retrieved from (80) (GEO dataset accession GSE73331). The CA20 score was calculated as for the TCGA dataset.

The transcriptomic alterations between non-ciliating mouse tracheal epithelial cells and those undergoing differentiation, through transition to an air-liquid interface culture (ALI), and harvested at four (ALI+4) or twelve (ALI+12) days, were retrieved from (81 ). Those probe-level transcriptomic alterations were mean-summarised per gene.

Spearman’s correlation

Spearman’s correlations were performed using the cor.test R function (method =“spearman”) (82). The difference between two Spearman’s correlations was tested using the paired. r function from R package psych (83).

Unpaired two-sample statistical analyses

Wilcoxon rank-sum tests were performed using the wiicox.test R function (82).

Linear regression analyses

Multiple linear regression modelling was implemented using the Im function from R package limma (84). Covariate collinearity was tested using the corvif function from (85), in which all covariates had a variance inflation factor below 2.

We have normalised the genomic instability covariates using z-scores (number of standard deviations from the mean) to account for differences in the prevalence of aneuploidy, mutation burden, CNA and number of clones per cohort.

Fligner-Killeen test of homogeneity of variances

Fligner-Killeen test was implemented using the fligner.test R function (82).

Test of equal proportions

Proportions tests were performed using the prop.test R function (82).

Two-way analysis of variance (ANOVA)

Two-way ANOVA was done using the aov R function (82).

Hierarchical clustering analyses

Unsupervised hierarchical clustering of the multiple linear regression results per cancer type was performed using the heatmap.2 function from R package gplots (86). Gene Set Enrichment Analyses

Genes ranked according to the knock-down connectivity score were analysed for pathway enrichment using Gene Set Enrichment Analysis (41 ,42) with default parameters. We used a list of 299 cancer driver genes from (43), a manually curated list of centriole duplication factors (93 genes, including 10 from the CA20 signature), gene sets retrieved from the KEGG pathway database

(https://www.keaa.ip/) and the MSigDB’s Hallmark Gene Sets library (50). Those with a False Discovery Rate (FDR) lower than 5% were considered significant.

Survival analyses

Dividing patients into two subgroups by CA20 median value, the significance of differences in prognostic was estimated using Kaplan-Meier plots and log-rank tests, per cancer type, through R package survival (87).

Q-Q plot of p-values

To calculate the expected Spearman’s correlation coefficients and p-values used in the quantile- quantile (Q-Q) plot, we permutated 1000 times the drug-sensitivity (in AUC) of all compounds across cell lines and, for each permutated dataset, we calculated the respective CA20-AUC Spearman’s correlations. The expected values were obtained by median-summarizing the ranked 1000 permutations’ results.

Results

The pan-cancer landscape of centrosome amplification-associated gene expression

To estimate relative CA levels in human samples, we used CA20, a score based on the expression of 20 genes experimentally associated with CA (23), as a surrogate. We quantified CA20 across the transcriptomes (profiled by RNA-seq) of 9,721 tumour and 725 matched-normal samples spanning 32 cancer types from TCGA (Fig 1a). CA20 correlates with the predicted proliferation rates of TCGA tumour samples (24) (Spearman’s correlation coefficient, r = 0.4, p-value < 2.2e-16), as expected, given that some of the CA20 genes encode for proteins involved in cell proliferation. Cervical (CESC), testicular (TGCT) and oesophageal (ESCA) cancers show high CA20, contrasting with lower scores in kidney (KIRP, KICH and KIRC) and prostate (PRAD) cancers (Fig 1 b). Some cancer types, such as low-grade glioma (LGG) and breast invasive carcinoma (BRCA), exhibit high variability of CA20, concordantly with previous observations that the proportion of cells with CA in breast tumours ranges from 1 to 100% (7,25) depending on the tumour subtype (26).

We also observed significant differences in CA20 between specific cancer types with the same tissue of origin. Although all kidney cancers have low CA20 scores, kidney renal papillary cell carcinoma (KIRP) shows a lower score than the other types (p-value < 0.0001 , Wilcoxon rank-sum test; Similarly, glioblastoma multiforme (GBM), skin cutaneous melanoma (SKCM) and lung squamous cell carcinoma (LUSC) show higher CA20 than low-grade glioma (LGG), uveal melanoma (UVM) and lung adenocarcinoma (LUAD), respectively (p-value < 0.0001 for all comparisons, Wilcoxon rank-sum test). We note that squamous cell carcinomas have higher CA20 within cervical (CESC) and oesophageal (ESCA) cancers (p-value < 0.001 and < 0.01 , respectively, Wilcoxon rank-sum test), suggesting that the observed differences are indeed associated to the different cell types of origin and not only to differences between tissue of origin.

Since CA has been considered a hallmark of tumour cells (7), we tested the difference of CA20 between tumour and matched-normal samples. Indeed, tumour samples have higher CA20 levels in all 15 cancer types with both sample types available (at least 10 samples of each type; False Discovery Rate (FDR) < 0.0001 , Wilcoxon rank-sum test; Fig 1c). In addition, using linear regression analyses with proliferation rate as an additional covariate, we found that CA20 is higher in tumour samples, either when considering all cohorts together (linear regression p-value < 0.0001 , using cohort as an additional covariate) or per individual cohort (FDR < 0.0001 for all cohorts),

independently of proliferation rate, discarding the suggestion of CA20 being its mere surrogate. These results emphasise CA as a hallmark of cancer.

CA20 is associated with breast cancer clinical and molecular features

Breast cancer is one of the best studied cancer types, with large cohorts of clinically annotated tumour samples available (27,28), and where the CA20 score was developed (23). In addition, CA has been frequently correlated with aggressive features in breast cancer (6,25,26,29). Given that we observed high variability of CA20 in TCGA breast tumour samples, we sought to investigate in more detail the relationship between CA20 and different breast cancer molecular features in that cohort. CA20 is higher in tumours than in normal breast samples (p-value < 0.0001 , Wilcoxon rank-sum test; Fig 2a) and we found higher levels of CA20 in invasive tumours from ductal histologic subtype (the most common, accounting for 90% of tumours) (30) when compared with lobular ones (p-value < 0.0001 , Wilcoxon rank-sum test; Fig 2b). The difference between ductal and lobular subtypes is consistent in non-triple negative breast tumours (p-value < 0.0001 , Wilcoxon rank-sum test), as well as in samples from tumour stages II and III (p-value < 0.0001 and < 0.01 , respectively, Wilcoxon rank- sum test). We also tested the differences in CA20 between the different PAM50 molecular subtypes, derived based on a 50-gene classifier (31 ). Basal-like breast tumours have the highest CA20 scores (p-value < 0.0001 , p-value < 0.0001 , and p-value < 0.001 for contrasts with luminal A, luminal B, and HER2-enriched, respectively, Wilcoxon rank-sum test; Fig 2c). This is in line with our recent work experimentally showing that basal-like breast cancers have indeed more CA than luminal ones (26). We also observed a strong difference between luminal subtypes, with higher CA20 in luminal B samples (p-value < 0.0001 , Wilcoxon rank-sum test; Fig 2c). Moreover, we tested the association between CA20 and tumour stage, having found a significant CA20 increase from stage I to stage II (p- value < 0.0001 , Wilcoxon rank-sum test; Fig 2d), but no significant changes between subsequent stages (Fig 2d). All associations between CA20 and breast cancer histology, PAM50 molecular subtypes and tumour stage remain significant within both low and high proliferating tumours (samples divided by the median of estimated proliferation rates). All the aforementioned associations were validated in an independent cohort (Fig 2e-h), comprising 144 normal and 1 ,992 tumour breast samples from the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) (28). We still tested the association between CA20 and the METABRIC integrative clusters, 10 molecular subgroups defined based on joint clustering of copy number and gene expression data (28). CA20 varies across integrative clusters (p-value < 0.0001 , Fligner-Killeen test) and is particularly enriched in cluster 10 (FDR < 0.0001 , Wilcoxon rank-sum test, for comparisons with each of all the other clusters), characterized by high proportion of basal-like tumours, high genomic instability, high rate of TP53 mutations, chromosome arm 5q deletions and very poor prognosis in the short term (28).

We complementarily analysed the frequency of CA in human breast carcinomas from the different PAM50 molecular subtypes, comprising 29 luminal A, 3 luminal B, 3 HER2 and 13 basal-like tumours (Fig 2i). Concordantly with TCGA and METABRIC results, we observed a higher percentage of cells with supernumerary centrioles in luminal B (average of 27%) than in luminal A carcinomas (7%; p- value < 0.05, Wilcoxon rank-sum test; Fig 2j). Moreover, basal-like (25%) display higher levels of CA than luminal A tumours (p-value < 0.0001 , Wilcoxon rank-sum test). Despite the reduced number of luminal B samples, our patient data support CA20 as a good surrogate of CA levels and the suggestion that CA is more frequent in luminal B than in luminal A human breast carcinomas.

CA20 is associated with genomic instability features in cancer

CA and consequent multipolar mitoses have been associated with aneuploidy, genomic instability and tumourigenesis for more than a century (32,33). Using the available quantitative characterization of aneuploidy in TCGA (34), we found that CA20 is higher in samples with genome doubling (p-value < 0.0001 , Wilcoxon rank-sum test; Fig 3a) and positively correlated with their aneuploidy score

(measured as the total number of altered - gained or lost - chromosome arms; Spearman’s correlation coefficient, r = 0.44, p-value < 2.2e-16; Fig 3b). Although CA20 is positively correlated with both chromosomal deletions and amplifications (Spearman’s correlation coefficient, r = 0.41 and 0.36, p-value < 2.2e-16, respectively), it is more strongly associated with chromosomal deletions (p-value < 2.2e-16, t-test for z-transformed coefficients). Given the known association between loss of p53 and CA (6,7,35) and the recent observation that p53 null cells tend to have an enrichment of chromosome losses over gains (36), we tested the hypothesis that the observed association between CA20 and chromosomal deletions could be linked to TP53 mutations. However, the increase in the proportion of deletions per sample from low to high CA20 samples is consistent within both TP53 wild-type and mutated samples (p-value < 0.0001 and < 0.05, respectively, Wilcoxon rank-sum test), showing it is independent of TP53 mutations (two-way ANOVA p-value for interaction = 0.6).

Investigating the hypothesis that CA20-associated aneuploidy levels could vary between

chromosomes, we identified 20 chromosome arms whose deletion (10 arms) or amplification (10 arms) was enriched in tumour samples with higher CA20 (linear regression, FDR < 0.05; Fig 3c). The strongest associations were with loss of 5q, 16p and 7p. Interestingly, 5q deletion was previously associated with CA20-high basal-like breast tumours (27,37-40) and METABRIC integrative cluster 10 (28) (Fig 2c, g). The association between CA20 and 5q deletion remains when removing the breast cancer cohort (linear regression p-value < 2.2e-16). This observation raises the question if matched- normal samples of the analysed tumour samples have a CA20 signal predictive of those 5q, 16p and 7p deletions. We tested this hypothesis by comparing the CA20 levels between normal samples (with intact tested chromosomal arms) whose matched tumours lost 5q, 16p or 7p, with those with tumours with amplifications or no alterations in those chromosomal arms. We found that normal samples whose matched tumours lost 5q or 16p exhibit higher CA20 scores (p-value < 0.01 and < 0.05, respectively, Wilcoxon rank-sum test), therefore suggesting that a CA20 increase may precede those chromosomal abnormalities.

In addition to tumour aneuploidy, CA20 is positively correlated with mutation burden, number of somatic Copy Number Alterations (CNA) and number of clones per tumour (Spearman’s correlation coefficient, r = 0.48, 0.47 and 0.43, respectively, p-value < 2.2e-16 for all; Fig 3d-f). All these associations are independent of cell proliferation (linear regression p-values < 1e-8 for all). We found that the correlation with mutation burden holds for different types of mutations (silent, missense, splice site and nonsense), as well as for mutations shown to be pathogenic (data from ClinVar

https://www.ncbi.nlm.nih.gov/clinvar/) in all diseases and particularly in cancer. Since these genomic instability features are likely correlated between each other, we applied multiple linear regression analyses across 1050 tumour samples (from 12 different cancer types; minimum of 30 and average of 88 samples per cohort) with information for those 4 covariates. We identified independent positive associations between CA20 and all genomic instability features, with stronger association for CNAs (linear regression p-values = 1 3e-5, 7.2e-4, 5.3e-10 and 6.4e-3 for aneuploidy, mutation burden,

CNA and number for clones, respectively; Fig 3g). These associations remain significant when proliferation rate is used as an additional covariate in the regression (p-values = 2.3e-5, 7e-4, 2.4e-9 and 0.03 for aneuploidy, mutation burden, CNA and number for clones, respectivel). We performed similar analyses per TCGA cohort and identified a group of cancer types where CA20 is mostly associated with CNA and aneuploidy (prostate adenocarcinoma, glioblastoma multiforme, bladder urothelial carcinoma, and brain low-grade glioma; Fig 3g. Although CA has been globally associated with genomic instability, these results highlight CNA as the main associated feature and show that these associations differ between cancer types.

CA20 is associated with cancer’s mutational spectrum

Point mutations are one of the most common types of mutational events that impact the stability of a cancer genome. We examined the pan-cancer association between CA20 and somatic mutations in 14,589 genes and found 752 whose mutations are associated with CA20 (FDR < 0.05; Fig 4a). Most significant associations of mutated genes with the CA20 score are positive, consistently with its correlation with higher mutation burden (Fig 3d), and enriched in cancer driver genes (Gene Set Enrichment Analysis (GSEA) (41 ,42) p-value < 0.001 , using a list of 299 cancer driver genes derived from TCGA’s PanCancer analysis (43)). TP53 shows the strongest association (linear regression p- value < 0.0001 ; Fig 4a), with positive correlations for the majority of cancer types surveyed (10 out of 17 cancer types with at least 20 mutated samples; FDR < 0.05; Fig 4b), therefore putatively extending the reported association between loss of p53 and CA (6,7,35) to 10 different cancer types. The second strongest positive association is with tumour suppressor pRb (RB1), whose acute loss has been found to induce CA (44). Unexpectedly, the strongest negative association is with E-cadherin (encoded by CDH1), meaning CDH /-mutated samples have lower CA20 levels. Given its tumour suppressor role in cancer and the fact that its mutations mostly induce loss of function (45), this result suggests loss of E-cadherin is associated with lower CA in human tumours, which is contrary to what have been reported in epithelial cancer cells (46). GSEA on genes whose mutations are associated with CA20 found that they are enriched in cancer-associated pathways and Wnt/ -catenin signalling. As only a small fraction of somatic mutations represent driver events, we repeated the pan-cancer analysis of association between CA20 and somatic mutations using likely driver mutations from the Cancer Genome Interpreter (https://www.cancerqenomeinterpreter.org/mutations) (45). Within the tested 33 genes with at least 10 mutated samples, we found three ( TP53 , PIK3CA and EGFR) whose driver mutations are associated with CA20 (FDR < 0.05), TP53 being again the strongest association. Overall, we show that CA20 is associated with both passenger and driver mutational spectra in cancer, with particular enrichment in cancer driver genes and Wnt/ -catenin signalling.

CA has still been proposed as a driver of genomic instability (1 1). We thus wondered if the DNA mutation spectrum associated with CA was similar to specific signatures of somatic mutations caused by different mutational processes in cancer (47). We therefore retrieved the contribution of the 30 published mutational signatures for each TCGA tumour sample from mSignatureDB (48) and uncovered three of them positively associated with CA20: signature 3, associated with BRCA1/2 mutations; signature 13, attributed to APOBEC activity; and signature 4, characteristic of smoking’s mutational pattern (FDR < 0.05). As these signatures are likely confounded with genomic instability, we performed multiple linear regression on CA20 including, as independent variables, the mutational signature and the four aforementioned genomic instability features: aneuploidy, mutation burden,

CNA and number of clones per tumour. Signature 1 , linked with ageing and characterised by C>T substitutions, and its“reverse” (T>C substitution bias) Signature 12, found mainly in liver cancer, are respectively positively and negatively associated (FDR < 0.05) with CA20 (Fig 4c), independently of other types of genomic instability and even when proliferation rate is added as a variable (FDR =

0.051 for both signatures).

To evaluate the putative causality of CA20-associated mutations (Fig 4a), we interrogated the Connectivity Map (CMap) database of signatures (49) about the impact of each of the 3,799 gene knock-downs on the CA20 gene set in human cancer cell lines. The resultant connectivity scores, ranging from 100 (CA20 up-regulation) to -100 (CA20 down-regulation), were compared with the pan- cancer association between somatic mutations in the cognate genes and CA20 (Fig 4d). We thereby identified 6 genes with a putative causal effect on CA20 scores (|connectivity score| > 80; Fig 4d): P2RY12, RB1, ITSN1 and MYCBP2 are putative inhibitors of CA (their knock-down up-regulate CA20 genes), whereas ABCC5 and COPA are putative promoters of CA (their knock-down down-regulate CA20 genes). Although acute loss of pRb (encoded by RB1) has been found to induce CA (44), confirming pRb as a CA inhibitor, to our knowledge none of the remaining genes identified herein has been previously associated with CA. They are therefore interesting candidates for future functional studies. Genes from a manually curated list of centriole duplication factors (93 genes, including only 10 from the CA20 signature) are enriched in negative CMap knock-down scores (GSEA p-value < 0.001 ; Fig 4e), suggesting they are indeed needed for cells to express CA-associated genes. Using the MSigDB’s Hallmark Gene Sets library (50), we identified unfolded protein response and mitotic spindle as significantly enriched in genes whose knock-down showed negative scores, i.e. CA20 down-regulation (GSEA FDR < 0.05; Fig 4f). This association suggests that mitotic spindle components activate CA-associated genes and/or that cells highly expressing CA-associated genes may be less likely to survive when their mitotic spindle is perturbed.

CA20 is associated with prognosis, hypoxia and stromal infiltration in cancer

CA has been associated with poor patient prognosis in a variety of cancer types (7). We therefore tested CA20’s association with overall patient’s survival across 31 TCGA cancer types with more than 40 samples each, finding high CA20 significantly associated with worse prognosis in 8 different cancer types (FDR < 0.05, log-rank test; Fig 5a). This result supports the potential of CA20 for prognostic-based patient stratification.

Hypoxia is a potent microenvironmental factor promoting genetic instability and malignant progression (51-53). Given that hypoxia has been shown to enhance centrosome aberrations in breast cancer (54,55), we investigated whether CA20 is associated with the relative hypoxia levels in TCGA tumour samples, given by a previously established surrogate metagene expression signature (56). We found a positive correlation between CA20 and the hypoxia score (Spearman’s correlation coefficient, r = 0.61 , p-value < 2.2e-16; Fig 5b) that is independent of genomic instability (linear regression p-value = 7.8e-9). We further confirmed that this association is independent of estimated proliferation rates (linear regression p-value = 5.6e-7 when proliferation rate is added as a covariate to the regression). We also performed this linear regression analysis for each of the 12 TCGA cohorts with information for all covariates and identified three cancer types (glioblastoma multiforme, lung adenocarcinoma and bladder urothelial carcinoma) where hypoxia is positively associated (FDR < 0.05) with CA20 (Fig 5c).

Although a tumour is also composed by stromal and immune cells (57), the association between CA and tumour cellular composition has not been addressed yet. CA20 is associated with lower stromal (Spearman’s correlation coefficient, r = -0.52, p-value < 2.2e-16; Fig 5d) and immune (Spearman’s correlation coefficient, r = -0.34, p-value < 2.2e-16) cell infiltration in TCGA. However, pan-cancer linear regression analyses revealed that only the negative association with stromal infiltration is independent of genomic instability (linear regression p-value = 2.7e-6 and 0.24, for stromal and immune, respectively). The same was observed when including proliferation rate as an additional covariate (linear regression p-value = 1.2e-4 and 0.21 , respectively. We have also performed similar analyses for each of the 5 TCGA cohorts with information for all covariates and found that CA20 is significantly associated (FDR < 0.05) with lower stromal infiltration in head and neck and lung cancers (Fig 5e), with lower immune infiltration in glioblastoma, and with higher immune infiltration in head and neck cancer, all independently of genomic instability.

Identification of compounds that selectively kill cancer cells with high CA20

CA is a hallmark of cancer cells and hence an appealing target in cancer therapy. In order to identify compounds that could target cancer cells with such abnormality, we have employed CA20 to estimate relative CA levels in 823 human cancer cell lines from the Cancer Therapeutics Response Portal (CTRP) (58), for which both transcriptomic and drug-sensitivity profiles are publicly available.

Correlation analyses between CA20 and drug-sensitivity (in Area Under the dose-response Curve, AUC) for 354 compounds revealed 81 negatively correlated with CA20 (FDR < 0.05, Spearman’s correlation; Fig 6a), i.e. higher CA20 was associated with lower drug AUC and, therefore, higher drug activity. The enrichment of negative correlations may reflect the bias for cancer-targeting compounds in CTRP. These results suggest several candidate compounds to selectively kill cancer cells with CA, such as 3-CI-AHPC, CD-437, STF-31 , methotrexate, BI-2536 and clofarabine (Fig 6b). The first three are probes, methotrexate and clofarabine are U.S. Food and Drug Administration (FDA)-approved drugs for several cancer types (https://www.cancer.gov/about-cancer/treatment/drugs/methotr exate) and paediatric acute lymphoblastic leukemia (https://www.cancer.gov/about- cancer/treatment/drugs/fda-clofarabine), respectively. Interestingly, BI-2536 has been in clinical trials for several solid and liquid tumours

(https://clinicaltrials.gov/ct2/results?cond=&term=bi+25 36&cntry=&state=&city=&dist) and is an inhibitor of polo-like kinase 1 (PLK1), whose inhibition has already been associated with CA suppression (59,60).

Complementarily, we mined the CMap database to identify compounds that could impact the CA20 score and therefore putatively reduce/increase CA levels. We calculated the impact of 2,837 compounds on the CA20 transcriptomic levels in human cancer cell lines and identified some whose activity drove CA20 up-regulation (putative CA promoters), such as VEGFR2-kinase-inhibitor-IV, dienestrol (oestrogen receptor agonist) and sulforaphane (anticancer agent in clinical trials for Bladder, Breast, Lung and Prostate cancers;

https://clinicaltrials.gov/ct2/results?cond=sulforaphane& amp;Search=Apply&recrs=d&age_v=&gndr=&type =&rslt=). We also identified compounds that down-regulated CA20, such as two CDK inhibitors (purvalanol-a and aminopurvalanol-a), JAK3-inhibitor-VI, etoposide (topoisomerase and cell cycle inhibitor) and CD-437 (agonist of RARG, retinoic acid receptor gamma; Fig 6c).

For the 164 drugs tested in both datasets, we observed a positive correlation between their

CA20/sensitivity correlations in CTRP and their CMap scores (Spearman’s correlation coefficient, r = 0.26, p-value = 8.3e-4; Fig 6d), indicating that drugs selectively targeting cells with higher CA20 are reducing the expression of these genes, possibly by killing the abnormal cells in the tumour cell population. These complementary approaches uncovered RARG’s agonist CD-437 as the strongest candidate for targeting CA. Moreover, drugs targeting coagulation factor II (F2R), farnesyltransferase (FNTA and FNTB), ubiquitin isopeptidases (USP13 and USP5), DNA topoisomerase II alpha (TOP2A) and cyclin-dependent kinases (CDKs) are also promising candidates (Fig 6d). Given cell proliferation’s association with CA20, we have tested the association between its estimated rates across TCGA primary tumour samples and the expression of the 164 compounds’ predicted target genes (merging this information from the CTRP and CMap datasets, using linear regression analyses with cohort as additional covariate. The resultant coefficients are not correlated with CMap’s average scores of the respective compounds (Spearman’s correlation coefficient, r = 0.016, p-value = 0.84), but are correlated with their CTRP’s Spearman correlation coefficients (Spearman’s correlation coefficient, r = -0.26, p-value = 9e-04), i.e. compounds selective for cells with high CA20 are predicted to target genes positively associated with proliferation in TCGA tumour samples. Nevertheless, predicted target genes of several compound candidates from our analyses do not show strong association with proliferation. These results need to be considered when prioritizing candidate compounds for further experiments aiming to target cancer cells through CA.

CA is known to promote tumourigenesis but its molecular role therein remains elusive and, although it is also suggested to be a promising target for cancer therapy, CA’s prevalence in different types of cancer and therapeutic value in the clinic are still pretty much unprobed. Using the CA20 signature and TCGA RNA-seq data, we characterise the landscape of CA-associated gene expression in a broad range of cancer types, thereby demonstrating the potential of using gene expression-based signatures in multi-omic and clinical data integrative approaches to investigate the biological and medical relevance of their respective cellular and molecular processes.

Despite the lack of a full direct experimental validation of CA20 as a surrogate of CA levels, our observations are very consistent with known CA’s features, namely CA20’s upregulation in cancer (7) and in basal-like breast tumours (26), and its association with the knock-down of centriole duplication factors, genomic instability (11 ), loss of p53 (6,7,35) and pRB (44), hypoxia (54,55) and worse patient’s prognosis (7). In addition, we found that luminal B breast tumours have higher prevalence of CA than luminal A ones, concordantly with the observed differences in the CA20 score between the two molecular subtypes in two independent cohorts. Finally, we have analysed two transcriptomic datasets of multiciliogenesis, where cells escape centriole number regulation to generate hundreds of centrioles during differentiation (61), and found that CA20 increases during the centriole

overduplication stage, resuming basal levels afterwards, suggesting CA20 as a marker of active amplification. These observations vouch for the present proof-of-concept study to pave the way for more in-depth and bona fide findings when CA’s transcriptomic signature is experimentally refined. Moreover, here we already propose novel hypotheses that will trigger studies aiming at a more comprehensive understanding of the role of CA in cancer. We observed higher CA-associated gene expression in cancer samples of squamous cell origin than in adenocarcinomas, suggesting that their different cell types of origin can have different CA’s prevalence and/or ways to cope with this abnormality. Previous work has indeed shown that CA triggers spontaneous squamous cell carcinomas, lymphomas and sarcomas, but not

adenocarcinomas, in mice (11). We also show that breast invasive carcinoma samples have high variability on CA20, concordantly with previous observations (7,25), that is related to their distinct clinical and molecular features. We had recently shown that basal-like breast carcinomas have higher CA than luminal tumours (26), but here we report for the first time an upregulation of CA-associated genes in tumours from both invasive ductal histologic subtype and luminal B molecular subtype. We validated the CA20-based predictions by quantitatively analysing centrosome numbers in human breast carcinoma samples, where we found that indeed CA is more prevalent in luminal B than luminal A tumours, providing a novel insight into the differences between these two hormone-receptor positive molecular subtypes. Given the limited number of luminal B samples in our cohort, more extensive analyses are necessary to confirm this association. Our data show that centrosome amplification is associated with breast cancer clinical features and endorses the potential of using a gene-expression-based signature for patient stratification.

CA-associated gene expression upregulation is positively correlated with different types of genomic instability, like aneuploidy, mutation burden, CNA and tumour heterogeneity. In particular, CA20 is more strongly associated with chromosomal deletions than amplifications, independently of TP53 mutations. We speculate that this association may be due to the impact of CA in cellular genomic stability having non-random genomic“hot spots”. In fact, through a more detailed analysis, we found an association with alterations in specific chromosomal arms, that may be due to the localisation of genes encoding for regulators of CA20 genes therein and/or to those arms’ higher susceptibility to the genomic instability triggered by centrosome abnormalities. The latter is supported by recent work showing that human chromosome mis-segregation is not random and can be biased by inherent properties of individual chromosomes (62), and also by our observation that normal samples whose matched tumours lost 5q or 16p have higher CA20 predictive of those deletions. Moreover, we characterised the DNA mutation spectrum associated with CA20 and found it to be enriched in C>T mutations, a signature characteristic of ageing, with which centrosome aberrations have also been associated (63-67). Genes whose mutations are associated with CA20 are enriched in cancer driver genes, and particularly in Wnt^-catenin signalling. Wnt^-catenin signalling components interact with the centrosome (68) and a previous study has demonstrated that mutant b-catenin induces centrosome aberrations in normal epithelial cells and is required for CA in cancer cells (69). Our results extend this previous association to human cancer samples, suggesting mutations in b-catenin might contribute to the observed CA in cancer. Finally, we show the usefulness of a novel approach whereby we integrated information on genes whose somatic mutations are associated with CA20 in TCGA tumour samples with the impact of their knock-downs on the CA20 expression in human cancer cell lines, aiming at unveiling candidate molecular players in CA in cancer. Concordantly with previous work on CA (7), we observed that high CA20 is associated with poor patient’s survival in several cancer types. Furthermore, we found a positive correlation between CA20 and hypoxic levels in glioblastoma multiforme that is particularly interesting, due to its highly hypoxic microenvironment and HIF-1a levels (70), also shown to enhance migration and invasion of its tumour cells (71 ,72). Given the observed association between CA and invasion of tumour cells (15,17), an exciting hypothesis is hypoxia-induced invasion being mediated through CA. When looking at the tumour cellular composition, we found that tumours with high CA20 have lower stromal and immune cell infiltration, although the latter is not independent of tumour genomic instability and proliferation rate. Detailed studies aiming to decouple these effects could provide relevant molecular insights when considering immunotherapy, alone or in combination with genotoxic and/or anti-proliferative therapeutic approaches. Moreover, by pioneering the integration of drug sensitivity with drug perturbation profiles in human cancer cell lines, we identify candidate compounds for selectively targeting cancer cells exhibiting transcriptomic evidence for CA. These compounds could be particularly useful in the treatment of cancer types we identified as having high CA and to whose current therapy patients respond poorly. For instance, their potential in specifically targeting basal-like and luminal B breast tumours could be assessed taking advantage of resources like patient-derived tumour xenografts (73). The observed ability of cells carrying extra centrosomes to manipulate the surrounding tumour cells and promote their invasiveness (15, 17) suggests that targeting the former may be clinically more impactful. Given CA’s cancer-specificity, the compounds identified herein could underlie the development of novel targeted cancer therapeutic options.

Table 1

Table 2 References

1. Wang X, Tsai J-W, Imai JH, Lian W-N, Vallee RB, Shi S-H. Asymmetric centrosome inheritance maintains neural progenitors in the neocortex. Nature. 2009;461 (7266):947-55.

2. Stevens NR, Raposo AASF, Basto R, St Johnston D, Raff JW. From stem cell to embryo without centrioles. Curr Biol. 2007; 17(17): 1498-503.

3. Vinogradova T, Paul R, Grimaldi a. D, Loncarek J, Miller PM, Yampolsky D, et al. Concerted effort of centrosomal and Golgi-derived microtubules is required for proper Golgi complex assembly but not for maintenance. Mol Biol Cell. 2012;23(5):820-33.

4. Bettencourt-Dias M, Glover DM. Centrosome biogenesis and function: centrosomics brings new understanding. Nat Rev Mol Cell Biol. 2007;8(6):451-63.

5. Bornens M. The Centrosome in Cells and Organisms. Science (80- ). 2012;335:422-6.

6. Godinho SA, Pellman D. Causes and consequences of centrosome abnormalities in cancer. Philos Trans R Soc oB. 2014;369:20130467.

7. Chan JY. A clinical overview of centrosome amplification in human cancers. Int J Biol Sci.

201 1 ;7(8): 1122-44.

8. Silkworth WT, Nardi IK, Scholl LM, Cimini D. Multipolar Spindle Pole Coalescence Is a Major Source of Kinetochore Mis-Attachment and Chromosome Mis-Segregation in Cancer Cells. PLoS One. 2009;4(8):e6564.

9. Ganem NJ, Godinho SA, Pellman D. A mechanism linking extra centrosomes to chromosomal instability. Nature. 2009;460(7252):278-82.

10. Lingle WL, Barrett SL, Negron VC, Assoro ABD, Boeneman K, Liu W, et al. Centrosome amplification drives chromosomal instability in breast tumor development. Proc Natl Acad Sci.

2002;99(4): 1978-83.

11. Levine MS, Bakker B, Boeckx B, Moyett J, Lu J, Vitre B, et al. Centrosome amplification is sufficient to promote spontaneous tumorigenesis in mammals. Dev Cell. 2017;40:1-10.

12. Serein O, Larsimont J-C, Karambelas AE, Marthiens V, Moers V, Boeckx B, et al. Transient PLK4 overexpression accelerates tumorigenesis in p53-deficient epidermis. Nat Cell Biol.

2015; 18(1 ): 100—10.

13. Coelho PA, Bury L, Shahbazi MN, Liakath-ali K, Tate PH, Wormald S, et al. Over-expression of Plk4 induces centrosome amplification, loss of primary cilia and associated tissue hyperplasia in the mouse. Open Biol. 2015;5: 150209.

14. Marusyk A, Tabassum DP, Altrock PM, Almendro V, Michor F, Polyak K. Non-cell- autonomous driving of tumour growth supports sub-clonal heterogeneity. Nature. 2014 Jul 30;514:54.

15. Ganier O, Schnerch D, Oertle P, Lim RYH, Plodinec M, Nigg EA. Structural centrosome aberrations promote non-cell-autonomous invasiveness. EMBO J. 2018;37:e98576.

16. Godinho SA, Picone R, Burute M, Dagher R, Su Y, Leung CT, et al. Oncogene-like induction of cellular invasion from centrosome amplification. Nature. 2014;510(7503):167-71.

17. Arnandis T, Monteiro P, Adams SD, Bridgeman VL, Rajeeve V, Gadaleta E, et al. Oxidative Stress in Cells with Extra Centrosomes Drives Non-Cell-Autonomous Invasion. Dev Cell. 2018;47(4):409-424.e9.

18. Duensing S. Analysis of centrosomes in human cancer. In: Methods in Cell Biology. Elsevier; 2015. p. 51-60.

19. Farahani N, Braun A, Jutt D, Huffman T, Reder N, Liu Z, et al. Three-dimensional Imaging and Scanning: Current and Future Applications for Pathology. J Pathol Inform. 2017;8(36).

20. Chng WJ, Braggio E, Mulligan G, Bryant B, Remstein E, Valdez R, et al. The centrosome index is a powerful prognostic marker in myeloma and identifies a cohort of patients that might benefit from aurora kinase inhibition. Blood. 2008;11 1 (3):1603-10.

21. Chng WJ, Ahmann GJ, Henderson K, Santana-davila R, Greipp PR, Gertz MA, et al. Clinical implication of centrosome amplification in plasma cell neoplasm. Blood. 2006;107(9):3669-76.

22. Pannu V, Mittal K, Cantuaria G, Reid MD, Li X, Donthamsetty S, et al. Rampant centrosome amplification underlies more aggressive disease course of triple negative breast cancers. Oncotarget. 2015;6(12): 10487—97.

23. Ogden A, Rida PCG, Aneja R. Prognostic value of CA20, a score based on centrosome amplification-associated genes, in breast tumors. Sci Rep. 2017;7(1 ):262.

24. Diener C, Resendis-Antonio O. Personalized Prediction of Proliferation Rates and Metabolic Liabilities in Cancer Biopsies. Front Physiol. 2016;7(644).

25. Ogden A, Rida PCG, Aneja R. Centrosome amplification: a suspect in breast cancer and racial disparities. Endocr Relat Cancer. 2017;24(9):T47-64.

26. Marteil G, Guerrero A, Vieira AF, de Almeida BP, Machado P, Mendonga S, et al. Overelongation of centrioles in cancer promotes centriole amplification and chromosome missegregation. Nat Commun. 2018;9:1258.

27. The Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature. 2012;490(7418):61-70.

28. Curtis C, Shah SP, Chin S, Turashvili G. The genomic and transcriptomic architecture of 2 , 000 breast tumours reveals novel subgroups. Nature. 2012;486(7403):346-52.

29. Denu RA, Zasadil LM, Kanugh C, Laffin J, Weaver BA, Burkard ME. Centrosome

amplification induces high grade features and is prognostic of worse outcomes in breast cancer. BMC Cancer. 2016; 16(47).

30. Barroso-Sousa R, Metzger-Filho O. Differences between invasive lobular and invasive ductal carcinoma of the breast: results and therapeutic implications. Ther Adv Med Oncol. 2016;8(4):261-6.

31. Parker JS, Mullins M, Cheang MCU, Leung S, Voduc D, Vickery T, et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol. 2009;27(8):1160-7.

32. Holland AJ, Cleveland DW. Boveri revisited: chromosomal instability, aneuploidy and tumorigenesis. Nat Rev Mol Cell Biol. 2009;10(7):478-87.

33. Boveri T. Concerning the origin of malignant tumours by Theodor Boveri. Translated and annotated by Henry Harris. J Cell Sci. 2008;121 :1-84.

34. Taylor AM, Shih J, Ha G, Gao GF, Zhang X, Berger AC, et al. Genomic and Functional Approaches to Understanding Cancer Aneuploidy. Cancer Cell. 2018;33:676-89.

35. Lopes CAM, Mesquita M, Cunha Al, Cardoso J, Carapeta S, Laranjeira C, et al. Centrosome amplification arises before neoplasia and increases upon p53 loss in tumorigenesis. J Cell Biol. 2018;

36. Laks E, Zahn H, Lai D, McPherson A, Steif A, Brimhall J, et al. Resource: Scalable whole genome sequencing of 40,000 single cells identifies stochastic aneuploidies, genome replication states and clonal repertoires. bioRxiv. 2018.

37. Yu W, Kanaan Y, Baed Y, Gabrielson E. Chromosomal changes in aggressive breast cancers with basal-like features. Cancer Genet Cytogenet. 2009; 193(1 ):29-37.

38. Gillette MA, Clauser KR, Wang P, Wang X, Cao S, Petralia F, et al. Proteogenomics connects somatic mutations to signalling in breast cancer. Nature. 2016;534:55-62.

39. Silva GO, He X, Parker JS, Gatza ML, Carey LA, Hou JP, et al. Cross-species DNA copy number analyses identifies multiple 1q21-q23 subtype-specific driver genes for breast cancer. Breast Cancer Res Treat. 2015;152(2):347-56.

40. Weigman VJ, Chao H-H, Shabalin AA, He X, Parker JS, Nordgard SH, et al. Basal-like Breast cancer DNA copy number losses identify genes involved in genomic instability , response to therapy , and patient survival. Breast Cancer Res Treat. 2012;133:865-80.

41. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette M a, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005; 102(43): 15545-50.

42. Mootha VK, Lindgren CM, Eriksson K-F, Subramanian A, Sihag S, Lehar J, et al. PGC-1a- responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet. 2003;34(3):267-73.

43. Bailey MH, Tokheim C, Porta-Pardo E, Sengupta S, Bertrand D, Weerasinghe A, et al.

Comprehensive Characterization of Cancer Driver Genes and Mutations. Cell. 2018; 173(2):371- 385.e18.

44. lovino F, Lentini L, Amato A, Leonardo A Di. RB acute loss induces centrosome amplification and aneuploidy in murine primary fibroblasts. Mol Cancer. 2006;5(38).

45. Tamborero D, Rubio-Perez C, Deu-Pons J, Schroeder MP, Vivancos A, Rovira A, et al.

Cancer Genome Interpreter Annotates The Biological And Clinical Relevance Of Tumor Alterations. Genome Med. 2018; 10(25).

46. Rhys AD, Monteiro P, Smith C, Vaghela M, Arnandis T, Kato T, et al. Loss of E-cadherin provides tolerance to centrosome amplification in epithelial cancer cells. J Cell Biol. 2017 Nov 13.

47. Alexandrov LB, Stratton MR. Mutational signatures: The patterns of somatic mutations hidden in cancer genomes. Curr Opin Genet Dev. 2014;24(1):52-60.

48. Huang PJ, Chiu LY, Lee CC, Yeh YM, Huang KY, Chiu CH, et al. MSignatureDB: A database for deciphering mutational signatures in human cancers. Nucleic Acids Res. 2018;46(D1):D964-70.

49. Subramanian A, Narayan R, Corsello SM, Peck DD, Natoli TE, Lu X, et al. A Next Generation Connectivity Map: L1000 Platform and the First 1 ,000,000 Profiles. Cell. 2017; 171 (6): 1437-52.

50. Liberzon A, Birger C, Thorvaldsdottir H, Ghandi M, Mesirov JP, Tamayo P. The Molecular Signatures Database Hallmark Gene Set Collection. Cell Syst. 2015;1 (6):417-25.

51. Rankin EB, Giaccia AJ. Hypoxic control of metastasis. Science (80- ). 2016;352(6282):175- 80. 52. Luoto KR, Kumareswaran R, Bristow RG. Tumor hypoxia as a driving force in genetic instability. Genome Integr. 2013;4(5).

53. Petrova V, Annicchiarico-petruzzelli M, Melino G, Amelio I. The hypoxic tumour

microenvironment. Oncogenesis. 2018;7(10).

54. Mittal K, Choi DH, Ogden A, Donthamsetty S, Melton BD, Gupta M V., et al. Amplified centrosomes and mitotic index display poor concordance between patient tumors and cultured cancer cells. Sci Rep. 2017;7(October 2015):43984.

55. Mittal K, Choi D, Maganti N, Ogden A, Melton B, Kaur J, et al. Hypoxia induced centrosome amplification via HIF-1a/Plk4 signaling axis associates with poorer overall survival in TNBC [abstract]. In: Proceedings of the 2017 San Antonio Breast Cancer Symposium; AACR; Cancer Res 2018;78(4 Suppl):Abstract nr P1-01-23. 2018;

56. Winter SC, Buffa FM, Silva P, Miller C, Valentine HR, Turley H, et al. Relation of a Hypoxia Metagene Derived from Head and Neck Cancer to Prognosis of Multiple Cancers. Cancer Res.

2007;67(7):3441-50.

57. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Vol. 144, Cell. 201 1. p. 646-74.

58. Rees MG, Seashore-ludlow B, Cheah JH, Adams DJ, Price E V, Gill S, et al. Correlating chemical sensitivity and basal gene expression reveals mechanism of action. Nat Chem Biol.

2016;12(2):109-16.

59. Liu X, Erikson RL. Activation of Cdc2/cyclin B and inhibition of centrosome amplification in cells depleted of Plk1 by siRNA. Proc Natl Acad Sci. 2002;99(13):8672-6.

60. Izumi H, Matsumoto Y, Ikeuchi T, Saya H, Kajii T, Matsuura S. BubR1 localizes to centrosomes and suppresses centrosome amplification via regulating Plk1 activity in interphase cells. Oncogene. 2009;28(31 ):2806-20.

61. Vladar EK, Stearns T. Molecular characterization of centriole assembly in ciliated epithelial cells. J Cell Biol. 2007;178(1 ):31-42.

62. Worrall JT, Tamura N, Mazzagatti A, Vladimirou E, Foijer F, McClelland SE, et al. Non-random Mis-segregation of Human Chromosomes. Cell Rep. 2018;23(11 ):3366-80.

63. Chichinadze K, Tkemaladze J, Lazarashvili A. A New Class of RNAs and the Centrosomal Hypothesis of Cell Aging. Adv Gerontol. 2012;2(4):287-91.

64. Na H, Park J, Pyo J, Jeon H, Kim Y, Arking R, et al. Metformin inhibits age-related centrosome amplification in Drosophila midgut stem cells through AKT/TOR pathway. Mech Ageing Dev. 2015;149:8-18.

65. Park J, Pyo J, Na H, Jeon H, Kim Y, Arking R, et al. Biochemical and Biophysical Research Communications Increased centrosome amplification in aged stem cells of the Drosophila midgut. Biochem Biophys Res Commun. 2014;450(2):961-5.

66. Ohshima S. Centrosome Aberrations Associated with Cellular Senescence and p53

Localization at Supernumerary Centrosomes. OxidativeMedicine Cell Longev. 2012;2012:217594.

67. Manning JA, Kumar S. A potential role for NEDD1 and the centrosome in senescence of mouse embryonic fibroblasts. Cell Death Dis. 2010;1 (4):e35-8. 68. Bryja V, Cervenka I, Cajcinek L. The connections of Wnt pathway components with cell cycle and centrosome: side effects or a hidden logic? Crit Rev Biochem Mol Biol. 2017;52(6):614-37.

69. Bahmanyar S, Guiney EL, Hatch EM, Nelson WJ, Barth AIM. Formation of extra centrosomal structures is dependent on b-catenin. J Cell Sci. 2010;123(18):3125-35.

70. Jawhari S, Ratinaud MH, Verdier M. Glioblastoma, hypoxia and autophagy: A survival-prone “menage-a-trois.” Cell Death Dis. 2016;7(10): 1-10.

71. Joseph J V., Conroy S, Pavlov K, Sontakke P, Tomar T, Eggens-Meijer E, et al. Hypoxia enhances migration and invasion in glioblastoma by promoting a mesenchymal shift mediated by the HIF1a-ZEB1 axis. Cancer Lett. 2015;359(1 ): 107-16.

72. Monteiro A, Hill R, Pilkington G, Madureira P. The Role of Hypoxia in Glioblastoma Invasion. Cells. 2017;6(4):45.

73. Bruna A, Rueda OM, Greenwood W, Batra AS, Callari M, Batra RN, et al. A Biobank of Breast Cancer Explants with Preserved Intra-tumor Heterogeneity to Screen Anticancer Compounds. Cell.

2016; 167(1 ):260-74.

74. Li B, Dewey CN. RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics. 2011 ;12(1):323.

75. Law CW, Chen Y, Shi W, Smyth GK. Voom: precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 2014;15(2):1-17.

76. Mermel CH, Schumacher SE, Hill B, Meyerson ML, Beroukhim R, Getz G. GISTIC2.0 facilitates sensitive and confident localization of the targets of focal somatic copy-number alteration in human cancers. Genome Biol. 2011 ; 12(4): 1—14.

77. Andor N, Graham TA, Jansen M, Xia LC, Aktipis CA, Petritsch C, et al. Pan-cancer analysis of the extent and consequences of intratumor heterogeneity. Nat Med. 2016;22(1 ): 105—13.

78. Yoshihara K, Shahmoradgoli M, Martinez E, Vegesna R, Kim H, Torres-Garcia W, et al. Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun. 2013;4:2612.

79. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, et al. The cBio Cancer Genomics Portal: An open platform for exploring multidimensional cancer genomics data. Cancer Discov. 2012;2(5):401-4.

80. Mori M, Hazan R, Danielian PS, Mahoney JE, Li H, Lu J, et al. Cytoplasmic E2f4 forms organizing centres for initiation of centriole amplification during multiciliogenesis. Nat Commun.

2017;8:15857.

81. Hoh RA, Stowe TR, Turk E, Stearns T. Transcriptional Program of Ciliated Epithelial Cells Reveals New Cilium and Centrosome Components and Links to Human Disease. PLoS One.

2012;7(12):e52166.

82. R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: Foundation for Statistical Computing; 2017.

83. Revelle W. psych: Procedures for Personality and Psychological Research. R package version 1.7.8. 2017.

84. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. Limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43(7):e47.

85. Zuur A, leno EN, Walker N, Saveliev AA, Smith G. Mixed Effects Models and Extensions in Ecology with R. Springer, editor. Springer; 2009.

86. Warnes GR, Bolker B, Bonebakker L, Gentleman R, Huber W, Liaw A, et al. gplots: Various R Programming Tools for Plotting Data. R package version 3.0.1. 2016;

87. Therneau TM, Grambsch PM. Modeling Survival Data: Extending the Cox Model. New York: Springer; 2000.